complex networks luis miguel varela cost meeting, lisbon march 27 th 2013
TRANSCRIPT
Complex Networks
Luis Miguel VarelaCOST meeting, Lisbon
March 27th 2013
Complex Networks
Luis Miguel VarelaCOST meeting, Lisbon
March 27th 2013
Complex Networks
Luis Miguel Varela Cabo
Introduction
Main properties of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Introduction
Complex networks analysis in socioeconomic models
Statistical mechanics of complex networks
Computational algorithms
Applications - Market models - Regional trade and development - Other social network models of interest
Suggested trends
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Statistical mechanics of complex networks
Graph: an undirected (directed) graph is an object formed by two sets, V and E, a set of nodes (V={v1,…,vN}) and an unordered (ordered) set of links (E={e1…eK}).
Adjacency matrix:
- Contains most of the relevant information about the graph- A symmetric: undirected graph (a)- A non symmetric: directed acyclic graph (b)
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Measures on networks
Applications
social networks
financial networks
Economic networks
Small-worlds:Erdös-Bacon number
Degree distribution- Assortativeness- Preferential attachment
Clustering coefficient
Betweenness
Statistical mechanics of complex networks
Objects kindly provided by G. Rotundo
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social networks models of interest
Suggested trends
Statistical mechanics of complex networks
Small-worlds relatively short path between any two nodes, defined as the number of edges along the shortest path connecting them. The connectedness can also be measured by means of the diameter of the graph, d, defined as the maximum distance between any pair of its nodes. Networks do not have a “distance” : no proper metric space. Chemical distance between two vertices lij: number of steps from one point to the other following the shortest path.
In most real networks, < l > is a very small quantity (small-world)
In a square lattice of size N:
In a complex network of size N:
lji
ij llplNN
l )()1(
2
Nl
Nl log47.3
200
l
N
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
The diameter of the network is small if compared to the number of nodes ~(log(N))
Examples (social networks):• sociological experiment of Stanley Milgram (1967):
anybody can be contacted through at most d=6 intermediaries
• Hollywood actors (mean) d=3.65
• Co-authors in maths (mean) d =9.5
Photo of a poster in the metro station of Paris, advertising on music events
small world
Slide kindly provided by G. Rotundo
Kevin Bacon number:
Kevin Bacon.
Nick Nolte
CAPE FEAR Robert De Niro
GOODFELLAS Joe Pesci
JFK
Degree of separation of Nick Nolte: 3
Val Kilmer Tom Cruise Kevin Bacon
Number of intermediaries of Hollywood actors to have worked with Kevin Bacon (social game popular in 1994)
Degree of separation of Val Kilmer: 2
TOP GUN A FEW GOOD MAN
Example 1
Example 2
small world-Bacon number (analogously to Erdos for mathematicians)
Slide kindly provided by G. Rotundo
Statistical mechanics of complex networksCentrality, To go from one vertex to other in the network, following the shortest path, a series of other vertices and edges are visited. The ones visited more frequently will be more central in the network. Betweenness, number of shortest paths that passes through a given node for all the possible paths between two nodes. Measures the “importance” of a node in a network.
Number of shortest paths including vC(v)=
Total number of shortest paths
Betweenness (red=0,blue=max)
Example: node has the most high C(v)
Objects kindly provided by G. Rotundo
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Statistical mechanics of complex networksClustering coefficient, ratio between the number Ei of edges that actually exist between these ki nodes and the total number ki (ki-1)/2 gives the value of the clustering coefficient of node i. The clustering coefficient provides a measure of the local connectivity structure of the network
Clustering spectrum: Average clustering coefficient of the vertices of degree k
Average clustering coefficient
c Low c Large)1(
2
2
ii
i
i
ii kk
EkE
c
i
icNc
1
ii
kk ckNpkc
i )(
1)(
Clustering coefficient of real networks and random graphsR. Albert, A.-L. Barabási, Rev. Mod. Phys. 74, 47 (2002)
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Counting triangles: friends of friends are friends of mine, too?
clustering coefficient
Clustering coefficient =
N. triangles
N. Tringles less one links + +=
AB
C
D
A is friend of B, that is friend of A and C, but A and C are not friends
Transitivity property
Slide kindly provided by G. Rotundo
A network is called sparse if its average degree remains finite when taking the limit N>> . In real (finite) networks, <k> <<N
Average degree
Degree distribution: p(k) probability that a node has a definite amount of edges. In directed networks the in-degree and out-degree are defined.
AB
WS
random)1(
)(
k
e
ppk
N
kp k
kNk
i k
i kkpkN
k )(1
Statistical mechanics of complex networks
degree 1 degree 2 degree 3
Object kindly provided by G. Rotundo
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Fundamental concepts for network topology description
Two-vertex correlations Real networks are usually correlated: degrees of the nodes at the ends of a given vertex are not in general independent. P(k’ | k)= probability that a k-node points to a k’-node.
Uncorrelated network: independent of k
Correlated network: p(k’|k) depends on both k’ and k
Degree of detailed balanced condition: P(k) and P(k’ | k) are not independent, but are related by a degree detailed balance condition.
Consequence of the conservation of edges
Number of edges k k’ = number of edges k’ k
Statistical mechanics of complex networks
k
kpkkkp
)'(')|'(
)'|(')'()|'()(
)'|(')'()|'()(
kkpkkpkkkpkp
kkpkkNpkkkpkNp
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Fundamental concepts for network topology descriptionCorrelation measures
Average degree of the nearest neighbors of the vertices of degree. Alternative to p(k’|k)
knn(k) dependent on k: correlations
Assortative: knn(k) increasing function of kDisassortative: knn(k) decreasing function of k
Statistical mechanics of complex networks
Two-vertex correlations
'
)|'(')(k
nn kkpkkk
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Fundamental concepts for network topology description
- Motifs: A motif M is a pattern of interconnections occurring either in a undirected or in a directed graph G at a number significantly higher than in randomized versions of the graph, i.e. in graphs with the same number of nodes, links and degree distribution as the original one, but where the links are distributed at random.
- Community (or cluster, or cohesive subgroup) is a subgraph G(N,L), whose nodes are tightly connected, i.e. cohesive.
S. Boccaletti et al. Physics Reports 424 (2006) 175 –308
Statistical mechanics of complex networks
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Nodes are divided into groups - high internal connection - low connection to the other groups
community structure
Example:
Network of friends at school
Divided by younger age,
Older age,
White and black
Slide kindly provided by G. Rotundo
Bocaletti et al. Physics Reports 424, 175 – 308 (2006).
Directed networks and weighted networks
Weighted networks: strong and weak ties between individuals in social networks
NodesLinksweights
Weighted degree
Weighted clustering coeff.
Statistical mechanics of complex networks
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Statistical mechanics of complex networks
Ising model in networks: paradigm of order-disorder transitions in agent-based models
2D spin system
1D regular lattice (Ising, 1925)
2D regular lattice (onsager, 1944)
No phase transition!
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Statistical mechanics of complex networks
Ising model in networks: paradigm of order-disorder transitions in agent-based models
2D spin system
1D WS network (Viana-Lopes et al., 2004)
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Phase transition in 1D! (long-distance correlations dramatic increase in connectivity)
Statistical mechanics of complex networks
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Ising model in networks: paradigm of order-disorder transitions in agent-based models
2D spin system
Mean-field for 1D AB network (Bianconi, 2004)
Phase transition in 1D! (long-distance correlations dramatic increase in connectivity)
Watts-Strogatz algorithm
Start with order Randomize rewiring with probability p excluding self-connections and duplicate edges
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Computational algorithms
Duncan J. Watts, Steven H. Strogatz, Nature 393, 440 (1998)
Albert-Barabási algorithm
1. Network growth: start with a small number of nodes and at each time step add a new node that links to m already existing nodes
2. Preferential attachment (evolving network): the probability that a new node links to node i depends on the degree of the already existing node:
Albert-Barabási
Dorogovtsev- Mendes-Samukhin
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Computational algorithms
1. Potential degree distribution (extreme events, superspreader, hierarchies):
P(k) ~ k-g
2. Average path length shorter than in exponentially distributed networks.
3. Degree of correlation of the degree of the different nodes
4. Clusterization degree ~ 5 times greater than that of random networks.
Scale free networks (e.g. Albert-Barabasi)
75.0N
kCrand
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Computational algorithms
preferential attachment
New nodes are attached to the hubs (preferentially)
Scale-free networks
Social networks models
Some objects kindly provided by G. Rotundo
Real or hypothetical. Depends on the amount of data:
Intrinsic characteristics (e. g. classes). Full description (e. g. contact tracing). Building algorithm (e. g. Barabási-Albert). Sampling of the degree distribution (e. g. polls). Tools:
Standard statistical methods and software.Analysis and visualization interactive programs.
POPULATION ANALYSIS
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Computational algorithms
Pajek: http://pajek.imfm.si/doku.php
Others: Cytoscape (http://www.cytoscape.org/), UCINet.
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Computational algorithms
Much more time-consuming than ODE-based methods Automatization need:
Long unattended runs. Parallelization
Most convenient option: high-level language + network algorithm libraries.
Python (http://www.python.org)
NumPy: array treatment (MATLAB-like).Scipy: scientific functions on NumPy.RPy : integrates R in Python with NumPy.Parallelism, access to databases, text processing and binary files, user graphic interfaces, 2D/3D plots, geographical information systems...Windows distribution: Python(x,y) (http://www.pythonxy.com)
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Computational algorithms
General methods: Calculus sheets [catastrophic precission: G. Almiron et al., Journal of Statistical Software 34 (2010)]. Analysis environments: MATLAB, Mathematica, Octave, etc. Specific: R+statnet, Python+NetworkX
POSTPROCESSING
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Computational algorithms
SIMULATION
NetworkX: http://networkx.lanl.gov/,Included in Python(x,y).Generators, algebra, input/output, representation...Optimized algorithms, programmed in low level languages.Nodes can contain any type of data.Integration with NumPy.
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Computational algorithms
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Computational algorithms
SIMULATION
NetworkX: http://networkx.lanl.gov/ ,Included in Python(x,y).Generators, algebra, input/output, representation...Optimized algorithms, programmed in low level languages.Nodes can contain any type of data.Integration with NumPy.
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Computational algorithms
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Computational algorithms
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Applications: market models
GDP and other macroeconomic indicators
Miskiewicz and Ausloos
Gligor and Ausloos
Lambiotte and Ausloos
Redelico, Proto and Ausloos
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Applications: market models
Market correlations and concentrations. Tax evasion.
Rotundo and coauthors
Westerhoff and coauthors
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Applications: market modelsSpreading of innovations
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Applications: regional trade and development
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Applications: other social networks
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Suggested trends
- Directed diffusion of commodities and people: complex networks based description
- Nonlinear production processes: emergence of time and space patterns (enzyme model of production…)
J. D. Murray, Mathematical Biology, Springer, 2001.
Complex Networks
Luis Miguel Varela Cabo
Introduction
Statistical mechanics of complex networks
Computational algorithms
Applications- Market models
Regional trade and development
- Other social network models of interest
Suggested trends
Suggested trends-Go beyond spin ½ systems (Potts models) (richness of decision).
-Apply known statistical physics models (phase transitions, percolation, non-Markovian processes, linear response theory…
- Combine with dynamic processes for: a) Spreading of innovations and market models
b) Financial models
c) Companies/banks networks
d) ETC.
MANY THANKS FOR
YOUR ATTENTION
Complex Networks
Luis Miguel VarelaCOST meeting, Lisbon
March 27th 2013