Strong Kantorovich duality for quantum optimal transport with generic cost and optimal couplings on quantum bits

This paper establishes Kantorovich duality for a linearized non-quadratic quantum optimal transport problem, applies it to derive optimal solutions for qubits with specific cost operators, and uses these results to analytically prove the triangle inequality for the square of induced quantum Wasserstein divergences.

The Gist

Imagine you are running a massive logistics company, but instead of moving boxes of apples, you are moving "quantum states." In the quantum world, these states are like delicate, invisible clouds of probability that describe where a particle (like an electron) might be or how it is spinning.

This paper is about finding the cheapest, most efficient way to move one of these quantum clouds into the shape of another, without breaking the laws of quantum physics.

Here is the breakdown of their work using simple analogies:

1. The Big Problem: Moving Quantum Clouds

In the classical world (our everyday reality), if you have a pile of sand in one spot and want to move it to another spot, you can calculate the cost of moving every grain. This is called Optimal Transport. You want to spend the least amount of energy (or money) to get the job done.

In the quantum world, it's trickier. You can't just grab a quantum cloud and move it. You have to use a "quantum channel" (a special machine or process) to transform the first cloud into the second. The authors are trying to figure out: What is the absolute minimum "cost" to turn Quantum State A into Quantum State B?

2. The Two Ways to Solve It (The Primal and The Dual)

The paper tackles this using a famous mathematical trick called Kantorovich Duality. Think of this as looking at a problem from two different angles to make sure you get the right answer.

• Angle 1: The "Primal" View (The Truck Driver)
  Imagine you are the truck driver. You are looking at all possible routes and all possible ways to shuffle the quantum particles. You are trying to find the single best "transport plan" (a specific coupling of the two states) that minimizes the cost.

  • The Paper's Twist: The authors realized that the original way people tried to calculate this cost was too complicated (non-linear). They created a simplified, linear version of the problem. It's like saying, "Instead of trying to solve a 3D puzzle with moving parts, let's flatten it out into a 2D grid where the math is easier."

• Angle 2: The "Dual" View (The Inspector)
  Imagine you are an inspector trying to prove that the truck driver cannot do it cheaper than a certain price. You set up a system of "prices" or "potentials" for every possible state. If your prices add up correctly, you can prove that no matter what route the driver takes, they can't beat your price.

  • The Paper's Achievement: They proved that for their simplified problem, the "Truck Driver's" best cost is exactly equal to the "Inspector's" best proof. This is called Strong Duality. It means they have found the perfect, unbreakable answer.

3. The Specific Case: The Quantum Bit (Qubit)

To show their theory works, they zoomed in on the simplest quantum object: the Qubit (a quantum bit, like a coin that can be heads, tails, or a blur of both).

They tested this with two specific scenarios:

• Scenario A: The Symmetric Cost. Imagine the cost of moving the cloud depends on how much it spins in any direction (up, down, left, right). They found a neat, closed-formula "map" for the cheapest way to move these clouds.
• Scenario B: The Single-Direction Cost. Imagine the cost only matters if the cloud spins up or down (ignoring left/right). They found another specific formula for this.

4. The "Triangle Inequality" Surprise

In geometry, the Triangle Inequality says that if you go from Point A to Point B, and then from B to C, the total distance is always longer than or equal to going directly from A to C. (You can't get somewhere faster by taking a detour).

In many quantum transport theories, this rule breaks down. Sometimes, going A $\to$ B $\to$ C actually costs less than going A $\to$ C directly, which makes no sense for a real "distance."

The Paper's Result:
Using their new formulas for the Qubit, the authors proved that for these specific quantum states, the triangle inequality holds true, even when you square the distance (which is a common way to measure quantum "energy").

• Analogy: They proved that in this specific quantum universe, you can't cheat the system by taking a detour. The direct path is always the most efficient (or at least, never more expensive than a detour).

5. A Warning: Sometimes the "Perfect" Plan Doesn't Exist

The paper also points out a weird quirk. In some very specific, rare cases (like when one cloud is perfectly pure and the other is mixed), there might not be a single "perfect" transport plan that hits the theoretical minimum cost. It's like trying to find the absolute lowest point in a valley that has a flat bottom; you can get infinitely close to the bottom, but you might never land on a single, unique "best" spot.

Summary

The authors built a new, simplified mathematical framework to measure the "distance" between quantum states. They proved that their simplified math is perfectly accurate (Strong Duality), used it to solve the puzzle for the simplest quantum objects (Qubits), and showed that for these objects, the rules of geometry (like the triangle inequality) still hold up, even in the strange quantum world.

Technical Summary

1. Problem Statement and Motivation

The paper addresses the Quantum Optimal Transport (QOT) problem, a non-commutative generalization of the classical Monge-Kantorovich problem. While classical optimal transport minimizes the cost of moving probability distributions, QOT seeks to transport quantum states (density matrices) using quantum channels.

• Context: The authors build upon the framework introduced by De Palma and Trevisan, where transport between states $\rho$ and $\omega$ is realized via quantum channels, inducing a "quantum coupling" $\Pi$.
• The Gap: Previous work often focused on quadratic cost functions. This paper investigates a non-quadratic generalization of the transport problem with generic cost operators.
• Core Challenge: The original formulation of the QOT problem with non-quadratic costs is non-linear in the coupling variable $\Pi$. This non-linearity complicates the application of standard duality theorems.
• Objective: The authors aim to:
  1. Formulate a linear relaxation of the non-linear QOT problem.
  2. Prove Strong Kantorovich Duality for this linearized problem.
  3. Determine explicit optimal couplings and potentials for qubits (2-level systems) under specific cost operators.
  4. Prove the triangle inequality for the square of the induced quantum Wasserstein divergences under specific commutativity constraints.

2. Methodology

2.1. Mathematical Framework

The authors utilize the formalism of quantum mechanics on a separable Hilbert space $\mathcal{H}$.

• States: $\mathcal{S}(\mathcal{H})$ denotes the set of density operators.
• Couplings: Following De Palma and Trevisan, a coupling $\Pi$ of $\rho$ and $\omega$ is a state on $\mathcal{H} \otimes \mathcal{H}^*$ such that its marginals are $\omega$ and $\rho^T$ (the transpose of $\rho$).
• Cost Operator: For a collection of observables $A = \{A_1, \dots, A_K\}$ and a classical cost function $c$, the quantum cost operator $C_c^{(A)}$ is defined via spectral measures.

2.2. Linear Relaxation

The primary methodological innovation is the transition from a non-linear primal problem to a linear one:

• Non-linear Primal (Original): Minimize $\text{tr}[\Pi^{\otimes K} C_c^{(A)}]$ where $\Pi$ is a single coupling. The cost function is non-linear in $\Pi$.
• Linear Primal (Relaxation): The authors introduce a variable $\Gamma$ on the tensor product space $(\mathcal{H} \otimes \mathcal{H}^*)^{\otimes K}$. The constraints require that the marginals of $\Gamma$ on specific subsystems match $\omega$ and $\rho^T$.
  • Crucial Distinction: In the linear relaxation, the couplings on different subsystems are allowed to be correlated, whereas in the original non-linear problem, they must be identical (independent copies of a single coupling).
  • The authors prove that the minimum of the linear relaxation is strictly smaller than or equal to the non-linear problem, providing a lower bound.

2.3. Duality Proof

The authors prove Strong Kantorovich Duality (equality between the primal minimum and dual maximum) for the linear relaxation using:

• Fenchel-Rockafellar Duality: They define convex functionals $\Theta$ (related to the cost constraint) and $\Xi$ (related to the marginal constraints) on the space of self-adjoint operators.
• Legendre-Fenchel Transform: By computing the conjugates of these functionals, they establish the dual problem.
• The Dual Problem: Maximize $\sum_{k=1}^K (\text{tr}[\omega Y_k] + \text{tr}[\rho X_k])$ subject to the operator inequality $\sum_{k=1}^K I^{\otimes(k-1)} \otimes (Y_k \otimes I + I \otimes X_k^T) \otimes I^{\otimes(K-k)} \leq C_c^{(A)}$.

3. Key Contributions and Results

3.1. Strong Duality Theorem

Theorem 1 establishes that for generic cost operators and the linear relaxation, the infimum of the primal problem equals the supremum of the dual problem. This is a foundational result allowing the use of dual potentials to verify optimality.

3.2. Existence of Optimal Solutions (Counter-Example)

The paper highlights a subtle but important phenomenon: Optimal Kantorovich potentials (dual maximizers) may not exist, even in low-dimensional cases (qubits).

• The authors provide a specific example where the primal problem has a solution, but the dual problem only has a supremum, not a maximum. This occurs when the cost operator and states lead to a situation where the constraint boundary cannot be touched by a bounded operator.

3.3. Explicit Solutions for Qubits

The authors apply the duality theory to derive closed-form solutions for Quantum Bits (Qubits, $\mathcal{H}=\mathbb{C}^2$) under two specific cost scenarios:

• Case A: Symmetric Cost (3 Observables)

  • Setup: $K=3$, observables are Pauli matrices $\{\sigma_x, \sigma_y, \sigma_z\}$, cost $c(x,y) = \|x-y\|_p^p$.
  • Result: For commuting qubits (states diagonal in the same basis), the authors derive an explicit formula for the $p$-Wasserstein distance $D_{symm,p}$.
  • Optimal Coupling: They construct an explicit matrix form for the optimal coupling $\Pi(\alpha, \beta)$ between states $\rho(\alpha)$ and $\rho(\beta)$.
  • Triangle Inequality: They prove that the square of the induced quadratic Wasserstein divergence ($d_{symm,2}^2$) satisfies the triangle inequality for commuting qubits. This is significant because standard quantum Wasserstein distances often fail the triangle inequality or the identity of indiscernibles.

• Case B: Single Observable Cost

  • Setup: $K=1$, observable $\sigma_z$, cost $c(x,y) = |x-y|^p$.
  • Result: For qubits orthogonal to $\sigma_z$ (Bloch vectors in the $xy$-plane), they derive the closed form for $D_{z,p}$.
  • Triangle Inequality: Similar to Case A, they prove that $d_{z,2}^2$ satisfies the triangle inequality for states in this subspace.

3.4. Comparison of Linear vs. Non-Linear

Proposition 4 demonstrates that the linear relaxation (Problem 1) can yield a strictly lower minimum than the original non-linear problem (Problem 9). This confirms that the relaxation is not trivial and that correlations between subsystems in the relaxed problem can reduce the transport cost below what is achievable with a single independent coupling.

4. Significance and Impact

1. Theoretical Rigor: The paper provides the first rigorous proof of strong Kantorovich duality for a non-quadratic quantum optimal transport problem with generic cost operators. This extends the applicability of QOT beyond the quadratic case (which is often easier to handle due to connections with quantum Fisher information).
2. Metric Properties: A major open question in quantum optimal transport is whether the induced distances satisfy the triangle inequality. The authors show that while the raw distance might not, the squared divergence (a modified metric designed to fix self-distances) does satisfy the triangle inequality for commuting states and specific subspaces of qubits. This supports the conjecture that these divergences are genuine metrics on quantum state spaces.
3. Computational Utility: By providing explicit closed-form solutions for qubits, the paper offers practical tools for calculating transport costs in quantum information tasks, such as state discrimination, quantum thermodynamics, and machine learning on quantum data.
4. Duality Insights: The demonstration that dual maximizers may not exist in the quantum setting (unlike the classical compact case) adds depth to the understanding of the functional analytic properties of quantum transport.

Conclusion

This work bridges the gap between abstract non-commutative geometry and concrete quantum information theory. By linearizing the problem and proving strong duality, the authors unlock the ability to compute optimal transport costs and verify metric properties for quantum states, specifically resolving the triangle inequality for squared divergences in commutative and specific qubit scenarios.