Folded optimal transport and its application to separable quantum optimal transport

This paper introduces "folded optimal transport," a unified framework that extends cost functions from extreme boundaries to entire convex sets using Choquet theory, thereby generalizing classical optimal transport and enabling the construction of a separable quantum Wasserstein distance on density matrices derived from pure states.

The Gist

The Big Picture: Moving Things from Simple to Complex

Imagine you have a set of pure, perfect ingredients (like a single grain of salt, a drop of water, or a pure color). In the world of physics, these are called "pure states." You also have mixtures of these ingredients (like a pinch of salt mixed with pepper, or a shade of gray). These are called "mixed states."

The paper asks a fundamental question: If we know the "distance" or "cost" to move one pure ingredient to another, how do we calculate the cost to move a whole mixture to another mixture?

Usually, in classical physics (like moving boxes of apples), this is easy because the mixtures are just simple averages. But in quantum physics, things get weird. The mixtures can be "entangled" (twisted together in ways that don't exist in our daily life), making the math of moving them incredibly difficult.

This paper introduces a new mathematical tool called "Folded Optimal Transport" to solve this problem.

---

Analogy 1: The "Folding" Map

Think of a convex set (a shape where if you draw a line between any two points inside, the line stays inside) as a folding map.

• The Edges: The "extreme boundary" of this map represents the pure states. These are the corners of the shape.
• The Middle: The inside of the shape represents the mixed states. These are just combinations of the corners.

In standard math, if you want to move from one point in the middle of the map to another, you usually have to invent a new rule. This paper says: "Don't invent a new rule. Just look at the corners."

The method works like this:

1. Lift: Imagine taking the mixed states and "unfolding" them back into all the possible ways they could have been made from the pure corners.
2. Transport: Calculate the cost of moving the pure corners to each other using standard rules.
3. Fold: "Fold" the map back up. The cost of moving the mixed states is the cheapest way to move the underlying pure corners that make them up.

The authors call this "Folded Optimal Transport" because it takes a complex, mixed situation, unfolds it to the simple edges, does the math, and folds it back up.

Analogy 2: The "Best Route" vs. The "Direct Route"

The paper distinguishes between two ways of measuring distance in this folded world:

1. The "Folded Kantorovich" Distance (The Direct Route):
   Imagine you want to move a pile of mixed sand (State A) to another pile (State B). You look at every single grain of sand in pile A and find the best match in pile B to minimize the total walking distance.

   • The Catch: Sometimes, if you take a direct route from A to B, the math doesn't add up perfectly. If you go A → B → C, the cost might not equal the cost of A → C plus C → B. It's like a map where the triangle inequality (the rule that the shortest path is a straight line) breaks down. This is called a semi-distance.

2. The "Folded Wasserstein" Distance (The Best Route):
   To fix the broken triangle rule, the authors say: "Okay, if the direct route is weird, let's allow you to take a detour."
   If you want to go from A to C, but the direct path is expensive or broken, you are allowed to go A → B → C. You calculate the cost of the whole chain and pick the absolute cheapest chain.

   • The Result: This creates a perfect, reliable distance (a "metric") that behaves exactly like the distances we use in everyday life (like driving from city to city).

The Quantum Application: Separable vs. Entangled

The paper applies this specifically to Quantum Mechanics.

• The Problem: In quantum physics, particles can be "entangled," meaning they are linked in a way that defies normal logic. Calculating the distance between two quantum states usually requires considering these weird entangled links, which is computationally a nightmare.
• The Solution (Separable Transport): The authors focus on "Separable" quantum transport. This means they only consider mixtures where the particles are not entangled with each other in a weird way. They are just simple mixtures.
• The Result: By using their "Folded" method, they successfully created a new, reliable way to measure the distance between quantum states (density matrices) based only on the distance between the pure states.

They found that their new "Folded Wasserstein" distance is:

• Reliable: It follows all the rules of geometry (triangle inequality).
• Continuous: Small changes in the quantum state lead to small changes in the distance.
• Connected to the Past: It turns out their method is very similar to a previous method proposed by other scientists (Beatty and Stilck-França), but their "Folded" approach explains why it works and fixes some of its mathematical quirks.

A Surprising Connection: The Semiclassical Bridge

The paper ends with a cool "Eureka" moment. They show that a famous, complex formula used by physicists Golse and Paul to compare quantum states with classical physics (called the Golse–Paul cost) is actually just a special case of their "Folded Optimal Transport."

In simple terms: They discovered that a very complicated quantum formula is actually just a specific type of "folding" of a simple cost function. This unifies three different worlds:

1. Classical (moving probability clouds).
2. Semiclassical (bridging quantum and classical).
3. Quantum (moving quantum states without entanglement).

Summary

The paper doesn't invent a new physical law or a new machine. Instead, it invents a new mathematical lens.

It says: "If you want to measure the distance between complex, mixed things (like quantum states), don't try to measure the mix directly. Unfold them to their pure components, measure the distance there, and fold the result back up."

This creates a unified, reliable framework that works for classical probability, semiclassical physics, and a specific type of quantum physics, making the math of "moving" quantum states much clearer and more consistent.

Technical Summary

Technical Summary: Folded Optimal Transport and Its Application to Separable Quantum Optimal Transport

Problem Statement
The paper addresses the challenge of extending cost functions or distances defined on the extreme boundary (pure states) of a convex set to the entire set (mixed states). In classical optimal transport, the state space is a simplex of probability measures, where the extreme boundary corresponds to Dirac masses. Standard Wasserstein distances naturally extend a metric on the underlying space to the space of measures. However, in quantum settings, the state space consists of density matrices (a convex set of operators), and the extreme boundary corresponds to pure states (rank-one projectors). Defining a meaningful "quantum Wasserstein distance" that respects the convex structure and extends a metric on pure states has been an open problem with various competing formulations (e.g., non-separable vs. separable couplings). The specific motivation is to provide a unified framework that extends a distance from pure states to mixed states using convex extension theory, specifically addressing the "separable" case where entanglement is not considered in the coupling.

Methodology
The author introduces a general mathematical construction termed Folded Optimal Transport, which relies on Choquet theory and standard optimal transport tools. The methodology proceeds in three main stages:

1. Convex Identification via Choquet Theory: The paper utilizes the Choquet–Bishop–De Leeuw theorem to identify a compact convex set $C$ with the set of Radon probability measures over its extreme boundary $E$, quotiented by the equivalence relation of sharing the same barycenter ($C \cong P(E)/\sim$).
2. Lifting and Gluing:
   • Lifting: A distance $d$ defined on the extreme boundary $E$ is lifted to the space of probability measures $P(E)$ using the standard Wasserstein-$p$ distance $W_p$.
   • Gluing: The lifted distance is projected back onto the convex set $C$ by minimizing over the equivalence classes. This yields a "folded Kantorovich semi-distance" $\bar{D}_p$.
   • Quotienting: To ensure the triangle inequality (subadditivity), which $\bar{D}_p$ may lack, the author defines the folded Wasserstein pseudo-distance $D_p$ as the quotient distance. This involves minimizing the sum of $\bar{D}_p$ costs over all possible chains connecting two points in $C$.
3. Application to Quantum Systems: The general construction is applied to the finite-dimensional quantum setting where $C$ is the set of density matrices $S_1^+$ and $E$ is the set of pure states $P_H$. The resulting distance is termed the quantum folded Wasserstein distance.

Key Contributions and Results

• Unified Framework: The paper establishes that standard optimal transport on probability measures is a particular case of this framework where the convex set is a simplex. It unifies classical, semiclassical, and separable quantum optimal transport under the single concept of folded optimal transport.
• Metric Properties:
  • The paper proves that the folded Wasserstein distance $D_p$ is a true distance (satisfying the triangle inequality) on the convex set $C$, provided the original distance $d$ on the boundary satisfies a lower bound condition relative to the norm ($d(x,y) \geq \|x-y\|$).
  • $D_p$ induces the natural topology on the convex set and, under specific conditions (geodesic boundary and $p>1$), the resulting space is geodesic.
  • The distance $D_p$ upper-bounds the norm distance on the convex set.
• Relation to Beatty–Stilck-França: The paper demonstrates that the Beatty–Stilck-França semi-distance (a known separable quantum transport metric) is exactly equivalent to the folded Kantorovich semi-distance $\bar{D}_p$.
  • A key finding is that the Beatty–Stilck-França semi-distance is a true distance if and only if it coincides with the folded Wasserstein distance ($D_p$).
  • The paper provides a sufficient condition for the subadditivity of the Beatty–Stilck-França semi-distance.
• Finite Support of Optimal Plans: Drawing on linear programming theory, the paper proves that optimal representing transport plans for the folded Kantorovich semi-distance exist and have finite support. Specifically, for quantum states in dimension $d$, the optimal plan has at most $2d^2 - 1$ atoms.
• Semiclassical Connection: The paper shows that the Golse–Paul semiclassical cost, used to compare quantum density matrices with classical probability densities, can be reformulated as a folded Kantorovich cost.

Significance and Claims
The paper claims to provide a rigorous theoretical foundation for "separable quantum optimal transport" by framing it as a convex extension problem. Its primary significance lies in:

1. Clarifying the relationship between different quantum transport formulations, specifically linking the Beatty–Stilck-França semi-distance to the broader theory of convex roofs and Choquet representations.
2. Resolving the metric status of the Beatty–Stilck-França semi-distance: it is shown to be a true distance only when it equals the folded Wasserstein distance, which enforces the triangle inequality via chain minimization.
3. Unification: It offers a single mathematical language (folded optimal transport) that encompasses classical transport, the Golse–Paul semiclassical cost, and separable quantum transport, suggesting that these are not disparate theories but specific instances of extending boundary metrics to convex sets.

The author notes that while the Beatty–Stilck-França semi-distance is computable (via finite support plans), the folded Wasserstein distance $D_p$ (which guarantees the triangle inequality) is less clear to compute directly, though the paper provides conditions under which they coincide. The work does not propose new experimental protocols but rather refines the mathematical definitions and properties of existing quantum transport metrics.