Abstract: Given a transportation cost $c: M \times \bar M \to \mathbf{R}$, optimal maps minimize the total cost of moving masses from $M$ to $ar M$. We find a pseudo-metric and a calibration form on $M \times \bar M$ such that the graph of an optimal map is a calibrated maximal submanifold. We define the mass of space-like currents in spaces with indefinite metrics.
Introduction
Optimal Transportation (OT) is the study of the most efficient way to transform one probability measure into another given a cost function $c(x, y)$. While the existence of optimal transport maps is well-understood under broad conditions, the question of their regularity—whether the map is smooth or exhibits discontinuous "shattering"—remains a central challenge in the field.
In the mid-2000s, Ma, Trudinger, and Wang identified a complex fourth-order differential condition (the MTW condition) necessary for the smoothness of these maps. However, this condition was initially viewed as a purely analytic requirement. This research provides a geometric foundation for OT by demonstrating that the graph of an optimal transport map is a "maximal submanifold" within a specially constructed pseudo-Riemannian space. This framework transforms the analytic problem of regularity into a question of geometric curvature and stability.
Geometric Construction
To bridge OT and geometry, the authors construct a pseudo-Riemannian metric on the product space $M \times \bar{M}$. Unlike standard Riemannian geometry, where distances are always positive, this metric has an indefinite signature, reminiscent of the Minkowski space used in General Relativity.
This metric is defined using the cost function $c$ and the densities of the source and target masses ($ ho$ and $ar{ ho}$). Specifically, the authors build upon the work of Kim and McCann, but introduce a conformal factor that incorporates the mass densities. This allows the geometry to account not just for the distance (cost) but for the volume of the mass being moved.
Calibrations and Maximal Surfaces
In classical geometry, a calibration is a closed $n$-form used to prove that a submanifold is volume-minimizing (a minimal surface). In the pseudo-Riemannian setting, the objective shifts: the goal is often to find a maximal surface. The authors adapt the concept of calibration to this indefinite setting, providing a tool to prove that certain surfaces maximize volume among all competitors in the same homology class.
Key Insights
1. Optimal Maps as Maximal Submanifolds
The central discovery of this work is that the graph of an optimal transport map is a calibrated maximal submanifold. In the constructed geometry, the "volume" of the graph is maximized precisely when the transport cost is minimized. This provides a powerful intuition: optimal transport maps are the "perfect" shapes of the $M \times \bar{M}$ space.
2. Geometric Interpretation of the MTW Condition
The paper demonstrates that the MTW condition for smoothness is equivalent to the non-negativity of certain sectional curvatures of the pseudo-Riemannian metric. This is a significant breakthrough, as it replaces a difficult-to-parse differential inequality with a standard geometric property. If the space is "curved" in the right way, the optimal transport map remains smooth.
3. Stability and Currents
By employing the language of space-like currents (a generalized notion of surfaces from geometric measure theory), the authors extend their results to maps that may not be perfectly smooth. They prove that the space-like current defined by an optimal transport map is homologically mass maximizing (Theorem 3.10). This implies a form of stability: small perturbations of the map cannot increase the volume of the graph, mirroring the stability of the optimal transport solution itself.
Results and Implications
The research yields several formal conclusions:
- Theorem 3.10: Among all surfaces connecting the same boundaries in the constructed space, the graph of the optimal transport map has the largest possible volume.
- Corollary 3.11: At any point where the optimal transport map is sufficiently smooth, its graph has zero mean curvature, satisfying the Euler-Lagrange equation for the volume functional in this pseudo-Riemannian manifold.
- Conformal Equivalence: The metric used is shown to be conformally equivalent to the Kim-McCann metric, but the inclusion of mass densities provides a more complete physical and geometric picture of the transport process.
Limitations and Future Directions
The current framework relies on several assumptions:
- Smoothness and Non-degeneracy: The cost function $c$ must be smooth and satisfy non-degeneracy conditions (A1 and A2).
- Cut Locus Constraints: The results generally apply away from the "cut locus," where the cost function might lose its smoothness.
- Metric Space Generalization: While the theory is robust for smooth manifolds, optimal transportation on general metric spaces can be significantly more chaotic. Extending this geometric calibration theory to non-smooth or discrete settings remains an open challenge.
By providing a geometric "home" for optimal transportation, this work clarifies the relationship between the efficiency of mass movement and the underlying curvature of the space in which that movement occurs.
Summary updated