Scale-free network

Scale-free network
Name	Scale-free network
Type	Network model
Field	Complex systems
Introduced	1999
Notable	Barabási–Albert model

Scale-free network A scale-free network is a class of network structure characterized by a heavy-tailed degree distribution in which a few nodes have very high degree while many nodes have low degree, producing robustness and fragility patterns observed across multiple empirical settings. Originating from empirical studies and theoretical models in the late 1990s, the concept connects work from researchers associated with University of Notre Dame, Boston University, Harvard University, Princeton University, and Santa Fe Institute and has influenced analysis in domains as diverse as World Wide Web, Internet protocol suite, Metropolitan area networks, Human brain, and Protein Data Bank studies. The framework intersects with landmark developments in Erdős–Rényi model, Watts–Strogatz model, and network science literature produced by scholars affiliated with Cornell University, Columbia University, Stanford University, Massachusetts Institute of Technology, and California Institute of Technology.

Definition and characteristics

The defining property is a degree distribution that follows a power law P(k) ~ k^-γ over a broad range, a signature identified in datasets from World Wide Web, Arpanet, AIDS epidemic, European airline network, Hollywood film industry, and Citation index corpora. Characteristic features include hubs that dominate connectivity, a small average shortest-path length noted in analyses of the Erdős number and Six Degrees of Kevin Bacon phenomena, and a high heterogeneity captured in measures used by researchers at Max Planck Society, Royal Society, and National Academy of Sciences. Empirical exponent γ often lies between 2 and 3 in studies reported by teams at Los Alamos National Laboratory, Bell Labs, and IBM Research.

Mathematical models and properties

The Barabási–Albert model introduced preferential attachment and growth to generate power-law degree distributions and was developed by scholars linked to University of Notre Dame and Boston University. Variants and rigorous results connect to branching processes studied at Institut des Hautes Études Scientifiques, extreme value theory applied in work of Princeton University statisticians, and percolation theory advanced by groups at ETH Zurich and University of Cambridge. Spectral properties of adjacency matrices have been examined by mathematicians at Institute for Advanced Study, Imperial College London, and University of Chicago, while degree correlations and assortativity were analyzed in research from Yale University, University of Michigan, and University of Oxford. Limit theorems link to work by Paul Erdős, Alfréd Rényi, and collaborators in probabilistic combinatorics.

Generation mechanisms

Mechanisms proposed include preferential attachment, copying models used to explain the World Wide Web growth studied by teams at University of California, Berkeley and University of Washington, optimization and trade-off models developed at Santa Fe Institute, and fitness models introduced in collaborations involving Los Alamos National Laboratory and University of Rome La Sapienza. Evolutionary models comparing biological networks reference datasets from European Molecular Biology Laboratory, GenBank, and research groups at Howard Hughes Medical Institute. Economic network formation explanations draw on models tested by researchers at London School of Economics, Princeton University, and University of Chicago.

Empirical examples and applications

Reported examples include the topology of the World Wide Web, the Internet protocol suite at the router and Autonomous System level analyses by teams from Cisco Systems and AT&T Labs, metabolic networks catalogued in the Protein Data Bank, and social networks measured in studies at Facebook, Twitter, and projects led by Stanford University and University of Pennsylvania. Applications range from epidemic modeling applied by researchers at Centers for Disease Control and Prevention and World Health Organization to resilience analyses used by Department of Defense and infrastructure planning by Federal Aviation Administration and European Commission teams. Financial networks studied by Federal Reserve Bank, International Monetary Fund, and scholars at Columbia Business School also report heavy-tailed connectivity patterns.

Criticisms and alternatives

Critiques have arisen from empirical reanalysis by groups at Microsoft Research, Google Research, and Max Planck Institute for Software Systems who emphasized finite-size effects, sampling bias, and alternative distributions such as the log-normal, stretched exponential, and exponentially truncated power law identified in studies at University of California, Santa Barbara and New York University. Alternatives like the configuration model, exponential random graph models developed by teams at Duke University and University of Pennsylvania, and stochastic block models advanced at Carnegie Mellon University and University of Toronto provide complementary explanations. Methodological debates have been highlighted in commentaries by scholars associated with Nature, Science, and Proceedings of the National Academy of Sciences.

Dynamical processes on scale-free networks

Dynamical behaviors such as epidemic spreading, synchronization, cascading failures, and diffusion have been widely studied by groups at Imperial College London, Los Alamos National Laboratory, Princeton University, University of California, Los Angeles, and École Polytechnique Fédérale de Lausanne. Classic results include vanishing epidemic thresholds for simple contagion in uncorrelated networks reported by researchers at Santa Fe Institute and Boston University, and resilience-vulnerability trade-offs applied in studies by RAND Corporation and National Aeronautics and Space Administration. Control strategies and intervention design informed by network controllability theory were developed in collaborations involving Cornell University, University of California, San Diego, and Massachusetts General Hospital.

Category:Network theory