Fractional Combinatorial Games F. Giroire 1 N. Nisse 1 S. Prennes 1 - - PowerPoint PPT Presentation

fractional combinatorial games
SMART_READER_LITE
LIVE PREVIEW

Fractional Combinatorial Games F. Giroire 1 N. Nisse 1 S. Prennes 1 - - PowerPoint PPT Presentation

Fractional Combinatorial Games F. Giroire 1 N. Nisse 1 S. Prennes 1 R. P. Soares 1 , 2 1 COATI, Inria, I3S, CNRS, UNS, Sophia Antipolis, France 2 ParGO Research Group, UFC, Fortaleza, Brazil EURO 2013, stream Graph Searching Roma, July 4th, 2013


slide-1
SLIDE 1

1/17

Fractional Combinatorial Games

  • F. Giroire1
  • N. Nisse1
  • S. Pérennes1
  • R. P. Soares1,2

1 COATI, Inria, I3S, CNRS, UNS, Sophia Antipolis, France 2 ParGO Research Group, UFC, Fortaleza, Brazil

EURO 2013, stream Graph Searching

Roma, July 4th, 2013

Giroire et al. Fractional Combinatorial Games

slide-2
SLIDE 2

2/17

Cops & robber games [Nowakowski and Winkler; Quilliot, 83]

Initialization:

1

C places the cops;

2

R places the robber. Step-by-step: each cop traverses at most 1 edge; the robber traverses at most 1 edge. Robber captured: A cop at same node as robber. Goal: Cop-number=minimum number of cops

Giroire et al. Fractional Combinatorial Games

slide-3
SLIDE 3

2/17

Cops & robber games [Nowakowski and Winkler; Quilliot, 83]

Initialization:

1

C places the cops;

2

R places the robber. Step-by-step: each cop traverses at most 1 edge; the robber traverses at most 1 edge. Robber captured: A cop at same node as robber. Goal: Cop-number=minimum number of cops

Giroire et al. Fractional Combinatorial Games

slide-4
SLIDE 4

2/17

Cops & robber games [Nowakowski and Winkler; Quilliot, 83]

Initialization:

1

C places the cops;

2

R places the robber. Step-by-step: each cop traverses at most 1 edge; the robber traverses at most 1 edge. Robber captured: A cop at same node as robber. Goal: Cop-number=minimum number of cops

Giroire et al. Fractional Combinatorial Games

slide-5
SLIDE 5

2/17

Cops & robber games [Nowakowski and Winkler; Quilliot, 83]

Initialization:

1

C places the cops;

2

R places the robber. Step-by-step: each cop traverses at most 1 edge; the robber traverses at most 1 edge. Robber captured: A cop at same node as robber. Goal: Cop-number=minimum number of cops

Giroire et al. Fractional Combinatorial Games

slide-6
SLIDE 6

2/17

Cops & robber games [Nowakowski and Winkler; Quilliot, 83]

Initialization:

1

C places the cops;

2

R places the robber. Step-by-step: each cop traverses at most 1 edge; the robber traverses at most 1 edge. Robber captured: A cop at same node as robber. Goal: Cop-number=minimum number of cops

Giroire et al. Fractional Combinatorial Games

slide-7
SLIDE 7

2/17

Cops & robber games [Nowakowski and Winkler; Quilliot, 83]

Initialization:

1

C places the cops;

2

R places the robber. Step-by-step: each cop traverses at most 1 edge; the robber traverses at most 1 edge. Robber captured: A cop at same node as robber. Goal: Cop-number=minimum number of cops

Giroire et al. Fractional Combinatorial Games

slide-8
SLIDE 8

2/17

Cops & robber games [Nowakowski and Winkler; Quilliot, 83]

Initialization:

1

C places the cops;

2

R places the robber. Step-by-step: each cop traverses at most 1 edge; the robber traverses at most 1 edge. Robber captured: A cop at same node as robber. Goal: Cop-number=minimum number of cops

Giroire et al. Fractional Combinatorial Games

slide-9
SLIDE 9

2/17

Cops & robber games [Nowakowski and Winkler; Quilliot, 83]

Initialization:

1

C places the cops;

2

R places the robber. Step-by-step: each cop traverses at most 1 edge; the robber traverses at most 1 edge. Robber captured: A cop at same node as robber. Goal: Cop-number=minimum number of cops

Giroire et al. Fractional Combinatorial Games

slide-10
SLIDE 10

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

An Observer must ensure that a Surfer never reaches a dangerous node

Giroire et al. Fractional Combinatorial Games

slide-11
SLIDE 11

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

An Observer must ensure that a Surfer never reaches a dangerous node

Giroire et al. Fractional Combinatorial Games

slide-12
SLIDE 12

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

Turn by turn: Observer marks k = 2 nodes

Giroire et al. Fractional Combinatorial Games

slide-13
SLIDE 13

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

Turn by turn: Observer marks k = 2 nodes then Surfer may move on a adjacent node

Giroire et al. Fractional Combinatorial Games

slide-14
SLIDE 14

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

Turn by turn: Observer marks k = 2 nodes then Surfer may move on a adjacent node

Giroire et al. Fractional Combinatorial Games

slide-15
SLIDE 15

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

Turn by turn: Observer marks k = 2 nodes then Surfer may move on a adjacent node

Giroire et al. Fractional Combinatorial Games

slide-16
SLIDE 16

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

Turn by turn: Observer marks k = 2 nodes then Surfer may move on a adjacent node

Giroire et al. Fractional Combinatorial Games

slide-17
SLIDE 17

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

Turn by turn: Observer marks k = 2 nodes then Surfer may move on a adjacent node

Giroire et al. Fractional Combinatorial Games

slide-18
SLIDE 18

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

Turn by turn: Observer marks k = 2 nodes then Surfer may move on a adjacent node

Giroire et al. Fractional Combinatorial Games

slide-19
SLIDE 19

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

Turn by turn: Observer marks k = 2 nodes then Surfer may move on a adjacent node

Giroire et al. Fractional Combinatorial Games

slide-20
SLIDE 20

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

Turn by turn: Observer marks k = 2 nodes then Surfer may move on a adjacent node

Giroire et al. Fractional Combinatorial Games

slide-21
SLIDE 21

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

Turn by turn: Observer marks k = 2 nodes then Surfer may move on a adjacent node

Giroire et al. Fractional Combinatorial Games

slide-22
SLIDE 22

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

Turn by turn: Observer marks k = 2 nodes then Surfer may move on a adjacent node

Giroire et al. Fractional Combinatorial Games

slide-23
SLIDE 23

3/17

Surveillance game

[Fomin,Giroire,Mazauric,Jean-Marie,Nisse 12]

In this example, all nodes are marked Victory of the Observer using 2 marks per turn

Giroire et al. Fractional Combinatorial Games

slide-24
SLIDE 24

4/17

Model: another Two players game

a Surfer starts from safe homebase v0 in G, a dangerous graph a Guard with some amount k of bullets Turn by turn:

1 the guard secures ≤ k nodes; 2 then, the Surfer may move to an adjacent node.

Defeat: Surfer in unsafe node Victory: G safe Minimize amount of bullets to win for any Surfer’s trajectory Surveillance number of G (connected) from v0: sn(G, v0)

Giroire et al. Fractional Combinatorial Games

slide-25
SLIDE 25

5/17

Two Players Combinatorial Games

Two players play a game on a graph. Game is played turn-by-turn. Players play by moving and/or adding tokens on vertices

  • f the graph.

Optimization problem: minimizing number of tokens to achieve some goal

Giroire et al. Fractional Combinatorial Games

slide-26
SLIDE 26

6/17

All these games are hard

Cops and Robber: k cops are enough? PSPACE-complete in general graphs [Mamino, 2012]. Surveillance Game: k marks per turn are enough? k = 2 NP-complete for Chordal/Bipartite Graphs [Fomin et al, 2012]. k = 4 PSPACE-complete for DAGS [Fomin et al, 2012]. Angels and Devils: Does an Angel of power k wins? 1-Angel loses in (infinite) grids [Conway, 1982]. 2-Angel wins in (infinite) grids [Máthé, 2007]. Eternal Dominating set. Eternal Vertex Cover. NP-hard [Fomin et al, 2010].

Giroire et al. Fractional Combinatorial Games

slide-27
SLIDE 27

7/17

New tools/approaches are required

Several questions remain open Meyniel conjecture: cn(G) = O(√n) in any n-node graphs? Polynomial-time Approximation algorithms? less difficult but still intriguing number of cops to capture fast robber in grids? cost of connectivity in surveillance game? etc. Here, we present preliminary results of our new approach

Giroire et al. Fractional Combinatorial Games

slide-28
SLIDE 28

8/17

Fractional Combinatorial Game

Fractional games: both players can use “fractions” of tokens. c

c

c

c

c

c

r

r

r

r

r

r

Semi-Fractional games:

  • nly one player (Player C) can use fractions of tokens. c

c

c

c

c

c

Integral games: classical games, token are unsplittable

Giroire et al. Fractional Combinatorial Games

slide-29
SLIDE 29

9/17

Example: Fractional Cops and Robber

integral game: cop-number = 2 1 2 3 4 5 6 7 8 Player C can use fractions of tokens.

Giroire et al. Fractional Combinatorial Games

slide-30
SLIDE 30

9/17

Example: Fractional Cops and Robber

integral game: cop-number = 2 semi fractional: cop-number ≤ 3/2 1 2 3 4 5 6 7 8

c c c

3 half cops are placed on the graph.

Giroire et al. Fractional Combinatorial Games

slide-31
SLIDE 31

9/17

Example: Fractional Cops and Robber

integral game: cop-number = 2 semi fractional: cop-number ≤ 3/2 1 2 3 4 5 6 7 8

c c c

r Robber takes a position.

Giroire et al. Fractional Combinatorial Games

slide-32
SLIDE 32

9/17

Example: Fractional Cops and Robber

integral game: cop-number = 2 semi fractional: cop-number ≤ 3/2 1 2 3 4 5 6 7 8

c c

r

c

Cop moves.

Giroire et al. Fractional Combinatorial Games

slide-33
SLIDE 33

9/17

Example: Fractional Cops and Robber

integral game: cop-number = 2 semi fractional: cop-number ≤ 3/2 1 2 3 4 5 6 7 8

c c

r

c

Robber moves.

Giroire et al. Fractional Combinatorial Games

slide-34
SLIDE 34

9/17

Example: Fractional Cops and Robber

integral game: cop-number = 2 semi fractional: cop-number ≤ 3/2 1 2 3 4 5 6 7 8

c c

r

c

Cop moves.

Giroire et al. Fractional Combinatorial Games

slide-35
SLIDE 35

9/17

Example: Fractional Cops and Robber

integral game: cop-number = 2 semi fractional: cop-number ≤ 3/2 1 2 3 4 5 6 7 8

c c

r

c

Robber moves.

Giroire et al. Fractional Combinatorial Games

slide-36
SLIDE 36

9/17

Example: Fractional Cops and Robber

integral game: cop-number = 2 semi fractional: cop-number ≤ 3/2 1 2 3 4 5 6 7 8

c

c Arrested! Cop moves capturing the robber.

Giroire et al. Fractional Combinatorial Games

slide-37
SLIDE 37

9/17

Example: Fractional Cops and Robber

integral game: cop-number = 2 semi fractional: cop-number ≤ 3/2 Remark: by definition: semi-fractional ≤ integral gap? relationship with fractional? 1 2 3 4 5 6 7 8

c

c Arrested! Cop moves capturing the robber.

Giroire et al. Fractional Combinatorial Games

slide-38
SLIDE 38

10/17

Preliminary results

Fractional games general framework: fractional relaxation of turn-by-turn games important property: “convexity" of winning states Semi-fractional = fractional (properties of robber’s moves) solutions of fractional games provides lower bounds for integral games Algorithm A to decide which player wins tools: linear programming techniques. Bad news: one step of A is exponential (exponent: length of the game) Hope: use specifities of games to reduce time-complexity Integrality gap Bad news: fractional cop-number ≤ 1 + ǫ for any graph and any ǫ > 0 Hope: surveillance game: fractional game gives a probabilistic log n-approximation

Giroire et al. Fractional Combinatorial Games

slide-39
SLIDE 39

11/17

States of the Game

In n-node graph c ∈ Rn

+ represents the tokens of Player C.

r ∈ Rn

+ represents the tokens of Player R.

(c, r) ∈ R2n

+ represents the state of the game.

1 2 3

c

c

c

r r

c = (0.7, 0.2, 0.1) and r = (0, 0.5, 0.5); set of states = polytope Examples: cops and robber:

i≤n ri = 1 and i≤n ci = k (# of cops)

surveillance game:

i≤n ri = 1

Giroire et al. Fractional Combinatorial Games

slide-40
SLIDE 40

12/17

Winning states and moves

winning states = convex subset of states cops and robber: {(c, r) | ci ≥ ri, i = 1 · · · , n} surveillance game: {(c, r) | ci ≥ 1, i = 1 · · · , n} moves slide tokens along edges = multiplication by stochastic matrix in   [αi,j]1≤i,j≤n

  • ∀1 ≤ i, j ≤ n, αi,j ≥ 0, and

∀j ≤ n,

1≤i≤n αi,j = 1, and

if {i, j} / ∈ E(G) then αi,j = 0    if ij ∈ E, an amount αi,j of the token in vj goes to vi mark nodes = add to c a vector in {(m1, · · · , mn) |

  • i≤n

mi ≤ k}

Giroire et al. Fractional Combinatorial Games

slide-41
SLIDE 41

13/17

Main Idea of Algorithm 1/2

Ri−1: states from which Player C always wins in at most i − 1 rounds, when Player R is the first to play. Ci = {(c, r) | ∃move, (move(c), r) ∈ Ri−1}

Ri−1

Giroire et al. Fractional Combinatorial Games

slide-42
SLIDE 42

13/17

Main Idea of Algorithm 1/2

Ri−1: states from which Player C always wins in at most i − 1 rounds, when Player R is the first to play. Ci = {(c, r) | ∃move, (move(c), r) ∈ Ri−1} Ci: states from which Player C always wins in at most i rounds when playing first Ci = polytope, computable from Ri−1, polynomial-size but in higher dimension Problem: projection

Ri−1 Ci

Giroire et al. Fractional Combinatorial Games

slide-43
SLIDE 43

14/17

Main Idea of Algorithm 2/2

Ci: states from which Player C always wins in at most i rounds when playing first Ri = {(c, r) | ∀move, (c, move(r)) ∈ Ci}

Ci

Giroire et al. Fractional Combinatorial Games

slide-44
SLIDE 44

14/17

Main Idea of Algorithm 2/2

Ci: states from which Player C always wins in at most i rounds when playing first Ri = {(c, r) | ∀move, (c, move(r)) ∈ Ci} Ri: states from which Player C always wins in at most i rounds, when Player R is the first to play. Ri = polytope, computable from Ci, polynomial-size. Trick: “just" have to reenforce each constraint

Ci Ri

Giroire et al. Fractional Combinatorial Games

slide-45
SLIDE 45

15/17

Bad news and Good news

Fractional cop-number is one :( Strategy: place f1 = 1/n cop per node At step i,

1 hi =

j≤i fj cop “follows" the robber.

1 − hi cop remains

2 place fi+1 = 1−hi

n

cop per node. hi →i→∞ 1 Meyniel conjecture seems safe... approximation for surveillance number? :)

Surveillance game: inequality defining the polytopes are similar to set cover Proof based on approximation of set cover for the moment: only probabilistic strategy

Giroire et al. Fractional Combinatorial Games

slide-46
SLIDE 46

16/17

Conclusion and Future Work

Promising framework (we hope) Lot of work remains: polynomial algorithm? (it is if length of game is bounded) approximation? fast robber? careful analysis of polytopes in each game etc.

Giroire et al. Fractional Combinatorial Games

slide-47
SLIDE 47

17/17

Thank you

Giroire et al. Fractional Combinatorial Games