Emergent Detailed Balance in Human Mobility under Temporal Coarse-Graining
Lei Dong · 2026
Read Emergent Detailed Balance in Human Mobility under Temporal Coarse-Graining by Lei Dong (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-21552.txt.
Introduction
Dong asks whether human mobility can look like an equilibrium system when viewed at coarser time scales. Using five years of intercity flow data covering millions of travelers, Dong finds short-term flows are strongly asymmetric. As flows are averaged over longer windows, over half of city pairs settle toward balance. Directional imbalance decays as a power law. The rest show steady drift or a crossover between regimes. A stochastic model splitting drift from correlated fluctuations reproduces all three behaviors.
Catalog record
| CREATOR | Lei Dong |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 22011 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.21552
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Emergent Detailed Balance in Human Mobility under Temporal Coarse-Graining Lei Dong Abstract A fundamental question in nonequilibrium statistical physics is whether effective equilibrium behavior can emerge at coarse-grained scales in strongly driven systems. Here, we investigate this question in the context of human mobility by analyzing five years of intercity flow data covering millions of travelers. While short-term flows are highly asymmetric, temporal coarse-graining reveals that over half of all city pairs converge toward effective flow balance, with normalized directional imbalance decaying as a power law. The remaining pairs either exhibit persistent drift-dominated currents or a crossover between these two extremes. A stochastic model decomposing mobility into directional drift and correlated fluctuations quantitatively captures the coexistence of all three regimes. Directly measured variance scaling of the fluctuation process confirms near-diffusive behavior with regime-dependent deviations. These results demonstrate that large-scale mobility networks exhibit a scale-dependent transition from broken to restored flow symmetry, with direct implications for modeling transport and spreading dynamics. Introduction.— A central question in nonequilibrium statistical physics is whether and how effective equilibrium behavior can emerge from coarse-graining of strongly driven dynamics. In Markovian systems, detailed balance — the condition \pi_{i}P_{ij}=\pi_{j}P_{ji} relating stationary distributions to transition rates — is the hallmark of thermodynamic equilibrium 28 ; 30 . Its violation signals irreversibility and is intimately linked to entropy production 15 ; 26 ; 5 ; 16 . Recent theoretical work has shown that the detection and magnitude of detailed-balance violation depend sensitively on the level of coarse-graining, whether spatial or temporal 13 ; 12 ; 31 ; 9 . Yet empirical demonstrations of scale-dependent transitions between nonequilibrium and effective equilibrium behavior in real-world complex systems remain scarce. Human mobility provides a natural setting to probe this question. Intercity population flows are shaped by economic gradients, infrastructure constraints, and policy interventions 23 ; 8 ; 4 ; 32 ; 3 , generating directed fluxes characteristic of nonequilibrium transport. Yet many city pairs sustain intense bidirectional exchange through commuting, tourism, and recurrent travel 22 ; 29 , raising a fundamental question: do the directional asymmetries observed in mobility data represent persistent nonequilibrium currents, or transient fluctuations that vanish under temporal aggregation? We emphasize that in this Letter, we use “detailed balance” in an effective, flow-level sense: the condition that forward and reverse population fluxes between city pairs are equal. This is a necessary consequence of, but not identical to, the strict Markovian detailed balance condition \pi_{i}P_{ij}=\pi_{j}P_{ji} defined on transition probabilities weighted by stationary populations. In networks where the stationary population distribution \pi_{i} is approximately constant over the timescales considered, as is the case for intercity flows on weekly-to-annual scales, the two conditions become equivalent. We discuss this equivalence explicitly in the Supplemental Material (SM). Distinguishing transient fluctuations from persistent currents is challenging, as it requires mobility data spanning sufficiently long timescales together with a large number of city pairs to resolve heterogeneity across links. This distinction has direct implications: if flows approach balance under coarse-graining, they admit an effective equilibrium description 7 ; 36 ; if directional biases persist, the system retains genuine nonequilibrium currents 11 . Such persistent currents can bias spreading dynamics on spatial networks 2 ; 6 ; 18 ; 10 and influence spatial pattern formation processes including urban growth and regional development 33 ; 27 ; 35 . In this Letter, we analyze five years of intercity mobility data and demonstrate that flow balance is an emergent, scale-dependent property. A stochastic model with directly measured parameters, including the diffusion exponent of mobility fluctuations, quantitatively captures the coexistence of three distinct transport regimes. Figure 1: Weekly mobility flows between Beijing and Baoding. (a) The bidirectional flows exhibit similar temporal patterns, with clear disruptions during the COVID-19 and holiday peaks after 2023. (b) The net flow reveals strong fluctuations, frequently reaching 30–40% of the total bidirectional volume. Data.— We analyze five years (2021–2025) of daily intercity mobility flows from the Baidu Qianxi project, which infers movements from mobile phone geolocation data 1 . The data cover nearly all prefecture-level cities in China, providing a comprehensive view of large-scale population transport. To smooth out weekly periodicities, we aggregate the raw data into weekly flows and restrict our analysis to city pairs with an average weekly flow of at least 100 individuals, ensuring statistical reliability. This yields approximately 1.2\times 10^{4} city pairs for subsequent analysis (see SM for details). For each city pair (i,j) , let F_{ij}(t) denote the number of individuals traveling from i to j during week t . If flow balance holds, we would observe F_{ij}(t)=F_{ji}(t) for all pairs. In practice, short-term flows are highly asymmetric. Figure 1 illustrates this point for Beijing and Baoding, two neighboring cities with intense exchange. While the bidirectional flows exhibit similar temporal trends, the net imbalance F_{ij}(t)-F_{ji}(t) is substantial, frequently reaching 30–40% of the total bidirectional traffic (Fig. 1 b). This motivates a scale-dependent measure of imbalance. Imbalance measure.— To quantify the directional asymmetry, we define the instantaneous net flow X_{ij}(t)=F_{ij}(t)-F_{ji}(t) and the total bidirectional traffic Y_{ij}(t)=F_{ij}(t)+F_{ji}(t) . Aggregating over a window of length \tau yields the cumulative imbalance D_{ij}^{(\tau)}(t)=\sum_{s=t}^{t+\tau-1}X_{ij}(s) and total traffic S_{ij}^{(\tau)}(t)=\sum_{s=t}^{t+\tau-1}Y_{ij}(s) . The normalized imbalance R_{ij}^{(\tau)}(t)=\frac{|D_{ij}^{(\tau)}(t)|}{S_{ij}^{(\tau)}(t)}, (1) measures the relative violation of flow balance between city pairs at temporal scale \tau . For each city pair, we compute the time-averaged \langle R_{ij}^{(\tau)}(t)\rangle over all starting weeks t in the five-year record. The ensemble average \langle\langle R^{(\tau)}\rangle\rangle across all pairs decreases systematically with \tau from one week to nearly one year, following an approximate power law \langle\langle R^{(\tau)}\rangle\rangle\sim\tau^{\alpha} with exponent \alpha\approx-0.30 (Fig. 2 a). This decay suggests that short-term directional fluctuations tend to partially cancel under coarse-graining. The ensemble average, however, masks a profound heterogeneity. Figure 2 b shows individual trajectories R_{ij}^{(\tau)} for a representative sample of city pairs: some decay continuously, others saturate to a finite plateau, and yet others exhibit a clear bend between these behaviors. This diversity indicates that the mobility network is not a monolithic nonequilibrium system but a composite of dynamically distinct links. Figure 2: (a) Ensemble-averaged normalized imbalance \langle\langle R^{(\tau)}\rangle\rangle as a function of aggregation window \tau . The dashed line indicates a power-law fit with exponent -0.30 . (b) Normalized imbalance trajectories for a representative sample of city pairs, illustrating heterogeneous behavior. Figure 3: Three empirical regimes of directional imbalance revealed by the scaling behavior. (a) Regime-averaged normalized imbalance \langle\langle R^{(\tau)}\rangle\rangle as a function of the aggregation window \tau for city pairs classified into three regimes: fluctuation-dominated decay (blue), crossover to plateau (orange), and drift-dominated saturation (red). Shaded regions indicate the interquartile range (25–75%). The dashed line shows the theoretical prediction for uncorrelated fluctuations (slope = -0.5). (b) Distribution of decay exponents for fluctuation-dominated links. (c) Distribution of goodness-of-fit, measured by R^{2} , for the power-law fits. (d) Growth of cumulative bidirectional traffic with aggregation window \tau , showing the approximately linear scaling S_{ij}^{(\tau)}\propto\tau . Flows are normalized by their maximum value for comparison across city pairs. (e) Distribution of plateau values of R^{(\tau)} for drift-dominated and crossover links. (f) Distribution of mean weekly flow for the three regimes. Empirical transport regimes.— To systematically classify these patterns, we analyze the scaling behavior of each pair by estimating local exponents \alpha=\frac{d\log R_{ij}^{(\tau)}}{d\log\tau} over small-