This preview shows pages 1–7. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: IE521 Advanced Optimization Lecture 5 Dr. Zeliha Akc ¸a October 2011 1 / 22 Reading I Bertsimas 3.13.3. 2 / 22 Recall I There is a onetoone correspondence between the extreme points of a polyhedron and the basic feasible solutions . I Constructing a basic solution : I Choose a basis B of m linearly independent columns of A . I Set the remaining columns to 0, x N = (nonbasic solutions). I Solve the system Bx B = b to obtain the values of the basic variables. I Can move between adjacent (nondegenerate) basic solutions by removing one column of the basis and replacing it with another. I In the presence of degeneracy , we might stay at the same extreme point. I These are the building blocks we need to construct algorithms for solving LPs. 3 / 22 Iterative Search Algorithms I Start from an initial point, move to a new point in a determined search direction . I A search direction is defined as a feasible (inside the feasible region) and improving (improve the objective function value). I Stop when a feasible and improving direction can not be found (a local optimal is found). I Many optimization algorithms are iterative . I For linear programs, local optimal means the global optimal solution . I This is because we are optimizing a convex function over a convex set. I This makes linear programs ’easy’ to solve. 4 / 22 Feasible and Improving Directions I Assume we have point ˆ x ∈ P . We want to move in a direction d , but want to stay inside P . Definition Let ˆ x be an element of a polyhedron P . A vector d ∈ R n is said to be a feasible direction if there exists θ ∈ R + such that ˆ x + θ d ∈ P . Definition Consider a polyhedron P and the associated linear program min x ∈P c &gt; x for c ∈ R n . A vector d ∈ R n is said to be an improving direction if c &gt; d &lt; . I When we move along an improving direction d from ˆ x , we would get a better objective function value c &gt; ˆ x + c &gt; d . I Move along that the feasible and improving direction as far as possible . I Recall that we are interested in extreme points . I The algorithm moves between adjacent extreme points using improving directions. 5 / 22 Constructing Feasible Search Directions I Consider a BFS ˆ x ....
View
Full
Document
 Fall '11
 ZelihaAkça
 Optimization

Click to edit the document details