⚛️ quantum physics

Wasserstein Distances on Quantum Structures: an Overview

This review consolidates the scattered literature on quantum Wasserstein distances to provide a comprehensive overview of the current state of the art, highlight key applications, and outline open problems and future directions for researchers in both classical optimal transport and quantum information theory.

Original authors: Emily Beatty

Published 2026-04-21

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Emily Beatty

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a logistics manager. Your job is to move piles of sand from one location to another. In the classical world, you have a map, a starting pile, a destination pile, and a cost for moving a single grain of sand from point A to point B. Your goal is to find the cheapest way to move the whole pile. This is the Optimal Transport problem.

Now, imagine you are trying to do this, but the "sand" isn't just sand. It's quantum sand. This quantum sand has weird properties: it can be in two places at once (superposition), it can be entangled with other sand (spooky action at a distance), and you can't look at it without changing it.

This paper, "Wasserstein Distances on Quantum Structures," is a massive guidebook written by Emily Beatty to help us figure out how to measure the "distance" between two piles of this quantum sand.

Here is the breakdown in simple terms:

1. The Problem: Why Can't We Just Copy-Paste?

In the normal world, we have a perfect ruler called the Wasserstein Distance. It tells us exactly how much "effort" it takes to turn one distribution of sand into another.

But when scientists tried to copy this ruler into the quantum world, it broke.

The Gluing Problem: In the classical world, if you have a map from A to B and a map from B to C, you can easily glue them together to get a map from A to C. In the quantum world, because of entanglement, you often can't glue these maps together. It's like trying to connect two puzzle pieces that were designed for different boxes.
The "True" Distance: Because of these quantum weirdnesses, there is no single "True" quantum ruler yet. Different scientists have built different rulers, each sacrificing a different feature to make it work.

2. The Three Schools of Thought (The Three Rulers)

The paper organizes all the current attempts into three main "schools" or approaches, like three different architects designing a bridge:

School A: The Coupling Architects (The "Matching" Team)

The Idea: They try to match every grain of quantum sand in pile A to a grain in pile B, just like the classical method.
The Catch: Because of quantum rules, they often have to accept that the "distance" between a pile and itself isn't zero (which is weird!), or they have to give up the rule that says "the shortest path between two points is a straight line" (the triangle inequality).
The Metaphor: Imagine trying to match socks from two different drawers. In the quantum drawer, some socks are glued together. To make a match, you might have to cut a sock in half or accept that a sock doesn't quite match itself perfectly.
What they use it for: Studying how quantum systems cool down or how particles move in large groups.

School B: The Dynamical Architects (The "Flow" Team)

The Idea: Instead of looking at the start and end piles, they look at the flow of water between them. They imagine the sand turning into a liquid, flowing through a pipe, and solidifying again.
The Catch: This only works if the "pipe" (the quantum system) follows very specific, smooth rules. It's hard to build a pipe for every type of quantum system.
The Metaphor: Instead of counting how many steps it takes to walk from your house to the store, they calculate the energy it takes to drive a car there, assuming the road is perfectly smooth.
What they use it for: Proving that quantum systems stabilize quickly and understanding how heat moves in quantum computers.

School C: The Lipschitz Architects (The "Smoothness" Team)

The Idea: They don't measure the distance directly. Instead, they measure how "wiggly" or "smooth" the functions describing the sand are. If a function changes too quickly, the distance is big.
The Catch: This is great for specific types of quantum computers (like those using bits and qubits), but it's hard to apply to other weird quantum shapes.
The Metaphor: Instead of measuring the distance between two cities, they measure how much your speedometer jumps if you drive between them. If the speedometer jumps a lot, the cities are "far apart" in terms of difficulty.
What they use it for: Checking how well quantum computers can learn patterns and how noisy they are.

3. Why Do We Need This? (The Applications)

Why are we building these weird rulers?

Better AI: Just like classical "Wasserstein GANs" (a type of AI) are great at generating realistic images, Quantum GANs could generate realistic quantum states. This is crucial for designing new drugs or materials.
Fixing Noisy Computers: Quantum computers are very fragile. These distances help us measure how much "noise" (errors) is ruining the calculation, helping engineers build better error-correction codes.
Understanding the Universe: It helps physicists understand how quantum systems evolve over time, from the Big Bang to the inside of a black hole.

4. The Big Picture: A Patchwork Quilt

The author concludes with a beautiful metaphor.
Imagine the theory of Quantum Wasserstein Distances is a half-finished patchwork quilt.

We have lots of patches (the different definitions by different scientists).
Each patch covers a hole (solves a specific problem).
But... there is very little thread connecting the patches.

We have many different rulers, but we don't have a single thread that ties them all together into one perfect theory. We don't know if the "Matching" ruler is the same as the "Flow" ruler.

The Takeaway

This paper is a map for explorers. It says: "We are in a new, strange land (Quantum Optimal Transport). We have built three different compasses to navigate it. They all point in slightly different directions, and none of them is perfect. But together, they are helping us build a better future for quantum computing, AI, and physics. Now, we need to find the thread to stitch them all together."

1. Problem Statement

The paper addresses the challenge of generalizing Optimal Transport (OT) theory, specifically Wasserstein distances, from classical probability measures to the non-commutative setting of quantum states (density operators).

The Gap: While classical OT has profound applications in economics, machine learning, statistical physics, and concentration of measure, there is no consensus on a "true" quantum Wasserstein distance.
The Obstacle: Direct translation of classical definitions fails due to non-commutativity and entanglement.
- The Gluing Lemma Failure: The classical proof of the triangle inequality relies on the "gluing lemma" (constructing a joint distribution for three variables from two marginals). In quantum mechanics, the Quantum Marginal Problem prevents this; not all pairs of bipartite states can be extended to a tripartite state with the correct marginals.
- Cost Matrix Issues: Defining a cost matrix that satisfies properties like faithfulness (zero distance only for identical states) and the triangle inequality simultaneously is difficult. Many proposals sacrifice one property to gain another.
Goal: To unify the scattered literature on quantum Wasserstein distances, categorize existing approaches, analyze their properties, and identify open problems and future directions.

2. Methodology

The paper employs a comparative review methodology, structuring the field into three distinct frameworks based on how they generalize classical OT concepts:

Coupling Approach: Generalizes the Monge-Kantorovich primal formulation (minimizing transport cost over couplings).
Dynamical Approach: Generalizes the Benamou-Brenier formulation (viewing transport as a gradient flow of entropy over a Riemannian manifold).
Lipschitz (Dual) Approach: Generalizes the Kantorovich dual formulation (maximizing the difference in expectations over Lipschitz functions).

The author analyzes each framework by:

Defining the specific mathematical construction.
Evaluating them against a set of desirable properties (e.g., triangle inequality, faithfulness, data processing inequality, continuity, computational tractability).
Reviewing specific applications where each framework has been successfully applied.

3. Key Contributions

A. Taxonomy of Quantum Wasserstein Distances

The paper systematically categorizes existing definitions:

Coupling Formulations:
- Position-Momentum (Golse et al.): Uses cost operators based on position ( $Q$ ) and momentum ( $P$ ). Useful for mean-field limits but lacks faithfulness (non-zero self-distance).
- Quadratures with Partial Transpose: Uses partial transpose to link couplings to stochastic maps.
- Modular Transport Plans: Uses Tomita-Takesaki theory (modular automorphisms) to recover the triangle inequality and faithfulness.
- Asymmetric Projector: Uses the projector onto the antisymmetric subspace. Satisfies data processing but lacks the triangle inequality in general dimensions.
- Transport Plans (Beatty): A flexible definition based on decomposing states into pure states, allowing arbitrary metrics on the projective Hilbert space.
Dynamical Formulations:
- Riemannian Metric (Carlen, Maas, etc.): Defines a metric on the space of density matrices such that the quantum Markov semigroup is the gradient flow of relative entropy. This approach successfully recovers Ricci curvature bounds and implies a chain of concentration inequalities (MLSI $\to$ TC $\to$ Poincaré).
Lipschitz (Dual) Formulations:
- Spectral Distance (Connes): Based on spectral triples and commutators with a Dirac operator.
- Hamming-based (De Palma et al.): Generalizes the Hamming distance for $n$ -qudit systems. It satisfies the triangle inequality and data processing but is computationally hard (SDP) for large $n$ .
- Resource-based: Defines Lipschitz constants relative to a set of "resources" (e.g., Pauli gates), linking distance to circuit complexity.

B. Analysis of Properties

The paper provides a comprehensive matrix (Tables 2, 3, 4) evaluating each definition against key criteria:

Triangle Inequality: Only the Modular and Dynamical approaches (and some specific Hamming cases) guarantee this.
Faithfulness: Many coupling-based definitions fail this (non-zero distance for identical states).
Data Processing Inequality: Crucial for information theory; satisfied by some but not all definitions.
Computational Complexity: Most are NP-hard or require solving large Semidefinite Programs (SDPs), though some admit efficient approximations for small systems.

C. Applications Survey

The review highlights where these distances have yielded results:

Concentration of Measure: Dynamical and Lipschitz approaches have successfully established quantum analogues of Transportation-Cost (TC) inequalities, Log-Sobolev inequalities, and Gaussian concentration.
Quantum Machine Learning (qGANs): Wasserstein distances are used to define loss functions for Quantum Generative Adversarial Networks to mitigate "barren plateaus" (vanishing gradients) and mode collapse.
Quantum Complexity: Lipschitz distances are linked to the circuit complexity of quantum channels and the hardness of simulating open quantum systems.
Statistical Mechanics: Coupling approaches (Golse et al.) have been used to prove convergence rates for the Schrödinger equation to the Hartree and Vlasov equations in the mean-field limit.
Error Correction: Used to analyze the decay of errors in GKP (bosonic) codes.

4. Key Results

No "One True" Distance: The paper concludes that due to the fundamental incompatibility of non-commutativity with classical geometric structures (like the gluing lemma), there is likely no single definition that satisfies all desirable properties of classical Wasserstein distances.
Trade-offs: Every proposal involves a trade-off. For example, the Dynamical approach yields powerful analytic tools (Ricci curvature, concentration inequalities) but is restricted to systems with specific detailed balance conditions. The Coupling approach is flexible but often fails the triangle inequality or faithfulness.
Concentration Inequalities: The dynamical framework has been the most successful in replicating the classical chain of implications (e.g., Modified Log-Sobolev $\implies$ Transportation Cost $\implies$ Gaussian Concentration) for quantum Markov semigroups.
Barren Plateaus: Quantum Wasserstein GANs (qWGANs) utilizing Lipschitz constraints (specifically the Hamming-based distance) show promise in overcoming the barren plateau problem in variational quantum algorithms.

5. Significance and Future Directions

Unification: This review serves as a critical resource for bridging the gap between classical optimal transport theorists and quantum information scientists, clarifying the landscape of fragmented definitions.
Open Problems:
- Linking Frameworks: Establishing quantitative equivalences (inequalities) between the different definitions (e.g., relating the Coupling distance to the Dynamical distance).
- Triangle Inequality: Determining if the triangle inequality can be recovered for broader classes of coupling-based distances.
- Rapid Mixing: Extending the classical path-coupling method for Markov chain mixing times to the quantum setting (currently an open problem).
- Numerical Computation: Developing efficient algorithms to compute these distances for high-dimensional systems, as current methods are often intractable.
- General Geometries: Moving beyond discrete Hamming metrics or specific Riemannian manifolds to more general physical geometries.

Conclusion: The field of quantum optimal transport is described as a "half-finished patchwork quilt." While individual patches (definitions) cover specific holes (applications like concentration or GANs), there is currently little "thread" (unifying theory) connecting them. The paper argues that the future lies not in finding a single perfect definition, but in understanding the relationships between these diverse frameworks to unlock broader applications in quantum physics and computing.