Coherent versus stochastic error injection on a repetition-code logical qubit in superconducting hardware

This study experimentally investigates the impact of coherent versus stochastic error injection on a superconducting repetition-code logical qubit but fails to observe the theoretically predicted fidelity differences, hypothesizing that qubit frequency drifts effectively convert coherent errors into stochastic noise.

The Gist

The Big Picture: Two Ways to Break a Code

Imagine you are trying to send a secret message across a noisy room. To protect your message, you use a "repetition code." Instead of sending the word "Yes" once, you send it three times: "Yes, Yes, Yes." If the room is noisy and one "Yes" gets garbled into "No," the listener can still guess the original message was "Yes" because the other two agree.

In the world of quantum computers, this "room" is filled with different types of noise (errors). The scientists in this paper wanted to test a specific theory: Does it matter how the noise messes up the message?

They compared two types of noise:

1. Stochastic Noise (The "Random Coin Flip"): Imagine a mischievous gremlin who randomly flips a switch. Sometimes it changes a "Yes" to a "No," and sometimes it leaves it alone. It's purely random, like rolling a die.
2. Coherent Noise (The "Synchronized Dance"): Imagine a wind that gently but consistently pushes every "Yes" slightly toward "No." It's not random; it's a smooth, predictable rotation. If you push it just right, it might turn "Yes" into a weird mix of "Yes" and "No" at the same time.

The Theory: Computer simulations suggested that these two types of noise should affect the quantum computer differently. The "Synchronized Dance" (coherent) noise was predicted to be much more dangerous and harder to fix than the "Random Coin Flip" (stochastic) noise. The scientists expected to see a clear gap in performance between the two.

The Experiment: The Quantum Playground

The researchers built a small quantum computer using superconducting circuits (called transmons) to act as their testbed. They created a "repetition code" with 3 and 5 quantum bits (qubits).

To test the theory, they had to inject errors into the system:

• For Coherent Noise: They simply added a tiny, precise rotation to the quantum gates (like intentionally turning a steering wheel 1 degree too far). This is easy to do.
• For Stochastic Noise: They couldn't just "turn a wheel" because that's still a smooth motion. Instead, they had to create a scenario where errors happened randomly. Since their computer couldn't generate truly random errors in real-time, they used a clever trick called subset sampling.

The "Subset Sampling" Analogy:
Imagine you want to know how a car handles driving on a road with 100 different potholes. Instead of driving the car 100 times and hoping to hit every pothole randomly, you drive the car 100 times, but each time you intentionally hit exactly 1, then 2, then 3 potholes in a specific pattern. Afterward, you use math to combine all those results to predict what would happen if the potholes were truly random. This allowed them to simulate random noise without needing a super-fast random number generator.

The Surprise: The Gap Didn't Appear

The scientists ran the experiment and compared the results to their computer simulations.

• What they expected: The simulations showed a clear gap. The "Synchronized Dance" (coherent) noise should have made the quantum computer fail much more often than the "Random Coin Flip" (stochastic) noise.
• What they found: There was no gap. The quantum computer performed almost exactly the same way for both types of noise. The "dangerous" coherent noise didn't seem to be any worse than the random noise.

Why Did the Theory Fail? The "Drifting Tuning Fork"

The researchers had to figure out why the real world didn't match the math. They hypothesized that their quantum computer had a hidden flaw: frequency drift.

The Analogy:
Imagine you have a tuning fork that is supposed to vibrate at a perfect note. However, the room temperature is slowly changing, causing the tuning fork to drift slightly out of tune over time.

• In the simulation, the tuning fork was perfect and stayed in tune.
• In the real experiment, the tuning fork was slowly drifting.

This drift introduced a subtle, invisible "phase error" (a timing mismatch). The researchers believe this drift acted like a "twirler." It took the smooth, synchronized "dance" of the coherent noise and spun it around so much that it looked like random noise by the time the computer tried to fix it. The natural instability of the machine accidentally "stochastified" the coherent errors, hiding the difference the scientists were looking for.

They tested this idea by adding "drift" to their simulations, and it matched the real-world results much better.

The Conclusion

The paper concludes that while theory says coherent noise should be a unique and dangerous beast, in a real, imperfect quantum computer, the machine's own natural instability (like drifting frequencies) tends to turn that coherent noise into random noise.

Because of this, the "coherent-stochastic gap" (the difference in performance) disappeared in their experiment. They suggest that to see this gap clearly in the future, scientists will need to build quantum computers that are incredibly stable and don't drift, or use more complex codes that can handle these phase errors better.

In short: They tried to prove that "smooth" errors are worse than "random" errors, but the quantum computer's own slight instability smoothed out the difference, making them look the same.

Technical Summary

Technical Summary: Coherent versus Stochastic Error Injection on a Repetition-Code Logical Qubit in Superconducting Hardware

Problem Statement
Theoretical studies of Quantum Error Correction (QEC) often rely on stochastic depolarizing noise models, which are efficiently simulable via the Gottesman-Knill theorem. However, physical quantum systems experience coherent errors (unitary rotations) that differ fundamentally from stochastic Pauli errors. While theoretical analyses suggest that coherent errors can be more detrimental than stochastic errors—potentially lowering the error threshold for codes like the repetition code or surface code—experimental verification of a "coherent-stochastic gap" (a measurable difference in logical performance between the two noise types) has been lacking. Furthermore, it is unclear whether the "stochastification" of coherent errors by parity-check measurements, as predicted by theory, holds true in realistic experimental settings with baseline noise.

Methodology
The authors utilized a transmon quantum processor (Surface-17 architecture) to implement bitflip repetition codes at distances $d=3$ and $d=5$. The study employed a memory experiment where errors were injected at the start of each QEC round.

• Error Injection: The team compared two noise injection strategies:
  1. Coherent: Implemented via deterministic unitary gates ($R_X(\theta)$) causing over- or under-rotations.
  2. Stochastic: Implemented via pre-compiled randomized circuits containing random bitflip errors.
• Sampling Technique: To overcome the time overhead of real-time random pulse injection and the limitations of Monte Carlo sampling for circuit compilation, the authors adapted a subset sampling technique. They executed circuits with a deterministic number $k$ of injected errors and classically reconstructed the logical error probability $P_L(p_{inj})$ for any injection probability $p_{inj}$ using a binomial weighting scheme.
• Simulation: Two simulation approaches were used for comparison:
  1. Free-fermion simulator: Scalable and efficient, used to simulate large distances and benchmark against experimental data using a phenomenological bitflip noise model.
  2. Density-matrix simulator: Used to model richer, circuit-level noise (including relaxation, dephasing, leakage, and coherent evolution of $|f\rangle$ states) to understand discrepancies between theory and experiment.
• Decoding: A Minimum-Weight Perfect Matching (MWPM) decoder with uniform weights on time- and space-like edges was used to ensure consistency across all simulation sets and experimental benchmarks.

Key Results

• Absence of Observed Gap: Contrary to free-fermion simulation predictions, the experiment did not observe a significant difference in logical infidelity between coherent and stochastic error injection for either $d=3$ or $d=5$. The measurement uncertainties ($\sim 2\%$) were comparable to the predicted gap size ($1-2\%$), but no distinct divergence was resolved.
• Simulation vs. Experiment:
  • The free-fermion simulator predicted a small coherent-stochastic gap ($1-2\%$) even when matching the experimental baseline detection probability.
  • Density-matrix simulations incorporating more realistic noise models (e.g., amplitude damping, circuit-level noise) showed a reduction in the predicted gap, bringing it closer to experimental observations, particularly for $d=3$.
• Hypothesis on Discrepancy: The authors hypothesize that the lack of an observed gap is due to small, unmitigated drifts in qubit frequencies. These drifts introduce phase-coherent noise ($R_Z$ errors) that effectively "stochastifies" the injected coherent errors before they can interfere constructively or destructively to produce a distinct logical signature.
  • Numerical tests confirmed that accumulated phase errors can asymmetrically affect logical infidelity when combined with coherent rotations (potentially reducing the error via destructive interference) but symmetrically affect stochastic noise.
  • The ensemble average of frequency fluctuations in a real experiment likely washes out the specific interference effects required to observe the gap.

Significance and Claims
The paper contributes to the understanding of how coherent errors affect experimental QEC by providing the first direct experimental comparison of coherent versus stochastic noise injection on repetition codes.

• Methodological Contribution: The work demonstrates the utility of the subset sampling technique for efficiently characterizing logical error probabilities without the overhead of generating unique random circuits for every data point.
• Scientific Insight: The study highlights that while theoretical models predict a coherent-stochastic gap, experimental realities—specifically baseline noise and frequency drifts—may obscure this effect. The authors argue that the baseline noise in current devices acts as a natural dephasing or twirling mechanism, converting coherent errors into stochastic ones.
• Future Directions: The authors modestly conclude that resolving the coherent-stochastic gap experimentally will likely require addressing residual phase errors, such as by tracking frequency drifts between runs or improving baseline qubit coherence. They suggest that extending this framework to surface codes, which offer topological protection against phase errors, might provide a more robust platform for observing these differences.

The paper does not claim to have definitively proven the absence of the gap, but rather that it was not observed under current experimental conditions, likely due to the "stochastifying" effect of uncontrolled phase noise.