Orbit dimensions in linear and Gaussian quantum optics

This paper introduces a framework for computing the dimension of the state manifold (orbit) reachable by quantum states under linear or Gaussian dynamics, revealing fundamental limitations on state transformations, providing a witness for non-Gaussianity, and clarifying the expressivity constraints of bosonic variational circuits across both discrete and continuous-variable settings.

The Gist

Imagine you are in a giant, infinite-dimensional dance hall called Hilbert Space. This is where all possible quantum states of light live. You are holding a specific "dance move" (a quantum state), and you want to know: How many different directions can I dance if I'm only allowed to use a specific set of moves?

This paper, titled "Orbit dimensions in linear and Gaussian quantum optics," by Eliott Z. Mamon, is essentially a map of that dance floor. It figures out exactly how much "freedom" a specific quantum state has when you are restricted to using only standard, linear, or Gaussian optical tools (like beam splitters, mirrors, and phase shifters).

Here is the breakdown using simple analogies:

1. The Dance Floor and the Rules

In the quantum world, light is made of "modes" (like different colors or paths a photon can take).

• The Goal: Ideally, a quantum computer should be able to turn any dance move into any other dance move. This is called "universality."
• The Problem: Most quantum optics experiments are limited. They can only use "Linear" or "Gaussian" tools. Think of these tools as a specific playlist of dance moves.
  • Linear/Gaussian tools: These are like standard steps: spinning, sliding, or shifting your position. They are smooth and predictable.
  • The Limitation: If you only have these standard moves, you can't suddenly learn a complex breakdance move (a "non-Gaussian" state) just by spinning faster. You are stuck in a specific zone of the dance floor.

2. What is an "Orbit"?

Imagine you start at a specific spot on the dance floor (your initial quantum state). You start dancing using only the allowed moves from your playlist.

• The Orbit is the entire path you trace out on the floor. It's the collection of every single new position you can reach.
• Orbit Dimension: This is the paper's main discovery. It asks: How many independent directions can I move in?
  • If you can only move forward and backward, your dimension is 1.
  • If you can move forward, backward, left, right, up, and down, your dimension is 6.
  • In this quantum dance hall, the "dimension" tells you how "rich" or "complex" the set of reachable states is.

3. The Big Surprises (The "Aha!" Moments)

Surprise #1: Bunching Doesn't Help
You might think that if you pack more photons (particles of light) into the same mode (like squeezing more dancers into one spot), you would gain more freedom to explore the dance floor.

• The Paper's Finding: No! If you have a "Fock state" (a specific number of photons), adding more photons to the same mode does not increase the number of directions you can explore.
• Analogy: Imagine a group of dancers holding hands in a tight circle. Whether there are 2 dancers or 100 dancers in that circle, if they are all doing the exact same move, they still only have the same number of ways to move as a group. The "bunching" doesn't create new possibilities.

Surprise #2: A New Way to Spot "Non-Gaussian" States
In quantum computing, "Gaussian" states are the easy, standard ones. "Non-Gaussian" states are the special, powerful ones needed for advanced computing.

• The Paper's Finding: You can tell if a state is "special" (non-Gaussian) just by counting its orbit dimension.
• Analogy: If you are dancing in a standard ballroom (Gaussian), you have a specific number of steps you can take. If you suddenly find yourself able to take more steps than the rules of the ballroom allow, you know you've broken the rules and are doing something "non-Gaussian." The paper says: If the dimension is higher than the maximum possible for a Gaussian state, the state is definitely non-Gaussian.

Surprise #3: The "No-Go" Theorem
The paper uses this dimension count to prove that certain things are impossible.

• Example: They show that you cannot build a specific logic gate (the CNOT gate, essential for computing) using only standard Gaussian tools on a specific type of encoding.
• Analogy: It's like trying to build a skyscraper using only a hammer and a screwdriver. You might be able to build a shed, but if you try to build a skyscraper, the "dimension" of your tools just isn't big enough. The math proves you'll fail before you even start.

4. How Do We Measure This?

The paper doesn't just do math on paper; it suggests how to measure this in a real lab.

• For Pure States: You can use "homodyne" or "heterodyne" measurements (basically, shining a laser and measuring the interference patterns) to figure out the dimension. It's like taking a snapshot of the dance floor and counting the unique directions the dancers are moving.
• For Mixed States (Messy states): If the state is noisy or mixed, you need two copies of the state and a "SWAP test" (a quantum trick to compare two things) to figure out the dimension.

5. Why Does This Matter?

This is crucial for Quantum Machine Learning and Quantum Computing.

• Expressivity: If you are training a quantum AI, you want it to be able to explore as many states as possible. This paper gives you a ruler to measure how "expressive" your quantum circuit is.
• Resource Management: It tells engineers, "Don't waste time trying to make a Gaussian machine do a non-Gaussian job. You need to add specific non-Gaussian ingredients (like special measurements or non-linear crystals) to expand your orbit."

Summary

Think of this paper as a topological map for quantum light.

• It tells you the size of the playground you can reach with your current tools.
• It reveals that crowding particles together doesn't expand the playground.
• It provides a simple test (counting dimensions) to see if you are doing something truly "quantum" (non-Gaussian) or just standard stuff.
• It helps researchers understand why some quantum computers can't solve certain problems and what extra tools they need to add to break through those limits.

In short: It's about knowing exactly how far you can go with the tools you have, and realizing when you need to upgrade your toolbox to go further.

Technical Summary

Here is a detailed technical summary of the paper "Orbit dimensions in linear and Gaussian quantum optics" by Eliott Z. Mamon.

1. Problem Statement

Quantum optical platforms (both discrete and continuous-variable) are fundamental to quantum computing and information processing. However, practical implementations are often restricted to sub-universal sets of operations, such as:

• Passive Linear Quantum Optics (PLO): Beam splitters and phase shifters.
• Displaced Passive Linear (DPLO): PLO plus displacements.
• Active Linear (ALO): PLO plus squeezing.
• Gaussian Quantum Optics (GO): PLO plus displacements and squeezing.

These groups of unitaries ($G$) are finite-dimensional, whereas the underlying Hilbert space (Fock space) is infinite-dimensional. Consequently, a specific quantum state $|\psi\rangle$ cannot explore the entire Hilbert space under these dynamics; it is confined to a specific subset called its orbit ($\text{Orb}_G(|\psi\rangle) = \{U|\psi\rangle \mid U \in G\}$).

The Core Challenge: There is a lack of simple, computable invariants to characterize the structure of these orbits. Existing invariants (like stellar rank or covariance matrix spectra) do not always fully capture the topological properties of the reachable state space. The paper asks: How many independent directions in the Hilbert space can a given state explore under a specific sub-universal regime?

2. Methodology

The author employs Lie algebraic methods and differential geometry to analyze the structure of orbits.

• Orbit Dimension Definition: The paper defines the dimension of an orbit as the dimension of its tangent space at a given state. For a unitary group $G$ with Lie algebra basis $\{H_1, \dots, H_d\}$, the dimension is calculated by counting the number of linearly independent vectors generated by the action of the Lie algebra on the state.
  • For pure states $|\psi\rangle$: $\dim(\text{Orb}_G(|\psi\rangle)) = \text{rank}_{\mathbb{R}}(\{H_1|\psi\rangle, \dots, H_d|\psi\rangle\})$.
  • For mixed states $\rho$: $\dim(\text{Orb}_G(\rho)) = \text{rank}_{\mathbb{R}}(\{[H_1, \rho], \dots, [H_d, \rho]\})$.
• Gram Matrix Formulation: To make these calculations practical, the author reformulates the rank calculation using Gram matrices. The entries of these matrices are expectation values of symmetrized products of Hamiltonians (e.g., $\langle \psi | \{H_I, H_J\} | \psi \rangle$ for pure states). This allows the dimension to be computed via standard linear algebra on real symmetric matrices.
• Representations: The framework is shown to be representation-independent. Calculations can be performed in:
  • Fock basis: Direct matrix operations.
  • Phase-space representations: Specifically the Stellar representation (for pure states) and Wigner representation (for mixed states), where Hamiltonians act as differential operators on phase-space functions.
• Experimental Estimation: The paper proposes protocols to estimate these dimensions experimentally using homodyne/heterodyne measurements (for pure states) and SWAP tests on two copies of the state (for mixed states).

3. Key Contributions

A. Theoretical Framework

• Manifold Structure: Proves that orbits under these quantum optical groups admit a natural smooth manifold structure (Theorem 1).
• Computable Invariants: Establishes that the orbit dimension is a robust topological invariant. If two states have different orbit dimensions, they cannot be converted into one another by any unitary in $G$.
• Variational Circuit Bounds: Derives a fundamental upper bound on the expressivity of Bosonic Variational Quantum Circuits (VQCs). The number of directions a VQC can explore is strictly bounded by the orbit dimension of the initial state (Theorem 2). This implies that "barren plateaus" or limited expressivity can be predicted solely from the input state's orbit dimension.

B. Analytical Results

• Explicit Formulas: Provides closed-form formulas for the orbit dimensions of various states (Table II), including:
  • Fock states: The dimension depends on the number of unoccupied modes ($u$) but not on the specific photon numbers, provided they are non-zero.
  • NOON states: Shows that for $N \ge 3$, the orbit dimension is constant, unlike the $N=1,2$ cases.
  • Superpositions: Analyzes how superpositions affect the dimension (e.g., superpositions over a constant number of modes yield limited increases).
• Genericity: Proves that for "generic" states (randomly drawn from a finite-dimensional energy shell), the orbit dimension achieves the maximal possible value for that group (Proposition 1). States with sub-maximal dimensions are the exception, often due to high structural symmetry.

C. Non-Gaussianity Witness

• The paper establishes that the orbit dimension under Gaussian operations ($G_{GO}$) serves as a witness for quantum non-Gaussianity.
• All Gaussian pure states share the same orbit dimension ($m(m+3)$). Any pure state with a dimension greater than this is necessarily non-Gaussian. The author conjectures that all non-Gaussian pure states have a strictly higher orbit dimension.

4. Key Results

1. Boson Bunching Limit: The dimension of the orbit for Fock basis states does not increase if bosons are "bunched" into the same mode (e.g., $|n, 0, \dots\rangle$ vs $|1, 1, \dots\rangle$). The dimension depends only on the number of occupied modes, not the photon distribution within them.
2. Impossibility of CNOT: The framework rigorously proves the impossibility of implementing a deterministic CNOT gate in dual-rail encoding using only Gaussian unitaries. The orbit dimension of the input state $|+0\rangle_L$ differs from the output state $|\Phi^+\rangle_L$ under $G_{GO}$, proving they lie in disjoint orbits.
3. Robustness: Orbit dimensions are lower semi-continuous. Small perturbations (noise) to a state cannot decrease its orbit dimension; they can only increase it or keep it the same. This ensures that experimental noise does not falsely suggest a state is "less expressive" than it truly is.
4. Experimental Feasibility: The paper details how to estimate the necessary Gram matrix entries using $O(m^4)$ polynomial observables, achievable via homodyne measurements with 5 distinct phase angles per mode (Proposition 3).

5. Significance and Impact

• Resource Theory: The work provides a new tool for resource theories in quantum optics, specifically for quantifying non-Gaussianity and expressivity. It offers a rigorous way to determine if a state is "useful" for tasks requiring non-Gaussian resources.
• Quantum Machine Learning (QML): By bounding the local dimension of bosonic VQCs, the paper helps practitioners understand the limitations of bosonic QML models. It explains why certain initial states lead to limited model capacity and guides the selection of initial states for better trainability.
• Experimental Verification: The proposed measurement protocols allow experimentalists to verify the "reachability" of a state space without full state tomography, which is infeasible for high-dimensional bosonic systems.
• Fundamental Limits: It clarifies the fundamental limits of linear and Gaussian optics, showing that while they can manipulate the phase space, they cannot generate the full complexity of the infinite-dimensional Hilbert space without non-Gaussian resources.

In summary, Mamon introduces orbit dimension as a powerful, computable, and experimentally accessible metric that reveals the topological structure of reachable state spaces in quantum optics, bridging the gap between abstract Lie theory and practical quantum information protocols.