🔢 mathematics

p-Wasserstein distances on networks and 3D to 1D convergence

This paper investigates transport distances on metric graphs representing gas networks by reviewing dynamic formulations with and without vertex mass storage, proving the convergence of static Wasserstein distances from 3D domains to 1D graphs via $c$ -cyclically monotone optimal transport plans, and validating these findings through numerical examples.

Original authors: Martin Burger, Ariane Fazeny, Gilles Mordant, Jan-Frederik Pietschmann

Published 2026-01-22

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Martin Burger, Ariane Fazeny, Gilles Mordant, Jan-Frederik Pietschmann

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: From Pipes to Graphs

Imagine you are managing a massive, complex gas network. In the real world, these pipes have thickness; they are 3D cylinders with a specific diameter. However, for engineers and mathematicians, it is often much easier to think of these pipes as infinitely thin lines (1D) connected at junctions. This is called a metric graph.

This paper asks a fundamental question: Is it safe to treat these thick, 3D pipes as thin, 1D lines?

Specifically, the authors study how "expensive" it is to move gas from one place to another (a concept called Wasserstein distance, which measures the minimum work needed to rearrange a pile of sand from one shape to another). They want to prove that if you take a 3D pipe network and shrink the pipes until they are essentially lines, the cost of moving the gas doesn't suddenly break or behave strangely. It converges smoothly to the cost calculated on the thin line model.

The Two Main Stories in the Paper

1. The "Traffic Jam" at the Junctions (Dynamic Transport)

The first part of the paper looks at how gas moves over time.

The Analogy: Imagine a busy highway system. You can model traffic by looking at the cars moving along the road (the edges). But what happens at the intersections (the nodes)?
The Two Approaches:
1. The "No Parking" Rule: In some models, gas cannot stop at a junction. Whatever gas flows in must immediately flow out. This is like a strict traffic light where cars can't wait; they must keep moving.
2. The "Parking Lot" Rule: In other models, gas can actually sit and wait at a junction (like a gas tank or a storage node). This allows for more complex behaviors, like gas flowing in, waiting, and then flowing out later.
The Connection to Physics: The authors show that these mathematical models of moving gas are actually the same as "gradient flows." Think of a ball rolling down a hill to find the lowest point. In this case, the "hill" is an energy function, and the "ball" is the gas distribution. The gas naturally flows in a way that minimizes energy, just like water flowing downhill. They prove that a specific real-world gas equation (called the ISO3 model) is mathematically identical to this "rolling down the hill" process on a network.

2. The "Thick-to-Thin" Transition (3D to 1D Convergence)

The second, and perhaps most important, part of the paper tackles the 3D vs. 1D question.

The Setup: Imagine a 3D network of pipes with a small but real thickness ( $\epsilon$ ). As $\epsilon$ gets smaller and smaller, the pipes look more and more like 1D lines.
The Problem: In a 3D world, if two pipes meet at a T-junction, a particle of gas can take a "shortcut" by cutting across the corner of the junction. In a 1D line model, the gas must travel all the way to the center of the junction and then turn.
The Discovery: The authors prove that even though the 3D gas can take these tiny shortcuts, as the pipes get thinner and thinner, the cost of moving the gas in the 3D world converges to the cost in the 1D world.
The "Branching" Mystery: The paper highlights a tricky issue: In a 1D network, a path might split (branch) at a junction. If you start at point A and go toward a junction, you might not know which way the gas will go until it gets there. This makes it hard to predict a single, unique path for every drop of gas.
- The Analogy: Imagine a river splitting into two streams. If you drop a leaf in the river, you can't say for sure which branch it will take until it reaches the fork. In the 3D model, the leaf might drift slightly to the left or right before the fork, making its path unique. In the 1D model, the path is ambiguous.
- The Result: Despite this ambiguity, the authors prove that the total cost of moving all the gas remains consistent. The "messiness" of the 3D shortcuts disappears as the pipes shrink, and the math holds up.

Key Takeaways for the General Reader

Mathematical Validation: The paper provides a rigorous mathematical proof that simplifying complex 3D gas networks into 1D line graphs is a valid approach. You don't lose the "physics" of the transport cost when you make the pipes infinitely thin.
Optimal Paths are Tricky: In networks with junctions, the "best" way to move things isn't always a single, straight line. Sometimes, the optimal strategy involves splitting and merging flows in complex ways (cyclical monotonicity).
Real-World Relevance: This work helps justify why engineers can use simple, fast computer models (1D graphs) to simulate complex, real-world gas networks (3D pipes) without worrying that the results will be fundamentally wrong.

What the Paper Does Not Do

It does not propose a new way to build gas pipes.
It does not offer a new software tool for gas companies (though it supports the math behind them).
It does not discuss climate change or energy policy.
It focuses strictly on the mathematics of distance and movement on these networks, proving that the "thin line" approximation is mathematically sound as the pipes get smaller.

In short, the paper is a "quality control" check for mathematicians and engineers, confirming that their simplified maps of gas networks accurately reflect the physics of the real, thick pipes they represent.

Technical Summary: $p$ -Wasserstein Distances on Networks and 3D to 1D Convergence

1. Problem Statement

The paper addresses the mathematical modeling of gas transport networks using optimal transport theory. Specifically, it investigates two interconnected problems:

Dynamic Transport on Metric Graphs: Extending the dynamic formulation of the Wasserstein distance (Benamou-Brenier) to metric graphs representing gas networks. This involves handling mass conservation on edges and defining appropriate coupling conditions at vertices (pipe junctions). The authors consider two distinct physical scenarios:
- Kirchhoff Conditions: Mass is conserved globally at nodes with no storage (classical pipe junctions).
- Node Mass Storage: Mass can be stored at nodes, modeled via additional continuity equations and generalized Kirchhoff laws.
Dimensional Reduction (3D to 1D): The authors seek to rigorously justify the common engineering practice of modeling gas pipes as one-dimensional entities (metric graphs) rather than three-dimensional cylinders. They investigate the asymptotic behavior of the static Wasserstein distance on a 3D network domain as the pipe diameter ( $\varepsilon$ ) converges to zero, proving that the 3D transport cost converges to the 1D metric graph transport cost.

2. Methodology

Dynamic Formulation on Metric Graphs

The authors generalize the continuity equation to metric graphs. Instead of a velocity vector field, they utilize momentum (mass flux) to ensure linearity in the continuity equation and convexity in the optimization problem.

Mass Storage Case: They define a continuity equation where the time derivative of the node density ( $\gamma_v$ ) equals the net flux entering the node from adjacent edges. This leads to a dynamic Wasserstein distance defined via an action functional involving a perspective function $h(a,b) = |a|^p/b^{p-1}$ , penalizing negative densities and non-zero fluxes at zero density.
Kirchhoff Case: They utilize a quotient space approach where edges are glued at nodes. The continuity equation is formulated using test functions continuous across the entire graph, implicitly enforcing Kirchhoff's law (sum of fluxes equals zero).
Gradient Flows: The paper reviews how specific gas flow models arise as gradient flows in these Wasserstein spaces. Notably, the ISO3 model for gas flow is identified as a $W_{K,3}$ -gradient flow of a specific energy functional, while drift-diffusion equations correspond to $W_{K,2}$ -gradient flows.

3D to 1D Convergence Analysis

To prove the convergence of the Wasserstein distance from a 3D domain ( $N_\varepsilon$ ) to a 1D metric graph ( $N_0$ ), the authors employ the following steps:

Cost Function Definition: They define a cost function $c_\varepsilon(x,y)$ on the 3D domain as the infimum of the action integral over curves constrained to stay within the 3D network. This cost incorporates an indicator function $\iota_{N_\varepsilon}$ to penalize paths leaving the network.
Optimal Transport Plans: They analyze the optimal transport plans ( $\pi_\varepsilon$ ) for the 3D problem. A key challenge identified is the potential non-uniqueness of optimal maps due to branching geodesics at network junctions (where multiple paths of equal length exist).
Cyclical Monotonicity: The proof relies heavily on the characterization of optimal transport plans as $c$ -cyclically monotone sets. By establishing that the cost functions $c_\varepsilon$ converge to the 1D cost $c_0$ and utilizing the stability of $c$ -cyclical monotonicity, they demonstrate the convergence of the optimal transport costs and plans.

3. Key Contributions and Results

Theoretical Results on Metric Graphs

Well-posedness: The paper establishes the existence of absolutely continuous curves in the $p$ -Wasserstein space on metric graphs, linking them to weak solutions of the continuity equation with finite $p$ -energy (Theorem 1).
Gradient Flow Identification: It confirms that the ISO3 gas flow model is a gradient flow with respect to the $W_{K,3}$ distance, providing a variational interpretation of the physical model.

Convergence Results (3D to 1D)

Convergence of Costs: The primary result (Proposition 1) states that for sequences of absolutely continuous probability measures converging weakly to atomless measures on the 1D graph, the optimal transport cost $OT(\mu_\varepsilon, \nu_\varepsilon, c_\varepsilon)$ converges to $OT(\mu_0, \nu_0, c_0)$ as the pipe diameter $\varepsilon \to 0$ .
Convergence of Plans: Theorem 3 proves that if the optimal transport plan for the 1D limit is unique, the sequence of 3D optimal plans converges weakly to this unique limit. If the limit plan is not unique, every convergent subsequence of 3D plans converges to an optimal 1D plan.
Stability with Topology: Theorem 4 provides bounds on how the optimal transport cost changes when the network topology is modified (e.g., removing or adding edges), showing the cost difference is bounded by the maximum difference in the cost functions weighted by the transported mass.

Numerical Illustrations

The paper includes numerical examples using a space-discretized version of the problem. These illustrate that geodesics in the network can branch and collapse at junctions, visually confirming the theoretical difficulty in establishing unique transport maps in the presence of branching structures.

4. Significance and Claims

The paper claims to provide a rigorous mathematical foundation for the use of metric graphs in modeling gas networks.

Justification of 1D Models: By proving the convergence of the Wasserstein distance from 3D to 1D, the authors justify the "averaging argument" used in engineering to treat pipes as 1D entities, specifically within the context of optimal transport and gradient flows.
Unified Framework: The work unifies the dynamic formulation of Wasserstein distances with gas network physics, explicitly linking the ISO3 model to gradient flows in the $W_{K,3}$ space.
Handling Non-Uniqueness: The authors explicitly acknowledge and address the issue of non-uniqueness in transport plans caused by branching geodesics, a phenomenon often overlooked in standard Euclidean settings. They suggest that space-time lifting (Section 3.5) may be required to derive unique transport maps in these "well-behaved" but branching examples.

The authors remain modest regarding the dynamic formulation's convergence; they state that the proven convergence of the static distance serves as a "starting point for further investigations" into the convergence of the dynamic formulation and Wasserstein gradient flows.