miniconference on the mathematics of computation

38
1 Burning a graph as a model of social contagion Anthony Bonato Ryerson University 11th Workshop on Algorithms and Models for the Web Graph

Upload: julia-ross

Post on 08-Jan-2018

222 views

Category:

Documents


4 download

DESCRIPTION

Complex networks in the era of Big Data web graph, social networks, biological networks, internet networks, … Graph burning - Anthony Bonato

TRANSCRIPT

Page 1: Miniconference on the Mathematics of Computation

1

Burning a graph as a model of social contagion

Anthony BonatoRyerson University

11th Workshop on Algorithms and Models for the Web Graph

Page 2: Miniconference on the Mathematics of Computation

2

Complex networks in the era of Big Data

• web graph, social networks, biological networks, internet networks, …

Graph burning - Anthony Bonato

Page 3: Miniconference on the Mathematics of Computation

Graph burning - Anthony Bonato 3

Friendship networks• network of friends (some real, some virtual) form

a large web of interconnected links

Page 4: Miniconference on the Mathematics of Computation

Emotions are contagious

Graph burning - Anthony Bonato 4

(Kramer,Guillory,Hancock,14):• study of emotional or social

contagion in Facebook• the underlying network is

an essential factor• in-person interaction and

nonverbal cues are not necessary for the spread of the contagion

Page 5: Miniconference on the Mathematics of Computation

Modelling social influence• general framework:

– nodes are active or inactive– active nodes are introduced and influence the activity

of their neighbours

Graph burning - Anthony Bonato 5

Page 6: Miniconference on the Mathematics of Computation

Models• various models:

– (Kempe, J. Kleinberg, E. Tardos,03)– competitive diffusion (Alon, et al, 2010)

• literature in graph theory:– domination– firefighting

Graph burning - Anthony Bonato 6

Page 7: Miniconference on the Mathematics of Computation

Memes• memes:

– an idea, behavior, or style that spreads from person to person within a culture

Graph burning - Anthony Bonato 7

Page 8: Miniconference on the Mathematics of Computation

Quantifying meme outbreaks• meme breaks out at a node, then spreads to its

neighbors over time

• meme also breaks out at other nodes over discrete time-steps

• how long does it take for all nodes to receive the meme in the network?

Graph burning - Anthony Bonato 8

Page 9: Miniconference on the Mathematics of Computation

Burning number• G a connected, simple graph • there are discrete rounds• each node is either burning or non-burning

– if a node is burning, then it remains in that state • every round, choose an additional non-burning node to burn

– once a node is burning in round t, in round t + 1, each of its non-burning neighbors becomes burning

– chosen nodes: activators• process ends when all nodes are burning

• the burning number of a graph G, written by b(G), is the minimum number of rounds for all nodes to be burning– well-defined, as bounded above by |V(G)| (even (G)+1)

Graph burning - Anthony Bonato 9

Page 10: Miniconference on the Mathematics of Computation

Example: cliques

• b(Kn) = 2

Graph burning - Anthony Bonato 10

Page 11: Miniconference on the Mathematics of Computation

Paths

• burning sequence: (v3,v7,v9)– sequence of activators

Theorem (Bonato,Janssen,Roshanbin,14)

b(Pn) =

Graph burning - Anthony Bonato 11

1 2 32 2 3333

v1 v2 v3 v4 v5 v6 v7 v8 v9

Page 12: Miniconference on the Mathematics of Computation

Proof of lower bound• suppose (x1,…,xk) is a burning sequence for Pn

• then: Nk-1[x1] Nk-2[x2] N0[xk] = V(G) (1)

• as |Ni(x)| ≤ 2i+1 for all nodes x, we have by (1) that:

= 2k(k-1)/2 + k = k2 ≥ n

Graph burning - Anthony Bonato 12

Page 13: Miniconference on the Mathematics of Computation

Trees• rooted tree partition of G:

Graph burning - Anthony Bonato 13

collection of rooted trees which are subgraphsof G, with the property that the node sets of the trees partition V(G)

• x1, x2, x3 are activators

Page 14: Miniconference on the Mathematics of Computation

Trees

Theorem (BJR,14)b(G) ≤ k iff there is a rooted tree partition with trees

T1,T2,…,Tk of height at most

k-1, k-2, …,0 (respectively)such that for all i, j, the roots of Ti and Tj are distance at least |i-j|.

Graph burning - Anthony Bonato 14

Page 15: Miniconference on the Mathematics of Computation

Trees• note: if H is a spanning subgraph of G, then

b(G) ≤ b(H)– a burning sequence for H is also one for G

Corollary (BJR,14)b(G) = min{b(T): T is a spanning tree of G}

Graph burning - Anthony Bonato 15

Page 16: Miniconference on the Mathematics of Computation

BoundsLemma (BJR,14) If H is an isometric subgraph of G, then

b(H) ≤ b(G).– hence, burning number is monotone on subtrees

Corollary (BJR,14)1. b(Cn) =2. If G has a Hamiltonian path, then b(G) ≤

Graph burning - Anthony Bonato 16

Page 17: Miniconference on the Mathematics of Computation

Aside: spider graphs

SP(3,5):

Lemma (BJR,14) b(SP(s,r)) = r+1.

Graph burning - Anthony Bonato 17

Page 18: Miniconference on the Mathematics of Computation

Bounds

Theorem (BJR,14)If G has diameter d and radius r, then

≤ b(G) ≤ r+1.

• tight: – upper bound: spider graphs– lower bound: paths

Graph burning - Anthony Bonato 18

Page 19: Miniconference on the Mathematics of Computation

Coverings

Theorem (BJR,14)If C1,C2,…,Ct cover G, and each Ci is connected of radius at most k, then

b(G) ≤ t + k.

• (G): k-distance domination number

Corollary (BJR,14) i}

Graph burning - Anthony Bonato 19

Page 20: Miniconference on the Mathematics of Computation

How large can the burning number be?

Conjecture (BJR,14): b(G) ≤ .

• by using corollary on we have that:

b(G) ≤ 2-1.

Graph burning - Anthony Bonato 20

Page 21: Miniconference on the Mathematics of Computation

Graph burning - Anthony Bonato 21

Iterated Local Transitivity (ILT) model(Bonato, Hadi, Horn, Prałat, Wang, 08)

• key paradigm is transitivity: friends of friends are more likely friends; eg (Girvan and Newman, 03)– iterative cloning of closed neighbour sets– deterministic; – local: nodes often only have local influence; – evolves over time, but retains memory of initial graph

Page 22: Miniconference on the Mathematics of Computation

Graph burning - Anthony Bonato 22

ILT model

• begin with a graph G = G0

• to form the graph Gt+1 for each vertex x from time t, add a vertex x’, the clone of x, so that xx’ is an edge, and x’ is joined to each neighbor of x

• order of Gt is 2tn0

Page 23: Miniconference on the Mathematics of Computation

Graph burning - Anthony Bonato 23

G0 = C4

Page 24: Miniconference on the Mathematics of Computation

Graph burning - Anthony Bonato 24

Properties of ILT model• average degree increasing to ∞ with time• average distance bounded by constant and

converging, and in many cases decreasing with time; diameter does not change

• clustering higher than in a random generated graph with same average degree

• bad expansion: small gaps between 1st and 2nd eigenvalues in adjacency and normalized Laplacian matrices of Gt

Page 25: Miniconference on the Mathematics of Computation

Burning ILT

• although ILT generates graphs with exponential order/size, the burning number is constant:

Theorem (BJR,14) For all t, b(Gt) ≤ b(G0)+1.

Graph burning - Anthony Bonato 25

Page 26: Miniconference on the Mathematics of Computation

Cartesian grids

Graph burning - Anthony Bonato 26

Page 27: Miniconference on the Mathematics of Computation

Cartesian grids

Theorem (BJR,14)If 1 ≤ m ≤ n, and G is the m x n Cartesian grid, then we have the following:1. If m = O(), then b(G) = 2. If m = ), then b(G) =

Graph burning - Anthony Bonato 27

Page 28: Miniconference on the Mathematics of Computation

Sketch of proof• consider upper bound in the case

m = O() • idea: using a covering by t closed balls of radius

r (diamonds), with r to be determined– gives upper bound for b(G) of t+r by covering theorem

Graph burning - Anthony Bonato 28

Page 29: Miniconference on the Mathematics of Computation

Sketch of proof

• now let r = • (Pralat,14+): for the n x n grid, b(G) =

Graph burning - Anthony Bonato 29

Page 30: Miniconference on the Mathematics of Computation

Complexity

Burning number problem:

Instance: A graph G and an integer k ≥ 2.Question: Is b(G) ≤ k?

Graph burning - Anthony Bonato 30

Page 31: Miniconference on the Mathematics of Computation

Burning a graph is hardTheorem (BJR,14+) The Burning number problem is NP-hard.

Further, it is NP-hard when restricted to any one of the following graph classes:

– planar graphs – disconnected graphs– bipartite graphs

• reduction from planar 3-SAT

Graph burning - Anthony Bonato 31

Page 32: Miniconference on the Mathematics of Computation

Burning a graph is hard

Theorem (BJR,14+) The Burning number problem is NP-hard when restricted to trees of maximum degree 3.

• reduction from a partition problem

Graph burning - Anthony Bonato 32

Page 33: Miniconference on the Mathematics of Computation

Random burning

• select activators at random– we consider uniform choice with replacement

Graph burning - Anthony Bonato 33

Page 34: Miniconference on the Mathematics of Computation

Cost of drunkeness• bR(G): random variable associated with the first

time all vertices of G are burning

• b(G) ≤ bR(G)

• C(G) = bR(G)/b(G): cost of drunkenness

Graph burning - Anthony Bonato 34

Page 35: Miniconference on the Mathematics of Computation

Drunkeness on paths

Theorem (BJPR,14+)

C(Pn) =

– first and second moment methods

Graph burning - Anthony Bonato 35

Page 36: Miniconference on the Mathematics of Computation

Other random burning models• choose activators

1. without replacement2. from non-burning vertices

• for (1), cost of drunkenness on paths is unchanged, asymptotically

• for (2), cost of drunkenness is constant

Graph burning - Anthony Bonato 36

Page 37: Miniconference on the Mathematics of Computation

Future directions

• conjecture: b(G) ≤

• burning in grids – strong, hexagonal, triangular– 3-dimensional

• burning in graph products – Cartesian, strong, categorical

37Graph burning - Anthony Bonato

Page 38: Miniconference on the Mathematics of Computation

Future directions• random graphs and cost of drunkenness

– binomial, regular, geometric random graphs– drunkenness in hypercubes

• graph bootstrap percolation– vertices burn if joined to r >1 burning vertices

• burning in models for complex networks– preferential attachment, copying, geometric models

Graph burning - Anthony Bonato 38