A long-time limit of world subway networks

INTRODUCTION

• About 25% of cities above one million people have a subway, compared with 50% above two million and all cities above ten million.The passage presents these figures as evidence that larger cities are more likely to have mass transit.Source: INTRODUCTION · p. 1
• Subway networks are presented as self-organizing structures shaped by planning, population and economic change, geography, and historical paths.The authors conjecture that these factors can produce general features alongside local peculiarities.Source: INTRODUCTION · pp. 1–2
• The paper addresses a lack of tools for studying the temporal evolution of spatial networks, where stations are nodes and rail connections are links.It proposes measures for these time-dependent networks.Source: INTRODUCTION · p. 2
• The analysis focuses on the largest subway networks in major world cities, excluding smaller developing systems and longer-distance urban rail.The selected systems have at least 100 stations.Source: INTRODUCTION · p. 2

Data

• Network histories combine 2009 maps, which establish line order and connections, with historical opening dates for lines and stations.Opening stations are connected to their contiguous neighbors to build each year's network.Source: Data · p. 2
• For each year, the authors assemble open lines and stations into separate line topologies, then combine them into a subway graph.A station closed on a particular line is excluded from that line's topology even if it is open on another line.Source: Data · p. 2
• Pre-2009 snapshots can miss stations that closed before 2009, but the authors expect limited topological impact because closures are uncommon and usually affect degree-two stations.The passage gives the British Museum station as an example of a station omitted by this approach.Source: Data · p. 3

EXPLORING STATIC PROPERTIES

• The comparison uses a regular graph with the same number of stations, area, and average degree as the subway network.The metropolitan area is used for the area measure, and ℓreg denotes the regular graph's total route length.Source: EXPLORING STATIC PROPERTIES · pp. 3–4
• T varies from 0.08 to 0.88, averages of order 0.29, and displays essentially three outliers; Osaka, Madrid, and Seoul have very large values.Chicago and NYC have much smaller values (≈0.1), signaling a more heterogeneous structure, probably due to strong geographical constraints.Source: EXPLORING STATIC PROPERTIES · p. 3
• The study focuses on spatial and topological properties, leaving population density, land use, and traffic for possible later analysis.The authors note that small network sizes make distributions noisy, motivating globally structured indicators.Source: EXPLORING STATIC PROPERTIES · p. 3

NETWORK DYNAMICS

• The fraction of years with no stations built averages about 58%, and typical average building velocity is 1.4–3.7 stations per year outside rapidly expanding Seoul and Shanghai.For all networks, larger velocities occur earlier, with large year-to-year fluctuations.Source: NETWORK DYNAMICS · p. 4
• As networks become large, new stations represent an increasingly small percentage of existing stations, consistent with incremental growth rather than abrupt additions.The average number of stations by years since creation shows a linear increase, which the authors interpret as growth becoming relatively small over time.Source: NETWORK DYNAMICS · p. 4, Figure 3 · p. 5
• For network sizes above the most numerous range of 25–100 stations, fewer cities are available, preventing definitive conclusions for large N.The number of cities is almost 15 in the 25–100 station range.Source: NETWORK DYNAMICS · p. 5

• Despite geographic and economic differences, large subway networks converge toward a common long-time shape with a core and radiating branches.Source: Characterization of the core and… · p. 5
• In most large urban systems, a ring-delimited core connects to quasi-one-dimensional branches reaching farther urban areas; the ring need not be a single line.In New York City, lines coalesce to form a pseudo-core that enables circumferential movement.Source: Characterization of the core and… · p. 5
• Branches are defined by iteratively adding stations from degree-1 tails through degree-2 paths and qualifying degree-3 forks.At a fork, growth continues if at least one path from it consists of stations of degree 2 or less.Source: Characterization of the core and… · p. 5
• Branch size is tracked by β(t), using the numbers of stations on branches and in the core.The schematic identifies the core and branches as the two components of the structure.Source: Characterization of the core and… · p. 5, Figure 5 · p. 6
• Branch spatial extension is tracked by η(t), using average distances of core and branch stations from the geographic barycenter.The barycenter is the average location of all stations.Source: Characterization of the core and… · p. 5
• Core structure is characterized by average degree as a measure of density and by f2, the fraction of core nodes with degree 2, as an indicator of interconnection.The average core degree is also connected to the planar-network index γ(t).Source: Characterization of the core and… · p. 6

Time evolution of β, ⟨kcore⟩, f2 and η

• The comparisons are made against station count N because network growth is uneven over time and differs across cities.Some networks undergo major development in particular years and little or none in others.Source: Time evolution of β, ⟨kcore⟩, f2 and η · p. 6
• β averages appear to stabilize slowly within [0.35, 0.55], while η averages around 2 among the largest networks, with fluctuations that seem to decrease with N.The β dispersion also appears to decrease with N, although the authors caution that statistics are less reliable for large networks.Source: Time evolution of β, ⟨kcore⟩, f2 and η · pp. 6–7, Figure 6 · p. 7, Figure 7 · p. 7
• ⟨kcore⟩ varies moderately around 2.4 and increases slowly, while f2 rises again for networks with N > 100.The authors interpret the increase in f2 as densification without new interconnections; the core diameter appears to plateau for most cities.Source: Time evolution of β, ⟨kcore⟩, f2 and η · pp. 6–7, Figure 8 · p. 7
• For large networks, the fraction of connecting stations appears to decrease, which the authors suggest may reflect the sufficiency of a small fraction for easy navigation.They note an associated cost for large values of k > 2.Source: Time evolution of β, ⟨kcore⟩, f2 and η · p. 8
• β∞≈45%, η∞≈2, and approximately 80% of core stations are non-connecting in the proposed long-time limiting shape.The authors note non-negligible fluctuations and say future evolution is needed to confirm the convergence.Source: Time evolution of β, ⟨kcore⟩, f2 and η · pp. 7–8

Number of branches

• NB ∼ N^0.6 (r2 = 0.85), a fitted scaling consistent with the authors’ argument.The plotted fit uses data from the subway networks considered in the study.Source: Figure 10 · p. 8, Number of branches · p. 8
• NYC and Moscow span η ≈ 1.4, f2 ≈ 85% and η ≈ 3.3, f2 ≈ 45%, respectively.The authors characterize Moscow as highly ramified with a very dense core.Source: Number of branches · p. 8
• For the largest subway systems, f2 is above 60% and relatively independent of η.Here η characterizes branch extension relative to the core, while f2 is the percentage of degree-2 core nodes.Source: Number of branches · p. 8, Figure 11 · p. 8
• Beijing, Berlin, Shanghai, and Seoul cluster near f2 = 80% ± 5% and η = 2.84 ± 0.1, corresponding to greater ramification and fewer core interconnections.The authors note that this cluster includes rapidly evolving networks such as Beijing and Shanghai.Source: Number of branches · p. 8
• London, Paris, and Madrid cluster near f2 = 70% ± 5% and η ≈ 2, corresponding to denser networks that are less ramified and have more core interconnections.The authors suggest this cluster may reflect denser city-center structure.Source: Number of branches · p. 8

SPATIAL ORGANIZATION OF THE CORE AND BRANCHES

• NYC and Chicago are excluded from this spatial analysis because their barycenters move or their core geometry is constrained by geography.The authors note that both still fit the broader core-and-branches picture, but their cores lack clear spatial meaning in this analysis.Source: SPATIAL ORGANIZATION OF THE CORE AND… · p. 9
• The core radius rC is defined by N(r = rC) = NC, using the measured number of core stations and the station-count profile around the barycenter.This definition assumes an isotropic core, which the authors say applies to most networks except the excluded Chicago and NYC cases.Source: SPATIAL ORGANIZATION OF THE CORE AND… · p. 9, Table III · p. 9
• For r < rC, N(r) grows approximately as ρCπr2, consistent with uniform core density; at large distances, N(r) saturates.Between these regimes, networks show differing behavior as distance increases beyond the core size.Source: SPATIAL ORGANIZATION OF THE CORE AND… · p. 9, Figure 12 · p. 9
• The intermediate regime is interpreted through the core-and-branches model, with branch-station spacing increasing with distance from the barycenter.In Eq. (6), NB is the number of branches, NC the number of core stations, and Δ(r) the average branch-station spacing at distance r.Source: SPATIAL ORGANIZATION OF THE CORE AND… · pp. 9–10
• For Moscow’s long branches, N(r) ∼ NBr for r > rC, consistent with roughly constant interstation spacing.For other networks, spacing shows an increasing trend, but τ is difficult to estimate accurately.Source: SPATIAL ORGANIZATION OF THE CORE AND… · p. 10
• The intermediate regime is short and spacing is noisy in most networks, so Eq. (6) fits are only rough consistency checks.The available distance range spans no more than one decade; Paris gives τ ≈ 0.5 over that decade (r2 = 0.74), consistent with.Source: SPATIAL ORGANIZATION OF THE CORE AND… · p. 10, Figure 13 · p. 10

DISCUSSION

• The largest subway networks show decreasing fluctuations around stable values and converge toward a similar long-term structure of a core with branches.The authors describe this shape as relatively independent of the cities’ historical idiosyncrasies.Source: DISCUSSION · p. 10
• About 45% of stations are in branches, branch spatial extensions are about twice those of the core, and the number of branches scales roughly as the square root of station count.The core’s average degree is about 2.3–2.5 for networks with approximately 100 or more stations, and over 60% of its nodes have degree 2.Source: DISCUSSION · p. 10
• Spatial measurements of station counts by distance support the core-and-branches picture and its increasing spacing between consecutive branch stations.The authors say this geometry provides a natural interpretation of those measurements.Source: DISCUSSION · p. 10
• The authors interpret unique long-term topological and spatial features as a universal signature of mechanisms independent of historical and geographical differences.They describe the study as exploratory and say population density, land use, and traffic-flow data could enrich it.Source: DISCUSSION · pp. 10–11