[Paper Review] Contribution of hidden modes to nonlinear epidemic dynamics in urban human proximity networks

Naoya Fujiwara, Abhijeet R. Sonawane+2 more|arXiv (Cornell University)|Dec 7, 2015

COVID-19 epidemiological studies48 references3 citations

TL;DR

This paper proposes a spectral decomposition framework to analyze epidemic dynamics on urban human proximity networks derived from real mobility data, revealing that hidden modes—eigenmodes with small eigenvalues but large dynamical contributions—dominate epidemic spread, challenging the conventional assumption that only high-eigenvalue modes matter. The method enables accurate prediction of final epidemic size by solving a transcendental equation for infection probability using selected dominant modes.

ABSTRACT

Recently developed techniques to acquire high-quality human mobility data allow large-scale simulations of the spread of infectious diseases with high spatial and temporal resolution.Analysis of such data has revealed the oversimplification of existing theoretical frameworks to infer the final epidemic size or influential nodes from the network topology. Here we propose a spectral decomposition-based framework for the quantitative analysis of epidemic processes on realistic networks of human proximity derived from urban mobility data. Common wisdom suggests that modes with larger eigenvalues contribute more to the epidemic dynamics. However, we show that hidden dominant structures, namely modes with smaller eigenvalues but a greater contribution to the epidemic dynamics, exist in the proximity network. This framework provides a basic understanding of the relationship between urban human motion and epidemic dynamics, and will contribute to strategic mitigation policy decisions.

Motivation & Objective

• To address the limitations of existing theoretical models in predicting epidemic spread from network topology in realistic urban mobility networks.
• To investigate why conventional epidemic models based on high-eigenvalue modes fail to capture actual epidemic dynamics in empirical human proximity networks.
• To develop a spectral decomposition-based analytical framework that identifies the true drivers of epidemic spread beyond top eigenmodes.
• To quantify the contribution of low-eigenvalue modes (hidden modes) to epidemic dynamics and validate their dominance through stochastic simulations.
• To provide a foundation for improved mitigation strategies by linking urban human mobility patterns to epidemic outcomes.

Proposed method

• Constructs time-varying proximity networks from real human mobility data (People Flow dataset) by connecting individuals within a 1,000 m spatial threshold.
• Applies spectral decomposition to the adjacency matrix of the proximity network to extract eigenmodes and eigenvalues, representing collective motion patterns.
• Derives a transcendental equation for the asymptotic recovery probability by projecting the SIR dynamics onto the eigenbasis of the network adjacency matrix.
• Identifies dominant modes not by eigenvalue size but by their contribution to the epidemic dynamics, using a perturbative expansion and mode selection based on dynamical impact.
• Solves the resulting equation iteratively to estimate the final epidemic size without simulating the full agent-based model.
• Validates the framework by comparing predictions against stochastic SIR simulations on the same network, using parameters β = 1.5×10⁻⁵ min⁻¹ and μ = 2.0×10⁻⁴ min⁻¹.

Experimental results

Research questions

• RQ1 Do modes with small eigenvalues contribute more to epidemic dynamics than high-eigenvalue modes in realistic urban proximity networks?
• RQ2 Can a spectral decomposition framework accurately predict the final epidemic size without relying on high-eigenvalue modes?
• RQ3 How does the inclusion of hidden dominant modes improve the accuracy of epidemic predictions compared to traditional heterogeneous mean-field theory?
• RQ4 What is the relationship between the network’s structural modes and the actual spread of infection in a time-resolved, real-world mobility context?
• RQ5 To what extent do low-eigenvalue modes govern the final size of epidemics in heterogeneous, urban contact networks?

Key findings

• Hidden dominant modes—eigenmodes with small eigenvalues but large dynamical contributions—play a crucial role in determining the final epidemic size in urban proximity networks.
• The proposed spectral framework accurately predicts the final epidemic size by solving a transcendental equation derived from the eigenmode decomposition, outperforming conventional approaches.
• The framework reveals that the largest eigenvalue mode does not necessarily dominate the epidemic dynamics, contradicting common assumptions in network epidemiology.
• The method identifies the top M modes based on their dynamical contribution, not eigenvalue magnitude, and achieves high prediction accuracy with a reduced set of modes.
• The analysis shows that the final size of the epidemic is more strongly influenced by collective motion patterns encoded in low-eigenvalue modes than previously assumed.
• The framework provides a quantitative link between urban human mobility patterns and epidemic outcomes, enabling better-informed public health policy decisions.