Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mobility Trajectories from Network-Driven Markov Dynamics

David A. Meyer, Asif Shakeel

TL;DR

The paper addresses generating mobility trajectories from time-dependent Markov dynamics defined on a spatial, hierarchical network to enable privacy-preserving analysis of time-elapsed flows. It prescribes a structured Markov process on a hub-corridor overlay informed by gravity-type kernels, center-periphery asymmetries, and schedule-driven directional biases, and realizes synthetic, memoryless trajectories that align with aggregated OD data. A key contribution is proving internal consistency between trajectory realizations and the prescribed dynamics via multi-step transition comparisons and establishing a unique periodic invariant distribution to initialize the system. The framework bridges aggregate flow descriptions and trajectory-level realizations without relying on individual-level behavior, offering a scalable, privacy-friendly tool for studying mobility structure in networked settings and enabling data-informed extensions.

Abstract

We present a generative model of human mobility in which trajectories arise as realizations of a prescribed, time-dependent Markov dynamics defined on a spatial interaction network. The model constructs a hierarchical routing structure with hubs, corridors, feeder paths, and metro links, and specifies transition matrices using gravity-type distance decay combined with externally imposed temporal schedules and directional biases. Population mass evolves as indistinguishable, memoryless movers performing a single transition per time step. When aggregated, the resulting trajectories reproduce structured origin-destination flows that reflect network geometry, temporal modulation, and connectivity constraints. By applying the Perron-Frobenius theorem to the daily evolution operator, we identify a unique periodic invariant population distribution that serves as a natural non-transient reference state. We verify consistency between trajectory-level realizations and multi-step Markov dynamics, showing that discrepancies are entirely attributable to finite-population sampling. The framework provides a network-centric, privacy-preserving approach to generating mobility trajectories and studying time-elapsed flow structure without invoking individual-level behavioral assumptions.

Mobility Trajectories from Network-Driven Markov Dynamics

TL;DR

The paper addresses generating mobility trajectories from time-dependent Markov dynamics defined on a spatial, hierarchical network to enable privacy-preserving analysis of time-elapsed flows. It prescribes a structured Markov process on a hub-corridor overlay informed by gravity-type kernels, center-periphery asymmetries, and schedule-driven directional biases, and realizes synthetic, memoryless trajectories that align with aggregated OD data. A key contribution is proving internal consistency between trajectory realizations and the prescribed dynamics via multi-step transition comparisons and establishing a unique periodic invariant distribution to initialize the system. The framework bridges aggregate flow descriptions and trajectory-level realizations without relying on individual-level behavior, offering a scalable, privacy-friendly tool for studying mobility structure in networked settings and enabling data-informed extensions.

Abstract

We present a generative model of human mobility in which trajectories arise as realizations of a prescribed, time-dependent Markov dynamics defined on a spatial interaction network. The model constructs a hierarchical routing structure with hubs, corridors, feeder paths, and metro links, and specifies transition matrices using gravity-type distance decay combined with externally imposed temporal schedules and directional biases. Population mass evolves as indistinguishable, memoryless movers performing a single transition per time step. When aggregated, the resulting trajectories reproduce structured origin-destination flows that reflect network geometry, temporal modulation, and connectivity constraints. By applying the Perron-Frobenius theorem to the daily evolution operator, we identify a unique periodic invariant population distribution that serves as a natural non-transient reference state. We verify consistency between trajectory-level realizations and multi-step Markov dynamics, showing that discrepancies are entirely attributable to finite-population sampling. The framework provides a network-centric, privacy-preserving approach to generating mobility trajectories and studying time-elapsed flow structure without invoking individual-level behavioral assumptions.
Paper Structure (42 sections, 32 equations, 6 figures, 2 tables)

This paper contains 42 sections, 32 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. The Trajectory Generation Model
  3. Spatial discretization and base graph
  4. Centers, hubs, and corridor structure
  5. Node potentials and directional orientation
  6. Population Evolution and Trajectory Realization
  7. Time-dependent Markov matrices
  8. Population evolution
  9. Gravity-type interaction kernel.
  10. Structural edge weights.
  11. Time-dependent bias factors.
  12. Construction of the transition matrices
  13. Center--periphery structure and flow objectives
  14. Module G: Directional bias schedules
  15. Example.
  16. ...and 27 more sections

Figures (6)

  • Figure 1: H3 spatial discretization and undirected base graph for the study region (Atlanta shown for geographic scale only). Hexagonal tiles indicate H3 cells at resolution 6, while nodes mark the corresponding cell centers. Adjacent cell centers are separated by approximately $6.44\,\mathrm{km}$ on average, and edges represent immediate H3 adjacency.
  • Figure 2: Corridor overlay network constructed on top of the H3 base grid. Nodes correspond to H3 cells, with the reference center shown as a black node and hub locations shown as larger markers. Highlighted edges define a hierarchical routing structure: green edges indicate hub--backbone connections linking each hub to the central anchor, while purple edges denote feeder paths connecting non-hub cells to hubs. Where uniquely defined, feeder direction is indicated by darker terminal segments. Metro links between hubs are drawn with increased line width.
  • Figure 3: Periodic fixed point population distribution $\mathbf{p}^\ast$ of the daily Markov matrix $Q = M_T \cdots M_1$. Node color intensity represents the invariant population mass associated with each H3 cell, with darker colors indicating higher population as shown in the colorbar. This distribution is uniquely determined (up to normalization) by the OD transition probabilities encoded in the matrices $\{M_t\}$, and reflects the intrinsic spatial organization implied by the network-driven dynamics rather than any imposed initial condition. The Atlanta region is shown for geographic scale only.
  • Figure 4: Single time-step OD flows during the 09:00--09:30 time-interval, restricted to edges oriented inward toward the reference center. Edge color intensity represents the magnitude of flow along each directed edge, with darker colors indicating higher flow as shown in the colorbar. Flows reflect the directional structure encoded in the transition matrix $M_t$ for this time step.
  • Figure 5: Single time-step OD flows during the same 09:00--09:30 time-interval as Figure \ref{['fig:flow-inward']}, restricted to edges oriented outward from the reference center. Edge color intensity represents the magnitude of flow along each directed edge, with darker colors indicating higher flow as shown in the colorbar.
  • ...and 1 more figures