IV Latin-American Algorithms, Graphs and Optimization Symposium - 2007

Post on 14-Jan-2016

19 views 0 download

Tags:

description

IV Latin-American Algorithms, Graphs and Optimization Symposium - 2007. Puerto Varas - Chile. The Generalized Max-Controlled Set Problem. Carlos A. Martinhon Fluminense Fed. University Ivairton M. Santos - UFMT Luiz S. Ochi – IC/UFF. Contents. 1. Basic definitions. - PowerPoint PPT Presentation

transcript

IV Latin-American Algorithms, Graphs and Optimization Symposium - 2007

Puerto Varas - Chile

The Generalized Max-Controlled The Generalized Max-Controlled Set ProblemSet Problem

Carlos A. MartinhonFluminense Fed. University

Ivairton M. Santos - UFMTLuiz S. Ochi – IC/UFF

2

ContentsContents

1. Basic definitions1. Basic definitions

2. The Generalized Max-Controlled Set 2. The Generalized Max-Controlled Set ProblemProblem

3.3. a) A 1/2-Approximation procedurea) A 1/2-Approximation procedure

b) A Based LP Prog. Heuristicb) A Based LP Prog. Heuristic

c) A combined heuristicc) A combined heuristic

5. Comp. results and final comments 5. Comp. results and final comments

4. Tabu Search Procedure4. Tabu Search Procedure

3

ContentsContents

1. Basic definitions1. Basic definitions

2. The Generalized Max-Controlled Set 2. The Generalized Max-Controlled Set ProblemProblem

3.3. a) A 1/2-Approximation procedurea) A 1/2-Approximation procedure

b) A Based LP Prog. Heuristicb) A Based LP Prog. Heuristic

c) A combined heuristicc) A combined heuristic

5. Comp. results and final comments 5. Comp. results and final comments

4. Tabu Search Procedure4. Tabu Search Procedure

4

Basic definitionsBasic definitions Consider G=(V,E) a non-oriented graph and

MV.

Definition: v is controlled by MV |NG[v]M| |NG[v]|/2

ExampleM

v1

v2

v3 v4

v5

v6 v7Cont(G,M)

5

Basic definitionsBasic definitions

• ContCont((G,MG,M) ) → set of vertices controlled by M→ set of vertices controlled by M..

• MM defines a defines a monopolymonopoly in in GG ContCont((G,MG,M) = ) = V.V.

0 1 2

3 4 5

M

Given G=(V,E) and MV:

6

Basic definitionsBasic definitions Sandwich Graph

0 1 2

3 4 5

G1=(V,E1)

0 1 2

3 4 5

G=(V,E) where E1E E2

0 1 2

3 4 5

G2=(V,E2)

7

Basic definitionsBasic definitions

Monopoly Verification Problem – MVP

• Given GGiven G11(V,E(V,E11), G), G22(V,E(V,E22) and M) and MV, V,

G=(V,E) s.t. E1 E E2 and M is monopoly nopoly in in GG ? ?

• Solved in polynomial time (Makino, Yamashita, Solved in polynomial time (Makino, Yamashita, Kameda, Kameda, AlgorithmicaAlgorithmica [2002]). [2002]).

8

Basic definitionsBasic definitions

- Max-Controlled Set Problem – MCSP• If the answer to the MVP is If the answer to the MVP is NO,NO, we have the we have the

MCSP!MCSP!

• In the MCSP, we hope to maximize the In the MCSP, we hope to maximize the

number of vertices controlled by M.number of vertices controlled by M.

• The MCSP is NP-hard !! (Makino The MCSP is NP-hard !! (Makino et alet al..

[2002]).[2002]).

9

3

Basic definitionsBasic definitions MCSP

0 1 2

5

4 6

M

Fixed EdgesOptional Edges

Not-controlled vertices

Controlled vertices

10

ContentsContents

1. Basic definitions1. Basic definitions

2. The Generalized Max-Controlled Set 2. The Generalized Max-Controlled Set ProblemProblem

3.3. a) A 1/2-Approximation procedurea) A 1/2-Approximation procedure

b) A Based LP Prog. Heuristicb) A Based LP Prog. Heuristic

c) A combined heuristicc) A combined heuristic

5. Comp. results and final comments 5. Comp. results and final comments

4. Tabu Search Procedure4. Tabu Search Procedure

11

GMCSPGMCSP

f-controlled vertices

• A vertex A vertex iiVV is is -controlled by -controlled by MMV V iffiff, |, |

NNGG[[ii]]MM|-||-|NNGG[[ii]]UU| | i i , , withwith i i ZZ and and UU==V V \ \ M.M.

Vertices not -controlled by M-controlled vertices by M

0 1 2

3 4 5

M(0) (4) (1)

(3) (-2) (4)

f i fixed gaps (for i V)

12

GMCSPGMCSP

We also add positive weights

0 1

4 52 3

M

(0)[2] (0)[3]

(0)[5] (0)[7] (0)[10](0)[1]

Fixed EdgesOptional Edges

Vertices not -controlled-controlled vertices

13

GMCSPGMCSP

Generalized Max-Controlled Set Problem

• INPUT:INPUT: Given G Given G11(V,E(V,E11), G), G22(V,E(V,E22) and M) and MV V (with fixed gaps and positive weights).(with fixed gaps and positive weights).

• OBJECTIVE:OBJECTIVE: We want to find a sandwich We want to find a sandwich graph graph G=(V,E), in order to maximize the sum of the weights of all vertices f-controlled by M.

14

GMCSPGMCSP Reduction Rules:

We fix alloptional edges

We deleteall optional edges

M U=V\M

15

GMCSPGMCSP Reduction Rules

0 1 2

3 4 5

M(0)[1] (0)[1] (0)[1]

(0)[1] (0)[1] (0)[1]

E1D(M,M) E E1D(M,M)D(U,M)

Fixed EdgesOptional Edges

Vertices not -controlled-controlled vertices

16

GMCSPGMCSP Reduction Rules

• Consider the following partition of Consider the following partition of VV::

– MMACAC and and UUAC AC vertices always vertices always -controlled -controlled

– MMNCNC and and UUNC _NC _ vertices never vertices never -controlled -controlled

– MMRR and and UUR R vertices vertices -controlled or not.-controlled or not.

17

GMCSPGMCSP

Reduction Rules

MAC

MR

MNC

UAC

UR

UNC

M U

18

GMCSPGMCSP

Reduction Rules

MAC

MR

MNC

UAC

UR

UNC

M U

optional edges

fixed edges

19

PMCCGPMCCG Reduction Rules

0 1 2

3 4 5

M(0)[1] (0)[1] (0)[1]

(0)[1] (0)[1] (0)[1]

MSC={1}

UNC={5}

Fixed EdgesOptional Edges

Vertices not -controlled by M-controlled vertices by M

20

ContentsContents

1. Basic definitions1. Basic definitions

2. The Generalized Max-Controlled Set 2. The Generalized Max-Controlled Set ProblemProblem

3.3. a) A 1/2-Approximation procedurea) A 1/2-Approximation procedure

b) A Based LP Prog. Heuristicb) A Based LP Prog. Heuristic

c) A combined heuristicc) A combined heuristic

5. Comp. results and final comments 5. Comp. results and final comments

4. Tabu Search Procedure4. Tabu Search Procedure

21

GMCSPGMCSP ½-Approximation algorithm - GMCSP

• Algorithm 1Algorithm 1

1: 1: WW11 Summation of all weights for Summation of all weights for EE==EE11

2: 2: WW22 Summation of all weights for Summation of all weights for EE==EE22

3: 3: zzH1H1 maxmax{{WW11,,WW22}}

22

M(0)[5] (0)[1] (0)[3]

(0)[2] (0)[1] (0)[3]

GMCSPGMCSP

½-approximation for the GMCSP

Not -controlled vertices

f-controlled verticesFixed EdgesOptional Edges

0 1 2

3 4 5

W1=9

W2=7

23

ContentsContents

1. Basic definitions1. Basic definitions

2. The Generalized Max-Controlled Set 2. The Generalized Max-Controlled Set ProblemProblem

3.3. a) A 1/2-Approximation procedurea) A 1/2-Approximation procedure

b) A Based LP Prog. Heuristicb) A Based LP Prog. Heuristic

c) A combined heuristicc) A combined heuristic

5. Comp. results and final comments 5. Comp. results and final comments

4. Tabu Search Procedure4. Tabu Search Procedure

24

GMCSPGMCSP LP formulation

• Consider Consider KK=|=|VV|+|+maxmax{|{|ii| s.t. | s.t. iiVV}}

Vi

ii zpz maxmax

VizK

fxaxa

iMj Uj

iijijijij

,1

Subject to:

1),(,1 Ejixij Vixii ,1

12 \),(},1,0{ EEjixij

Vizi },1,0{

P~

25

GMCSP GMCSP

• ConsiderConsider RRMj

iUj

ijijijiji UMifxaxab

,

M(2)

M

(1)

bi=3 bi=3

1),(,1 Ejixij

Vixii ,1

26

PMCCGPMCCG Stronger LP Formulation

Vi

ii zpz maxmax

RRii

Mj Ujiijijijij

UMizb

fxaxa

,1

Subject to:

1),(,1 Ejixij

Vixii ,1

12 \),(},1,0{ EEjixij

Vizi },1,0{

ACACi UMiz ,1

NCNCi UMiz ,0

P

27

Theorem Theorem : Let and the optimum : Let and the optimum

values of and respectively. Then:values of and respectively. Then:

GMCSPGMCSP

max~z

maxz

P~

P

maxmax~ zz

max~z

maxz

Z*=? Optimum objective value

What about the feasible solutions?

max

28

GMCSPGMCSP

Theorem:Theorem: Consider a relaxed solution of Consider a relaxed solution of

with with .. and . and .

If for some (i,j)If for some (i,j)EE22, then there exists , then there exists

another relaxed solution withanother relaxed solution with

and and

),( zx P

2),(],1,0[ Ejixij Vizi ],1,0[

)1,0(ijx

)ˆ,ˆ( zx

2),(},1,0{ˆ Ejixij Vizi ],1,0[ˆ

29

PMCCGPMCCG Feasible solution based in the Linear

Relaxation

0 1 2

3 4

M

0,5

0,5

0,5

0,5

0 1 2

3 4

M

10

0

1

12 \),(,5,0 EEjixij 12 \),(},1,0{ˆ EEjixij

Fixed edgesOptional edges

Not-controlled vertices

Controlled vertices

30

Integer solution obtained from our stronger Linear Programming formulation.

• Algorithm 2Algorithm 2

– Given a relaxed solution for .Given a relaxed solution for .

– Define as Define as -controlled all vertice -controlled all vertice iiV V with with

, and not , and not -controlled if-controlled if . .

GMCSPGMCSP

),( zx P

1iz 1iz

31

Quality of upper and lower bounds

generated by our stronger formulation P

32

ContentsContents

1. Basic definitions1. Basic definitions

2. The Generalized Max-Controlled Set 2. The Generalized Max-Controlled Set ProblemProblem

3.3. a) A 1/2-Approximation procedurea) A 1/2-Approximation procedure

b) A Based LP Prog. Heuristicb) A Based LP Prog. Heuristic

c) A combined heuristicc) A combined heuristic

5. Comp. results and final comments 5. Comp. results and final comments

4. Tabu Search Procedure4. Tabu Search Procedure

33

MCSPMCSP

• Combined HeuristicCombined Heuristic - CH- CH

• 1) 1) zz11 ½-approximation ½-approximation

• 2) 2) zz22 Based LP Heuristic Based LP Heuristic

• 3) z 3) z max{ max{zz11 , , zz22}}

((Martinhon&Protti, Martinhon&Protti, LNCCLNCC[2002]) [2002])

4,)1(2

1

2

1

nn

n

MCSP Similar combined heuristic with ratio:

34

ContentsContents

1. Basic definitions1. Basic definitions

2. The Generalized Max-Controlled Set 2. The Generalized Max-Controlled Set ProblemProblem

3.3. a) A 1/2-Approximation procedurea) A 1/2-Approximation procedure

b) A Based LP Prog. Heuristicb) A Based LP Prog. Heuristic

c) A combined heuristicc) A combined heuristic

5. Comp. results and final comments 5. Comp. results and final comments

4. 4. Tabu Search ProcedureTabu Search Procedure

35

ContentsContents

1. Basic definitions1. Basic definitions

2. The Generalized Max-Controlled Set 2. The Generalized Max-Controlled Set ProblemProblem

3.3. a) A 1/2-Approximation procedurea) A 1/2-Approximation procedure

b) A Based LP Prog. Heuristicb) A Based LP Prog. Heuristic

c) A combined heuristicc) A combined heuristic

5. Comp. results and final comments 5. Comp. results and final comments

4. Tabu Search Procedure4. Tabu Search Procedure

36

Computational ResultsComputational Results Tabu Search solutions for instances with

50, 75 and 100 vertices.

37

THANK YOU !!

38

GMCSPGMCSP Reduction Rules

• Rule 3Rule 3: Add to : Add to EE11 all edges of D( all edges of D(MMACACMMNCNC, U, URR).).

• Rule 4Rule 4: Remove from : Remove from EE22 the edges the edges

DD((MMRR,U,UACACUUNCNC).).

• Rule 5Rule 5: Add or remove at random the edges : Add or remove at random the edges

D(D(MMACACMMNCNC, U, UACACUUNCNC).).

MAC

MR

MNC

UAC

UR

UNC

M U

39

GMCSPGMCSP

Reduction Rules

• Given two graphs Given two graphs GG11 e e GG22, and 2 subsets , and 2 subsets A,BA,BVV, ,

we define: we define:

DD((A,BA,B)={()={(i,ji,j))EE22\\EE11 | | iiAA, , jjBB}}

• Rule 1Rule 1:: Add to Add to EE11 the edges the edges DD((M,MM,M).).

• Rule 2Rule 2:: Remove from Remove from EE22 the edges the edges DD((U,UU,U).).