Mahmoud Abdelgalil,Tryphon T. Georgiou
Abstract:The theory of Monge-Kantorovich Optimal Mass Transport (OMT) has in recent years spurred a fast developing phase of research in stochastic control, control of ensemble systems, thermodynamics, data science, and several other fields in engineering and science. Specifically, OMT endowed the space of probability distributions with a rich Riemannian-like geometry, the Wasserstein geometry and the Wasserstein $\mathcal W_2$-metric. This geometry proved fruitful in quantifying and regulating the uncertainty of deterministic and stochastic systems, and in dealing with problems related to the transport of ensembles in continuous and discrete spaces.
We herein introduce a new type of transportation problems. The salient feature of these problems is that particles/agents in the ensemble, that are to be transported, are labeled and their relative position along their journey is of interest. Of particular importance in our program are closed orbits where particles return to their original place after being transported along closed paths. Thereby, control laws are sought so that the distribution of the ensemble traverses a closed orbit in the Wasserstein manifold without mixing. This feature is in contrast with the classical theory of optimal transport where the primary object of study is the path of probability densities, without any concern about mixing of the flow, which is expected and allowed when traversing curves in the Wasserstein space.
In the theory that we present, we explore a hitherto unstudied sub-Riemannian structure of Monge-Kantorovich transport where the relative position of particles along their journey is modeled by the holonomy of the transportation schedule. From this vantage point, we discuss several other problems of independent interest.
What problem does this paper attempt to address?
The paper attempts to address the issue of mixing phenomena in Optimal Mass Transport (OMT). Specifically, the authors focus on whether particles return to their initial positions when probability distributions move along closed orbits in the Wasserstein manifold. Traditional optimal transport theory mainly studies the changes in the path of probability densities, without special attention to the relative positions of particles and the mixing phenomena along the path.
### Main Issues of the Paper:
1. **Relative Position of Particles**: When a collection of particles is transported, do the labels and relative positions of the particles remain unchanged?
2. **Mixing Phenomena on Closed Orbits**: When a collection of particles is transported along a closed path and returns to the origin, do the particles return to their initial positions? If not, how can this mixing phenomenon be described and controlled?
### Background and Motivation:
- **Optimal Mass Transport (OMT)**: OMT theory has been widely applied in fields such as stochastic control, system control, thermodynamics, and data science in recent years. It endows the space of probability distributions with a rich Riemannian geometric structure, namely the Wasserstein geometry.
- **Sub-Riemannian Geometry**: The authors introduce a new type of transport problem where the relative positions of particles are important during the transport process. Specifically, they focus on transport along closed orbits, aiming to find control laws that allow a collection of particles to move along closed orbits in the Wasserstein manifold without mixing.
### Specific Issues:
- **Transport on Closed Orbits**: How to ensure that particles return to their initial positions without mixing when transported along a closed path?
- **Isoholonomic Problem**: How to find the optimal transport path under a given parallel transport mapping?
### Key Concepts:
- **Wasserstein Manifold**: The space of probability distributions, equipped with the Wasserstein metric.
- **Holonomy**: Describes the change in the relative positions of particles during the transport process.
- **Parallel Transport**: How to transport particles from one position to another in the Wasserstein manifold without changing their relative positions.
### Main Contributions:
- **Sub-Riemannian Structure**: The authors explore a new sub-Riemannian structure to model the changes in the relative positions of particles during the transport process.
- **Isoholonomic Problem**: They propose a new optimal transport problem aimed at achieving specific parallel transport mappings and holonomy.
- **Control Law Design**: They study how to design control laws that allow a collection of particles to move along closed orbits in the Wasserstein manifold without mixing.
### Application Prospects:
- **Control Theory**: In control theory, especially in the control of ensemble systems, this new type of transport problem can provide more refined control methods.
- **Data Science**: In data science, this theory can be used to handle the transport and mixing of probability distributions, such as in data augmentation and generative models in machine learning.
In summary, by introducing sub-Riemannian geometry and the isoholonomic problem, this paper explores the mixing phenomena in optimal transport and proposes new control law design methods to ensure that particles maintain their relative positions during transport. This provides new research directions and tools for fields such as control theory and data science.