Quantum optimal control of the Dicke manifold in Rydberg atom arrays

This paper introduces a novel "irrep distillation" (IRD) method combined with quantum optimal control algorithms to effectively mitigate leakage errors caused by finite-range dipole interactions, enabling the efficient generation of highly entangled symmetric states like GHZ and Dicke states in Rydberg atom arrays using only linear-scaling computational resources.

The Gist

The Big Picture: Herding Quantum Cats

Imagine you have a room full of N cats (these are your quantum bits, or "qubits"). Your goal is to get all of them to perform a perfect, synchronized dance routine. In the quantum world, this "dance" is called a Dicke state.

If you have just a few cats, it's easy to choreograph their moves. But as you add more cats, the number of possible ways they can move together explodes exponentially. It becomes impossible to control every single cat individually without a supercomputer.

Usually, scientists try to solve this by assuming the cats are all identical and move in perfect unison (a "symmetric" dance). This simplifies the math. However, in the real world, the cats don't just move in a vacuum; they interact with their neighbors. If the room is small, the cats far apart can't "see" each other to coordinate. This causes the perfect dance to break down, and the cats start wandering off into a chaotic mess. This is called leakage.

The Problem: The "Leaky" Dance Floor

The researchers are working with Rydberg atoms (super-excited atoms that act like giant magnets). These atoms are arranged in a 2D grid (like a honeycomb). They interact with each other through electric forces, but this force gets weaker the further apart they are.

Because the force isn't infinite (it's "finite-range"), the atoms at the edge of the grid don't feel the same pull as the ones in the middle. This breaks the perfect symmetry. If you try to force them into a specific quantum state (like a GHZ state, which is a super-entangled "all-or-nothing" state), the atoms "leak" out of the desired formation into a chaotic, messy state.

The Solution: "Irrep Distillation" (The Filter)

To fix this, the authors developed a new method called Irreducible Representation Distillation (IRD).

Think of the quantum system as a complex machine with many gears.

1. The Main Gear: This is the perfect, symmetric dance (the Dicke manifold).
2. The Gears that Break: These are the messy, chaotic states the atoms leak into.

Instead of trying to simulate the entire machine (which is too big for computers), IRD acts like a smart filter. It identifies exactly which "broken gears" are most likely to catch the main gear. It then builds a simplified model that includes the Main Gear plus just those specific broken gears that are most likely to cause trouble.

By ignoring the rest of the infinite chaos, they can run their calculations on a computer that is small enough to handle, but still accurate enough to predict the real-world behavior.

The Strategy: The "Leakage Penalty"

The team used a technique called Quantum Optimal Control (specifically an algorithm called GrAPE). Imagine you are a coach trying to teach the cats the dance. You can only shout commands to the whole room at once (a "global" command), not whisper to individual cats.

They tested three coaching strategies:

1. Naive Coaching: The coach only cares if the cats look like they are dancing at the end. They ignore the fact that some cats are wandering off. Result: The dance looks good on paper, but in reality, it's a disaster because the "wandering" cats ruin the final pose.
2. Leakage-Aware Coaching: The coach adds a rule: "If you see a cat wandering off, you get a penalty." The algorithm learns to adjust the commands to keep the cats in the formation, even if it means the dance takes a slightly different path. Result: High success rates.
3. Local Coaching: The coach tries to whisper to specific cats to pull them back. Result: This helps a little bit for very complex dances, but it's hard to do and doesn't help much for simpler ones.

The Results: What They Achieved

Using their "Leakage-Aware" method, the team successfully simulated creating complex quantum states for up to 19 atoms.

• GHZ States (The "All-or-Nothing" Dance): They achieved very high accuracy (fidelity), beating what you would get if you just let the atoms interact naturally without any control.
• Dicke States (The "Specific Count" Dance): They could create states where a specific number of atoms are in one state and the rest in another. It got harder as the number of atoms in the "wrong" state increased, but it still worked well.
• Extremal Quantum States (The "Super-Complex" Dance): They tried to create the most complex, "quantum" state possible. Here, the method hit a wall. Even with their best tricks, the atoms leaked too much. This suggests that with only global commands and finite-range interactions, you simply cannot reach every possible quantum state perfectly.

The Bottom Line

The paper shows that you can control a large group of quantum atoms to perform complex, synchronized tasks without needing to control every single atom individually. By using a mathematical "filter" (IRD) to ignore the impossible-to-simulate chaos and focusing only on the likely errors, they can design control pulses that keep the atoms in line.

However, there is a limit. If the quantum state is too complex (like the Extremal Quantum State), the "leakage" is too strong for global commands to fix. In those cases, the atoms simply cannot be herded into that specific formation using only the tools available in this setup.

Key Takeaway: You can herd quantum cats into a perfect line using a megaphone (global control) if you have a smart filter to ignore the chaos, but if you ask them to do something too complicated, the megaphone just isn't enough.

Technical Summary

Technical Summary: Quantum Optimal Control of the Dicke Manifold in Rydberg Atom Arrays

Problem Statement
The central challenge addressed is the engineering and control of quantum states within the fully symmetric subspace (the Dicke manifold) of $N$ qubits. While this subspace is crucial for quantum sensing, metrology, and qudit-based information processing, generic state preparation is infeasible due to the exponential growth of the full Hilbert space ($2^N$). A promising physical platform for realizing collective control is an array of Rydberg atoms interacting via electric dipoles. However, unlike idealized all-to-all coupling models (e.g., in cavity QED), realistic Rydberg arrays in 2D lattices exhibit finite-range interactions (scaling as $1/r^3$). This finite range breaks the perfect permutation symmetry, causing the system's Hamiltonian to induce "leakage" from the computational symmetric subspace into the exponentially large, non-symmetric Hilbert space. Standard quantum optimal control (QOC) algorithms become computationally intractable when applied to the full Hilbert space for medium-to-large $N$, and naive control strategies that ignore leakage fail to produce high-fidelity states when implemented on the actual physical system.

Methodology
The authors propose a framework combining Quantum Optimal Control (QOC) with a novel truncation technique called Irreducible Representation Distillation (IRD).

1. Irreducible Representation Distillation (IRD):

   • The total Hilbert space is decomposed into irreducible representations (irreps) of $SU(2)$, where the symmetric subspace corresponds to the maximum spin $J=N/2$.
   • The finite-range interaction Hamiltonian ($\hat{H}_{int}$) is treated as a perturbation to a "gapped" One-Axis Twisting (OAT) Hamiltonian ($\hat{H}_{gOAT}$), which possesses $U(1)$ symmetry and exact $J$ quantum numbers.
   • IRD identifies the specific "distilled" irreps that couple to the symmetric subspace via this perturbation. By solving for the coupling vectors (using generalized spherical tensor operators), the method constructs a truncated Hilbert space ($H_D^{(x)}$) that includes the target symmetric subspace and the specific leakage subspaces (auxiliary irreps) relevant at a given perturbative order.
   • Crucially, the dimension of this truncated space scales linearly with $N$ (e.g., $3(N-1)$ for first-order IRD), rendering the problem computationally tractable for $N \approx 19$.

2. Control Protocols (GrAPE):

   • The authors employ Gradient Ascent Pulse Engineering (GrAPE) to optimize time-dependent control waveforms (global Rabi frequency $\Omega(t)$ and phase $\phi(t)$).
   • Three control strategies are compared:
     • Naïve Control: Minimizes infidelity without penalizing population in leakage subspaces.
     • Global Control with Leakage Penalty: Minimizes a cost function $C = 1 - F + P_x$, where $P_x$ penalizes population in the "escape" (leakage) subspace of the truncated model.
     • Local Control with Leakage Penalty: Introduces a weak, inhomogeneous local field $\omega(t)$ coupled to the interaction strength of individual atoms to actively pump leaked population back into the symmetric subspace.

Key Contributions

• Development of IRD: The paper introduces IRD as a method to define a linearly scaling truncated Hilbert space that captures the dynamics of leakage from the Dicke manifold due to finite-range interactions. This allows for the simulation of QOC on systems where exact diagonalization is impossible.
• Leakage-Aware Control: The authors demonstrate that incorporating a leakage penalty into the cost function is essential. Naïve control yields artificially high fidelities in truncated models but fails catastrophically when tested against exact dynamics or higher-order IRD models.
• Benchmarking Resource States: The study systematically benchmarks the preparation of Greenberger-Horne-Zeilinger (GHZ) states, Dicke states ($|D_k\rangle$), and Extremal Quantum States (EQS) in Rydberg arrays with $N$ up to 19.

Results

• GHZ and W-States ($|D_1\rangle$): Using global control with a leakage penalty on a first-order IRD model, the authors achieve high fidelities ($F \geq 0.975$) for GHZ and W-states with $N=14$ and $N=19$. These results are competitive with gate-based circuit protocols and significantly outperform passive evolution under the interaction Hamiltonian.
• Higher-Order Dicke States ($|D_k\rangle, k>1$): As the complexity of the target state increases (higher $k$), the infidelity grows with system size. While global control struggles, the addition of local addressing (local control) provides marginal but significant improvements for difficult states like $|D_3\rangle$, reducing infidelity from $\approx 6 \times 10^{-2}$ to $\approx 3 \times 10^{-2}$ for $N=14$.
• Extremal Quantum States (EQS): The control of EQS proves intractable. The cost function fails to converge to zero, and significant population loss occurs in the extended Hilbert space. This suggests that global control with finite-range dipolar interactions cannot achieve universal controllability on the symmetric subspace for highly entangled, non-polarized states.
• Quantum Speed Limits (QSL): The presence of leakage and finite-range interactions increases the QSL (time and complexity required) compared to ideal all-to-all models. However, the scaling remains linear with system size ($VT \propto N$), indicating that the fundamental speed limit is preserved despite the added complexity of combating leakage.
• Physical Feasibility: The authors calculate a speed figure of merit ($\eta$) for various Rydberg platforms (Rb, Cs). They conclude that for typical experimental parameters, the finite lifetime of Rydberg states (leakage out of the qubit manifold) is the dominant error source, but amplitude damping (bit-flips) is negligible. The required control times are achievable within the collective lifetime of the ensemble.

Significance and Claims
The paper claims that IRD provides a rigorous, perturbative framework to simulate and control quantum many-body systems with finite-range interactions, bypassing the exponential scaling of the full Hilbert space. The primary significance lies in demonstrating that leakage-aware global control can generate high-fidelity resource states (GHZ, W-states) in realistic Rydberg arrays without requiring the experimentally challenging implementation of full individual atom addressing.

The authors modestly note that while their method enables the preparation of many desirable states, it does not achieve universal control over the entire symmetric subspace, particularly for highly complex states like EQS. They attribute this limitation to the topological constraints of the unitary group generated by the finite-range Hamiltonian, which differs from the ideal all-to-all case. The work suggests that for highly detailed resource states, some form of local addressing may be necessary, but for many practical applications, global control augmented by leakage penalties is sufficient. The paper concludes that as experimental platforms scale to 3D lattices or higher dimensions (approaching the strong long-range regime), the efficacy of these IRD-based control protocols will likely improve further.