Fundamental diagram constrained dynamic optimal transport via proximal splitting methods

Anqi Dong, Karl H. Johansson, Johan Karlsson · 2025

Read Fundamental diagram constrained dynamic optimal transport via proximal splitting methods by Anqi Dong, Karl H. Johansson, Johan Karlsson (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-20717.txt.

Introduction

Anqi Dong, Karl Johansson, and Johan Karlsson build, in a 2025 paper, a traffic flow model on the math of optimal transport. That math moves goods or crowds at least cost and has solved many large flow problems. Yet plain versions miss a rule traffic engineers rely on: the link between density and flow, which caps how much traffic a road can carry.

Their model adds that rule as a bound on flow that tightens as density grows, so crowding and saturation appear in the math. They solve it with splitting methods that need only simple steps and can handle barriers or changing limits. Tests show flow that spreads with crowding in mind, reroutes sensibly, and settles as the method runs.

Catalog record

CREATORAnqi Dong, Karl H. Johansson, Johan Karlsson
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE77 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-20717

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Fundamental diagram constrained dynamic optimal transport via proximal splitting methods Thanks: This research has been supported in part by the Swedish Research Council (VR) under grant 2020-03454, KTH Digital Futures, Swedish Research Council Distinguished Professor Grant 2017-01078, and Knut and Alice Wallenberg Foundation Wallenberg Scholar Grant. Anqi Dong, Karl H. Johansson, Johan Karlsson Abstract Optimal transport has recently been brought forward as a tool for modeling and efficiently solving a variety of flow problems, such as origin-destination problems and multi-commodity flow problems. Although the framework has shown to be effective for many large scale flow problems, the formulations typically lacks dynamic properties used in common traffic models, such as the Lighthill–Whitham–Richards model. In this work, we propose an optimal transport framework which includes dynamic constraints specified be the fundamental diagram for modeling macroscopic traffic flow. The problem is cast as a convex variant of dynamic optimal transport for arbitrary dimensionality, with additional nonlinear temporal-spatial inequality constraints of momentum, modeled after the fundamental diagram from traffic theory. This constraint imposes a density-dependent upper bound on the admissible flux, capturing flow saturation and congestion effects, and thus leaves space for kinetic optimization. The formulation follows the Benamou–Brenier transportation rationale, whereby average kinetic energy over density and momentum fields is optimized subject to the mass conservation law. We develop first-order splitting algorithms, namely, Douglas–Rachford and Chambolle–Pock algorithms that exploit the separable structure of the constraint set and require only simple proximal operations, and may accommodate additional (time-varying) spatial restrictions or obstacles. Numerical experiments illustrate the impact of the constraint on transport behavior, including congestion-aware spreading, rerouting, and convergence. The framework establishes a connection between optimal transport and macroscopic traffic flow theory and provides a scalable, variational tool for modeling congestion-constricted (or saturation-aware) Wasserstein gradient flow. Keywords: Optimal transport, fundamental diagram, traffic flow, first-order method, flux-density constraint 1 Introduction Ninety years ago, Greenshields’ 1935 empirical study of vehicular flow [ 30 ] introduced what is now known as the fundamental diagram, which has since been refined through decades of experimental and theoretical developments [ 55 , 54 , 41 ] . These studies consistently reveal the principle quantified by the fundamental diagram – “as traffic density \rho increases, the velocity v of individual vehicles decreases from the free-speed limit v_{0} , approaching zero as \rho nears the jam density \hat{\rho} ” . Consequently, the flow-rate (also, momentum or flux) m=\rho v initially increases, reaches a maximum at a critical density, and then decreases as congestion appears at \hat{\rho} . This relation, conceptualized by m=\mathcal{Q}(\rho) , captures the fundamental trade-off between density and flux. Similar phenomena have also been observed in pedestrian dynamics [ 32 ] , where increasing crowding gradually inhibits motion. Twenty years later, Lighthill and Whitham [ 39 ] , and a few months later, Richards’ independent work [ 48 ] , introduced what is now known as the Lighthill–Whitham–Richards (LWR) model [ 18 , 1 , 60 ] . This model treats traffic as a compressible one-dimensional continuum governed by the continuity equation \partial_{t}\rho+\nabla_{x}\cdot m=0 , and the fundamental diagram equilibrium m=\mathcal{Q}(\rho) yields a scalar conservation law with concave flux \partial_{t}\rho+\nabla_{x}\cdot(\mathcal{Q}(\rho))=0. Under appropriate regularity conditions [ 4 ] , this equation admits a unique solution, which can be approximated numerically with models such as the Cell Transmission Model (CTM) [ 20 , 12 ] . The LWR model is a structurally simple first-order model and has been widely used in traffic flow modeling, however, it also has several limitations. First, the specification m=\mathcal{Q}(\rho) assumes that vehicles adjust their speed instantaneously to local density, leaving no room for transient or non-equilibrium behavior. As a result, the model struggles with sustained stop-and-go waves or capacity drops, both of which depend on dynamic interactions between flow and density. A related relaxation arises in second-order models, such as the Aw–Rascle–Zhang (ARZ) system [ 3 , 61 ] , where momentum evolves dynamically. Moreover, because momentum is uniquely determined by density, control of the flow are restricted to boundary inflows, outflows, or structural perturbations of the flux function. There is no mechanism to adjust the interior momentum field independently. Finally, it can be noted that although the one-dimensional setting is generally well-behaved, basic properties such as uniqueness of solutions may not be preserved for extensions to multi-lane settings or two-dimensional domains [ 1 ] . Contemporaneous with LWR model, Kantorovich’s linear programming relaxation of Monge’s classical mass relocation problem [ 35 , 59 , 47 ] seeks an optimal transportation map between an initial and a target distribution. The Monge–Kantorovich framework of optimal mass transport (OMT) is further developed to a dynamical version by Benamou and Brenier [ 8 , 53 ] , which interprets transport cost as the kinetic energy of a continuum flow, linking OMT to fluid mechanics and geodesic paths in the Wasserstein space. These developments have led to widespread applications in physics and imaging [ 40 , 14 , 36 ] , as well as in machine learning [ 46 , 31 ] and large-scale optimization [ 49 , 57 ] . Meanwhile, physical or structural constraints along transport map or trajectory has been considered. For instance, hard bounds are considered on flux field [ 37 , 38 ] , congestion-aware penalization of high-density flows [ 13 , 53 ] , stochastic and mean-field interactions [ 24 ] , and porous-media-type behavior through Wasserstein gradient flows [ 43 ] . More recent works have imposed pointwise flow constraints to model local bottlenecks, toll regions, or throughput limits [ 22 , 56 ] . However, most OMT formulations in literature treat flow-rate and density as decoupled variables, and do not impose nonlinear physical relations that govern their interaction in real systems. Although some recent works have adapted OMT to traffic modeling, they typically reinterpret transport cost, velocity, or density within the existing metric structure, rather than enforcing the flow–density coupling as a structural constraint. Contributions. We introduce a convex optimization framework for modeling traffic evolution based on a dynamic optimal transport problem with a flow constraint motivated by the fundamental diagram. Specifically, we consider a model where the flow-rate can be controlled as long as it does not exceed the limit specified by the fundamental diagram. This leads to a relaxed fundamental diagram constraint m\leq\mathcal{Q}(\rho) that allows the flow-rate to remain strictly below capacity in free-flow regimes, consistent with empirical observations. Unlike the LWR model, which enforces a temporal-spatial equilibrium relation m=\mathcal{Q}(\rho) , our formulation treats flow-rate (also, momentum or flux) as an independent variable subject to a nonlinear upper bound, which permits non-equilibrium transport behavior and allows for kinetic optimization. This formulation remains convex, scalable, and extensible, offering both a principled modeling framework and a direct optimization tool for congestion-aware transport. The main contributions of this paper can be summarized as follows: i) We formulate a model where the constraints from the fundamental diagram and the mass conservation law are decoupled, allowing both to be imposed independently . The uniqueness of the optimal flow is established in arbitrary dimensions. The fundamental inequality diagram enforces capacity limits in congested regions while permitting undersaturated flow elsewhere. ii) The formulation uses time-dependent density and momentum (or velocity) fields as optimization variables, allowing direct integration of

Read the complete text