High-fidelity neutral atom gates leveraging low-rank Hessian optimization

This paper presents a Hessian-based calibration method that leverages the low-rank structure of quantum-control landscapes to efficiently optimize high-dimensional waveforms for neutral atom gates, experimentally demonstrating rapid convergence and high fidelity (0.99902) for a robust controlled-Z gate on 171Yb qubits.

The Gist

Imagine you are trying to teach a very precise dance to a pair of atoms. In the world of quantum computing, these atoms are the "dancers" (qubits), and the dance steps are the logic gates that perform calculations. To make them dance perfectly, scientists use laser pulses to guide their movements.

The problem is that the lasers aren't perfect. They wobble, they distort, and the "music" (the control waveform) they play is often messy. If you try to fix a messy dance by randomly tweaking the music, you have to search through millions of possible changes. It's like trying to find a specific needle in a haystack the size of a city, and you might never find it.

The Big Idea: The "Low-Rank" Shortcut

This paper introduces a clever shortcut. The researchers discovered that even though the laser waveform has millions of possible ways to be distorted, only a handful of those distortions actually ruin the dance.

Think of the laser waveform as a giant, complex piece of clay. You could squeeze it, stretch it, or twist it in infinite ways. However, the researchers found that the "dance" (the quantum gate) only cares about five to ten specific ways of squeezing that clay. All the other ways you could twist the clay are "invisible" to the dance; they don't change the outcome at all.

They call this the "Low-Rank Hessian Optimization."

• Hessian: A fancy math word for a map that shows which directions are sensitive (ruin the dance) and which are not.
• Low-Rank: The map shows that only a tiny number of directions (the "principal space") matter.

How They Did It

Instead of guessing randomly, the team used this map to find the "sensitive directions."

1. Identify the Trouble Spots: They calculated which specific distortions in the laser pulse would cause the atoms to make mistakes (like falling out of the dance floor or stepping on each other's toes).
2. Focus Only on Those: They ignored the millions of irrelevant changes and only adjusted the laser along those few critical directions.
3. Closed-Loop Feedback: They ran the experiment, measured how well the atoms danced, and used that result to nudge the laser in the right direction. Because they were only looking at the few important knobs, the system learned incredibly fast.

The Results

They tested this on a specific type of atom (Ytterbium) and a specific dance move (a Controlled-Z gate).

• Speed: The optimization converged (found the perfect setting) very quickly, taking just a few steps instead of thousands.
• Accuracy: They achieved a success rate of 99.59% (and 99.9% if they ignored the rare cases where an atom got lost).
• Robustness: The best part? Even if they turned the laser power up or down by 20% (a huge change), the dance still worked perfectly. The optimized pulse was so well-tuned that it didn't care about small mistakes in the laser's strength.

Why It Matters

This method is like having a GPS that tells you exactly which few roads lead to your destination, rather than letting you drive randomly through every street in the country.

The paper claims this approach is:

• Efficient: It solves the problem of calibrating complex quantum gates without needing millions of experiments.
• Physically Motivated: It's based on the actual physics of how errors happen (leakage and phase errors), not just random guessing.
• Broadly Applicable: While they tested it on neutral atoms, the logic applies to many other types of quantum computers.

In short, they found a way to tune a very complex, high-dimensional quantum machine by focusing only on the few "knobs" that actually matter, resulting in a highly accurate and robust quantum gate.

Technical Summary

Technical Summary: High-fidelity neutral atom gates leveraging low-rank Hessian optimization

Problem Statement
Quantum optimal control offers a pathway to fast and robust multi-qubit gates by shaping time-dependent control fields. However, experimentally calibrating the resulting high-dimensional waveforms is a significant bottleneck. Direct searches over large parameter spaces converge slowly, and existing optimization approaches (e.g., parameterized ansatzes, machine learning, genetic algorithms) often require excessive experimental samples, fail to converge to optimal fidelity, or depend sensitively on independent characterizations of the system Hamiltonian or device transfer functions. Furthermore, the accuracy of optimized pulses is limited by the fidelity of the model Hamiltonian used to design them and the physical transfer function of the control hardware.

Methodology: Low-Rank Hessian Optimization
The authors develop a calibration strategy based on the observation that the fidelity landscape around an optimal gate is robust, meaning the gate fidelity is insensitive to most perturbations. They propose that the space of control waveform perturbations affecting gate fidelity is low-rank.

1. Theoretical Framework:

   • The gate waveform $\Omega(t)$ is decomposed into a high-dimensional basis of real-valued functions.
   • The gate infidelity ($1-F$) is expressed as a quadratic form involving the Hessian matrix $H$ of the fidelity with respect to control parameters: $1 - F \approx \frac{1}{2} \delta\vec{\Omega}^T H \delta\vec{\Omega}$.
   • The authors demonstrate that the rank of $H$ is determined not by the dimension of the control basis, but by the number of accessible error channels in the target unitary. Specifically, the rank is bounded by the number of independent leakage channels (transitions to non-computational states) and coherent error channels (phase errors and mixing within the computational subspace).
   • For a symmetric Rydberg-mediated CZ gate, the relevant control space rank is shown to be only 5 (for a simplified three-level model) or 10 (for a more complex four-level model involving unwanted Rydberg states).
   • The eigenvectors corresponding to non-zero eigenvalues define a "principal space." Perturbations orthogonal to this space (the null space) do not affect fidelity to leading order.

2. Experimental Protocol:

   • The optimization proceeds by iteratively scanning the coefficients of the control waveform along the principal eigenvectors of the Hessian.
   • This is a closed-loop process using experimental feedback (gate fidelity measurements) rather than relying on a pre-characterized model of the hardware distortions.
   • The method corrects the projection of the gate error onto the principal space while ignoring the null space, significantly reducing the dimensionality of the search.

Experimental Implementation and Results
The authors applied this method to an amplitude-robust (AR) controlled-Z (CZ) gate on metastable-state $^{171}\text{Yb}$ nuclear-spin qubits.

• System: Qubits are encoded in the $3P_0$ metastable manifold. The gate utilizes Rydberg blockade with a specific manifold ($\nu=52.3$) where Zeeman splitting constraints introduce an unwanted coupling to a second Rydberg state, increasing the Hessian rank to 10.
• Verification of Low-Rank Structure: The team measured the sensitivity of the gate fidelity along 10 principal directions and 4 random null-space directions. Experimental sensitivities matched theoretical predictions, clearly distinguishing the high sensitivity of the principal eigenvectors from the negligible sensitivity of the null-space modes.
• Optimization Performance:
  • Starting with waveforms significantly distorted by the acousto-optic modulator (AOD) bandwidth, the optimization protocol converged to the expected gate error after a single iteration.
  • Raw Fidelity: The optimized gate achieved a raw fidelity of $F = 0.9959(2)$.
  • Post-selected Fidelity: After post-selecting on events with no detected atom loss, the fidelity increased to $F_{ps} = 0.99902(7)$.
  • Robustness: The gate performance remained essentially unchanged under laser power variations of up to 20%, demonstrating the robustness of the amplitude-robust design.
  • Stability: The optimized gate maintained stable performance over a 10-hour period without re-optimization.
• Hamiltonian Error Correction: The authors demonstrated that the same Hessian-sensitive directions could correct for errors in the Hamiltonian model. By deliberately altering the Zeeman splitting (a parameter error), they showed that a single optimization cycle along the original Hessian eigenvectors recovered most of the lost fidelity, even though the optimal waveform shape changed significantly.

Significance and Claims
The paper establishes low-rank Hessian optimization as an efficient and physically motivated calibration strategy for high-dimensional optimal-control gates. The key claims are:

1. Efficiency: By restricting optimization to the principal space defined by the accessible error channels, the method avoids the prohibitive cost of searching high-dimensional spaces, enabling rapid convergence.
2. Physical Motivation: The number of required optimization directions is explicitly linked to physical error mechanisms (leakage and coherent errors), providing a clear theoretical bound on the complexity of the calibration.
3. Generality: The approach is broadly applicable to many qubit types and is particularly relevant for systems where control signals are distorted (e.g., solid-state qubits with long cryogenic chains) or where Hamiltonian parameters vary due to fabrication uncertainty.
4. Complementarity: The authors suggest this method complements pre-compensation techniques; low-rank Hessian optimization can efficiently correct residual errors after a distorted pulse has been brought into the perturbative regime.

The work does not claim to solve all calibration challenges (e.g., it notes that large initial distortions may require multiple cycles or that the method relies on leading-order perturbation theory), but it provides a rigorous model for reducing the dimensionality of the experimental optimization problem for multi-qubit gates.