Executive Summary

TL;DR: Researchers from JPMorganChase and Quantinuum have successfully executed two hallmark quantum algorithms—QAOA (optimization) and HHL (linear systems)—using entirely fault-tolerant components on the H2 and Helios trapped-ion processors. By leveraging the Steane [[7,1,3]] code and sophisticated dynamic circuits, they achieved near-break-even performance across up to 12 logical qubits.

Background: This is not just another "qubit count" milestone. It represents a shift from component-level benchmarks (testing a single gate) to system-level algorithmic benchmarking. It places the community at the doorstep of the "Early Fault Tolerant" era, where error correction actually supports meaningful computation.

1. The Bottleneck: Why Algorithms are Harder than Primitives

In contemporary Quantum Error Correction (QEC), we have seen "break-even" for single logical qubits. However, running an algorithm like HHL or QAOA requires a massive stack:

1. Magic State Distillation: Injecting non-Clifford gates (like the T gate).
2. Dynamic Feedback: Mid-circuit measurements to decide whether to "Repeat-Until-Success."
3. QEC Cycles: Active syndromes to prevent error propagation across multiple logical qubits.

Prior work often skipped one of these, using non-FT magic states or "bare" circuits. This paper attacks the "end-to-end" pipeline, ensuring every single mid-circuit error is, in theory, correctable.

2. Methodology: Optimized FT Gadgets

The core innovation lies in the refinement of Steane Code gadgets. The Steane code is a CSS code that encodes 1 logical qubit into 7 physical ones.

The Logical T-Gate Breakthrough

Because of the Eastin-Knill theorem, non-Clifford gates cannot be transversal. The authors used a state-injection protocol involving a specialized $|H\rangle$ state preparation circuit.

• Infidelity: 2.6(4) × 10⁻³ per T-gate (a ~5x improvement over previous non-FT methods).
• Technique: They utilized "T-swap" and "T-bare" gadgets, prioritizing the T-swap variant which effectively "decodes away" physical errors during measurement.

Figure 4: T-gate teleportation and Steane QEC gadgets used in the experiment.

Dynamic "Repeat-Until-Success" (RUS)

Instead of simply discarding a shot when a syndrome measurement fails (post-selection), the authors used the real-time control capabilities of the Helios processor to branch. If the magic state preparation fails, the system resets and retries during the execution. This reduced the discard rate to nearly zero without significantly hurting the logical fidelity.

3. Results: Algorithmic Resilience

QAOA: Scaling Beyond Depth-1

The Quantum Approximate Optimization Algorithm (QAOA) was tested on the Low Auto-Correlation Binary Sequences (LABS) problem.

• Observation: Even as the physical gate count grew substantially (from 399 to 551 gates for N=5), the success probability increased as the logical T-gate budget was expanded.
• Scale: They reached a 12-logical-qubit implementation for portfolio optimization, utilizing 97 of the 98 available qubits on the Helios system.

Figure 11: Comparison of active QEC strategies. The Steane-swap protocol showed superior entanglement fidelity over traditional flag-FT.

HHL: The Linear System Stress Test

HHL is notoriously difficult due to its deep circuits (QFT and Phase Estimation).

• Result: Adding 2 active QEC cycles in the middle of the HHL circuit improved logical state fidelity to ~0.97.
• Insight: They identified "compilation artifacts"—software bottlenecks that caused higher RUS limits to perform worse than expected—proving that application-level benchmarking is vital for uncovering system-level bugs that isolated tests miss.

4. Critical Analysis & Future Outlook

Takeaway

We are no longer just asking "Can we make a logical qubit?" We are now asking "How do we coordinate 100 logical gates?" The success on the 12-qubit portfolio problem demonstrates that the overhead of QEC is finally becoming manageable.

Limitations

• Breakeven Gap: While performance is "near-break-even," the unencoded versions of these tiny circuits still often edge out the encoded ones in absolute fidelity. The Steane code's distance (d=3) is small.
• Idling Errors: In the QCCD architecture, moving ions creates "idling" time for other qubits. As circuits get deeper, managing these "memory errors" via dynamical decoupling remains a priority.

The Future

The path forward clearly involves Correlated Decoding (treating circuit-level noise holistically) and scaling to higher-distance codes (d=5 or d=7). However, this paper provides the blueprint for how a financial institution like JPMC might actually use a 2026-era quantum computer for real optimization tasks.

Reference: Perlin, M. A., et al. "Fault-tolerant execution of error-corrected quantum algorithms." JPMorganChase & Quantinuum Technical Report (2025/2026).