Scribd
Upload
113 views24 pages

Percolation: Theory and Applications: Daniel Genin, NIST

This document provides an overview of percolation theory and its applications. It begins with an introduction to percolation problems involving porous rocks and disease spread. The basic bond and site percolation models on 2D grids are described. Key results discussed include the existence of distinct subcritical, critical, and supercritical phases depending on the occupation probability p, with a critical probability pc separating non-percolating and percolating regimes. Quantities of interest like cluster size, percolation probability, and mean cluster size are defined. The document concludes with an example application to modeling network robustness against random or targeted node removal.

Uploaded by

Rejaul Karim
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
113 views24 pages

Percolation: Theory and Applications: Daniel Genin, NIST

This document provides an overview of percolation theory and its applications. It begins with an introduction to percolation problems involving porous rocks and disease spread. The basic bond and site percolation models on 2D grids are described. Key results discussed include the existence of distinct subcritical, critical, and supercritical phases depending on the occupation probability p, with a critical probability pc separating non-percolating and percolating regimes. Quantities of interest like cluster size, percolation probability, and mean cluster size are defined. The document concludes with an example application to modeling network robustness against random or targeted node removal.

Uploaded by

Rejaul Karim
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 24

Percolation: Theory and

Applications
Daniel Genin, NIST
October 17, 2007

OUTLINE

Introduction/Setup

Basic Results

Example of Application

Introduction
Original problem: Broadbent and Hammersley(1957)
Suppose a large porous rock is submerged under water for a long time, will the water reach
the center of the stone?
Related problems:
How far from each other should trees in an orchard (forest) be planted in order to minimize
the spread of blight (fire)?
How infectious does a strain of flu have to be
to create a pandemic? What is the expected
size of an outbreak?
2

Setup: 2D Bond Percolation

Stone: a large two dimensional grid of channels (edges). Edges in the grid are open or
present with probability p (0 p 1) and
closed or absent with probability 1 p.

Pores: open edges and p determines the


porosity of the stone.

A contiguous component of the graph of open


edges is called an open cluster. The water will
reach the center of the stone if there is an open
cluster joining its center with the periphery.
Similarly, in the orchard example, p is the probability that blight will spread to an adjacent
tree and minimizing the spread corresponds to
minimizing the size of the largest open cluster.
3

Setup: 2D Bond Percolation

p=0.25

p=0.48

p=0.52

p=0.75
4

Setup: Bond Percolation


General Bond Percolation Model
The space of the model is Zn or any infinite
graph.
The edges are open or closed with probability p, which may depend on the properties
of the edge (e.g. degree).
Open cluster is a connected component of
the open edge graph.
The network is said to percolate if there
is an infinite open cluster containing the
origin.
If the graph is translation invariant there is no
difference between the origin and any other
vertex.
5

Setup: Site Percolation


Site Percolation Model
The space of the model is Zn or any infinite
graph.
The vertices are open or closed with probability p, which may depend on the properties of the vertex (e.g. degree).
Open cluster is a connected component of
the open vertex graph.
The network is said to percolate if there
is an infinite open cluster containing the
origin.
Every bond percolation problem can be realized as a site percolation problem (on a different graph). The converse is not true.
6

Setup: Why Percolation?

Percolation provides a very simple model


of random media that nevertheless retains
enough realism to make its predictions relevant in applications.

It is a test ground for studying more complicated critical phenomena and a great source
of intuition.

Basic Results: Quantities of Interest

|C| the size of the open cluster at 0,


where C stands for the open cluster itself;

(p) percolation probability, defined as


(p) = Pp(|C| = );

Basic Results: Percolation Probability


Exact shape of (p) is not known but it is believed to be a continuous function of p
( p)
1

p c (d)

Percolation thus has three distinct phases


1) subcritical if p < pc
2) critical if p = pc
3) supercritical if p > pc
9

Basic Results: Quantities of Interest

|C| the size of the open cluster at 0,


where C stands for the open cluster itself;

(p) percolation probability, defined as


(p) = Pp(|C| = );

pc(d) critical probability, defined as


pc(d) = sup{p : (p) = 0};

10

Basic Results: Critical Probability


Theorem. If d 2 then 0 < pc(d) < 1.
The exact value of pc(d) is known only for a
few special cases:
(1) = psite
pbond
c (1) = 1
c
pbond
(2) = 1/2, psite
c
c (2) .59
(triangular lattice) = 2 sin(/18)
pbond
c
(hexagonal lattice) = 1 2 sin(/18)
pbond
c
Theorem. Probability that an infinite open cluster exists is 0 if p < pc(d) and 1 if p > pc(d).
It is known that no infinite open cluster exists
for p = pc(d) if d = 2 or d 19.
11

Basic Results: Critical Probability


Some bounds on the critical probability are
known
Theorem. If G is an infinite connected graph
and maximum vertex degree < . The critical probabilities of G satisfy
1
) .
1 (1 pbond
psite
pbond
c
c
c
1
In particular, pbond
psite
and strict inequality
c
c
holds for a broad family of graphs.

12

Basic Results: Quantities of Interest


|C| the size of the open cluster at 0,
where C stands for the open cluster itself;
(p) percolation probability, defined as
(p) = Pp(|C| = );
pc(d) critical probability, defined as
pc(d) = sup{p : (p) = 0};
(p) the mean size of the open cluster
at the origin, defined as
(p) = Ep[|C|];
f (p) the mean size of the finite open
cluster at the origin, defined as
f (p) = Ep[|C| : |C| < ];
13

Basic Results: Subcritical Phase


If p < pc all open clusters are finite with probability 1.
Theorem. Probability of a cluster of size n at 0
decreases exponentially with n. More precisely,
there exists (p) > 0, (p) as p 0 and
(pc) = 0 such that
Pp(|C| = n) en(p) as n

This also implies that (p) is finite for all p in


the subcritical region.
Theorem. Probability distribution for cluster
radii decays exponentially with the radius, i.e.
Pp(0 B(r)) er/(p)
where (p) the characteristic length of exponential decay is the mean cluster radius.
14

Basic Results: Supercritical Phase


If p > pc, with probability 1 at least one infinite
open cluster exists.
Theorem. The infinite open cluster is unique
with probability 1.
Theorem. Probability of a finite open cluster
of size n at 0 decreases exponentially with n.
More precisely, there exist functions 1(p) and
2(p), satisfying 0 < 2(p) 1(p) < , such
that
exp(1(p)n(d1)/d) Pp(|C| = n)
exp(2(p)n(d1)/d)
Because (p) is infinite for p > pc the truncated
mean f (p) over finite clusters only is
considered.

15

Basic Results: (p)


The general shape of (p) is believed to be as
follows

( p)

( p)

p c (d)

16

Application: Network Robustness and


Fragility
Problem: How many random nodes can be removed before a network looses connectivity?
How many of the highly connected nodes can
be removed before the network looses connectivity?
Use site percolation model on a random graph
with a given degree distribution pk and vertex occupation probability qk depending on the
vertex degree.
Allowing qk to vary with k allows to study various types of attacks: random if qk = q is independent of k, targeted deletion of high degree
nodes if qk = H(kmax k).
17

Application: Network Robustness and


Fragility
Using formalism of generating functions it can
be shown that the generating function H0 of
cluster size |C| at a random vertex satisfies
H0(x) = 1 F0(1) + xF0(H1(x))
1
H1(x) = 1 (F0 (1) + xF0 (H1(x)))
z
F0(x) =

pk q k x k

k=0

and z is the mean graph degree, and


(q) = H0 (1)

18

Application: Network Robustness and


Fragility
Although closed form solutions to the above
equations do not exist in general, it is possible
to compute H0 to any degree of accuracy by iterating equations for H1 and then substituting
into the equation for H0.
In the case qk = q (uniform distribution) it can
be shown that (q) diverges at
qc =

1
G(1)

where G = 1z k pk xk . This is the percolation


threshold probability.
P

19

Application: Network Robustness and


Fragility

pk =

0
if k = 0
Ck ek/ if k 1

20

Application: Network Robustness and


Fragility
If the highest degree vertices are removed first,
qk = H(kmax k), the probability that a random vertex does not belong to the giant open
cluster is
S = 1 H0(1) = F0(1) F0(u)
where u solves
1
u = 1 (F (1) + F (u))
z
These equations can be solved numerically.

21

Application: Network Robustness and


Fragility

22

Bibliography
Reka Albert and Albert Laszlo Barabasi, Statistical mechanics of complex networks, Reviews of modern Physics, 74, Jan. 2002.
Duncan Callaway, M. E. J. Newman, Steven H.
Strogatz, and Duncan J. Watts, Network robustness and fragility: Percolation on random
graphs, arXiv:cond-mat/0007300, Oct. 2000.
Geoffrey Grimmet, Percolation, Grundlehren der
mathematischen Wissenschaft, vol 321, Springer,
1999.
M. E. J. Newman, The structure and function
of complex networks, arXiv:cond-mat/0303516v,
Mar. 2003.
23

You might also like