Defining the urban "local" with low dimensional manifolds of human mobility networks

Hezhishi Jiang, Liyan Xu, Tianshu Li, Jintong Tang, Zekun Chen, Yuxuan Wang, Haoran Liu, Hongmou Zhang, Huanfa Chen, Yu LiuarXiv 2025 · DOI · arXiv

Abstract

Recent research reveals that the complex networks of human mobility within cities reside on low-dimensional geometric manifolds, typically with a dimension of five or less. By mapping mobility data onto these manifolds, researchers can simplify complex urban problems—such as healthcare infrastructure planning and epidemic modeling—by revealing hidden symmetries that traditional geographic maps obscure.

Introduction: The Challenge of the "Local"

Urban science has historically struggled to reconcile "global" models, like the gravity model, with "local" heterogeneity. While global equations attempt to predict movement based on distance, they often fail because neighborhood-specific characteristics vary significantly. The core of this issue is the lack of a rigorous mathematical definition for "locality." Is it defined by a radius, a political boundary, or a social construct? This paper argues that locality is a topological feature of the human mobility network itself. Without a proper definition, the unique characteristics of city segments remain invisible to planners, leading to suboptimal resource allocation and inaccurate disease modeling.

Methodology: Topological Framework and TCIE Embedding

The researchers introduced a framework to detect and define "localities" within mobility networks through several key steps:

  1. Dichotomy Analysis: Testing whether a location's network can expand indefinitely or hits a boundary, proving that localities exist as distinct entities.
  2. Manifold Construction: Using percolation theory to demonstrate that network "open sets" path-connect to form a manifold in geographic space.
  3. Topologically Constrained Isometric Embedding (TCIE): This algorithm, a sophisticated evolution of Isomap, handles non-convex shapes (e.g., cities divided by rivers) by detecting boundaries and preventing "short-circuit" errors in the mapping process.
  4. Dimension Reduction: Mathematically proving that these mobility manifolds maintain a maximum dimension of five.
The transformation of geographic mobility data into a manifold representation, highlighting local intersection relationships.

Empirical Validation and Results

The study validated its framework across five diverse datasets: Shenzhen (China), an undisclosed city in Japan, Greater Mexico City, England, and Italy.

High Embedding Accuracy

The manifold embeddings demonstrated exceptional performance across all test sites. Pearson correlation coefficients between network weights and manifold distances ranged from 0.75 to 0.98, indicating that the low-dimensional representation accurately captures the underlying mobility dynamics.

Validation of the manifold embedding across five diverse global datasets, showing high correlation and consistent probability densities.

Simplified Optimization and Propagation

  • Facility Location: In tests for placing service centers, layouts that appeared chaotic on geographic maps formed perfectly uniform, grid-like patterns on the manifold.
  • Disease Modeling: During the June 2022 COVID-19 outbreak in Beijing, the virus spread irregularly in geographic space. However, on the mobility manifold, the spread followed a near-perfect isotropic concentric pattern, moving equally in all directions from the source.

Key Insights: The "Featureless Plain"

The most striking discovery is that projecting urban data onto these manifolds causes the "messiness" of physical geography—traffic, terrain, and zoning—to disappear. The manifold acts as a "featureless plain" where human movement becomes uniform. This transforms "locality" from a vague sociological concept into a rigorous geometric entity, suggesting a universal "grammar" for how humans occupy space regardless of specific city layouts.

Limitations and Future Work

The current proof relies on the discrete nature of human mobility data, which simplifies certain mathematical requirements into isometric embeddings. Future research will focus on Dynamic Manifolds to explore how these shapes change during different times of day, Real-time Applications for live traffic management, and Higher-Order Interactions to analyze complex motifs of human movement beyond simple origin-destination pairs.

Summary updated

BibTeX
@misc{jiang2025defining,
  author = {Hezhishi Jiang and Liyan Xu and Tianshu Li and Jintong Tang and Zekun Chen and Yuxuan Wang and Haoran Liu and Hongmou Zhang and Huanfa Chen and Yu Liu},
  title = {{Defining the urban "local" with low dimensional manifolds of human mobility networks}},
  year = {2025},
  eprint = {2510.17174},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2510.17174},
}
Figure