Computing with many encoded logical qubits beyond break-even

1. Problem Statement

Quantum error correction (QEC) is essential for scaling quantum computers to solve practical problems, but it traditionally incurs massive overheads in physical qubits and circuit depth. Most experimental demonstrations have been limited to small numbers of logical qubits that do not yet outperform unencoded (physical) computations ("break-even").

• The Challenge: Implementing high-rate QEC codes (encoding many logical qubits into few physical qubits) requires high-fidelity gates and long-range connectivity. Existing hardware often lacks the connectivity to implement these codes efficiently or the fidelity to demonstrate performance gains over unencoded operations.
• The Gap: While fully fault-tolerant (FT) algorithms are the ultimate goal, there is a significant complexity gap before they outperform unencoded ones. There is a need for "partially fault-tolerant" (pFT) techniques that can immediately improve performance and demonstrate "beyond break-even" results on near-term hardware.

2. Methodology

The authors utilized the Quantinuum Helios trapped-ion quantum processor, a 98-qubit device with all-to-all connectivity, which is ideal for implementing geometrically non-local codes.

A. Code Family: Concatenated Iceberg Codes

The core of the work is the implementation of Iceberg codes ($I_k$) and their concatenation.

• Iceberg QED ($d=2$): These are low-overhead Quantum Error Detection (QED) codes. An $I_k$ code encodes $k$ logical qubits into $k+2$ physical qubits (adding only 2 ancillas) to detect any single-qubit error.
• Concatenated QEC ($d=4$): By concatenating two iceberg codes ($I_{k_2} \circ I_{k_1}$), the authors created Quantum Error Correction (QEC) codes with distance $d=4$. This allows for the correction of single errors and detection of double errors.
• Scalability: The authors demonstrated codes with up to 94 logical qubits (QED) and 48 logical qubits (QEC) on the 98-qubit Helios processor.

B. Fault-Tolerant Gadgets and Protocols

To achieve "beyond break-even" performance, the team introduced new FT and pFT gadgets optimized for the Helios architecture:

• State Preparation (SPAM): Developed log-depth tree circuits for preparing logical $|0\rangle$ and GHZ states. They utilized Repeat-Until-Success (RUS) protocols and Decoherence-Free Subspaces (DFS) to mitigate coherent errors and increase acceptance rates.
• Syndrome Extraction: Introduced new gadgets using GHZ states for syndrome extraction, allowing for high parallelization (groups of 8 gates) matching Helios' gate zones. This replaced serial extraction methods, reducing time overhead.
• Logical Gates:
  • Implemented a universal gate set including non-FT intra-block $U_{ZZ}(\theta)$ and FT logical FANOUT gates.
  • Used Pauli Twirling and Cycle Benchmarking (CB) to estimate logical gate fidelities.
• Partial Fault Tolerance (pFT): For Hamiltonian simulations, they used pFT techniques where state preparation and syndrome extraction are FT, but the simulation gates (rotations) are not fully FT. This leverages the hardware's native gate bias (where dangerous errors are rare) to achieve high fidelity without full distillation overhead.

C. Benchmarking Strategy

The paper employs a "Mirror Benchmarking" (MB) protocol for Hamiltonian simulation. This involves running a forward time evolution and then a reverse evolution to measure the survival probability $P(t) = |\langle \psi_0 | U^\dagger U | \psi_0 \rangle|^2$. This allows for fidelity estimation without requiring classical simulation of the full quantum state (which is intractable for these system sizes).

3. Key Contributions

1. Beyond Break-Even Demonstration: The first experimental demonstration where encoded logical computations consistently outperform their unencoded physical counterparts across a wide range of benchmarks, including state preparation, gate operations, and complex simulations.
2. High-Rate Concatenated Codes: Successful implementation of concatenated iceberg codes with up to 48 logical qubits and distance $d=4$, proving that high-rate codes are viable on current hardware.
3. Novel FT Gadgets: Introduction of GHZ-based syndrome extraction and log-depth state preparation circuits that significantly reduce spacetime overhead compared to previous methods (e.g., Knill-style QEC).
4. pFT Simulation at Scale: Execution of a 3D XY model quantum magnetism simulation with 64 logical qubits, demonstrating that pFT schemes can extend the reach of quantum computers beyond classical simulation limits.
5. Postselection Scaling: Evidence that increasing code distance via concatenation suppresses postselection overheads exponentially, making high-distance codes practical for longer computations.

4. Key Results

• SPAM Performance: Encoded logical state preparation and measurement achieved >10x improvement in error rates compared to unencoded physical qubits. For the $d=4$ concatenated code, logical infidelity was bounded at $< 4 \times 10^{-5}$ with no logical errors recorded in 5,000 shots.
• Logical Gate Fidelity: Logical gate infidelities were measured between $(1.0 - 1.2) \times 10^{-4}$. This significantly outperforms the bare physical two-qubit gate error of $\sim 8 \times 10^{-4}$ on Helios.
• GHZ State Preparation: High-fidelity preparation of logical GHZ states was achieved. Notably, the $I_{8} \circ I_{6}$ code (48 logical qubits) and leakage-heralded $I_4$ codes (60 logical qubits) recorded zero logical errors over thousands of shots.
• Hamiltonian Simulation (XY Model):
  • Encoded simulations of a 3D antiferromagnet showed a ~30% reduction in effective two-qubit error rates compared to unencoded simulations.
  • The system successfully simulated dynamics that are likely intractable for classical computers due to the high dimensionality (4x4x4 lattice) and rapid entanglement spreading.
• Acceptance Rates: The acceptance rate for the $d=4$ concatenated code was 0.62(2), significantly higher than expected for uncorrected systems, demonstrating that error correction effectively filters out noise.

5. Significance and Outlook

• Viability of High-Rate Codes: This work proves that high-rate QEC/QED codes are not just theoretical constructs but are practical tools for near-term quantum advantage. They allow for encoding many logical qubits with minimal overhead.
• Path to Fault Tolerance: By demonstrating that concatenation improves acceptance rates without degrading fidelity, the paper provides a clear roadmap for scaling to fully fault-tolerant regimes.
• Classical Supremacy: The successful simulation of the 3D XY model with 64 logical qubits represents a step toward "beyond-classical-scale" computation, where the quantum computer performs tasks that classical supercomputers cannot feasibly simulate.
• Hardware Agnostic Potential: While demonstrated on trapped ions, the principles of high-connectivity and these specific code structures are applicable to other platforms like neutral atoms, accelerating the development of universal fault-tolerant quantum computers.

In summary, this paper marks a pivotal transition from small-scale error correction experiments to large-scale, high-rate logical computations that demonstrably outperform physical qubits, validating the feasibility of practical quantum computing in the near term.