Source-linked AI summary
Emergence of network features from multiplexity
Alessio Cardillo, Jesús Gómez-Gardeñes, Massimiliano Zanin, Miguel Romance, David Papo, Francisco del Pozo, Stefano Boccaletti
TL;DR
The paper examines how structural properties of the European Air Transportation Network emerge as airline layers are merged. Using data from 37 airlines, it finds that many properties arise from multilayer aggregation rather than appearing in individual layers, with major and low-cost layers producing qualitatively different aggregate networks.
Problem
The study asks whether topological properties of the European Air Transportation Network are present in individual airline layers or emerge from the multilayer system.
Method
The authors construct a multiplex network from 37 airlines and examine how structural measures change as single-airline layers are progressively merged.
Results
More than 80% of the final clustering value is reached by merging five layers, while major-airline layers produce a rich club and low-cost layers do not.
Takeaways & Limitations
Considering layers as relevant network entities can improve understanding and modeling of dynamical processes in aggregate networks.
Abstract
from arXiv · showhide
Many biological and man-made networked systems are characterized by the simultaneous presence of different sub-networks organized in separate layers, with links and nodes of qualitatively different types. While during the past few years theoretical studies have examined a variety of structural features of complex networks, the outstanding question is whether such features are characterizing all single layers, or rather emerge as a result of coarse-graining, i.e. when going from the multilayered to the aggregate network representation. Here we address this issue with the help of real data. We analyze the structural properties of an intrinsically multilayered real network, the European Air Transportation Multiplex Network in which each commercial airline defines a network layer. We examine how several structural measures evolve as layers are progressively merged together. In particular, we discuss how the topology of each layer affects the emergence of structural properties in the aggregate network.
RESULTS
The European ATN is modeled as 37 airline-specific layers over a common set of European airports. Aggregate and single-airline views distinguish redundant and unique connections, with major and low-cost airlines showing different network structures.
- RESULTS: The European ATN comprises 37 layers, each representing a different European airline.
- RESULTS: The aggregate view separates links appearing in multiple layers from unique links connected to at least one high-degree node.
- RESULTS: The figure contrasts aggregate-network redundancy and unicity with representative major-airline and low-cost-airline layers.
- RESULTS: Each layer contains the same European-airport node set, while the aggregate network combines connections across airlines.
A. Topological measures
The study characterizes aggregate and layer networks using degree distribution, clustering, connectivity, path length, and rich-club measures. These metrics capture connectivity, triangular structure, reachability, routing distance, and hub interconnection.
- A. Topological measures: The cumulative degree distribution P>(k) gives the probability that a node has degree at least k.It helps characterize structural and dynamical properties relevant to tolerance of attacks or failures.
- A. Topological measures: The average path length ⟨L⟩ measures the average number of hops between nodes, interpreted for ATNs as the average number of flights between destinations.When the network is disconnected, it is evaluated on the giant component.
- A. Topological measures: The clustering coefficient C measures the probability that two nodes sharing a neighbor are connected, indicating triangular motifs in the ATN.These triangles represent possible round trips of length 3.
- A. Topological measures: The giant-component size S is the largest fraction of nodes mutually connected by finite paths, representing reachable destination coverage.For an airline or combination of airlines, it estimates coverage from origins inside the giant component.
- A. Topological measures: The Rich-club coefficient R measures the relative abundance of links among nodes whose degree is at least k.A Rich-club is present when R(k) > 1 for some k.
B. Emergence of topological properties of the European ATN
Progressively merging airline layers changes the ATN’s structural measures, showing that several properties emerge through multiplex aggregation rather than existing within individual layers. The effects differ across degree heterogeneity, clustering, coverage, path length, and hub connectivity.
- B. Emergence of topological properties of the European ATN: The degree-distribution exponent α decreases from 1.84 for one layer to 1.39 for the aggregate ATN, indicating increasing heterogeneity as layers are added.The authors associate this with a richer-gets-richer phenomenon driven by adding layers rather than nodes.
- B. Emergence of topological properties of the European ATN: Merging five randomly chosen layers achieves more than 80% of the final clustering value.The result indicates that the ATN’s high triangle density is mainly produced by combining layers.
- B. Emergence of topological properties of the European ATN: Around 40% of European cities are covered after merging five randomly chosen layers, while coverage increases monotonically with additional layers.The giant component is already above the percolation threshold for a single layer.
- B. Emergence of topological properties of the European ATN: Average path length rises initially and then falls as layers are added.Early merging joins previously disconnected components; later layers add destinations and alternative links, balancing and eventually reducing path length.
- B. Emergence of topological properties of the European ATN: The aggregate ATN’s rich-club comparison contrasts real and degree-preserving randomized networks, with curves initially coinciding below k = 30.For k ∈[30, 60], the figure examines divergence between observed and randomized hub connectivity.
C. Major versus Low-cost layers
Major and low-cost airline layers exhibit distinct topologies, and merging them produces different structural trends. Major layers generate rich-club organization, while low-cost layers contribute clustering and alter path-length behavior.
- Degree distribution: Major-airline degree distributions shift from single-layer plateaus to continuous decay as layers merge, reflecting hubs of different sizes.Single major layers are centralized around a few highly connected hubs; merging combines hubs across national airlines.
- Degree distribution: Low-cost layers show progressively decaying degree distributions, with airports of different sizes coexisting within individual layers.
- Clustering and triangles: Major-layer clustering rises sharply and plateaus beyond m = 5, whereas aggregate-network clustering continues increasing as low-cost layers form new triangles.For major layers, saturation reflects a balance between newly added triangles and newly added connections, not an absence of new triangles.
- Path length: Major-layer average path length increases with merging, while low-cost layers show a rise-and-fall pattern; combining both types yields a saturated aggregate trend.Major-layer growth follows the addition of destinations available only through newly merged airlines.
- Rich-club effect: The rich-club effect appears when major layers are merged but is absent for low-cost layers, making it exclusively associated with major airlines in this analysis.Major-layer hubs form a connected highly connected core, whereas the more distributed low-cost topology prevents rich-club formation.
DISCUSSION
The study finds that important ATN properties generally emerge from multilayer aggregation rather than appearing in individual layers. Major and low-cost layers contribute differently to the global network's rich-club effect, redundancy, and small-worldness.
- DISCUSSION: Topological properties generally emerge through progressive layer merging and are not generally present in single airline layers.
- DISCUSSION: Major and low-cost layers produce qualitatively different aggregate networks when merged separately.
- DISCUSSION: The rich-club effect is mainly due to major-airline layers, path redundancy combines clustering from both layer types, and low-cost layers enhance small-worldness.
- DISCUSSION: Treating layers as relevant network entities can improve understanding and modeling of dynamical processes at the aggregate-network level.
Appendix A: Dataset
The dataset represents European airline operations as a multiplex of 37 airline layers, each sharing 450 airport nodes. It includes major and low-cost subsets selected from flights recorded by EUROCONTROL and the SESAR Work Package E dataset.
- Dataset composition: The dataset contains 37 airlines selected for having more than the average 32 destinations, representing 37 multiplex layers.
- Dataset composition: Each airline layer is a graph with the same 450 European airport nodes and airline-specific flight links.
- Dataset composition: The study separately analyzes subsets of 18 major-airline layers and 10 low-cost-company layers.
- Data source: The source data list airlines operating Instrumental Flight Rules flights between European airports on a certain day.
Appendix B: Topological indexes
The appendix defines the topological measures used to characterize aggregate and layer networks, then describes how these measures are extended to multiplex analyses and studied during layer merging.
- Degree distribution: P>(k) is the probability that a randomly chosen node has degree equal to or greater than k, computed from node counts across degrees.Using the cumulative distribution is useful because broad-distribution tails can be noisy when highly connected nodes are scarce.
- Path length: Average path length is the mean shortest-path distance between node pairs, while disconnected graphs require restriction to the giant component or an alternative distance average.The distance dij is the minimum number of hops from node i to node j; the ordinary average diverges when distances can be infinite.
- Clustering coefficient: The clustering coefficient averages each node’s ratio of links among its neighbors to the maximum possible number of such links.For node i, ci = 2ei/[ki(ki−1)], where ei counts mutually connected neighbors and ki is the node degree.
- Giant component: The giant component is the largest connected component, and its size is the proportion of network nodes belonging to it.For each node i, Ni denotes the number of nodes in the maximal connected subnetwork containing i.
- Multiplex extension and merging: Layer merging fixes a subset of layers and constructs a monoplex network, enabling the topological measures to be tracked as progressively more layers are combined.The Rich-club coefficient compares links among nodes of degree at least k with a degree-preserving randomized network; R(k) > 1 indicates a Rich-club.