Diffusion limited friendship network: A model for six degrees of separation
Extracted source text ยท arXiv cond-mat/0304047
\documentclass[twocolumn,showpacs,prl]{revtex4}
% Some other (several out of many) possibilities
%\documentclass[preprint,eqsecnum,aps]{revtex4}
%\documentclass[eqsecnum,aps,draft]{revtex4}
%\documentclass[prb]{revtex4}% Physical Review B
\usepackage{graphicx}%
\usepackage{dcolumn}
\usepackage{amsmath}
\usepackage{latexsym}
\begin {document}
\title
{Diffusion limited friendship network: A model for six degrees of separation}
\author
{
S. S. Manna
}
\affiliation
{
Satyendra Nath Bose National Centre for Basic Sciences
Block-JD, Sector-III, Salt Lake, Kolkata-700098, India
}
\begin{abstract}
A dynamic model of a society is studied where each person is an uncorrelated and
non-interacting random walker. A dynamical random graph represents
the acquaintance network of the society whose nodes are the individuals
and links are the pairs of mutual friendships. This network exhibits
a novel percolation like phase transition in all dimensions. On
introducing simultaneous death and birth rates in the population we
show that the friendship network shows the six degrees of separation
for ever after where the precise value of the network diameter depends
on the death/birth rate. A SIS type model of disease spreading shows that this
society remains always healthy if the population density is less than certain threshold value.
\end{abstract}
\pacs {05.10.-a,% Computational methods in statistical physics and nonlinear
05.40.-a,% Fluctuation phenomena, random processes, noise, and Brownian motion
05.50.+q,% Lattice theory and statistics
89.75.Hc % Networks and genealogical trees
}
\maketitle
Though at present the human population of the world has attained a
very large size, more than $6 \times 10^9$ precisely, it is believed that
any two randomly selected persons in the world are connected by a short
chain of intermediate acquaintances, typically of length 6. This
phenomenon is referred as the ``six degrees of separation''. The idea
originated from the famous letter distribution experiment of
Milgram in 1960s \cite {Milgram}. Since then any network of $N$ nodes
is said to display six degrees of separation if its diameter is small
and grows at most as $\log N$ \cite {WS, Barabasibook}.
Most human communications, especially the information exchanges, take
place directly between individuals when they are at close proximity
to one another. The spread of news, rumors, jokes and fashions all
take place by communications among individuals. More
importantly the infectious diseases also spread by person-to-person contact
and the structure of network of such contacts has important effects on
the nature of the epidemics. Naturally the speed of
spreading in general is faster for a network with small diameter.
There are important models of the social networks like Small-world
network (SWN) that displays the six degrees of separation \cite {WS}.
Also the process of spreading of epidemics is modeled by a
susceptible-infected-susceptible (SIS) model \cite {Anderson} in which a
non-equilibrium phase transition takes place from a healthy society to
an infected society at a critical value of the infection probability
\cite {Anderson, Marro, Bailey, Vesp}.
All these models of the social networks as well as for the spreading of the infectious
diseases consider a static picture of the society. More precisely
static individuals are positioned at the nodes of certain graphs and
a person interacts with only a fixed set of neighbours determined by the degree of the
node. Where as in actual society the number of acquaintances of a person
increases with time. Everyday a person goes to office, market, theaters, clubs etc. and therefore
gets acquainted with other people who were unknown to him. By the same movements
a person becomes exposed to infections by others or transfers his own infection to
others. Again not all the friends of an infected person has the chance of
getting infection, only those friends who come close to this person
has the risk of infection. In this paper we study this basic property of a dynamical society where
individuals are not static objects but move continuously and therefore comes
in contact with other people. To make a simple model we have considered
the diffusive motion of the individuals and modeled the society by a
set of random walkers. Specifically
in our model (i) unlike static models the number of acquaintances of a person evolves with
time (ii) irrespective of how many friends an infected
person has, he may infect only those friends who come to his close proximity, this
is unlike to the ordinary SIS type models. What we achieve are:
(i) with the introduction of a death/birth rate the society indeed shows
the six degrees of separation effect (ii) there is a threshold density of
population, below which the society is always healthy (iii) a very interesting
theoretical observation that the associated dynamical random graph has
a non-trivial dimension dependent critical behaviour.
Over last few years it is becoming increasingly evident that highly
complex structures of many social \cite {Newman}, biological \cite {Jeong,Sole}, electronic
communication systems \cite {web,Faloutsos} etc. can be modeled by simple graphs.
Erd\"os and R\'enyi
studied the well known random graphs (RG) of $N$ nodes where each pair of
nodes is connected with a probability $p$ and the graph shows a
continuous phase transition at $p_c=1/N$ \cite {Erdos}.
Scale-free networks (SFN) are characterized by the power law decay of the
nodal degree distribution function: $P(k) \sim k^{-\gamma}$. Two very
important networks in electronic communication system like World Wide Web \cite {web}
and the Internet \cite {Faloutsos} are observed to
possess the scale-free property. Barab\'asi and Albert (BA) proposed a
model for a growing SFN where nodes are linked with the preferential
attachment probability \cite {barabasi,albert}. Other routes, e.g., static \cite {Caldarelli}
and quasistatic \cite {Mukherjee} models to obtain SFNs are also studied. Assigning a Hamiltonian
correlations are studied in the optimized networks keeping biological
networks in mind \cite {Lassig}. SWNs with random walkers capable of making
long distance jumps are studied in \cite {Manrubia}.
In our model each member of the society executes a simple uncorrelated and non-interacting random
walk on a regular lattice. Initially the population of $N$ persons are released on the
square lattice of size $L \times L$ at randomly selected positions.
The system then starts evolving with time. At each time step each
person makes a jump to one of its neighbouring lattice positions with
equal probabilities. Each person represents a node of the growing
acquaintance network and a link is established between two nodes the moment
the corresponding pair of persons come in contact to each other at the same
position and at the same time \cite {sublattice}. Gradually the
number of links among the individuals grow. Thus the set of $N$ nodes and the
set of links among these nodes define our network called
as the Diffusion limited friendship network (DLFN) where as the associated
graph is referred as the Dynamical random graph (DRG).
%---------------------------------------------------------------------------
\begin{figure}[top]
\begin{center}
\includegraphics[width=6.5cm]{figure1.eps}
\end{center}
\caption{
Plot of the scaled characteristic times ${\cal T}_c$ (opaque symbols) and
${\cal T'}_c$ (filled symbols) vs. the system size $L$ for three different
populations: $N$ = 32 (circle), 64 (square) and 128 (triangle).
}
\end{figure}
%---------------------------------------------------------------------------
All persons which are at the same lattice site immediately become friends and
the associated subgraph with these people become a clique.
At each time many such cliques are formed at different sites.
All these cliques remain for ever, they never get destroyed, moreover they
grow in sizes as time proceeds. At the early times, the
number of links is small and the DRG has many different isolated components
of different sizes. The size of a component is determined by its number of nodes
and the giant component has the largest size. The giant component
not only grows by including new nodes into it but also by the process of merging equally large components.
After some slow initial growth the giant component grows very fast and
its size become proportional to $N$. This behaviour is just like the threshold phenomenon
in a continuous phase transition e.g., what happens in a random graph \cite {Erdos}.
The whole DRG ultimately reaches the
limiting stage of a giant $N$-clique when each node is linked to all other nodes.
%---------------------------------------------------------------------------
\begin{figure}[top]
\begin{center}
\includegraphics[width=6.5cm]{figure2.eps}
\end{center}
\caption{
The scaling of the order parameter $\psi(p,N)$ at the critical region $p_c(N)$
in the DRG: N=64 (circle), 128 (square) and 256 (triangle).
}
\end{figure}
%---------------------------------------------------------------------------
We first characterize DRG to compare with RG. DRG has
two characteristic time scales. ${\cal T}_c$ measures
the time required for the phase transition and is observed to vary like
$L^z/N^{\alpha}$. In a mean-field limit when the
density $\rho=N/L^d$ is small this variation is estimated in the following way:
If a person randomly walks a linear distance $R$ in $d$-dimension in time ${\cal T}_c$
then $R \sim {\cal T}^{\mu}_c$ and therefore around $L^d/R^d$ such $d$-dimensional spheres
are needed to cover the volume of size $L$ which is $N$ itself. This gives
${\cal T}_c \sim L^{1/\mu}/N^{1/\mu d}$ i.e., $z=1/\mu$ and $\alpha=1/\mu d$ in general
and therefore $z=2$ and $\alpha=1$ for ordinary random walks ($\mu=1/2$) in two dimension.
Fig. 1 shows the scaled plot of ${\cal T}_c$ for different $L$ and $N$ values and a good
collapse of the data is observed but for $\alpha \approx 0.89$ and $z \approx 2.25$.
We believe the difference in the exponents from the mean-field values
are due to finite size of the system.
At a second characteristic time ${\cal T'}_c \sim L^{z'}/N^{\alpha'} $ the DRG becomes an $N$-clique
where $z'$ and $\alpha'$ are estimated as 2.22 and -0.33 respectively for $\mu=1/2$ and $d$=2. The positive
value of $\alpha$ and the negative value of $\alpha'$ are consistent with intuition:
for a fixed $L$ but with increasing $N$, less number of steps per person are necessary
for the giant component to include all nodes, but a larger number of steps are
required to form the $N$-clique. The values of $z$ and $z'$ are likely to be the same.
The link density $p(t,N)$ at a time $t$ for an $N$ node network
is defined as the ratio of the number of links to its maximum
possible number $N(N-1)/2$. Numerically we find the following scaling form:
\begin {equation}
p(t,L) \sim {\cal F} (t/L^z)
\end {equation}
where the scaling function ${\cal F}(x) \sim x^{\alpha}$ and $\alpha$ and $z$ are approximately
found to be 0.89 and 2.25 again.
The order parameter $\psi(p,N)$ of this transition is the average fraction
of nodes in the giant component for a link density $p$. The critical link
density at the transition point $p_c$ is defined by
$\psi(p_c,N)=1/2$ and is observed to vary with $N$ as:
$p_c = b/N^{\alpha}$ with $b \approx 1.28$ and $\alpha \approx 0.89$ as before.
As the mean-field calculation gave $\alpha=1/\mu d$, we see that only
the ordinary random walks in two dimensions with $p_c =1/N$ correspond to
the random graphs \cite {Erdos} but for other walks with different $\mu$
and in different dimensions $p_c(N)$ have non-trivial dimension dependence.
A scaling plot for the order parameter is shown in Fig. 2 where we plot
$\psi(p,N)N^{-\beta/\nu}$ vs. $[p-p_c]N^{1/\nu}$. An excellent collapse of
the data shows that the order parameter has the following scaling form:
\begin {equation}
\psi(p,N) \sim N^{\beta/\nu}{\cal G}[(p-p_c)N^{1/\nu}]
\end {equation}
with $\nu \approx 0.8$ and $\beta \approx 0.02$.
%---------------------------------------------------------------------------
\begin{figure}[top]
\begin{center}
\includegraphics[width=8.0cm]{figure3.eps}
\end{center}
\caption{
The fluctuating diameters ${\cal D}(N,t)$ and ${\cal D}'(N,t)$ of the DLFN ($q$=0.43, $N$=64, $L$=64)
plotted in the upper and the lower curves with time. The data is averaged
over 100 configurations and have steady averages of $6.00 \pm 0.05$ and
$2.87 \pm 0.05$.
}
\end{figure}
%---------------------------------------------------------------------------
The topological distance between a pair of nodes is the number
of links on the shortest path connecting them and the diameter is
the maximum of such paths. The average diameter ${\cal D}(N)$ is
measured over many independent configurations. The configuration
average of the mean distance between an arbitrary pair of nodes is
denoted by ${\cal D}'(N)$. As the system evolves, both measures
first increase with time, reach their maxima and then decrease very slowly,
finally saturates to a fixed value for a long time. The maximum of the diameters
occur at the characteristic times ${\cal T}_c$. As expected the nodal degree distribution
of the giant component at the transition point is a Poisson distribution similar
to RG, since there is no preferential link attachment probability in this model as in the
scale-free networks.
The network described so far has a major drawback that it assumes each
individual as immortal. As a consequence the DRG becomes an $N$-clique at time ${\cal T'}_c$.
To make our model more realistic, we therefore introduce a probability of death and birth in the population
but with equal rates to keep the population conserved.
More precisely at each time step only one randomly selected individual is killed with
a probability $q$. As a consequence all links associated with the node
representing this individual are immediately deleted. This may result the
fragmentation of the particular component of the dynamical graph which belonged this node.
A fresh determination of the different components of the DLFN
especially the giant component is done immediately
before the system proceeds to the next time step. At the same time we assume
that a fresh individual has taken birth at the same position of the dead
individual so that the population conservation is maintained.
%---------------------------------------------------------------------------
\begin{figure}[top]
\begin{center}
\includegraphics[width=6.5cm]{figure4.eps}
\end{center}
\caption{
The variation of the activity $A(r)$ of the SIS model on DLFN
with the infection probability $r$ for $\rho=1.2$ and it vanishes at $r_c$. The inset shows
the variation of the critical infection probability $r_c(\rho)$ with
population density $\rho$.
}
\end{figure}
%---------------------------------------------------------------------------
When an individual dies, the deletion of all his links may severely affect the
distribution of distances between all pairs of nodes in the system. In
fact it is expected that in general the distance between an arbitrary
pair of nodes should increase due to the death of an individual, which
thus enhances the values of ${\cal D}(N,t)$ and ${\cal D}'(N,t)$. On the other hand
the newly born individual also start diffusing in the system and start
building up links of acquaintances with other individuals of the network.
Therefore the magnitudes of the diameters decrease again. As a result
the net effect of the two competitive processes of simultaneous death
and birth of individuals is to make the diameters fluctuate around their
steady averages whose magnitudes must depend on the rate $q$ of
death/birth processes. In Fig. 3 we show the time variation of these diameters and for
$L=64, N=64$ the diameter ${\cal D}(q,N,L)$ has a value very near to 6 for $q$ = 0.43
where as the ${\cal D}'(q,N,L)$ is around 2.85. As expected the diameters
increase with decreasing $q$.
Finally we study a susceptible-infected-susceptible model on the
DLFN. At any time a lattice site may be occupied by a number of persons.
If at least one of them is infected, each of the other healthy persons at that site
has a probability $r$ to become infected and with a probability $1-r$ it
remains healthy. An infected person at time $t$ becomes healthy at the next time step.
For a certain average density $\rho$, the average fraction of infected persons
in the system fluctuates but maintains a steady time independent average $A(r,\rho)$. In Fig. 4
we show that the average activity $A(r,\rho)$ vanishes for $r<r_c$ and
it continuously increases beyond $r_c$. The threshold $r_c$ is
the critical point of a phase transition from a completely healthy society
to an infected society. The $A(r,\rho)$ plays the role of the order parameter
in this transition. We also notice that $r_c$ is in general a function of
the population density $\rho$. In the inset of Fig. 4 we plot the variation of
$r_c(\rho)$ with $\rho$. The value of the critical infection probability
decreases with increasing the population density i.e., more the density
it is more likely that the infection really spreads. On the other hand,
below a certain density $\rho < \rho_c$ infection does not spread at all
even with the maximum possible infection probability $r_c=1$.
For the square lattice we estimate $\rho_c \approx 0.75$.
A number of different aspects of this model may be of interest. On average a human
being remains more or less localized up to his/her home, home city or home country. Therefore
perhaps it would be better to consider their motion as sub-diffusive ($R^2(t) \sim t^{2\mu}$ with $\mu<1/2$)
rather than normal diffusion. Secondly DLFN may be important to study the reaction
kinetic networks of two species reversible or irreversible chemical reactions $A+B \leftrightarrow C$.
To summarize, we have considered the evaluation of the mutual friendship network
in a dynamic model of a society. Each member of the society executes a diffusive motion.
Members of the society represent nodes of the network and their mutual friendships are
the links. The dynamical random graph associated with the network shows a novel dimension dependent phase
transition. With a certain death/birth probability the network displays the
six degrees of separation effect. We also observe that such a society remains always healthy
if the average population density is below certain threshold value, which should have very
important practical consequences.
I thank P. Sen for some initial discussions, D. Dhar for
some useful comments and Brian Hayes for pointing out the
paper by J. S. Kleinfeld \cite {Milgram}.
\begin{thebibliography}{90}
\bibitem {Milgram} S. Milgram, Psychology Today, {\bf 1}, 61 (1967); Also see
J. S. Kleinfeld in http://www.uaf.edu/northern/big\_world.html.
\bibitem {WS} D. J. Watts and S. H. Strogatz, Nature, {\bf 393}, 440 {1998};
D. J. Watts, {\it Small Worlds: The Dynamics of Networks Between order and Randomness},
(Princeton 1999).
\bibitem{Barabasibook} A.-L. Barab\'asi, {\em Linked: The New Science of Networks} (Perseus, Cambridge, 2002).
\bibitem{Anderson} R. M. Anderson and R. M. May, {\em Infectious diseases in humans}, (Oxford University Press, Oxford, 1992).
\bibitem{Marro} J. Marro and R. Dickman, {\em Non-equilibrium Phase Transitions in Lattice Models}
(Cambridge University Press, Cambridge, 1999).
\bibitem{Bailey} N. T. J. Bailey, {\em The Mathematical Theory of Infectious Diseases},
(Griffin, London, 1975).
\bibitem{Vesp} R. Pastor-Satorras and A. Vespignani, Phys. Rev. Lett. {\bf 86}, 3200 (2001);
R. Pastor-Satorras and A. Vespignani, Phys. Rev. E. {\bf 63}, 066117 (2001);
M. Bogu\~n\'a, R. Pastor-Satorras and A. Vespignani, Phys. Rev. Lett. {\bf 90}, 028701 (2003).
\bibitem {Newman} M. E. J. Newman, {\it Proc. Natl. Aca. Sci.}, {\bf 98}, 404 (2001);J. Stat. Phys. {\bf 101}, 819 (2000).
\bibitem {Jeong} H. Jeong, S. P. Mason, A.-L. Barab\'asi and Z. N. Oltvai, Nature, {\bf 411}, 41 (2001).
\bibitem {Sole} R. V. Sol\'e and J. M. Montoya, Proc. Royal Soc. London B, {\bf 268}, 2039 (2001);
J. Camacho, R. Guimer\'a and L. N. Amaral, Phys. Rev. Lett. {\bf 88}, 228102 (2002).
\bibitem {web} S. Lawrence and C. L. Giles, Science, {\bf 280}, 98 (1998); Nature, {\bf 400}, 107 (1999),
R. Albert, H. Jeong and A.-L. Barab\'asi, Nature, {\bf 401}, 130 (1999).
\bibitem {Faloutsos} M. Faloutsos, P. Faloutsos and C. Faloutsos, Proc. ACM SIGCOMM, Comput. Commun. Rev., {\bf 29}, 251 (1999).
\bibitem {Erdos} P. Erd\"os and A. R\'enyi, Publ. Math. Debrecen, {\bf 6}, 290 (1959).
\bibitem {barabasi} A.-L. Barab\'asi and R. Albert, Science, {\bf 286}, 509 (1999).
\bibitem {albert} R. Albert and A.-L. Barab\'asi, Rev. Mod. Phys. {\bf 74}, 47 (2002).
\bibitem {Caldarelli} G. Caldarelli, A. Capocci, P. De Los Rios and M.A. Munoz, Phys. Rev. Lett. {\bf 89}, 258702 (2002).
\bibitem {Mukherjee} G. Mukherjee and S.S. Manna, Phys. Rev. E {\bf 67}, 012101 (2003).
\bibitem {Lassig} J. Berg, Michael L\"assig, Phys. Rev. Lett. {\bf 89}, 228701 (2002).
\bibitem {Manrubia} S. C. Manrubia, J. Delgado and B. Luque, Europhys. Lett. {\bf 53}, 693 (2001).
\bibitem {sublattice} Here is a point of detail. Since each person jumps to one of its
neighbouring lattice point at each time step, individuals randomly distributed on a
square lattice are divided into two interpenetrating sublattices and there is no
friendship at all between these two socities. Therefore we drop individuals on one
sublattice initially.
\end{thebibliography}
\end{document}