Dynamic Optimal Transport with Optimal Preferential Paths
Marcello Carioni, Juliane Krautz, Jan-F. Pietschmann · 2025
Read Dynamic Optimal Transport with Optimal Preferential Paths by Marcello Carioni, Juliane Krautz, Jan-F. Pietschmann (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-2504-03285.txt.
Introduction
Marcello Carioni, Juliane Krautz, and Jan-F. Pietschmann study optimal transport where mass can hop between a region and a preferred curve inside it, in a mathematics paper from twenty twenty-five. Classic dynamic transport moves mass through a bulk domain at some cost. Adding a fast lane, a smooth curve, may cut the cost, and the lane itself might be worth reshaping.
The authors prove optimal flows exist, first with a fixed curve, allowing exchange with the bulk at a price and nonlinear mobility rules. They then let the curve vary too, adding an energy term that keeps it smooth and prevents self-crossing. Compactness arguments secure optimizers in both settings. Simulations with a primal-dual method picture the resulting paths.
Catalog record
| CREATOR | Marcello Carioni, Juliane Krautz, Jan-F. Pietschmann |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 135 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-03285
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Dynamic Optimal Transport with Optimal Preferential Paths Marcello Carioni, Juliane Krautz, Jan-F. Pietschmann Abstract We study a dynamic optimal transport type problem on a domain that consists of two parts: a compact set Ω ⊂ ℝ d \Omega\subset{\mathbb{R}}^{d} (bulk) and a non-intersecting and sufficiently regular curve Γ ⊂ Ω \Gamma\subset\Omega . On each of them, a Benamou-Brenier type dynamic optimal transport problem is considered, yet with an additional mechanism that allows the exchange (at a cost) of mass between bulk and curve. In the respective actions, we also allow for non-linear mobilities. We first ensure the existence of minimizers by relying on the direct method of calculus of variations and we study the asymptotic properties of the minimizers under changes in the parameters regulating the dynamics in Ω \Omega and Γ \Gamma . Then, we study the case when the curve Γ \Gamma is also allowed to change, being the main interest in this paper. To this end, the Tangent-Point energy is added to the action functional in order to preserve the regularity properties of the curve and prevent self-intersections. Also in this case, by relying on suitable compactness estimates both for the time-dependent measures and the curve Γ \Gamma , the existence of optimizers is shown. We extend these analytical findings by numerical simulations based on a primal-dual approach that illustrate the behaviour of geodesics, for fixed and varying curves. Mathematics Subject Classification 49Q22, 35R01, 60B05, 49M41 1 Introduction Originally formulated by Monge in the 18th century and later refined by Kantorovich in the 20th century, Optimal Transport offers a mathematical framework for comparing and transforming probability distributions by minimizing the cost of moving mass from one distribution to another. Given a compact set \Omega\subset{\mathbb{R}}^{d} and two probability measures \mu\in P(\Omega) and \nu\in P(\Omega) and choosing the Euclidean distance as the cost function results in the 2 -Wasserstein distance defined as \displaystyle W_{2}^{2}(\mu,\nu)=\sqrt{\inf_{\pi\in\Pi(\mu,\nu)}\int|x-y|^{2}d\pi(x,y)} (1) where \Pi(\mu,\nu) is the set of transport plans, that are all probability measures in the product space \Omega\times\Omega having \mu and \nu as marginals. While the 2 -Wasserstein distance focuses on static settings, many real-world problems require transport models that evolve over time. This need has motivated the study of dynamic optimal transport introduced in the pioneering work by Benamou and Brenier [ 5 ] . In this work it has been shown that the 2 -Wasserstein distance admits a reformulation as the following dynamical fluid-mechanics problem \displaystyle W_{2}(\mu,\nu)=\inf_{(\rho_{t},J_{t})}\int_{0}^{1}\int_{\Omega}\left|\frac{dJ_{t}}{d\rho_{t}}\right|^{2}\,d\rho_{t}dt (2) where the pair (\rho_{t},J_{t}) satisfies the continuity equation \displaystyle\partial_{t}\rho_{t}+{\rm div}(J_{t})=0 (3) in the distributional sense, with \rho_{0}=\mu and \rho_{1}=\nu . Such dynamic formulation of the Wasserstein distance has proven useful in a great variety of contexts and it is a fundamental ingredient to develop a gradient flow theory in Wasserstein spaces [ 35 , 28 , 2 ] . Successful examples of applications are consequences of the usefulness of an optimal transport prior in understanding dynamics and includes among others applications to crowd modelling and traffic congestion [ 32 , 12 , 40 ] , dynamic inverse problems [ 38 , 9 , 8 , 17 ] , optimal control models and algorithms in machine learning [ 34 ] . Moreover, variants of the Benamou-Brenier approach have been explored to include more general action functionals [ 20 , 31 , 14 ] , dissipations [ 15 , 30 , 29 ] and non-local interactions [ 39 ] . In this paper, we contribute to the above mentioned theory by studying a dynamic optimal transport type problem where mass can be transported both in a bulk domain \Omega as well as on a curve \Gamma\subset\Omega , which has co-dimension of at least one. In our model the mass between the bulk and the curve can be exchanged, while the total mass in the system remains a conserved quantity. The cost of transport is measured by means of an action functional that includes the kinetic energy of the time-dependent distribution in the bulk and on the curve, respectively, as well as the additional cost due to exchange of mass between bulk and curve. From an intuitive standpoint our mathematical framework models the optimal evolution of a population (for instance metropolitan traffic) when there exists a path (such as a highway) where the transport is faster (or cheaper), but it requires a cost to access the path. This could model highways where, due to a high traffic concentration at the toll booths, the access to the highway requires a non-negligible time investment. In our analysis we first consider the preferential curve \Gamma to be fixed, where the goal is to find the optimal evolution of the population given initial and final configurations. This setting is strongly related to previous works that coupled transport in a domain with transport on its boundary, [ 33 ] as well as recent approaches to transport on metric graphs, [ 21 , 10 ] . In a second step, we extend the formalism by allowing the curve to vary, making its shape part of the optimization problem. Such problems can be seen in the context of optimal planning, where the preferential path is optimized according to initial and final distributions and under the prior assumption that the population is moving following an optimal transport dynamic. We now describe the problem using the mathematical formalism we employ in the paper. We consider \Omega a compact, connected subset of {\mathbb{R}}^{d} , \Gamma=\gamma([0,1]) the image of a regular, non self-intersecting curve of class C^{1}([0,1];\Omega) . The distribution of mass on the bulk and in \Gamma is modelled via a pair of time-dependent measures (\rho_{t},\mu_{t}) , indexed by t\in[0,1] . Here, \rho_{t}\in M(\Omega) represents the mass distributions in the bulk while \mu_{t}\in M(\Gamma) denotes the distribution on the preferential path \Gamma . Since the total mass will be conserved in our model, we assume a compatibility condition for initial and final distributions, meaning that \rho_{0}(\Omega)+\mu_{0}(\Gamma)=1 and \rho_{1}(\Omega)+\mu_{1}(\Gamma)=1 . In what follows, this will be denoted by (\rho_{0},\mu_{0})\in\mathcal{P}_{\rm adm}(\Gamma) and (\rho_{1},\mu_{1})\in\mathcal{P}_{\rm adm}(\Gamma) . To model the transportation of mass, we denote by J_{t} and V_{t} the momentums associated to \rho_{t} and \mu_{t} . Together, (\rho_{t},\mu_{t},J_{t},V_{t}) are assumed to satisfy two continuity equations which are coupled in order to allow exchange between bulk and curve while preserving the total mass: \displaystyle\frac{\partial\rho_{t}}{\partial t}+\nabla\cdot J_{t}=0,\quad \displaystyle\text{on}\quad\Omega,\qquad\frac{\partial\mu_{t}}{\partial t}+\nabla_{\Gamma}\cdot V_{t}=f_{t}\quad\text{on}\ \ \Gamma, (4) \displaystyle J_{t}\cdot n_{\Gamma}=f_{t}\quad\text{in }\Gamma. We refer to Section 2.1 for a more rigorous definition of ( 4 ) given in a weak sense and also for a generalization of ( 4 ) where an exchange mechanism that acts at the end points of \Gamma is added. We already note that this formulation allows for mass being present on \Gamma and in \Omega at the same spatial location. All curves that satisfy the above for given initial and final data (\rho_{0},\mu_{0})\in\mathcal{P}_{\rm adm}(\Gamma) and (\rho_{1},\mu_{1})\in\mathcal{P}_{\rm adm}(\Gamma) are denoted by {\rm CE}(\Gamma) . Next we introduce an action functional which measures the time accumulated cost of transport in bulk and curve, respectively, but also the cost of mass transfer. Inspired by the works [ 20 , 31 ] we employ mobilities m_{\Omega} and m_{\Gamma} to regulate the dynamics on \Omega and \Gamma . The action functional is then defined as \displaystyle{\mathcal{A}}_{\Gamma}(J_{t},V_{t},\rho_{t},\mu_{t},f_{t}) \displaystyle=\int_{\Omega}\frac{|J_{t}|^{2}}{m_{\Omega}(\rho_{t})}\,d\lambda_{\Omega}+\alpha_{1}\int_{\Gamma}\frac{|V_{t}|^{2}}{m_{\Gamma}(\mu_{t})}\,d\lambda_{\Gamma}+\alpha_{2}\int_{\Gamma}\frac{|f_{t}|^{2}}{m_{\Gamma}(\mu_{t})}d\lambda_{\Gamma}. (BB)