Voter Master 2019

Voter Model
R. Holley and T. M. Liggett, Ann. Probab. 3, 643 (1975).

Clifford and Sudbury, Biometrika (1973)
Liggett T M 1985 Interacting Particle Systems (New York: Springer)

Liggett T M 1999 Stochastic Interacting Systems: Contact,
Voter and Exclusion Processes (New York:Springer)
C. Castellano, S. Fortunato, and V. Loreto, Rev. Mod. Phys. 81,

591 (2009).
P. L. Kaprisky, S. Redner and E. Ben Naim, 2010

A kinetic view of Statistical Physics CH. 8.2
(Cambridge University Press)
Updating rules and the voter model

Fernández-Gracia, Juan
Master Thesis , (2011)
Voter Model
Voter model dynamics in complex networks: Role of dimensionality,
Suchecki, Krzysztof; Eguíluz Víctor M.;San Miguel, Maxi
Physical Review E 72, 036132(1-8) (2005)
Conservation laws for the voter model in complex networks

Suchecki, Krzysztof; Eguiluz, Victor M.; San Miguel, Maxi
Europhysics Letters 69, 228-234 (2005)
Generic absorbing transition in coevolution dynamics

Vazquez, F.; Eguiluz, V. M.; San Miguel, M.
Physical Review Letters 100, 108702 (1-4) (2008)
Analytical Solution of the Voter Model on Uncorrelated Networks

Vazquez, F.; Eguiluz, V. M.
New Journal of Physics 10 No.6, 063011 (1-19) (2008)
Conservation laws for voter-like models on random directed networks

Serrano, M. Ángeles; Klemm, Konstantin; Vazquez, Federico; Eguíluz, Victor M.; San Miguel, Maxi
Journal of Statistical Mechanics: Theory and Experiment , P10024 (2009)
Update rules and interevent time distributions: Slow ordering vs. no ordering in the Voter Model
Fernández-Gracia, Juan;M.Eguíluz,Víctor;San Miguel,Maxi
Physical Review E 84, 015103 (2011))
A measure of individual role in collective dynamics: spreading at criticality

Klemm, Konstantin ; Serrano, M. Angeles; Eguiluz, Victor M. ;San Miguel,Maxi
Scientific Reports 2, 292 (2012)
Is the Voter Model a model for voters?

Fernandez-Gracia, J; Suchecki, K; Ramasco, JJ; San Miguel, M; Eguiluz, VM
Physical Review Letters 112, 158701 (2014)
Voter Model
Absorbing and Shattered Fragmentation Transitions in Multilayer Coevolution
Diakonova, Marina; San Miguel, Maxi; Eguiluz, Victor
Physical Review E 89, 06218 (2014)
Noise in Coevolving Networks

Diakonova, Marina; Eguiluz, Victor M.; San Miguel, Maxi
Irreducibility of multilayer network dynamics

Diakonova, Marina; Nicosia, Vincenzo; Latora, Vito; San Miguel, Maxi
New Journal of Physics 18, 023010 (2016)
Dynamical origins of the community structure of multi-layer societies

Klimek, Peter; Diakonova, Marina; Eguiluz, Victor M.; San Miguel, Maxi; Thurner, Stefan
The noisy voter model on complex networks

Carro, Adrián; Toral, Raúl; San Miguel, Maxi
Dynamics on networks: competition of temporal and topological correlations

Artime, Oriol; Ramasco, José J.; San Miguel, Maxi
Joint effect of ageing and multilayer structure prevents ordering in the voter model
Artime, Oriol; Fernández-Gracia, Juan; Ramasco, José J.; San Miguel, Maxi
Fragmentation transitions in a coevolving nonlinear voter model

Min, Byungjoon; San Miguel, Maxi
Zealots in the mean-field noisy voter model

Khalil, Nagi; San Miguel, Maxi; Toral, Raul
Physical Review E 97, (2018)
Voter Model
Stochastic pair approximation treatment of the noisy voter model

Peralta, A. F. ; Carro, A. ; San Miguel, M. ; Toral, R.
Aging-induced continuous phase transition

Artime, Oriol; Peralta, Antonio F.; Toral, Raúl; Ramasco, José J.; San Miguel, Maxi
Analytical and numerical study of the non-linear noisy voter model on complex networks
Peralta, A. F. ; Carro, A. ; San Miguel, M. ; Toral, R.
Chaos 28, 075516 (2018)
Coevolving nonlinear voter model with triadic closure

Raducha,Tomasz;Min,Byungjoon;San Miguel,Maxi
Europhysics Letters 124, 3001 (2018)
Multilayer coevolution dynamics of the nonlinear voter model

Min, Byungjoon; San Miguel, Maxi
New Journal of Physics (2019)
MODELS of SOCIAL CONSENSUS
Determine when and how the

dynamics of a set of interacting
CONSENSUS PROBLEMS:
agents that can choose among
several options (political vote,
opinion, cultural features,…) leads to
Dynamics of binary option
a consensus in one of these
opinions, or when a state with
several coexisting options prevails.
INTERACTIONS: Mechanisms (“rule”) and Network (with whom)
-VOTER MODEL MODELS -Imitation MECHANISMS

-SPIN FLIP KINETIC ISING MODEL (T=0) -Following majority. Social pressure
-AXELROD MODEL -Homophily
-GRANOVETTER’S MODEL -Threshold for social pressure
IMITATION
Herding
Behavior
Voter Model
Q?: If the only mechanism of interaction is pairwise imitation, when do
we collectively reach agreement, or else coexistence of states persist?
Ingredients of interacting agents simulation:
a) Mechanism: Ex.: Voter model (imitation)

Imperfect imitation: Noise, Zealots
Group interaction: nonlinear VM
b) Who interacts with whom?
Complex networks: Tie heterogeneity
Co-evolution : Ties are not persistent
Context: multilayer/multiplex/interconnected
c) When do elements interact? Interaction activity

Constant rate or Temporal Heterogeneity
Memory, Aging
Voter Model
-Definition and order parameter
-Results in regular and Complex Networks
-Voter model in fully connected network
-Mean field theory for uncorrelated networks
-Effects of network disorder and heterogeneity
-Activity patterns: aging, update rules and interevent time distributions
-Imperfect imitation: i) The noisy voter model. ii) Zealots
-Coevolution: i) Fragmentation transition, ii)Noise iii)Nonlinear VM
-Multiplex voter model. i)Aging, ii) Coevolution
-Data and Voter Model: Online games
-Data and Voter Model: Electoral processes

Voter Model: Imitation Dynamics
Clifford and Sudbury, Biometrika (1973)

Two options: Holley and Ligget, Ann. Probability (1975)
Interaction: copy the state of one of your neighbors at random (node update)
Question: When and how agreement is reached by imitation?

When and how one of the two absorbing states (consensus)
is reached?
Voter Model
Two possible positions (opinions) {-1=left,1=right} on a political issue.

Individuals (“voters”) blindly adopt the position of a random neighbor.
Initial state: density σ of – voters and 1-σ of + voters.
Dynamics: Node update
1) Pick a voter i with opinion xi at random.

2) Pick a neighbor j with opinion xj at random.
i adopts j 's opinion (xi → xi=xj).
3) Repeat ad infinitum.
http://ifisc.uib.es
MODELS of SOCIAL CONSENSUS
Prototype models with two excluding options:
- VOTER MODEL
- SPIN FLIP KINETIC ISING MODEL T=0
p?  B  3 / 4 Voter Model RANDOM IMITATION

p?  A  1 / 4
?
p?  B  1 Spin Flip
SOCIAL PRESSURE
p?  A  0 Kinetic Ising T=0
? Active
Option A
Option B
Voter Model
n 2(k  2n )
P(  )   
k k  N
m  2 / N
Node of degree k
si=-1, n+ + neighbors
Order Parameter: Average interface density

1  N 
     j ( i ) (1   i j ) 
2 N k  i 1 
ρ=0 in absorbing state
Interface: a link connecting nodes with different states.
Voter Model
Applet
Voter Model
-Imperfect imitation: i) The voter noisy model. ii) Zealots
-Coevolution: i) Fragmentation transition, ii) Noise iii)Nonlinear VM

Voter Model in regular networks
 t 1/ 2 , d  1  N 2, d  1, time to reach absorbing state

 
   ~  ln t  , d  2
1
 ~  N ln N , d  2, time to reach absorbing state
  bt  d / 2 , d 2  N , d  2, survival time of metastable state
 
d=1,2: Coarsening/Ordering
Unbounded growth of domains of absorbing states
d=1 SF Kinetic Ising T=0
d=2
Coarsening without surface tension:

Driven by interfacial noise
Voter Model
d>2 regular and complex networks
  ~   ( N )  N, survival time of metastable state
d>2: No Coarsening : Dynamical Metastability

Finite size fluctuations take the
Disordered states. system to an absorbing state
l  1
Characteristic size of    ~ e  t /  survival time
ordered domain
Voter Model: Applets
N-state voter model in 2d lattice
Voter Model in Complex Networks

Voter Model
ER RANDOM NETWORK <k>=8
<(0)>
S(t)=
survival
probability
S(t): Survival porbability   (t )  S (t )   * (t ) 

Voter Model
BA SCALE FREE NETWORK <k>=6
N=104
S(t)=
survival
probability
  (t )  S (t )   * (t ) 
Voter Model in Complex Networks
Small World Networks Scale Free Barabasi-Albert Nets
 
  ~ et /
  ~ et / <k>=6
Survival time
scales as in
regular d>2:
~N
 ~ NlnN
Castellano et al, Eur. Phys. Lett. 63,153(2003) Suchecki et al, Eur. Phys. Lett. 69,228(2005)
Voter Model in Scale Free Networks
K. Suchecki et al Eur. Phys. Lett. 69,228(2005)
Node update dynamics:

  N    0.880  0.003 Castellano et al. Phys. Rev. E (2005)
  N /lnN Sood-Redner , Phys. Rev. Lett. 2005, Vázquez et al, New J.of Phys. 2008
 i 1, N ki i
 m(t )     ;  
t   i 1, N ki
Average degree-weighted magnetization <> is conserved
Link update dynamics:


  N  =1 <k>=8
<> is conserved
Voter Model as a diffusion model
σ i =±1 Value of the state at site i. ki degree
∑ j Lij =0 Laplacian matrix

a ij adjacency
Conserved quantity:
Ensemble average weighted magnetization
Klemm et al, Sci. Rep. 2, 292 (2012)
Realization
Average
Voter Model

Voter Model
FULLY CONNECTED NETWORK
N+= number of + nodes Density of active links:
N-= number of - nodes
N  N 2NN 1  m 2 (t ) 1  m 2 (t )
Magnetization: m   (t )   
N ( N  1) 1 2
N 2 (1  )
N
Spin flip probability:
2( N   N   1)
Changes per flip: m(  )  2 / N  (  ) 
N ( N  1)
t / ( N )
d m   (t )  e   (0) 
0
dt  ( N )  ( N  1) / 2  
Voter Model
FULLY CONNECTED NETWORK   ( t )  e  t /  ( N )   ( 0 ) 
 (N )  N  
  ~ et /
N=103-104
N=5 103
Surviving N=104
realizations
S(t)=
survival
probability
Mean time to reach consensus   (t )  S (t )   * (t ) 

The voter model: mean-field & random walk
Slanina, Lavicka, EPJB 35, 279 (2003)
(N+, N-)
(0, N) (N, 0)
(N+-1, N-+1) (N++1, N--1)

N N N N
+- -- W ()  +- ++ W ( ) 
N N 1 N N 1
N N 1 N N 1
W ()  
N N 1 N N 1
1 m2
m   (t )
P ( m , t )

2

 1 m

2
 P ( m , t )  N
t m 2 
 N  1 m2
    0e 2t / N
2
Voter Model

Voter Model
F. Vazquez, V.M. Eguiluz, New J. Phys. 10 (2008) 063011
UNCORRELATED NETWORKS
n+ active links k-n+ active links

k-n+ inert links n+ inert links
n 2(k  2n )
P(  )   
k k  N
m  2 / N
Node of degree k
si=-1, n+ + neighbors si=+1
Coupled eqs. for <m> and <>
Mean field link approx. for Prob (k, -,n+) (pair approximation):
Neglect 2nd nearest-neighbor correlations

Voter Model
N N
  ;    B ( n  , k )  Prob. that node i ( S i   1) with degree k has n  active links
N N
Prob that node i –
Prob of node k being -
v = bias with n+ changes to + Change in m
Change in 
Pair approximation:
neglect 2nd nearest-neighbor correlations. k!
B (n , k )   n
(1   ) k  n 
n  ! ( k  n  )!
~ prob that a link from node i is active
Voter Model
Pair Approximation d  m 
 0
dt
d   (t )  2     2    
  (  k   1 )( 1  )  1
dt  k   1  m ( 0 ) 
2

No ordering:   ( t )   (  k  )( 1   m 2 ( 0 )  )
 k  2
 ( k  ) 
2 (  k   1)
Finite size fluctuations:   ( t )   (  k  )( 1   m 2 ( 0 )  ) e  t / 

(  k   1)  k  2
  N
2 ( k   2 )  k 
2
Barabasi-Albert, <k2>(N)   N / ln N
Voter Model in Uncorrelated Networks
Pair approximation: Single parameter theory

 k  2 (  k   1)  k  2
 s
     N
2 (  k   1) 2 ( k   2 )  k 
2
Network topology independence Barabasi-Albert Scale Free Networks
N=104
<k>= 8
Voter Model

Voter Model
Role of dimensionality
1D Scale free net?
Structured SF: SSF  <k>=8

SSF
Klemm and Eguíluz,
Phys. Rev. E 65,036123 (2002)
1d regular
  ~ t 1/ 2
SSF   N 2
1
Scale free but Dimensionality determines when voter
high clustering and 1d dynamics orders the system
P(k) k -3 Degree distribution or network disorder
LN C  N0 are not relevant
Voter Model
Suchecki et al., Physical Review E 72, 036132 (2005)
Network Disorder and Link Heterogeneity

Disorder: Rewiring parameter 0<p<1. d=1 random networks
p=0 0<p<1 p=1
1D regular Small-World: SW Random: RN / Single
EN SCALE
Structured SF: Small-World SF: Random SF: HUBS
SSF SWSF RSF / BA P(k)k -3
d=1 d=
p=1
p=0
p=0 p=1 P(k) k -3
P(k) k -3 ; LN
Voter Model and Complex Networks: Summary
 t /
Dimensionality: d=1 (SSF)   ~ t 1/ 2 ; d=   ~ e
Network Disorder: SSF SWSF RSF
p=0 p=1
1d SW RN
Shorter lifetimes:  SWSF   RSF  SW   RN
Smaller size of domains: lSWSF  l RSF lSW  l RN
Degree Heterogeneity: SW vs SWSF; RN vs RSF

Fluctuations more efficient with hubs:  SW   SWSF  RN   RSF
Hubs do not affect size of domains: lSW  lSWSF l RN  l RSF
Hubs change scaling law:   N  SW, RN:   1 ; RSF,SWSF:   1

Voter Model

Missing Factor: Activity patterns


Memory, Aging
Human Activity Patterns
Artime et al, Sci. Rep. 7:41627(2017)
Bursty interaction pattern between two twitter users

Inter-event time (IET) distribution
PNonpoissonian Similar inter and Temporal Correlations P(´)

intracommunities distributions
Questions:
1-Role of the Timing of Interactions. Updating processes
2-Temporal vs. topological correlations

Human Activity Patterns
Artime et al, Sci. Rep. 7:41627(2017)
Link update rule:
Times to reach consensus
-Assign to every link an activation time 
from the empirical IET. in finite systems
-Select  = min( and update state of
nodes of this link
-Subtract  to all links and assign new
drawn from the IET to the updated link
-Iterate
t+1 -Nonpoissonian IET speeds consensus

-Temporal correlations speeds it further
-No effect of topological correlations
(communities)
Voter Model
STANDARD UPDATE RULES
RANDOM ASYNCHRONOUS UPDATE (RAU) SEQUENTIAL ASYNCHRONOUS UPDATE (SAU)
One random agent updated per simulation step.
SYNCHRONOUS UPDATE (SU)

One agent updated per simulation step.
Agents always updated in same order.
On regular lattices periodic or chaotic solutions
All agents updated simultaneously.

Voter Model
STANDARD UPDATE RULES
Similar qualitative results for the three standard update rules on several interaction networks
Fully connected ER Random Network BA Scale-free
FC and ER
Scale-free
No ordering: No consensus reached for infinitely large N

Voter Model
STANDARD UPDATE RULES:
Interevent time distributions
M(t): interevent time distribution (ITD), distribution of times between consecutive changes of
state
C(t): Cumulative ITD
Fully connected ER Random Network BA Scale-free
Homogeneous activity patterns with a well

defined characteristic interevent time.
Human Activity Patterns: Aging
Fernandez-Gracia et al, Phys. Rev. E (2011); Artime et al, Sci. Rep. 7:7166(2017)
AGING: The longer you remain in a state, less probable to update it
UPDATE RULE: Each agent is characterized by state x and ‘internal time‘ 
1.with activation probability each agent i becomes active. Take

2. active agents update their state x according to voter model dynamical rule
3.  =  + 1
Limits: → SU; → RAU.
EXOGENOUS UPDATE
Active agents reset  after step 2.
ENDOGENOUS UPDATE
Only active agents that change state in step 2 reset
Activation prob. becomes a function of a persistence time AGING
EXOGENOUS: External clock
ENDOGENOUS: Updating is part of the dynamical model.

Coupled dynamics of state x and ‘internal time‘ 
Voter Model and Update Rules
2d lattice p(1 
Configurations Persistence times
RAU Update
EXOGENOUS
Update
ENDOGENOUS
Update
Fernandez-Gracia et al, Phys. Rev. E (2011); Artime et al, Sci. Rep. 7:7166(2017)
FULLY CONNECTED NETWORK ER RANDOM NETWORK <k>=6

Density of interfaces
Density of interfaces
Exogenous update
Endogenous update
No ordering.
Dynamical coexistence
Endogenous update
AGING
Ordering
Cumulative interevent time.
N=1000, 2000, 4000

Power law tail
C(t )  t  ,   1
Exogenous update
Endogenous
Aging
Exogenous
Aging societies more prone to agreement

Human Activity Aatterns: Aging
BA SCALE FREE NETWORKS <k>=6

Density of interfaces Cumulative persistence
update
Endogenous
Coarsening:
  (t )  t  ,   0.32
Persistence:
C(t )  t  ,   1
N=1000, 2000, 4000

Exogenous update
No Coarsening
C(t )  t  ,   1
Aging results in Heterogeneous Activity Patterns

Heterogeneous interevent time distributions
produce a qualitative change in the voter
model of social consensus:
From dynamical coexistence of opinions
to ordering dynamics
Beware of social simulations of interacting agents based
on a constant activity rate:
Human activity patterns need to be implemented as an

essential part of social simulation.
Voter Model

Voter Model


Memory, Aging
The noisy voter model
1958
1995
Ants, rationality and recruitment
1986
1993
Quarterly J. of Economics (1993)
1989
Part II – Herding behavior and financial markets

IMPERFECT IMITATION: The noisy voter model
Buy si =1
ki ?
Sell si =0

IMPERFECT IMITATION: The noisy voter model
h
∑ (1− s j )
−
ri = a +
Buy si =1 ki j ∈ nn(i)
Transition HERDING
ki ? rates
Sell si =0 + h
r = a +
i ∑ sj
ki j ∈ nn(i)
Idiosyncratic
Noise
Free will

The noisy voter model: Mean field
Idiosyncratic behavior > Herding behavior
N
h
a>
n n
N
n N
2
h h Idiosyncratic behavior
t < Herding behavior
a< a=
N N
P (n) a>
h N
Optimistic consensus
N
n
0 N /2 N 0 N /2 N 0 N /2 N h
n n n N a<
2
N
Finite-size Pessimistic consensus

noise induced 0
transition t
=
Carro et al. Sci. Rep. 6:24775 (2016)
Finite size noise induced transition in complex networks

Annealed approximation
for
uncorrelated networks
Weighted fully connected network
ac 𝐴 =𝑝
Prob that node i with degree 𝑘 is

connected to node j with degree 𝑘
σ 2 k 
2
k
Local ordering in complex networks Carro et al. Sci. Rep. 6:24775 (2016)
a>ac
a<ac
ER, <k>=6 noise
voter Noise destroys absorbing state:

Asymptotic state partially ordered
Mean field pair appr. Transition reflected only as a

crossover regime in the local
order
Imperfect imitation is only

Annealed approximation sensitive to network topology
for when it is not dominant
uncorrelated networks
.
ac
Stochastic pair approximation Peralta et al. New J. Phys 20, 103045 (2018)
Master equation Two methods of closure and solution of the dynamical equations
L number of active links

# agents in state 1
and degree k
Voter Model + Noise + Aging
Artime et al Phys. Rev. E 98, 032104 (2018)
Noisy Voter Model
Noise induced finite size CONTINUOUS

DISCONTINUOUS TRANSITION
TRANSITION
Voter Model + Noise + Aging
An aging-induced phase transition

Artime et al Phys. Rev. E 98, 032104 (2018)
Ising universality class
ER net 𝛽=1/2, 𝛾 1, 𝑑 =2 𝑑 =4
Ising exponents in d=2,3
Asymmetric aging
Mean internal times

Voter Model: Zealots
VOTER MODEL: Khalil et al. Phys. Rev. E 97, 012310 (2018)
• One zealot takes the system to its absorbing state
• Same number of opposed zealots: Dynamically active state Equivalent to:
General case:
N noisy voters (a,h)
forming M communities
M=N1 +…+NM
Community k connected to
zk+ zealots holding state +1
zk- zealots holding state -1
nk = Agents +1 in community k
Mean field transition rates:
Introducing Zealots Noisy voter model with heterogeneous agents (ak, hk )

Noisy Voter Model and Zealots
Khalil et al. Phys. Rev. E 97, 012310 (2018)
AU
Finite size
transitions
EA
AB
AB phase
only exists
for N1< N1,c
EA phase disappears
when z1+ = z1-
Voter Model

Voter Model


Memory, Aging
CO-EVOLUTION
Dynamics of Networks:
Rightwing view
1. Dynamics OF network formation: Structure created by
individual choices/actions
2. Dynamics ON the network: Actions of individuals constrained Leftwing view
by the social network
3. Co-evolution of agents and network :
Circumstances make men as much as men make circumstances
..new research agenda in which the structure of the network is no longer a given
but a variable.....explore how a social structure might evolve in tandem with the
collective action it makes possible (Macy, Am. J. Soc. 97, 808 (1991))
Final Goal: Understanding dynamical processes of group formation and

social differentiation: Emergence of social dynamical networks with
-Social structure
-Weak links (Granovetter)
-Community structure
NETWORK CO-EVOLUTION MODELS
Review paper: T. Gross and B. Blasius, J. R. Soc. Interface 5, 259 (2008)
Early papers on coevolution:
M. Zimmerman, V. M. Eguíluz and M. San Miguel in " Economics with Heterogeneous Interacting
Agents" Eds. A. Kirman and J. B. Zimmerman, Springer Verlag, Lecture Notes in Economics and
Mathematical Systems N°503, pp.73-86 (2001)
M. Zimmerman, V. M. Eguíluz and M. San Miguel, Phys. Rev. E. 69, 065102-6 (2004)
Key ingredients.
a) Going beyond dynamical models in which:
-Network evolution is decoupled from the evolution of agents actions
-Complete network redefined at each time step
b) Social plasticity as ratio of time scales of evolution of network and action
Generic result: Network fragmentation transition

(Independent of link conservation, rewiring rule, interaction….)
Imitating vs Choosing neighbors
COEVOLUTION:
Dynamics on the network coupled with dynamics of the network
M. Zimmerman, et al Lecture Notes in Economics and Mathematical Systems 503, (2001)
Social Imitation Breaking and..
changing
state
..establishing ties
rewiring
Voter Model Rewiring
Coevolving voter model: Non-persisting ties

Coevolving Voter Model
F. Vázquez, et al, Phys. Rev. Lett. 100, 108702 (2008)
Choosing neighbors
Imitation
Network Fragmentation Transition
Fragmentation due to
competition of time scales:
- evolution of the network
(link dynamics)
Transition
- evolution on the network
(node state dynamics)
Critical value of plasticity pc

F. Vázquez, et al, Phys. Rev. Lett. 100, 108702 (2008)
Initial: Degree-regular random graph with neighbors.

Nodes take state S= -1 or S= +1 with the same probability 1/2.
1.Pick a node i and a neighbor j at random.
2. If Si = Sj nothing happens.
3. If Si ≠ Sj then:
 Network dynamics: rewire
with probability p delete link i – j
and create link i – k (Si = Sk ).
 State dynamics: copy
with probability 1-p set Si = Sj .
4. Repeat ad infinitum.
Agents select interacting partner

according to their state
p gives a ratio of time scales of
evolution of state of nodes and network
Mean-Field pair approximation
Active links: + - Inert links: + +, - -
Absorbing phase transition in a coevolving network
Equation for the average density of active links in the N → ∞ limit:
d   (t )  2     2  
  (   1)(1  )  1
dt   1  m (0) 
2

Active - Frozen
Transition at
Active phase: Links continuosly being rewired and nodes flipping states
Frozen phase: Fixed network where connected nodes have the same state
Fragmentation transition in a FINITE coevolving network
Fragmentation
Transition Active links in surviving runs.
Size of largest network component.
p=pc
Convergence times
Active phase → Connected network (Smax/N = 1)
(N = ∞)
Frozen phase → Fragmented network (Smax/N ≈ 0.5)
p<pc : slow rewiring keeps network connected

until system fully orders and freezes in
a single component.
p>pc : fast rewiring leads to fragmentation of
network into two components before
system reaches full order.
Absorbing phase transition in coevolving networks
Coevolution in systems with two absorbing states induces a

fragmentation transition.
Active and frozen phases in infinite systems correspond to connected

and fragmented phases respectively in finite systems.
Fragmentation is explained as a competition between the rates at

which network and the state of the nodes evolve.
Coevolving Voter Model + Noise
Diakonova et al. Phys. Rev E 92, 032803(2015)
Choosing Idiosyncratic,“Free Will”
Social Imitation
neighbors
Voter Model Rewiring

Noise:
random change of state,
with probability ,
Coevolving Voter Model at end of update
Plasticity p
2 PARAMETERS
Noise intensity
Coevolving Voter + Homogenous Noise
Diakonova et al. Phys. Rev E 92, 032803(2015)
Noise destroys fragmentation transition. No absorbing states

Three regimes separated by finite-size noise induced transitions
magnetization size of largest

network component
Fully-Mixing
Bimodal Magnetization
1
 N  Dynamic
 c ( p, N )  1   Fragmentation
 2(1  p ) 
p=0 Transition voter+noise

A. Kirman,
Quarterly J. of Economics (1993)
Beyond Random Imitation: NONLINEARITY
? Active
Option A p?  B  3 / 4 Voter Model RANDOM IMITATION
Option B p?  A  1 / 4
?
p?  B  1 Spin Flip
SOCIAL PRESSURE
p?  A  0 Kinetic Ising T=0
Nonlinear voter model: Castellano et al PRE (2009); Schweitzer et al EPJB, 2009

Social impact theory, Nowak et al Psychological Rev.1990
ai=3
Flipping probability of node i: ki=4
q: Degree of nonlinearity
q=1 Voter Model Neutral situation:

Random imitation process
q>1 Probability below random imitation
q<1 Probability above random imitation
Coevolution in Nonlinear Voter Model
Min and SM, Sci. Rep. 7, 12864 (2017)
Discontinuous
fragmentation
transition
Absorbing
Discontinuous
absorbing Absorbing
transition
VM
Dynamically
active Continuous
absorbing
transition
Voter Model

Multilayer Voter Model
Diakonova et al., New J. Phys. 18, 023010 (2016)
Social Context: MULTIPLEXING
p1 Parameters:
• Degree of
t t+1 Multiplexing
p2
q=0.2
• Edge overlap 
Multilayer  • No ordering in random networks


 • Little effect of overlap
I Effective • Irreducibility of multilayer

single to single layer dynamics
II layer
Multilayer Voter Model + Aging
Artime et al, Sci. Rep. 7:7166 (2017)
Aging + multilayering prevents ordering
Two layers. Aging update.
FC
network
𝜌
𝜌 𝜌 >
𝜌~2𝑞 1 𝑞
Dynamically trapped states for q<q*: Symmetry breaking + shared nodes

Multilayer Voter Model + Aging
Artime et al, Sci. Rep. 7:7166 (2017)
Asymmetric aging sustains trapping
Differences in layer persistence times Critical multiplexing q*
N=10000
Finite-size scaling indicates q* ~ 0.273
Random ER networks
Multilayer Coevolving Voter Model
Diakonova et al. Phys. Rev. E 89, 062818 (2014)
Coevolution + Multiplexing
Context: MULTIPLEX
Social Imitation Homophily
Voter Model Rewiring • nodes represent agents, layers

represent contexts
• associate interlayer link with agent

Coevolving Voter Model existing in both contexts
Coevolving Voter Multiplex
Diakonova et al., Phys. Rev. E 89, 06218 (2014)
3 Main Parameters
Setup
Rewiring Probabilities
1 System:
ratio of link/node state updating random regular
___ networks with
Degree of Multiplexing N nodes and
2 (interlevel connectivity) average
degrees
identifies nodes that are the and
same in both layers
Evolution
1. At random pick a level 2. Evolve using the CVM 3. Synchronize state of
to evolve node update node across levels
1-
Transition Shift
Identical layers:
Absorbing Transition ( limit): Fragmentation Transition (finite size effect):

asymptotic interface density fraction of largest component
Active
Fully
interconnected
multiplex
monoplex
Frozen
Absorbing transition and Fragmentation transition

persist, coincide and shift with connectivity:
Multiplex prevents fragmentation

Fragmentation in the fully Asymmetric CVM
fully-dynamic
voter
Level 2 never fragments (p2=0 )

Changes of state in level 1 due to q
1
Distribution of sizes of three

largest components in level 1
(dynamic):
Only ever two significant 2
Sizes of components components
N = 250, M = 10 000 realisations
for varying rewiring asymmetry
only one component: no fragmentation

at least two equal-sized components: fragmentation
Varying asymmetry: for layer 1 (blue) and layer 2 (red)
non fragmented non fragmented

1
fragmented fragmented
For , becomes a step function at a critical degree of multiplexing q*(p1,p2):

Minimal interlevel connectivity required to stop the level 1 from fragmenting
decreasing asymmetry
q*(1,0)=1 q*(0.5,0.5)=0.5
Variation of with degree of multiplexing
Not
fragmented
Layer 1
Layer 2
Fragmented
q=0.5
p1=0.9
p2=0.1
Anomalous fragmentation is a general consequence of the rewiring asymmetry for q < q*(p1, p2)
Critical degree of multiplexing q* to prevent fragmentation of dynamic layer

Shattered fragmentation transition for q<q*
Voter Model

Voter Model Validation?
Voter Model: Beyond understanding mechanisms
Data?
1. Community structure of online games
Imitation + coevolution + multilayering
Klimek et al., New Journal of Physics 18, 083045 (2016)
2. Is the Voter Model

a
Model for Voters?
Imitation + imperfect imitation + mobility
Fernandez-Gracia et al. Phys. Rev. Lett.112,158701 (2014)

Community structure of multilayer societies
Klimek et al., New J. of Physics 18, 083045 (2016)
“..an award-winning graphical browser-based

MMORPG … set in a futuristic universe where
players interact and compete in space. It is a
persistent-universe, open-ended game with a player-
driven economy. Players travel through hundreds of
"sectors" or solar systems while trading, building or
battling with Non-Player Characters … and other
players.” https://www.pardus.at/
Online characters: 300
Pardus Multiplex Network Active characters: 8,712
Registered characters: 426,165
- Friendship F Links established by clicking.
- Communication C Private messages. Multiplex network constructed for each
month from Sep 2007 to Sep 2008
- Trade T Exchange of game money/commodities
Communication C fast
slow
F
Friendship
T fast
Different community structure from
Trade
different plasticities?
Lesson from data: Topology of networks with high triangular clustering
Modified coevolving model: Rewiring with triadic closure
Consequence of triadic closure:

Partial multiplexing not required for shatered fragmentation
Analysis of fully multiplexed Pardus data (q=1)
Data analysis: Mesoscopic structure of the layers exhibits significant

heterogeneity in size of largest community (S1) and relative
number of communities (Nc),
CONFIRMING MODEL PREDICTIONS FOR TWO FAST LAYERS
Communication C S1T > S1C

pT < pC
NcT < NcC
T
Trade
Comparison topology of fast and slow layers
F-C F-T
Jaccard coefficient
of edge sets
Degree correlation Degree rank correlation

Community size distribution: Data vs model

comm
Different form of shattering for different plasticities
F layer: Small plasticity implies

small number of very large communities (flat tail distribution)
T and C layers: Large plasticity implies
large number of small communities centered around n=50
Voter Model

Is the Voter Model a Model for Voters?
US presidential elections 1980-2012
Evolution of democrat shares
Global percentage of votes
for each party and turnout
Evolution of republican shares

Other
parties
Basically two option system

Blue: Democrat
Red: Republican
Statistical regularities of electoral data
US US
Irrespective
of the
winner!
IBM
Diffusive model?
SPATIAL
CORRELATIONS
~log decay
C. Borghesi et al.
Eur. Phys. J. B 75, 395-404 (2010)
PLoS ONE 7(5):e36289,05 (2012)
Social influence model of voting behavior
Ingredients of a social influence model:
a) Interaction mechanism: Imitation as basic manifestation of social influence.
THEORY
MODELLING
b) Social context: Set of all possible interactions of an individual with any

other peer. We model it as a network of interactions from census data for
population and mobility.
INPUT DATA
MODELS and DATA
THEORY: DATA
Mechanisms to be
Model understood
PARAMETERS
INPUT DATA
SOCIAL CONTEXT
DATA
to be
MODEL
INFERENCE reproduced
Voting population and mobility networks
US: geographical adjacency of populations Undirected,
unweighted
network.
N=3114
<k>~6
<C> = 0.431
<L> = 26.86
<C>Hex = 0.4
<L>Hex = 22.15
Degree
Link distances
Counties in US connected if they are adjacent.
Population and mobility networks
US: commuter network for human mobility
2001 census data Directed, weighted network.

Degree
N=3114
<k>~50
Commuters per link

Commuters per link
Populations
Link distances
Heterogeneous network in many characteristics

Metapopulation Voter Model
- N agents with a binary variable (state, opinion,...) with voter-like interaction

- There are Nsites (counties).
- Each agent is considered in two sites: where she lives and where she works.
Probability α: HOME interactions Probability 1-α: WORK interactions
Nji Nji Nij

Nij
Ni Nj N'i N'j
Nii Njj Nii Njj
Site i Site j
Site i Site j
-Nij = # of agents living in i and working in j.
-Ni = # number of agents living in i = Nii +j Nij
-N'i = # number of agents working in i= Nii + j Nji
An agent interacts with probability α with anyone in Ni: lives where she lives.
With probability 1-α interacts with anyone in N'i : works where she works.
Parameters (census) Variables

Nij: number of agents Vij: number of agents living in i
living in i and working in j. and working in j holding opinion +1.
Xi,Yi: location of city i. Correlations of densities
Transition rates
Langevin equation
US presidential elections: Metapopulation Voter Model
=1/2
Diffusion process:
→ correlations grow, share
distribution narrows.
Democrat share. Initial condition 2000 10 MCS
Extra ingredient needed for stationarity:

Imperfect imitation or External noise
Finite size (internal) + external noise
Parameters (census) Variables

Nij: number of agents Vij: number of agents living in i
living in i and working in j. and working in j holding opinion +1.
Xi,Yi: location of city i. Correlations of densities
Transition rates
Imperfect
imitation
Langevin equation
Vote share standard deviation

Noise
=1/2 after 1000 MC steps. calibration
For D=0.02 the st. dev. of the vote share distribution

remains stationary and fits the empirical value (data)
Fernandez-Gracia et al. Phys. Rev. Lett.112,158701 (2014)
Democrat share. Initial condition 2000 Calibrated Model

=1/2
Single fitted parameter: D=0.02
Time calibration:
10 MC steps =4 years=
1 election period
Vote share distribution Spatial correlations

Data Model Data Model
St. dev. remains constant. Logarithmic decay remains.

Electoral predictions?
Results for democrat party
Data-Model
Data Model (i.c. 2000)
Data-Model with <v>=0.5

Data with <v>=0.5 Model with <v>=0.5
Multiscale Predictions
COUNTIES CONGRESSIONAL DISTRICTS STATES
RANDOMIZED DATA
What have we learnt?
IBM implementation of a microscopic mechanism leading to diffusive

mesoscopic stochastic dynamics reproducing statistical regularities of
election data.
Data Based Modeling: Input parameters from census data for populations
and commuting fluxes.
Single calibrated model parameter: D, the noise intensity. Also calibration of
time scale.
What do we explain?
-Two generic features in the background of election results:
i)Stationarity of the dispersion of vote shares and ii) the time persistent logarithmic
decay of spatial correlations.
-Spatiotemporal fluctuations in electoral results at different length scales
-No attempt to predict electoral results
Alternatives?
-J.P. Bouchaud et al, J. Stat. Mech. (2014) P03010
Generality?
-W.S. Jo, B.J. Kim: Elections in Korea

Voter Master 2019

Caricato da

Informazioni sul documento

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Voter Master 2019

Caricato da

Copyright:

Formati disponibili

Voter Model

R. Holley and T. M. Liggett, Ann. Probab. 3, 643 (1975).

Liggett T M 1985 Interacting Particle Systems (New York: Springer)

C. Castellano, S. Fortunato, and V. Loreto, Rev. Mod. Phys. 81,

P. L. Kaprisky, S. Redner and E. Ben Naim, 2010

Updating rules and the voter model

Conservation laws for the voter model in complex networks

Generic absorbing transition in coevolution dynamics

Analytical Solution of the Voter Model on Uncorrelated Networks

Conservation laws for voter-like models on random directed networks

A measure of individual role in collective dynamics: spreading at criticality

Is the Voter Model a model for voters?

Noise in Coevolving Networks

Irreducibility of multilayer network dynamics

Dynamical origins of the community structure of multi-layer societies

The noisy voter model on complex networks

Dynamics on networks: competition of temporal and topological correlations

Fragmentation transitions in a coevolving nonlinear voter model

Zealots in the mean-field noisy voter model

Stochastic pair approximation treatment of the noisy voter model

Aging-induced continuous phase transition

Coevolving nonlinear voter model with triadic closure

Multilayer coevolution dynamics of the nonlinear voter model

Determine when and how the

INTERACTIONS: Mechanisms (“rule”) and Network (with whom)

-VOTER MODEL MODELS -Imitation MECHANISMS

Ingredients of interacting agents simulation:

a) Mechanism: Ex.: Voter model (imitation)

c) When do elements interact? Interaction activity

-Results in regular and Complex Networks

-Voter model in fully connected network

-Mean field theory for uncorrelated networks

-Effects of network disorder and heterogeneity

-Activity patterns: aging, update rules and interevent time distributions

-Imperfect imitation: i) The noisy voter model. ii) Zealots

-Coevolution: i) Fragmentation transition, ii)Noise iii)Nonlinear VM

-Multiplex voter model. i)Aging, ii) Coevolution

-Data and Voter Model: Online games

-Data and Voter Model: Electoral processes

Clifford and Sudbury, Biometrika (1973)

Question: When and how agreement is reached by imitation?

Two possible positions (opinions) {-1=left,1=right} on a political issue.

Initial state: density σ of – voters and 1-σ of + voters.

Dynamics: Node update

1) Pick a voter i with opinion xi at random.

Prototype models with two excluding options:

- SPIN FLIP KINETIC ISING MODEL T=0

p?  B  3 / 4 Voter Model RANDOM IMITATION

Order Parameter: Average interface density

-Results in regular and Complex Networks

-Voter model in fully connected network

-Mean field theory for uncorrelated networks

-Effects of network disorder and heterogeneity

-Activity patterns: aging, update rules and interevent time distributions

-Imperfect imitation: i) The voter noisy model. ii) Zealots

-Coevolution: i) Fragmentation transition, ii) Noise iii)Nonlinear VM

-Multiplex voter model. i)Aging, ii) Coevolution

-Data and Voter Model: Online games

-Data and Voter Model: Electoral processes

 t 1/ 2 , d  1  N 2, d  1, time to reach absorbing state

Coarsening without surface tension:

d>2: No Coarsening : Dynamical Metastability

N-state voter model in 2d lattice

Voter Model in Complex Networks