Connected Treewidth and Connected Graph Searching Pierre Fraigniaud - - PowerPoint PPT Presentation

connected treewidth and connected graph searching
SMART_READER_LITE
LIVE PREVIEW

Connected Treewidth and Connected Graph Searching Pierre Fraigniaud - - PowerPoint PPT Presentation

Connected Treewidth and Connected Graph Searching Pierre Fraigniaud 1 Nicolas Nisse 2 CNRS, LRI, Universit e Paris-Sud, France. LRI, Universit e Paris-Sud, France. LATIN 05, March 21 th , 2006 1/16 Pierre Fraigniaud, Nicolas Nisse


slide-1
SLIDE 1

1/16

Connected Treewidth and Connected Graph Searching

Pierre Fraigniaud1 Nicolas Nisse2

CNRS, LRI, Universit´ e Paris-Sud, France. LRI, Universit´ e Paris-Sud, France.

LATIN 05, March 21th, 2006

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-2
SLIDE 2

2/16

Graph Searching

Goal In a contaminated network, an invisible omniscient arbitrary fast fugitive ; a team of searchers ; We want to find a strategy that catch the fugitive using the fewest searchers as possible. Motivations network security, speleological rescue... game related to well known graphs’parameters : treewidth and pathwidth ;

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-3
SLIDE 3

2/16

Graph Searching

Goal In a contaminated network, an invisible omniscient arbitrary fast fugitive ; a team of searchers ; We want to find a strategy that catch the fugitive using the fewest searchers as possible. Motivations network security, speleological rescue... game related to well known graphs’parameters : treewidth and pathwidth ;

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-4
SLIDE 4

3/16

Search Strategy, Parson. [GTC,1978]

Sequence of three basic operations,. . .

1 Place a searcher at a vertex of the graph ; 2 Move a searcher along an edge of the graph ; 3 Remove a searcher from a vertex of the graph.

. . . that must result in catching the fugitive The fugitive is caugth when it meets a searcher at a vertex or in an edge of the graph. We want to minimize the number of searchers. Let s(G) be the smallest number of searchers needed to catch a fugitive in a graph G.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-5
SLIDE 5

3/16

Search Strategy, Parson. [GTC,1978]

Sequence of three basic operations,. . .

1 Place a searcher at a vertex of the graph ; 2 Move a searcher along an edge of the graph ; 3 Remove a searcher from a vertex of the graph.

. . . that must result in catching the fugitive The fugitive is caugth when it meets a searcher at a vertex or in an edge of the graph. We want to minimize the number of searchers. Let s(G) be the smallest number of searchers needed to catch a fugitive in a graph G.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-6
SLIDE 6

3/16

Search Strategy, Parson. [GTC,1978]

Sequence of three basic operations,. . .

1 Place a searcher at a vertex of the graph ; 2 Move a searcher along an edge of the graph ; 3 Remove a searcher from a vertex of the graph.

. . . that must result in catching the fugitive The fugitive is caugth when it meets a searcher at a vertex or in an edge of the graph. We want to minimize the number of searchers. Let s(G) be the smallest number of searchers needed to catch a fugitive in a graph G.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-7
SLIDE 7

3/16

Search Strategy, Parson. [GTC,1978]

Sequence of three basic operations,. . .

1 Place a searcher at a vertex of the graph ; 2 Move a searcher along an edge of the graph ; 3 Remove a searcher from a vertex of the graph.

. . . that must result in catching the fugitive The fugitive is caugth when it meets a searcher at a vertex or in an edge of the graph. We want to minimize the number of searchers. Let s(G) be the smallest number of searchers needed to catch a fugitive in a graph G.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-8
SLIDE 8

4/16

Simple Examples : Path and Ring

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-9
SLIDE 9

4/16

Simple Examples : Path and Ring

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-10
SLIDE 10

4/16

Simple Examples : Path and Ring

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-11
SLIDE 11

4/16

Simple Examples : Path and Ring

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-12
SLIDE 12

4/16

Simple Examples : Path and Ring

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-13
SLIDE 13

4/16

Simple Examples : Path and Ring

s(Path)=1

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-14
SLIDE 14

4/16

Simple Examples : Path and Ring

s(Path)=1

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-15
SLIDE 15

4/16

Simple Examples : Path and Ring

s(Path)=1

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-16
SLIDE 16

4/16

Simple Examples : Path and Ring

s(Path)=1

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-17
SLIDE 17

4/16

Simple Examples : Path and Ring

s(Path)=1

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-18
SLIDE 18

4/16

Simple Examples : Path and Ring

s(Path)=1 s(Ring)=2

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-19
SLIDE 19

5/16

Tree and Path Decompositions

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-20
SLIDE 20

5/16

Tree and Path Decompositions

a tree T and bags (Xt)t∈V (T)

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-21
SLIDE 21

5/16

Tree and Path Decompositions

a tree T and bags (Xt)t∈V (T) every vertex of G is at least in one bag ;

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-22
SLIDE 22

5/16

Tree and Path Decompositions

a tree T and bags (Xt)t∈V (T)

every vertex of G is at least in one bag ;

both ends of an edge of G are at least in one bag ;

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-23
SLIDE 23

5/16

Tree and Path Decompositions

a tree T and bags (Xt)t∈V (T)

every vertex of G is at least in one bag ; both ends of an edge of G are at least in one bag ;

For any vertex of G, all bags that contain it, form a subtree.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-24
SLIDE 24

5/16

Tree and Path Decompositions

a tree T and bags (Xt)t∈V (T)

every vertex of G is at least in one bag ; both ends of an edge of G are at least in one bag ; For any vertex of G, all bags that contain it, form a subtree.

Width = Size of largest Bag -1

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-25
SLIDE 25

5/16

Tree and Path Decompositions

a tree T and bags (Xt)t∈V (T)

every vertex of G is at least in one bag ; both ends of an edge of G are at least in one bag ; For any vertex of G, all bags that contain it, form a subtree.

Width = Size of largest Bag -1

treewidth of G tw(G), minimum width among any tree-decomposition

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-26
SLIDE 26

5/16

Tree and Path Decompositions

a path P and bags (Xt)t∈V (P)

every vertex of G is at least in one bag ; both ends of an edge of G are at least in one bag ; For any vertex of G, all bags that contain it, form a subpath.

Width = Size of largest Bag -1

pathwidth of G pw(G), minimum width among any path-decomposition

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-27
SLIDE 27

6/16

Relationship between search number and pathwidth

Ellis, Sudborough and Turner. [Inf. Comput.,1994] For any graph G, vs(G) ≤ s(G) ≤ vs(G) + 2

  • Kinnersley. [IPL.,1992]

For any graph G, vs(G) = pw(G) For any n-node graph G : pw(G) ≤ s(G) ≤ pw(G) + 2

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-28
SLIDE 28

7/16

Connected Graph Searching

Limits of the Parson’s model Searchers cannot move at will in a real network ; It would be better to let searchers be grouped.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-29
SLIDE 29

7/16

Connected Graph Searching

Limits of the Parson’s model Searchers cannot move at will in a real network ; It would be better to let searchers be grouped. Connected Search Strategy At any step, the cleared part of the graph must induced a connected subgraph. Let cs(G) be the connected search number of the graph G.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-30
SLIDE 30

8/16

Cost of connectedness : case of trees

Barri` ere, Flocchini, Fraigniaud and Santoro. [SPAA, 2002] Linear Algorithm Barri` ere, Fraigniaud, Santoro and Thilikos. [WG, 2003] For any tree T, s(T) ≤ cs(T) ≤ 2 s(T) − 2. Moreover, these bounds are tight.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-31
SLIDE 31

9/16

Cost of connectedness : case of arbitrary graphs

Seymour and Thomas. [Combinatorica, 1994] Bond Carving Fomin, Fraigniaud and Thilikos. [Technical repport, 2004] Using a branch-decomposition, polynomial constructive algorithm that computes a connected search strategy. For any connected graph G, cs(G) ≤ s(G) (2 + log |E(G)|).

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-32
SLIDE 32

10/16

Connected Treewidth

Connected e-cut of a tree-decomposition (T, X) The edge e is said connected if both G[T1(e)] and G[T2(e)] induced connected subgraphs of G. Connected tree-decomposition (T, X) For any e ∈ E(T), e is connected. Connected treewidth, ctw(G)

1

T (e)

e

T (e)

2

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-33
SLIDE 33

11/16

Connected Treewidth

For any connected graph G, ctw(G) = tw(G)

  • Golumbic. Algorithmic graph theory and perfect graphs

A “clique tree” of a minimal triangulation H of a connected graph G is an optimal tree-decomposition of G. Parra and Scheffler. [DAM 1997] A “clique tree” of a minimal triangulation H of a connected graph G is a connected tree-decomposition of G.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-34
SLIDE 34

12/16

Results (1)

Theorem 1 : new proof For any connected graph G, ctw(G) = tw(G) Constructive proof Given a tree-decomposition of width ≤ k of a connected graph G with n vertices, our algorithm computes a connected tree-decomposition of width ≤ k of G, in time O(n.k3).

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-35
SLIDE 35

13/16

Results (2)

Theorem 2 For any connected graph G, cs(G) ≤ s(G) (1 + log2 |V (G)|). Constructive proof Given a tree-decomposition of a graph G, our algorithm computes a connected search strategy for G, using at most tw(G) log |V (G)| searchers, in polynomial time.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-36
SLIDE 36

14/16

Squetch of the proof of Theorem 2

For any connected n-node graph G, cs(G) ≤ s(G) (1 + log2 n) proof by induction on n Robertson and Seymour. Graph Minors II. Algorithmic Aspects of Tree-Width. J. of Alg 7, 1986.

For any tree-decomposition (T, X) of a n-node graph G, there are one (or two adjacent vertices) of T such that : for any 1 ≤ j ≤ r, |G[Tj]| ≤ n/2 1 i

T T

...

T T

i+1 r 1 i r

T T T

... ... ...

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-37
SLIDE 37

15/16

Squetch of the proof of Theorem 2

Starting from a connected tree-decomposition of G

T

1

Ti Tr

... ...

For any 1 ≤ i ≤ r, G[Ti] is a connected subgraph with at most n/2 vertices.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-38
SLIDE 38

15/16

Squetch of the proof of Theorem 2

Starting from a connected tree-decomposition of G

< tw(G) (log(n/2)+1) searchers

T

1

Ti Tr

... ...

There is a connected search strategy for G[T1], using at most tw(G)(log(n/2) + 1) searchers.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-39
SLIDE 39

15/16

Squetch of the proof of Theorem 2

Starting from a connected tree-decomposition of G

< tw(G) (log(n/2)+1) searchers < tw(G) searchers

T

1

Ti Tr

to avoid recontamination

... ...

At most tw(G) searchers are required to protect G[T1] from recontamination from the remaining part of G.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-40
SLIDE 40

15/16

Squetch of the proof of Theorem 2

Starting from a connected tree-decomposition of G

< tw(G) (log(n/2)+1) searchers < tw(G) searchers

T

1

Ti Tr

to avoid recontamination

... ...

Then we can terminate the clearing of G[T1].

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-41
SLIDE 41

15/16

Squetch of the proof of Theorem 2

Starting from a connected tree-decomposition of G

< tw(G) searchers < tw(G) (log(n/2)+1) searchers

T

1

Ti Tr

to avoid recontamination

... ...

Then we can use our tw(G)(log(n/2) + 1) searchers to clear another subgraph G[Ti], and so on...

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-42
SLIDE 42

15/16

Squetch of the proof of Theorem 2

Starting from a connected tree-decomposition of G

< tw(G) + tw(G)(log(n/2)+1) searchers

T

1

Ti Tr

... ...

Connected search strategy using at most tw(G)(log n + 1)

  • searchers. Thus, cs(G) ≤ s(G)(log n + 1)

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching

slide-43
SLIDE 43

16/16

Conclusion and Further Work

Cost of connectedness new upper bound of the ratio cs(G)/s(G) constructive algorithm Open problems What is the optimal bound ? In trees : cs(T)/s(T) ≤ 2 and this bound is tight [Barri` ere et al.]. If the fugitive is visible : cs(G)/s(G) ≤ log n and this bound is tight. Is the problem of computing cs(G) NP-complete ? It is known to be NP-hard.

Pierre Fraigniaud, Nicolas Nisse Connected Treewidth and Connected Graph Searching