🔬 applied physics

Optimal designs of heterogeneous grid transit networks

This paper proposes a general Continuum Approximation model and a sequential geometric programming solution method to optimize flexible, heterogeneous grid transit networks, demonstrating that such designs significantly reduce generalized costs compared to conventional rigid models, particularly in cities with strong spatial demand heterogeneity.

Original authors: Wenbo Fan, Haoyang Mao, Li Zhen, Weihua Gu

Published 2026-05-25

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Wenbo Fan, Haoyang Mao, Li Zhen, Weihua Gu

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

The Big Picture: Designing a City's Bus System Like a Living Organism

Imagine a city as a giant chessboard. In most cities, the streets form a perfect grid, like the lines on graph paper. Traditionally, when city planners design bus routes for these cities, they treat the bus lines like rigid rulers. They draw straight lines across the board, spacing them out evenly, regardless of whether the people living in one corner of the city are desperate for a bus or if the people in another corner are barely using them.

This paper proposes a new way to design these bus networks. Instead of using rigid rulers, the authors suggest treating bus lines like flexible rubber bands.

The Problem: The "One-Size-Fits-All" Trap

The authors point out that real cities are messy. Demand for buses isn't spread out evenly.

The Old Way: Imagine a city where 90% of the people live in the downtown center, but the bus lines are spaced out evenly from the center to the far suburbs. You end up with empty buses in the suburbs and overcrowded, slow buses in the center.
The Limitation: Previous computer models could only handle "rigid" designs. They could make lines slightly closer together in busy areas, but they couldn't easily change the shape of the lines or let them merge and split dynamically without breaking the math.

The Solution: The "Heterogeneous Network" (HetNet)

The authors created a new mathematical model called HetNet. Think of this model as a smart, adaptive traffic controller.

Flexible Rubber Bands: In this model, bus lines aren't stuck to a single street. They can wiggle. If a bus line needs to pick up more passengers in a specific neighborhood, it can make a small "detour" (a lateral movement) to get there, then merge back into the main flow.
Merging and Splitting: Imagine a river. In some places, the river is wide and fast (many buses, frequent service). In other places, it narrows down (fewer buses, less frequent service). The HetNet model allows bus lines to merge together like tributaries joining a main river to create a high-frequency "trunk" line, and then split apart to serve specific neighborhoods.
The "Flow" Concept: The authors use a technique called Continuum Approximation. Instead of counting every single bus and every single street corner (which is like trying to count every drop of water in a river), they look at the "flow" of the water. This allows them to calculate the best design for the whole city without getting bogged down in tiny details.

The Math: Solving the Puzzle

Designing these flexible lines is incredibly hard. It's like trying to solve a 3D puzzle where the pieces keep changing shape.

The Challenge: The math involves "absolute values" (which represent the distance a bus has to detour) and complex conservation laws (making sure no buses disappear or appear out of nowhere). Standard math tools couldn't solve this efficiently.
The Trick: The authors developed a method called Sequential Geometric Programming (SGP).
- Analogy: Imagine trying to find the lowest point in a bumpy, foggy valley. You can't see the whole valley at once. So, you take a step, look at the ground right under your feet, approximate it as a smooth slope, and take another step. You repeat this process, getting closer and closer to the bottom.
- The SGP method does exactly this. It breaks the complex, bumpy math problem into a series of simpler, smooth problems that a computer can solve quickly and accurately.

The Results: Saving Time and Money

The authors tested their "flexible rubber band" design against three other types of designs:

Homogeneous: The old-school, perfectly even grid.
Hierarchical: A system with main "trunk" lines and smaller "local" feeder lines (but with rigid rules).
Partial: A mix that allows some spacing changes but not line merging/splitting.

What they found:

The Winner: The flexible HetNet design consistently saved the most money and time for both the bus company and the passengers.
The "Sweet Spot": The more uneven the demand was (e.g., a city with a few very busy hubs and many quiet areas), the bigger the advantage of the HetNet design. In these "checkerboard" demand scenarios, the new design saved 7% to 10% in total costs compared to the best existing methods.
Where it shines most: The benefits were biggest in large cities, cities with high demand, and cities where people have lower incomes (meaning every minute of waiting time or dollar of operating cost counts more).

The Bottom Line

This paper proves that we don't have to force bus lines to be straight and rigid. By allowing bus routes to be flexible—merging, splitting, and detouring slightly to match where people actually live and work—we can create a public transit system that is significantly more efficient. It's like switching from a rigid, pre-fabricated bridge to a suspension bridge that can sway and adjust to the weight of the traffic, ensuring everyone gets where they need to go faster and cheaper.

Problem Statement
Transit Network Design (TND) aims to determine the optimal physical layout and service frequencies of routes to meet travel demand. While significant research exists, traditional mathematical programming models often rely on discrete inputs (e.g., origin-destination matrices, specific street networks), leading to NP-hard problems that require heuristic solutions with no guarantee of global optimality. Continuum Approximation (CA) methods offer a tractable alternative by treating network variables as continuous functions, yet most existing CA models are limited to simple network layouts (e.g., unbranched orthogonal grids) or assume uniform or axisymmetric demand. Consequently, they struggle to address the spatially heterogeneous demand patterns found in real-world cities. Previous attempts to model heterogeneous demand, such as hierarchical networks with local lines or ring-radial designs, often impose rigid geometric constraints (e.g., equally spaced trunk lines) or fail to strictly enforce vehicle flow conservation.

Methodology
This paper proposes a general Continuum Approximation (CA) model for optimizing transit network designs in grid cities under spatially heterogeneous demand. The core innovation is the "Heterogeneous Network" (HetNet) framework, which allows transit lines to route flexibly by making lateral movements, merging, and diverging anywhere within the city, rather than being confined to rigid geometric configurations.

Model Formulation: The network is described by two continuous decision functions: transit line density ( $\delta_i(\vec{x})$ ) and service headway ( $h_i(\vec{x})$ ) for each direction $i$ . The model explicitly accounts for vehicle detour distances and the associated redistribution of line flows using a cumulative-flow representation. This allows for the characterization of arbitrary local merging and diverging decisions while satisfying vehicle flow conservation laws.
Cost Function: The objective is to minimize the system's generalized cost, comprising agency costs (infrastructure length, stops, vehicle distance, and time) and patron costs (access/egress time, wait time, in-vehicle time, and transfer penalties). The formulation includes terms for vehicle detour distance and patron in-vehicle detour time, the latter of which depends on an endogenous parameter $\alpha$ representing the fraction of vehicle detour that translates to additional passenger travel distance.
Solution Algorithm: The resulting optimization problem is nonconvex due to absolute-value operators in the detour distance constraints and equality constraints involving posynomials. To solve this, the authors develop a Sequential Geometric Programming (SGP) framework. This approach transforms the original model into a sequence of standard Geometric Programming (GP) problems by iteratively approximating non-conforming functions with local monomials. To ensure numerical robustness, the algorithm incorporates multiplicative trust-region constraints.

Key Results
Numerical experiments were conducted in a square city with a dense grid network under six distinct demand patterns, ranging from low-heterogeneity (monocentric, commute) to high-heterogeneity (checkerboard) distributions.

Model Accuracy: When comparing the CA model's estimated generalized costs against those computed from discretized network designs, the approximation error remained below 3% across all demand patterns. The model accurately captures operational characteristics while maintaining analytical tractability.
Performance vs. Benchmarks: The proposed HetNet design was compared against three benchmarks: Hierarchical HetNet (H-HetNet), Partial HetNet (P-HetNet), and Homogeneous Network (HomNet).
- Under low-heterogeneity demand, HetNet showed marginal improvements over benchmarks.
- Under high-heterogeneity demand (checkerboard patterns), HetNet reduced generalized costs by 7.03% to 10.43% compared to the best-performing benchmark.
- The cost savings were driven primarily by reductions in in-vehicle travel time and transfer times, achieved by better aligning line density and service structure with local demand concentrations.
Parametric Sensitivity: The advantages of the heterogeneous design were found to be most pronounced in scenarios characterized by:
- High total patron demand (economies of scale).
- Low patron value of time (where operational efficiency becomes more critical relative to time costs).
- Large city sizes (where spatial dispersion of demand is greater).

Significance and Contributions
The paper claims to make three primary contributions:

Generalized CA Framework: It extends CA theory to a more general class of transit network designs that can flexibly accommodate spatially non-uniform demand. This is the first CA-based TND model to represent overlapping bus lines that merge and diverge arbitrarily, moving beyond the rigid geometric constraints of previous heterogeneous models.
Computational Tractability: The development of the SGP solution framework provides an efficient method for solving complex CA-based TND problems that include nonlinear terms and absolute-value structures, achieving global or near-global optimality.
Practical Insights: The study demonstrates that fully heterogeneous network structures are particularly advantageous when patron demand exhibits strong spatial heterogeneity. It suggests that such flexible designs yield the greatest benefits in high-demand, low-wage, and large-area cities, offering a more efficient allocation of transit resources compared to conventional, geometrically restricted grid networks.

The authors acknowledge limitations, including the simplified representation of user route choice behaviors and the focus on grid networks, noting that future work could explore irregular city shapes, hybrid networks, and multiple transit modes.