1. The document describes a mathematical model of interacting particle systems or cellular automata on graphs and lattices.
2. In the model, sites can change color based on the colors of neighboring sites within a given range.
3. The document analyzes the long-term behavior of the systems, such as whether they reach consensus on one color or fluctuate between colors.
Circular Systems Emulating Dynamical Influence Processes
1. Moses Boudourides
University of Patras, Greece
4th Joint Japan-North America Mathematical Sociology Conference
Redondo Beach, CA, May 29 – June 1, 2008
2. • Sites are ether vertices of a finite or an
infinite (but locally finite) graph G or cells
in a lattice Zd (or a hypercube in Zd).
• Each site is in one of κ (≥ 2) colors. Let
ξt(x) denote the color of site x at time t.
• Initially, colors are distributed randomly
(uniformly and independently).
• Rule of interaction (color changes):
If ξt(x) = k, then ξt+1(x) = k + 1 mod κ,
provided that there exist at least θ sites in
the color k + 1 mod κ in the neighborhood
of x within range ρ.
3. Iterations occur according to one of
the following two scenarios:
• Random timing: At each time of a
rate 1 Poisson process, a randomly
chosen site is updated (Interacting
Particle System or IPS).
• Deterministic timing: At each time
(positive integer), all sites are
successively (and independently)
updated (Cellular Automaton or
CA).
4. When κ = 2 and θ = ρ = 1, we have the
(classic) voter model (introduced
independently by Clifford & Sudbury 1973,
Holley & Liggett 1975).
Asymptotic behavior of the IPS on Zd
(Holley & Liggett 1975):
• If d ! 2, then P(ξt(x) = ξt(y)) ! 1
(clustering or complete concensus).
• If d ! 3 and p is the density of the initial
product measure, then ξt
p converges in
distribution to ξ!
p, a one parameter family
of stationary distributions.
5. Let G be a graph of n vertices of one of
the following types:
• Erdös-Rényi random graph
• Barabási-Albert preferential attachment
graph
• Bollobás-Chung or Newman-Watts small-
world graph
Asymptotic behavior of the IPS on G
(Durrett 2007):
The consensus time for ξt
p is O(cpn) (i.e.,
the d ! 3 behavior).
6. IPS: Bramson & Griffeath 1989
CA: Fisch 1990
On Z1, when ρ = 1 and θ = 1,
if κ ! 4, ξt fluctuates,
if κ ≥ 5, ξt fixates,
where ξt fluctuates if, ! x ! Z1,
P(ξt(x) changes color at arbitrarily large t) = 1,
while ξt fixates if, ! x ! Z1,
P(ξt(x) changes color finitely often) = 1.
7.
8. Let δ be a fixed integer in:
1 ! δ ! κ – 1.
If ξt(x) = k, then ξt+1(x) is the
closest color in the set of all colors
at range δ from k:
{k + 1, k + 2, …, k + δ} (mod κ)
that are present in the
neighborhood of x within range ρ.
9. On Z1, when 1 ! δ ! κ – 1 and ρ
≥ 1,
•if θ > ρ, ξt fixates,
•if θ ! ρ, depending on κ and δ,
ξt either fluctuates in one color
(1-fluctuation) or fluctuates
alternating in two colors within
range δ (2-fluctuation).
10.
11.
12. • Square domains LxL,
• with wrap-around boundary
conditions.
• Two types of neighborhoods within
range ρ = 1:
•ρ = 1D, diamonds (von
Newmann nbhd),
•ρ = 1B, boxes (Moore nbhd).
13.
14.
15. • External Forcing: Color λ is pushed
‘externally’ at a site, in the sense
that an m-tuple of extra neighbors
(‘influentials’) in color λ are
appended to the site.
• Internal Propensity: Color µ is
promoted ‘internally’ at a site, i.e., it
is inserted at rank q within the δ
range at that site so that the more
lower q is, the more easily the site is
‘influenced’ by color µ.
16. • The effect of external forcing of any
m-tuples of a color is much lower
than that of internal propensity that
places the same color at the first
upper position.
• This verifies Watts’ & Dodds’
disaffirmation of the ‘influentials
hypothesis.’
17.
18. • The effect of internal propensity
fades away as the placement position
of the preferred color increases in
the δ range.
• Moreover, the preferred color of
internal propensity may become the
initiator of other colors that turn out
to be inadvertently influential.