Glassy dynamics of metropolitan human mobility frozen near the gravitational equilibrium

arXiv:2609.29057 · physics.soc-ph · Submitted 2026-09-24 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "Glassy dynamics of metropolitan human mobility frozen near the gravitational equilibrium".

Mira: Here's the gist of the paper. Glassy dynamics of metropolitan human mobility model?

Kai: asymptotically approaching the gravitational equilibrium establishes a connection between macroscopic urban commuting flows and thermal equilibrium The GHSM is as tractable as spin models, and may open urban problems to the broader interest of the physics community Model Foundation The research introduces the Gravity-based Home Swapping Model (GHSM), which applies Metropolis dynamics to urban commuting by treating individuals as interacting particles In this framework, total commuting time dictates the system energy, and temperature determines how strongly human mobility depends on energy-saving habits The stationary state of the GHSM is the gravity solution.

Mira: Equilibrium Description The study explores two routes to the most probable OD matrix of a metropolitan area The first is static, Wilson’s doubly constrained gravity model, which specifies the most probable OD matrix f∗ as the solution of a constrained entropy maximization The second is dynamical, the Gravity-based Home Swapping Model (GHSM), a Metropolis process whose stationary distribution concentrates on the same f∗.

Kai: Thermodynamic and Dynamical Dissociation The central result is a dissociation between thermodynamics and dynamics in the doubly constrained gravity model On the thermodynamic side, the Landau free energy is strictly convex on the constraint set, meaning there are no metastable states to trap the dynamics On the dynamical side, marginal-preserving swap dynamics fails to reach equilibrium because relaxation stalls at preparation-dependent configurations and two-time correlations age This identifies kinetic arrest rather than a thermodynamic glass transition.

Paper summary: Kai: So, to summarize what they're doing with the Gravity-based Home Swapping Model, they introduce a microscopic dynamics where pairs of commuters exchange residences while keeping every home and job location fixed.

Mira: And that pair swap move is described as a conserved-exchange move in the sense of Kawasaki, which means it preserves the marginals rather than just one global sum.

Lev: That specific type of move is what they call the elementary unit of rearrangement, where a cyclic exchange among k commuters breaks down into these pair swaps, and it’s related to a housing vacancy chain nineteen <ref:2609.29057#pg2,a cyclic exchange among k commuters>.

Kai: They establish three statements that fix this identification with the gravity model, which sets up the whole framework for analyzing how these commuting flows evolve.

Mira: The paper then shows a dissociation between thermodynamics and dynamics when they look at Wilson’s doubly constrained gravity model. On the thermodynamic side, their free energy is strictly convex on that constraint set, meaning there are no metastable states to trap the dynamics.

Lev: But on the dynamical side, their marginal-preserving swap dynamics fails to reach equilibrium because it stalls at configurations that depend on how you prepared it and two-time correlations age eighteen <ref:2609.29057#pg2>.

Kai: This suggests they aren't seeing a typical thermodynamic glass transition, but rather kinetic arrest caused by this stalling in preparation-dependent configurations.

Mira: The paper identifies the arrest mechanism as being carried by short-distance commuters exchanging trips that are nearly cost-degenerate. The residual between the observed city and its gravity equilibrium is concentrated among these short-distance commuters.

Conclusion: Kai: So, looking at this paper, "Glassy dynamics of metropolitan human mobility frozen near the gravitational equilibrium," it really centers on how kinetic arrest happens in urban systems rather than a simple thermodynamic phase transition.

Mira: The authors are using mobility data from over two million individuals across six Japanese cities to test their Gravity-based Home Swapping Model against the gravity solution.

Lev: It’s significant because they point out that the freezing they observe is actually a lower bound on the kinetic constraint operating in real cities, meaning it's a physical reality for what we see.

Kai: The title itself suggests this connection between macroscopic commuting flows and thermal equilibrium, which is what makes this model appealing to people outside of pure statistical mechanics.

Mira: It opens up the idea that urban problems might be better understood by looking at these dynamics in the gap between algorithmic optimization and actual human behavior.

Lev: If you listen just to this, it tells you that reducing commuting costs by just moving existing residents might not work dynamically, suggesting we need to change the landscape of where people live and work instead.

Department of Systems and Control Engineering, Institute of Science Tokyo · Department of Computer Science, Institute of Science Tokyo

physics.soc-ph

Submitted: 2026-09-24

Updated: 2026-11-12

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 89/100

The gist: Glassy dynamics of metropolitan human mobility model? asymptotically approaching the gravitational equilibrium establishes a connection between macroscopic urban commuting flows and thermal…

Key concepts

Gravity-based Home Swapping Model (GHSM)
This model treats people as interacting particles in a city. It uses Metropolis dynamics to simulate how people change their home locations based on commuting costs, where total commuting time acts as the system's energy.
Gravitational Equilibrium
This is the stable state of the urban system described by gravity models. The research shows that macroscopic commuting flows naturally gravitate toward this equilibrium point when considering energy minimization.
Kinetic Arrest
This refers to a situation where the dynamics stop moving even though a thermodynamic equilibrium might be possible. In this study, it happens because short-distance commuters get stuck in local, cost-minimizing arrangements that are hard to escape.

Terminology

Summary

Glassy dynamics of metropolitan human mobility model? p. 0

Where the paper says it

Glassy dynamics of a metropolitan human mobility model? asymptotically approaching the gravitational equilibrium Yixuan Y.

asymptotically approaching the gravitational equilibrium establishes a connection between macroscopic urban commuting flows and thermal equilibrium The GHSM is as tractable as spin models, and may open urban problems to the broader interest of the physics community

Model Foundation

The research introduces the Gravity-based Home Swapping Model (GHSM), which applies Metropolis dynamics to urban commuting by treating individuals as interacting particles In this framework, total commuting time dictates the system energy, and temperature determines how strongly human mobility depends on energy-saving habits The stationary state of the GHSM is the gravity solution.

Equilibrium Description

The study explores two routes to the most probable OD matrix of a metropolitan area The first is static, Wilson’s doubly constrained gravity model, which specifies the most probable OD matrix f∗ as the solution of a constrained entropy maximization The second is dynamical, the Gravity-based Home Swapping Model (GHSM), a Metropolis process whose stationary distribution concentrates on the same f∗.

Thermodynamic and Dynamical Dissociation

The central result is a dissociation between thermodynamics and dynamics in the doubly constrained gravity model On the thermodynamic side, the Landau free energy is strictly convex on the constraint set, meaning there are no metastable states to trap the dynamics On the dynamical side, marginal-preserving swap dynamics fails to reach equilibrium because relaxation stalls at preparation-dependent configurations and two-time correlations age This identifies kinetic arrest rather than a thermodynamic glass transition.

Kinetic Arrest Mechanism

The arrest is carried by short-distance commuters exchanging nearly cost-degenerate trips The residual between the observed city and its gravity equilibrium is concentrated among short-distance commuters, and as the system approaches equilibrium the accepted rearrangements become increasingly short-range The freezing we report is a lower bound on the kinetic constraint operating in real cities.

Calibration Results

Parameter calibration involves two procedures to find the best fit for observed data The first validates the cost form, finding that a stretched exponential deterrence function fits the data best across all six metropolitan areas The second determines the heat bath parameters (γ∗, T∗) by minimizing the total-variation distance between observed and model trip-length distributions.

Ergodicity Breaking

Aging is diagnosed by the two-time autocorrelation of the residential configuration, C(τ, tw), which shows a dependence on waiting time that decays slower at larger tw The freezing line Tg(γ) mapped in the (T, γ) plane is a dynamical crossover dependent on observation timescale rather than a thermodynamic phase boundary The arrest resides in this gap between algorithmic and behavioral optimization.

Conclusion on Intervention

The aggregate commuting cost of the city has therefore evolved close to its equilibrium value, and further reduction through the relocation of existing residents is dynamically suppressed, suggesting substantial gains require changing the landscape itself rather than spontaneous rearrangement The mechanism has a direct urban reading where the arrest resides in this gap between algorithmic and behavioral optimization.

Data Summary

The analysis uses mobility data covering more than two million individuals in six Japanese metropolitan areas over a full year Home and workplace locations are identified at a 100 m grid resolution The average commuting time in the Tokyo metropolitan area is about 39.9 minutes.

Parameter Findings

The calibrated values for the six cities cluster in a narrow range, γ∗ = 0.48–0.63 and T∗ = 3.04–5.74 The fitted Tokyo parameters are (γ∗, T∗) = (0. p. 6

Where the paper says it

Calibration of the deterrence function, shown for the Tokyo metropolitan area. (a) Mean commuter flow fij versus commuting time tij (5-minute intervals; error bars, interquartile range), compared with the doubly constrained gravity model (9) fitted by Poisson maximum likelihood under the three deterrence kernels.

48, 3.04).

Simulation Details

All runs use the elementary move with the acceptance rule, and for the largest system studied (Tokyo), one sweep is 1.49 × 107 attempted moves Relaxation and ageing use 10 12 steps, or 6. p. 7

Where the paper says it

Relaxation and ageing (Sec. III B) use 1012 steps, or 6.

7 × 10 4 sweeps. p. 7
Where the paper says it

Simulation protocols All runs use the elementary move (12) with the acceptance rule (14). Time is measured in Monte Carlo sweeps, one sweep being M attempted swaps, so that a sweep corresponds on average to one attempted relocation per commuter.

The freezing line is a dynamical boundary, not a thermodynamic one. p. 9
Where the paper says it

On the dynamical side, the marginal-preserving swap dynamics that samples this equilibrium fails to reach it: relaxation stalls at preparation-dependent configurations, cooling and heating produce history-dependent branches, and two-time correlations age. The freezing line Tg(γ) mapped in the (T, γ) plane is a dynamical crossover, dependent on the observation timescale as laboratory glass-transition temperatures depend on cooling rate [52], not a thermodynamic phase boundary, which convexity excludes.

Model Generalization

The construction generalizes to any maximum-entropy model whose constraints admit constraint-preserving elementary moves This applies to gravity models of trade and migration immediately, and inverse-Ising descriptions of biological populations in principle. The two canonical scenarios for glassy slowdown—a fractured free-energy landscape with exponentially many metastable states, and trivial thermodynamics with arrest generated by the dynamical rules alone—produce nearly identical phenomenology. p. 9

Where the paper says it

This identifies the phenomenon as kinetic arrest rather than a thermodynamic glass. The two canonical scenarios for glassy slowdown—a fractured free-energy landscape with exponentially many metastable states [53, 54], and trivial thermodynamics with arrest generated by the dynamical rules alone [14, 16]—produce nearly identical phenomenology, and hysteresis, aging, and growing relaxation times do not discriminate between them [18, 55].

Data Sources

The GPS data are available from Agoop Corporation, a Japanese private company that provides location big data acquired from smartphone applications Zone definitions use the 1 km national grid squares from the National Land Numerical Information dataset provided by MLIT. The specific product is “Pointo-gata ryoudou-jinkou data” (point-type population data). p. 1

Where the paper says it

II G) specify the runs; details are in Appendixes A and C. Data We analyze six major Japanese metropolitan areas, Tokyo, Osaka, Nagoya, Fukuoka, Sapporo, and Sendai, each with a population above 2 million [23], partitioned into 1 km × 1 km grid cells (approximately 13,000 cells for Tokyo).

The GPS data cannot be shared publicly because they are available only on request from a third party. p. 9
Where the paper says it

Aging is compatible with two causes—a rough landscape, or a simple landscape the dynamics cannot traverse [16]—and Proposition 1 excludes the first: the aging here is kinetic arrest, not a thermodynamic glass transition. For comparison, aging in the Schelling model has been attributed to long-lived structural clusters [13]; that mechanism is not available on a convex landscape.

Appendix Details

Appendix A details the construction of the mobility data, including how home and work locations are identified based on stay thresholds Appendix B collects the corresponding results for all six metropolitan areas Appendix C specifies simulation protocols and implementation details Appendix D derives concentration statements used in Section III A regarding cost density grouping. The L1 (matrix-level) calibration objective and its residual decomposition for all six cities are provided in Section S3 of the manuscript. p. 7

Where the paper says it

Second, D1 sums absolute errors over all ∼ N2 cells, most of which carry very small flows, so it is dominated by the sparse periphery of the matrix and its minimum in (γ, T) is shallow. We retain D1 as a consistency check; its landscape and the residual decomposition at the optimum are given in SM Sec.

Table Summary

Table III lists the fitted parameters of the stretched-exponential deterrence function f(t) ∝ exp(−tγ/T) for each metropolitan area. Table IV compares the three deterrence functions across metropolitan areas, showing that the stretched exponential attains the lowest AIC in all six areas. p. 6

Where the paper says it

Figure 3(a) shows the outcome for Tokyo: the stretched exponential follows the observed decay of fij with tij over the full range of commuting times, whereas the power law overestimates and the exponential underestimates the long-time tail. The same ordering holds in all six metropolitan areas, where the stretched exponential attains the highest CPC and the lowest AIC, with fitted exponents confined to γ ∈ [0.

The simulation protocols are summarized in Table V. p. 1
Where the paper says it

II F) returns the city parameters (γ ∗ , T ∗ ) used throughout Secs. , and the simulation protocols (Sec.

The summary of commuting OD data by metropolitan area is provided in Table II. p. 1
Where the paper says it

5% of the data. The average commuting time in the Tokyo metropolitan area is about 39.

Figure Summaries

Figure 4 shows four reference states of the commuting origin–destination (OD) matrix, including the current, gravity, random, and ground states Figure 5 illustrates ergodicity breaking and aging of the relocation dynamics at the fitted Tokyo parameters Figure 6 shows hysteresis and nonequilibrium phase diagram with arrows indicating sweep direction. Figure 7 shows energy and free-energy landscapes of the relocation dynamics at the fitted Tokyo parameters. p. 6

Where the paper says it

Figure 3(a) shows the outcome for Tokyo: the stretched exponential follows the observed decay of fij with tij over the full range of commuting times, whereas the power law overestimates and the exponential underestimates the long-time tail. The same ordering holds in all six metropolitan areas, where the stretched exponential attains the highest CPC and the lowest AIC, with fitted exponents confined to γ ∈ [0.

Figure 9 shows arrest diagrams of the six metropolitan areas, where each panel displays the protocol-history dependence ϕheating −ϕcooling over the (γ, T) plane. p. 6
Where the paper says it

Figure 3(a) shows the outcome for Tokyo: the stretched exponential follows the observed decay of fij with tij over the full range of commuting times, whereas the power law overestimates and the exponential underestimates the long-time tail. The same ordering holds in all six metropolitan areas, where the stretched exponential attains the highest CPC and the lowest AIC, with fitted exponents confined to γ ∈ [0.

Figure 10 shows heat-bath calibration of the six metropolitan areas, computing the trip-length-distribution objective Dtld(γ, T) over the (γ, log T) grid. p. 6
Where the paper says it

(22) The vector {pb[f]} B b=1 is the trip-length distribution (TLD) of the matrix f: the projection of the OD matrix onto the travel-time axis. Writing p obs b = pb[f obs] and p grav b (γ, T) = pb[f grav(γ, T)], the calibration objective is the total-variation distance between the two distributions, Dtld(γ, T) = 1 2 X B b=1 p obs b − p grav b (γ, T) ∈ [0, 1], (23) where the factor 1/2 sets the range from 0 (identical distributions) to 1 (disjoint support).

Figure 12 shows microscopic anatomy of the arrest for Tokyo at fitted parameters. p. 6
Where the paper says it

Figure 3(a) shows the outcome for Tokyo: the stretched exponential follows the observed decay of fij with tij over the full range of commuting times, whereas the power law overestimates and the exponential underestimates the long-time tail. The same ordering holds in all six metropolitan areas, where the stretched exponential attains the highest CPC and the lowest AIC, with fitted exponents confined to γ ∈ [0.

References

[1] E. Schneidman, M. J. Berry II, R. Segev, and W. Bialek, Weak pairwise correlations imply strongly correlated network states in a neural population, Nature 440, 1007 (2006)

[2] J. Harte, Maximum Entropy and Ecology: A Theory of Abundance, Distribution, and Energetics (Oxford University Press, Oxford, 2011)

[3] A. G. Wilson, A statistical theory of spatial distribution models, Transp. p. 0

Where the paper says it

Within this line the gravity model is the most widely used description, and its doubly constrained form imposes three conditions on the commuting flow: the row sums are the numbers of homes, the column sums are the numbers of jobs, and the total travel cost is fixed [3, 27]. Wilson derived this form from the maximum-entropy principle [5]: among all OD matrices meeting the three conditions, the observed one is the most probable, which in statistical mechanics is the state of lowest free energy [34, 39].

Res. 1, 253 (1967)

[4] J. Tinbergen, Shaping the World Economy: Suggestions for an International Economic Policy (Twentieth Century Fund, New York, 1962)

[5] E. T. Jaynes, Information theory and statistical mechanics, Phys. p. 0

Where the paper says it

Within this line the gravity model is the most widely used description, and its doubly constrained form imposes three conditions on the commuting flow: the row sums are the numbers of homes, the column sums are the numbers of jobs, and the total travel cost is fixed [3, 27]. Wilson derived this form from the maximum-entropy principle [5]: among all OD matrices meeting the three conditions, the observed one is the most probable, which in statistical mechanics is the state of lowest free energy [34, 39].

Rev. 106, 620 (1957)

[6] O. Sagarra, C. J. P´erez Vicente, and A. D´ıaz-Guilera, Statistical mechanics of multiedge networks, Phys. p. 0

Where the paper says it

Within this line the gravity model is the most widely used description, and its doubly constrained form imposes three conditions on the commuting flow: the row sums are the numbers of homes, the column sums are the numbers of jobs, and the total travel cost is fixed [3, 27]. Wilson derived this form from the maximum-entropy principle [5]: among all OD matrices meeting the three conditions, the observed one is the most probable, which in statistical mechanics is the state of lowest free energy [34, 39].

Rev. E 88, 062806 (2013)

[7] H. Touchette, Equivalence and nonequivalence of ensembles: Thermodynamic, macrostate, and measure levels, J. p. 8

Where the paper says it

With strict concavity established, ensemble equivalence follows from Ref. [7] at all three levels of description: thermodynamic (its Proposition 2, the Legendre– Fenchel pair s ↔ φ), macrostate (its Theorem 7, E e = Eβ at β = s ′ (e)), and measure (its Theorems 10 and 12).

Stat. Phys. 159, 987 (2015)

[8] T. Squartini, J. de Mol, F. den Hollander, and D. Garlaschelli, Breaking of ensemble equivalence in networks, Phys. p. 1

Where the paper says it

Second, the doubly constrained gravity solution is the microcanonical description of the same system, with the total travel cost in the role of the energy. Third, the two descriptions agree: the Landau free energy is strictly convex on the constraint set and ensemble equivalence follows [6–8].

Rev. Lett. 115, 268701 (2015)

[9] B. Harris and A. G. Wilson, Equilibrium values and dynamics of attractiveness terms in production-constrained spatial-interaction models, Environ. p. 0

Where the paper says it

In this framework, total commuting time dictates the system energy, and temperature determines how strongly human mobility depends on energy-saving habits, so that urban commuting can be analyzed through an evaluable free energy and simulated in almost the same manner as physical systems of matter, with the maximum-entropy doubly constrained gravity model as the stationary state. We first find that the residential dynamics is glassy: the system exhibits the signatures of kinetically constrained models, namely ageing, hysteresis, and freezing near yet short of equilibrium.

Plan. A 10, 371

[10] E. Crosato, R. Nigmatullin, and M. Prokopenko, On critical dynamics and thermodynamic efficiency of urban transformations, R. p. 0

Where the paper says it

Zheng∗ Department of Systems and Control Engineering, Institute of Science Tokyo, Yokohama, Kanagawa 226-8502, Japan Hideki Takayasu† Department of Computer Science, Institute of Science Tokyo, Yokohama, Kanagawa 226-8502, Japan Misako Takayasu‡ Department of Systems and Control Engineering, Institute of Science Tokyo, Yokohama, Kanagawa 226-8502, Japan and Department of Computer Science, Institute of Science Tokyo, Yokohama, Kanagawa 226-8502, Japan (Dated: September 25, 2026) This research establishes a connection between macroscopic urban commuting flows and thermal equilibrium. Driven by massive mobility data tracking over two million individuals across six Japanese cities for a full year, we introduce the Gravity-based Home Swapping Model (GHSM), which applies Metropolis dynamics to urban commuting by treating individuals as interacting particles.

Soc. Open Sci. 5, 180863 (2018)

[11] L. Dall’Asta, C. Castellano, and M. Marsili, Statistical physics of the Schelling model of segregation, J. p. 1

Where the paper says it

Our results connect statistical physics and complex social systems in three ways. First, they give the doubly constrained gravity model a dynamical foundation, promoting a static inference to the stationary state of an explicit microscopic dynamics.

Stat. Mech. (2008) L07002

[12] L. Gauvin, J.-P. Nadal, and J. Vannimenus, Schelling segregation in an open city: A kinetically constrained Blume–Emery–Griffiths spin-1 system, Phys. p. 0

Where the paper says it

Simulations initialized from arbitrary states evolve toward the observed state, indicating that urban constraints largely shape the commuting flow. The GHSM is as tractable as spin models, and may open urban problems to the broader interest of the physics community.

Rev. E 81, 066120 (2010)

[13] D. Abella, M. San Miguel, and J. J. Ramasco, Aging effects in Schelling segregation model, Sci. p. 9

Where the paper says it

Aging is compatible with two causes—a rough landscape, or a simple landscape the dynamics cannot traverse [16]—and Proposition 1 excludes the first: the aging here is kinetic arrest, not a thermodynamic glass transition. For comparison, aging in the Schelling model has been attributed to long-lived structural clusters [13]; that mechanism is not available on a convex landscape.

Rep. 12, 19376 (2022)

[14] G. H. Fredrickson and H. C. Andersen, Kinetic Ising model of the glass transition, Phys. p. 9

Where the paper says it

Aging is compatible with two causes—a rough landscape, or a simple landscape the dynamics cannot traverse [16]—and Proposition 1 excludes the first: the aging here is kinetic arrest, not a thermodynamic glass transition. For comparison, aging in the Schelling model has been attributed to long-lived structural clusters [13]; that mechanism is not available on a convex landscape.

Rev. Lett. 53, 1244 (1984)

[15] J¨ackle and S. Eisinger, A hierarchically constrained kinetic Ising model, Z. p. 1

Where the paper says it

Second, they provide the first empirical identification of kinetic arrest with provably trivial thermodynamics in a real human system, extending the kinetically constrained paradigm beyond condensed matter. Third, the construction generalizes: any maximum-entropy model whose constraints admit constraint-preserving elementary moves can be equipped with the same dynamics and subjected to the same reachability test—gravity models of trade and migration immediately, and inverse-Ising descriptions of biological populations in principle.

Phys. B 84, 115 (1991)

[16] F. Ritort and P. Sollich, Glassy dynamics of kinetically constrained models, Adv. p. 0

Where the paper says it

In this framework, total commuting time dictates the system energy, and temperature determines how strongly human mobility depends on energy-saving habits, so that urban commuting can be analyzed through an evaluable free energy and simulated in almost the same manner as physical systems of matter, with the maximum-entropy doubly constrained gravity model as the stationary state. We first find that the residential dynamics is glassy: the system exhibits the signatures of kinetically constrained models, namely ageing, hysteresis, and freezing near yet short of equilibrium.

Phys. 52, 219 (2003)

[17] D. Chandler and J. P. Garrahan, Dynamics on the way to forming glass: Bubbles in space–time, Annu. p. 7

Where the paper says it

Defining an arrest boundary against a fixed observation window is standard for systems without a thermodynamic transition: the laboratory glass transition temperature is itself set by a conventional relaxation time rather than by a singularity [18, 52]. RESULTS These four states anchor all subsequent analyses, which examine how the relocation dynamics connect them over time.

Rev. Phys. Chem. 61, 191 (2010)

[18] L. Berthier and G. Biroli, Theoretical perspective on the glass transition and amorphous materials, Rev. p. 7

Where the paper says it

Other parts use 1010 steps, or 671 sweeps. Defining an arrest boundary against a fixed observation window is standard for systems without a thermodynamic transition: the laboratory glass transition temperature is itself set by a conventional relaxation time rather than by a singularity [18, 52].

Mod. Phys. 83, 587 (2011)

[19] K. Kawasaki, Diffusion constants near the critical point for time-dependent Ising models. p. 9

Where the paper says it

The mean energy ⟨E⟩(T) and the heat capacity are smooth over the entire temperature range, including the calibrated point T ∗ . On the dynamical side, the marginal-preserving swap dynamics that samples this equilibrium fails to reach it: relaxation stalls at preparation-dependent configurations, cooling and heating produce history-dependent branches, and two-time correlations age.

I, Phys. Rev. 145, 224 (1966)

[20] H. C. White, Chains of Opportunity: System Models of Mobility in Organizations (Harvard University Press, Cambridge, MA, 1970)

[21] H. J. Ryser, Combinatorial properties of matrices of zeros and ones, Can. p. 8

Where the paper says it

For λ ∈ (0, 1), define the mixture fλ = (1 − λ)f1 + λ f2. (28) We verify three properties of fλ in turn.

J. Math. 9, 371 (1957)

[22] P. Diaconis and B. Sturmfels, Algebraic algorithms for sampling from conditional distributions, Ann. p. 9

Where the paper says it

(d)–(h) show the distance |ϕ(t)−ϕeq| from equilibrium: exponential decay with a finite relaxation time on the ergodic side (γ = 0. 1, or T = 100), algebraic decay over more than two decades—no characteristic timescale—on the arrested side (γ = 0.

Stat. 26, 363 (1998)

[23] Statistics Bureau of Japan, Population estimates, e-Stat portal, https://www.e-stat.go.jp/ (accessed 2026-XX-XX)

[24] Statistics Bureau of Japan, 2023 Housing and Land Survey / Employment Status Survey, commuting statistics (verify exact source), https://www.stat.go.jp/ (ac-

[51] J.-P. Bouchaud, L. F. Cugliandolo, J. Kurchan, and M. M´ezard, Out of equilibrium dynamics in spin-glasses and other glassy systems, in Spin Glasses and Random Fields, edited by A. p. 9

Where the paper says it

In structural glasses the question remains open because the thermodynamic side is accessible only through approximation. Here it is settled by a theorem: the Hessian of the free energy is positive definite, so no metastable-state structure exists to trap the dynamics, and the traps encountered by the swap dynamics are properties of the move set, not of the equilibrium measure.

P. Young (World Scientific, Singapore, 1998), pp. 161–223

[52] C. A. Angell, Formation of glasses from liquids and biopolymers, Science 267, 1924 (1995)

[53] G. Parisi, A sequence of approximated solutions to the S-K model for spin glasses, J. p. 0

Where the paper says it

Wilson derived this form from the maximum-entropy principle [5]: among all OD matrices meeting the three conditions, the observed one is the most probable, which in statistical mechanics is the state of lowest free energy [34, 39]. Its structure is as simple as that of a spin model, and it has served as an adjustable mathematical tool in applied transport research ever since [32, 38].

Phys. A 13, L115 (1980)

[54] T. R. Kirkpatrick, D. Thirumalai, and P. G. Wolynes, Scaling concepts for the dynamics of viscous liquids near an ideal glassy state, Phys. p. 0

Where the paper says it

In this framework, total commuting time dictates the system energy, and temperature determines how strongly human mobility depends on energy-saving habits, so that urban commuting can be analyzed through an evaluable free energy and simulated in almost the same manner as physical systems of matter, with the maximum-entropy doubly constrained gravity model as the stationary state. We first find that the residential dynamics is glassy: the system exhibits the signatures of kinetically constrained models, namely ageing, hysteresis, and freezing near yet short of equilibrium.

Rev. A 40, 1045 (1989)

[55] G. Biroli and J. P. Garrahan, Perspective: The glass transition, J. p. 7

Where the paper says it

Other parts use 1010 steps, or 671 sweeps. Defining an arrest boundary against a fixed observation window is standard for systems without a thermodynamic transition: the laboratory glass transition temperature is itself set by a conventional relaxation time rather than by a singularity [18, 52].

Chem. Phys. 138, 12A301 (2013)

[60] Ministry of Land, Infrastructure, Transport and Tourism (MLIT), Japan, National Land Numerical Information: 1 km Grid Square Statistics

[61] Z. Zhong, H. Takayasu, and M. p. 0

Where the paper says it

asymptotically approaching the gravitational equilibrium Yixuan Y. Zheng∗ Department of Systems and Control Engineering, Institute of Science Tokyo, Yokohama, Kanagawa 226-8502, Japan Hideki Takayasu† Department of Computer Science, Institute of Science Tokyo, Yokohama, Kanagawa 226-8502, Japan Misako Takayasu‡ Department of Systems and Control Engineering, Institute of Science Tokyo, Yokohama, Kanagawa 226-8502, Japan and Department of Computer Science, Institute of Science Tokyo, Yokohama, Kanagawa 226-8502, Japan (Dated: September 25, 2026) This research establishes a connection between macroscopic urban commuting flows and thermal equilibrium.

Takayasu, Renormalization of human mobility based on a revised electric circuit model and a new gravity relation, Phys. p. 0
Where the paper says it

com ‡ takayasu@comp. jp; Corresponding author networks [31, 63] to a revised electric circuit model in which human flow is treated as a current and renormalized across spatial resolutions [61].

Rev. Research 7, 013235 (2025)

[62] R. K¨olbl and D. p. 0

Where the paper says it

At the level of the OD matrix, commuters behave much as particles do: they move under fixed rules and under a bound on the energy they spend. K¨olbl and Helbing found that daily travel energy follows a canonical distribution with a mean near 615 kJ per person, obtained by maximizing entropy at fixed average energy [62].

Helbing, Energy laws in human travel behaviour, New J. p. 0
Where the paper says it

K¨olbl and Helbing found that daily travel energy follows a canonical distribution with a mean near 615 kJ per person, obtained by maximizing entropy at fixed average energy [62]. This correspondence has supported a physical line of work on human mobility, from scaling laws in spatial ∗ yixuan.

Phys. 5, 48 (2003)

[63] M. Barth´elemy, The statistical physics of cities, Nat. p. 1

Where the paper says it

The equilibrium assertion thus becomes testable: a city is at equilibrium when its empirical OD matrix coincides with the stationary state of the GHSM. Our results connect statistical physics and complex social systems in three ways.

Rev. Phys. 1, 406 (2019)

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Kinetic arrest detection in urban systems: The improved AI system can identify when urban commuting flows exhibit signatures of kinetically constrained models, namely ageing, hysteresis, and freezing near yet short of equilibrium by analyzing two-time correlation functions like C(τ, tw).

  2. Dynamical model validation: The system can distinguish between thermodynamic phase transitions and kinetic arrest by examining the behavior of the free energy landscape; specifically, it can confirm that the Hessian of the free energy is positive definite, so no metastable-state structure exists to trap the dynamics.

  3. Parameter estimation for urban equilibrium: The AI can calibrate city parameters using a distribution-based objective function, such as Dtld(γ, T), to find the unique pair (gamma star and T star) that best reproduces the observed trip-length distribution across all commuting times.

  4. Prediction of relaxation timescales: By analyzing the crossover between exponential and algebraic decay in order parameter deviations phi(t) - phieq, the AI can predict whether a system will relax ergodically or exhibit algebraic decay over more than two decades—no characteristic timescale—on the arrested side.

  5. Identification of short-range backbone: The system can pinpoint the mechanism causing arrest by analyzing the energy distribution of accepted swaps, confirming that the residual relaxation is carried by short-distance commuters trading nearly costdegenerate trips—the frozen short-range backbone of the arrest.

Abstract

The doubly constrained gravity model is one of the most widely used descriptions of urban commuting flows, yet whether residential relocation can actually drive a city to this equilibrium has not been tested. We address this question by connecting urban commuting to thermal equilibrium. Using mobility data from around 30 million records across six Japanese cities over one year, we introduce the Gravity-based Origin Swapping Model (GOSM), which applies Metropolis dynamics to urban commuting by treating individuals as interacting particles. Within this framework, total commuting time defines the system energy, and temperature controls how strongly mobility responds to cost savings. Consequently, urban commuting can be analyzed through an evaluable free energy and simulated similarly to physical systems of matter, with the maximum-entropy doubly constrained gravity model serving as the stationary state. We find that the residential dynamics exhibit glassy characteristics: calibration to observed data reveals signatures of kinetically constrained models, specifically aging, hysteresis, and freezing near equilibrium. These results suggest that static OD snapshots may misestimate the effect of structural intervention: substantial improvements may require changing the urban landscape itself, through transport investment that reshapes travel times or housing development that reshapes residential capacities, rather than waiting for spontaneous relocation to reach the theoretical optimum.

Related papers