Primal Dual

Post on 25-Jun-2015

5.270 views 0 download

Tags:

description

Primale duale

Transcript of Primal Dual

Carlo Lombardi, June 2008 Theoretical Computer Science

Primal-Dual Algorithms

A brief survey of Primal-Dual Algorithms

as an approximation technique for optimization problems

Scribe:Carlo Lombardilombardi.carlo@gmail.com

1

Carlo Lombardi, June 2008 Theoretical Computer Science 2

Overview

Introduction

The Minmum Weighted Vertex Cover Problem (WVC)

WVC as a ILP:

Solving WVC by rounding up a fractional solution

Solving WVC by Primal-Dual Strategy:

Duality: Background theoretic propertiesAlgorithmAnalysis

Example (on the blackboard)

Carlo Lombardi, June 2008 Theoretical Computer Science 3

Introduction

• We have seen many algorithms based on Linear Program (LP), typically involving the following strategy:

•We arise the initial difficult of the problem by relaxing it•We sacrifice the optimal solution to find a good approximate solution by solving the relaxed problem

Carlo Lombardi, June 2008 Theoretical Computer Science 4

Minimum Weighted Vertex Cover

Vertex Cover Problem“Each edge is covered by at least one node”

+Weighted Verteces

“Each vertex has a weight”

+Minimization of total weight“Minimize the total weight”

=

Minimum Weighted Vertex Cover (WVC)

Carlo Lombardi, June 2008 Theoretical Computer Science 5

WVC: ILP and LP formulation

We formulate the WVC as an Integer Linear Program (ILP) defining a variable xi for each vertex (xi=1 if vertex i belongs to the cover, 0 otherwise).

ILP FORMULATION

LP FORMULATION

by relaxing integrality constraints

Carlo Lombardi, June 2008 Theoretical Computer Science 6

WVC: Rounding the LP solution

Primal-Dual Method

We need to solve LP formulation…it can be

expensive for problems having many constraints!!!

Can we do something clever?

Carlo Lombardi, June 2008 Theoretical Computer Science 7

A different approach to LP relaxations:Primal-Dual strategy

Main idea:

!!! Don’t solve LP totally !!!Obtain a feasible integral solution to the LP (Primal) from scratch using a related LP (Dual) to guide your decision.

!!! Don’t solve LP totally !!!Obtain a feasible integral solution to the LP (Primal) from scratch using a related LP (Dual) to guide your decision.

LP Primal

LP Dual

Good approximated solution

“Solve me”

“I’ll be your guide”

Carlo Lombardi, June 2008 Theoretical Computer Science 8

P-D strategy: Background theoretic properties (1/2)

PRIMAL DUAL

(Weak Duality) For any feasible Primal-Dual solution pair (x,y):

= if (x,y) is optimal

(Strong Duality) If either the Primal or Dual have bounded optimal solution, the both of them do. Moreover, their objective functions values are qual. That is:

(Complementary Slackness) Let (x,y) be a solutions to a primal-dual pair of LPs with bounded optima. Then x and y are both optimal iff all of the following hold

Carlo Lombardi, June 2008 Theoretical Computer Science 9

P-D strategy: Background theoretic properties (2/2)

(Weak Duality) For any feasible Primal-Dual solution pair (x,y):

The dual solution is a lover bound for primal solution

= if (x,y) is optimal

(Strong Duality) If either the Primal or Dual have bounded optimal solution, the both of them do. Moreover, their objective functions values are qual. That is:

At the optimum the evaluation of solutions coincides(Complementary Slackness) Let (x,y) be a solutions to a primal-dual pair of LPs with bounded optima. Then x and y are both optimal iff all of the following hold

Only If a dual constraints is tight the corresponding primal variables can be greater than 0 (it can participate to the primal solution)

Carlo Lombardi, June 2008 Theoretical Computer Science 10

Primal-Dual strategy

Carlo Lombardi, June 2008 Theoretical Computer Science 11

WVC : The D-P Algorithm

Primal Dual

1. Maintains an integer solution x of ILP and a feasible solution y for DLP

2. Examines x and y3. Derives a ‘more feasible’ solution

x and a ‘better’ solution y4. Ends when the integer solution

becomes feasible5. Evaluates the integer solution

comparing it with dual solution

Carlo Lombardi, June 2008 Theoretical Computer Science 12

Analysis of Program 2.7

Note that for every it holds:

(1)

The o.f. is infact

From the (1)

Because we are considering all vertices in V

Each edge in E is taken two times

(Weak Duality)

Carlo Lombardi, June 2008 Theoretical Computer Science 13

References

• G. Ausiello, P. Crescenzi, G. Gambosi, V. Kann, A. Marchetti Spaccamela, M. Protasi, Complexity and Approximation, Springer, 1998, Chapter 2

•Michel X. Goemans, David P. Williamson, The primal dual method for approximations algorithms and its application to network design problems, PWS Publishing Co.,1997, Chapter 4