💻 computer science

Multi-Scale Equilibrium under Variable Indicator Dimensionality: Faithful Reduction of Dynamic Attractors in Urban Mobility Systems

This paper establishes a formal framework for analyzing disrupted urban mobility systems through a dynamic multi-layer equilibrium attractor, deriving conditions under which lower-dimensional indicator projections faithfully preserve the system's structural properties and decision-making power while providing a principled basis for city authorities to determine essential monitoring indicators.

Original authors: Ali Ghoroghi, Yacine Rezgui, Afrouz Ghaemi, Cristina De Nardi, Andrei Hodorog

Published 2026-07-24

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Ali Ghoroghi, Yacine Rezgui, Afrouz Ghaemi, Cristina De Nardi, Andrei Hodorog

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to understand how a giant, complex machine works. In the world of science, this machine is a city's transportation system, and the "gears" keeping it running are things like traffic flow, bus schedules, and energy usage. Scientists use math to build a "map" of how this machine behaves, hoping to predict what happens when things go wrong, like a massive storm or a power outage. This field is called dynamical systems, and it's all about finding the "steady state"—the point where everything settles down and runs smoothly again after a shake-up.

Usually, to get a perfect map, you need to measure every single gear. But in the real world, cities are messy. Some cities have fancy sensors everywhere; others only have a few basic counters. Sometimes, during a disaster, data stops coming in. This creates a problem: if you try to draw a map using only half the gears, will your map still tell the truth? Will the city actually behave the way your simplified map says it will? This is the big question: Can we throw away some data and still trust our predictions?

This paper tackles that exact puzzle. The authors, a team of mathematicians and engineers, built a super-detailed mathematical model of a city's mobility system. They didn't just look at traffic; they looked at how traffic, energy, and even human behavior all talk to each other. They wanted to know: if a city can only measure a few things (like traffic speed and stress levels) instead of everything, is it safe to use that limited data to decide if the city is "antifragile"?

"Antifragile" is a cool word that means "getting stronger when things get tough." It's the opposite of fragile, which breaks easily. The authors created a special test to see if a city is antifragile: they check if the city bounces back better than it was before the disaster. But here's the catch: their math shows that if you don't measure enough things, your test might give you a misleading picture.

The paper proves four main things through computer simulations (which are like virtual experiments):

The "Hidden Gear" Trap: If you remove too many measurements, your simplified map might look stable, but it could be hiding a slow, dangerous problem. Imagine a car that looks fine from the outside, but the engine is actually overheating because you aren't measuring the temperature. The authors found that if you drop too many data points, your map might say the city is "fixed" when it's actually still struggling.
The "Teamwork" Breakdown: The city's systems (traffic, energy, etc.) are like a sports team. If you only watch the forwards and ignore the defense, you might think the team is winning, but they could be losing the game. The paper shows that if you cut out too many data points, the "teamwork" between the different layers of the city breaks down in the math, and the city might lose its unique, stable state.
The "Blind Spot" in the Test: The test for "antifragility" needs specific data to work. The authors found that if you don't have the right data (like information about fairness or how people adapt), your test loses its power. It's like trying to hear a whisper in a noisy room; if you don't have a good microphone (enough data), you might miss the signal entirely.
The "Optimism Bias": This is the most important finding. The paper proves that if you use a simplified map with fewer data points, you will generically think the city is recovering faster than it really is, but you will never think it's taking longer. In the specific mathematical regime where the reduction works correctly, the bias is one-sided: reduced monitoring can only understate the recovery time, never overstate it. It's like looking at a clock that only has an hour hand; you might think it's 1:00 when it's actually 1:55, but you won't think it's 12:30. The paper warns city leaders: "If you see a quick recovery in your limited data, be careful! The real recovery might be much slower."

In short, the paper doesn't say you need to measure everything to understand a city. Instead, it gives city leaders a mathematical rulebook. It tells them exactly which data points are the "must-haves" to keep their maps honest. If they have those specific points, they can trust their simplified models. If they don't, they risk making decisions based on a map that is too optimistic, potentially leaving the city vulnerable when it thinks it's safe. The authors ran these ideas through a virtual test with three different "fake cities" to show that their rules work, proving that you can't just guess which data to keep—you have to do the math to find the right balance.

Technical Summary: Multi-Scale Equilibrium under Variable Indicator Dimensionality

1. Problem Statement

Urban mobility systems are increasingly modeled as high-dimensional dynamic attractors defined over a vector of normalized indicators (e.g., throughput, redundancy, equity, stress). However, data availability varies sharply across cities and disruption contexts; some municipalities lack specific data streams (e.g., behavioral compliance or energy-system support) required to compute the full indicator vector.

The core mathematical problem addressed is the well-posedness of a high-dimensional dynamic attractor when only a subset, projection, or aggregation of its indicators is observable. Existing literature offers mature results on local dimension reduction (center manifold theory, singular perturbation), multi-layer network aggregation, and composite indicator sensitivity. However, none of these frameworks directly address the simultaneous preservation of three distinct structures required for urban mobility analysis:

Asymptotic Stability: The projected system must preserve the stability class of the performance layer.
Fixed-Point Interpretation: The reduced performance and strategic layers must retain the same hierarchical fixed-point interpretation (coupling traffic assignment, modal split, and capacity learning).
Statistical Decision Power: The reduced indicator set must retain sufficient information to distinguish between recovery, resilience, and antifragility via a statistical decision rule.

Standard reduction methods often focus on trajectory reconstruction error or spectral preservation but fail to guarantee that an external decision rule tied to the indicator support remains valid. This paper formalizes the conditions under which a lower-dimensional indicator projection is "faithful" to the full equilibrium structure.

2. Methodology and Framework

2.1 The Multi-Layer Equilibrium Attractor

The authors define a dynamic attractor for disrupted urban mobility comprising two layers:

Fast Performance Layer: A state variable $E(t)$ relaxes toward a target function $\bar{E}(x)$ defined over a 12-dimensional indicator vector $x$ (including throughput, redundancy, adaptation velocity, entropy, stress, equity, etc.). The dynamics follow a relax-to-target ordinary differential equation: $\dot{E} = -\kappa(E - \bar{E}(x))$ .
Slow Strategic Layer: A hierarchical fixed point combining user-equilibrium traffic assignment (Beckmann program), modal split (multinomial logit), and a capacity-learning update. This layer supplies the slow-timescale baselines that the performance layer tracks.
Antifragility Decision Rule: A statistical test comparing a post-event performance ratio to a baseline-dependent threshold. Antifragility is declared if the lower confidence limit of the ratio exceeds a requirement threshold and a noise threshold.

2.2 Theoretical Framework for Reducibility

The paper treats variable indicator dimensionality as a quotient problem. Given a full state $x \in \mathbb{R}^n$ and a linear reduction $\Pi: \mathbb{R}^n \to \mathbb{R}^m$ ( $m < n$ ), the authors define faithfulness relative to the attractor rather than just trajectory reconstruction.

The framework establishes four theoretical pillars:

Projectability and Stability: A projection is "performance-faithful" if the vector field is projectable (the reduced dynamics do not depend on unobserved coordinates) and the reduced Jacobian is Hurwitz. For non-exact projectability, a "projectability defect" ( $\varepsilon_\Pi$ ) is defined, and a bound on the projection error is derived based on the ratio of the defect to the reduced contraction rate ( $\alpha$ ).
Coupled Fixed-Point Preservation: The paper proves that the joint fixed point of the performance-strategic map is preserved under reduction if the reduced coupling matrix remains a contraction ( $\rho(K_\Pi) < 1$ ). Crossing this boundary leads to a loss of uniqueness and hyperbolic growth in solver iteration counts.
Measurement Model and Fisher Information: A linear-Gaussian measurement model connects observable urban Key Performance Indicators (KPIs) to the latent state variables. The authors derive the retained Fisher information ( $I_m$ ) and residual variance ( $\sigma^2_{res}$ ) in closed form. They prove that detection power for the antifragility decision rule is monotone with respect to nested indicator supports.
Restoration-Time Bias: The paper proves that restoration time estimated from a reduced indicator set never exceeds, and generically understates, the restoration time of the full system. This creates a one-sided optimism bias where reduced monitoring hides slow recovery modes.

3. Key Contributions

The paper makes five specific theoretical contributions:

Formal Attractor Definition: A self-contained multi-layer equilibrium attractor for disrupted urban mobility, integrating a relax-to-target law, a hierarchical strategic fixed point, and a statistically testable antifragility rule.
Stability Preservation Theorem: Necessary and sufficient conditions for preserving local asymptotic stability under linear indicator projection, including a theorem for "practical faithfulness" that bounds trajectory error when projectability is approximate.
Fixed-Point Preservation Theorem: A proof of coupled fixed-point preservation for two-layer maps, identifying the contraction boundary ( $\rho(K_\Pi) = 1$ ) where uniqueness is lost.
Decision-Rule Degradation Analysis: A measurement model deriving the retained Fisher information of any indicator support. It proves that detection power is monotone in nested supports and provides a closed-form condition for the minimum admissible indicator support required to achieve a target statistical power.
Restoration-Time Bias Result: A proof that reduced monitoring introduces a one-sided bias, causing cities to report restoration times that are optimistically biased (shorter than reality) because slow modes in omitted coordinates are invisible.

4. Numerical Results and Simulation

The theoretical results are illustrated via a simulation study on three stylised pilot-city configurations representing different data availabilities:

Bratislava (Full): 12 indicators.
Larissa (Standard): 6 indicators (removing adaptation, behavioral compliance, demand management, and infrastructure support).
Thessaloniki (Minimal): 3 indicators (throughput, stress, disturbance).

Key Findings from Experiments:

E1 (Practical Faithfulness): Observed projection errors remained below the analytic bound ( $\varepsilon_\Pi/\alpha$ ). The minimal tier crossed the tolerance threshold at lower coupling strengths than the standard tier, confirming that omitted variables (like redundancy) can destabilize reduced models if they feed back into retained variables.
E2 (Restoration Time): In projectable linear regimes, the minimal and standard tiers consistently reported restoration times shorter than the full system, validating the one-sided bias. The gap grew monotonically with the fraction of the state vector lying in the kernel of the projection.
E3 (Fixed-Point Boundary): As the spectral radius of the reduced coupling matrix approached 1, the iteration count for the strategic solver grew hyperbolically, and the dispersion of limit points increased, signaling the onset of non-uniqueness.
E4 (Decision Power): The minimal tier (3 indicators) failed to meet the target statistical power (0.80) for detecting antifragility, while the standard tier (6 indicators) succeeded. This demonstrates that information loss falls disproportionately on the classification of post-disruption behavior rather than stability screening.
E5 (Minimum Admissible Support): A greedy selection algorithm based on innovation gain identified that achieving the target power required at least one channel feeding a coordinate outside the minimal tier (e.g., redundancy or equity), confirming that routine operator data alone is insufficient for antifragility classification.

5. Significance and Claims

The paper claims to provide a principled basis for city authorities to decide which indicators must be maintained. It argues that minimum data requirements should be defined mathematically based on the specific equilibrium property being asserted (stability, fixed-point, or decision power) rather than administratively.

For Stability Screening: A reduced model is acceptable if the projectability defect is small relative to the contraction rate.
For Joint Equilibrium: The reduced coupling matrix must remain within the contraction boundary.
For Antifragility Classification: The retained channels must satisfy a specific admissibility condition derived from Fisher information; otherwise, the decision rule is statistically inadmissible.

The authors explicitly state that the paper does not claim empirical validation, full indicator coverage, or contributions from generative agent modeling. Instead, it establishes the formal conditions under which partial indicator coverage is admissible for equilibrium analysis. The framework distinguishes between reductions that are merely numerically convenient and those that are mathematically faithful to the underlying dynamical and decision-theoretic structures.