⚛️ quantum physics

The Rotation Gap Is Not An Error: Ternary Structure in IBM Quantum Hardware

This paper demonstrates that IBM Quantum hardware exhibits structured, sub-Poissonian ternary transitions rather than purely random errors, revealing that standard quantum error correction protocols degrade performance by miscorrecting these valid states and proposing a classifier decoder that selectively abstains from correcting them to significantly reduce logical error rates.

Original authors: Selina Stenberg

Published 2026-04-15

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Selina Stenberg

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to keep a house clean. You have a very strict rule: "If you see a speck of dust, you must immediately vacuum it."

For years, scientists building quantum computers have operated under this exact rule. They assume that every time their computer's "sensors" (called syndromes) flash a red light indicating a mistake, it's just random noise—a speck of dust that needs to be wiped away immediately. They believe that the more aggressively they vacuum, the cleaner the house (the computer) will be.

This paper argues that this rule is wrong for a specific type of quantum computer.

The author, Selina Stenberg, suggests that on IBM's quantum computers, some of those "red lights" aren't dust at all. They are actually intentional, organized patterns—like a dancer doing a specific move. If you try to "vacuum" (correct) these moves, you aren't cleaning the house; you are actually tripping the dancer and breaking the dance.

Here is the breakdown of the paper's discovery using simple analogies:

1. The "Too Quiet" Crowd (Sub-Poissonian Statistics)

In a normal noisy room, people cough randomly. Sometimes two people cough at once, sometimes three, sometimes none. This is called a "Poisson" distribution (random noise).

However, when the author looked at IBM's quantum computers, the "coughing" was strangely organized. The errors were too spaced out. It was as if the people in the room had a secret agreement: "If I cough, you wait a moment before you cough."

In physics, this is called sub-Poissonian statistics. It means the errors aren't random chaos; they are cooperative. The computer's hardware is doing something structured that looks like noise to our current tools, but is actually a valid state of the system.

2. The "Wrong Direction" Vacuum (The Mistake)

Because the standard software (the decoder) thinks every red light is a mistake, it tries to fix them all.

The Binary View: The computer only sees two states: "Good" (0) and "Bad" (1).
The Reality: The IBM hardware has a hidden third state (a "Ternary" state) that the sensors can't quite see clearly. It looks like a mistake, but it's actually a valid "dance move."

When the standard software sees this "dance move," it tries to force it back to "Good." But since it was already "Good" (just in a different, valid way), the software accidentally breaks it. It's like a teacher telling a student to stop dancing because they think the student is fidgeting, only to realize the student was performing a specific routine.

3. The "Smart Janitor" (The Regime Classifier)

The author proposes a new kind of software called a Regime Classifier. Think of this as a "Smart Janitor."

Instead of vacuuming every speck of dust, the Smart Janitor looks at the pattern:

Is this a random speck? (Yes? -> Vacuum it.)
Is this part of a structured dance pattern? (Yes? -> Do nothing.)

The paper shows that by doing nothing for about 14% of the "errors," the computer actually works better. By not fixing the things that didn't need fixing, they avoided making new mistakes.

4. The Proof: IBM vs. Google

To prove this wasn't just a glitch, the author compared IBM's computers to Google's new "Willow" processor.

IBM (The Hexagonal House): The errors were organized (sub-Poissonian). The "Smart Janitor" worked great, improving performance by up to 19%.
Google (The Square House): The errors were chaotic and random (super-Poissonian). When the "Smart Janitor" tried to skip cleaning here, it didn't help at all.

This proved that the "dance moves" are specific to IBM's unique hardware layout (the heavy-hex lattice), not a universal rule for all quantum computers.

5. The Big Takeaway: "Less is More"

The most shocking part of the paper is the inversion of the standard logic of error correction.

Old Logic: More correction = Better performance.
New Logic: When the hardware has hidden cooperative structures, less correction = Better performance.

The paper suggests that what we call "noise" might actually be a signal we are measuring in the wrong way. By trying to "fix" valid quantum states, we are actively destroying the information we are trying to protect.

Summary Analogy

Imagine a choir singing a complex song.

The Old Way: A conductor hears a note that sounds slightly off and immediately yells, "Stop! Fix that!" This disrupts the harmony because the note was actually part of a special harmony the choir was trying to create.
The New Way: A new conductor listens closely. They realize that some "off" notes are actually intentional harmonies. They let those notes sing and only correct the truly random mistakes. The result? The choir sounds much better.

In short: IBM's quantum computers have a hidden "third gear" that looks like a mistake but isn't. The paper proves that if we stop trying to fix that third gear, our quantum computers will run faster and more accurately.

1. Problem Statement

The foundational assumption of Quantum Error Correction (QEC) is that all syndrome activations (error signals) represent random noise (binary errors) that must be identified and corrected to preserve logical information. Standard decoders operate on the premise that "more correction is better."

This paper challenges that premise, proposing that on specific hardware architectures (specifically IBM's heavy-hex lattice), a fraction of syndrome activations are not errors but structured cooperative transitions (ternary states). When standard binary decoders attempt to "correct" these valid ternary states, they inadvertently introduce new errors, thereby degrading logical performance. The core problem is the mismatch between the binary nature of standard QEC decoders and the ternary cooperative structure inherent in certain quantum hardware.

2. Methodology

The study employs a multi-pronged approach combining empirical data analysis, cross-platform comparison, and novel decoding algorithms:

Data Sources:
- IBM Eagle r3: Analysis of 756 QEC runs across three processors (ibm brisbane, ibm kyoto, ibm osaka) over 14 days, covering code distances $d=3, 5, 7$ .
- Google Willow: A control dataset of 420 experiments on a 105-qubit grid processor to test if the observed phenomena are generic to surface codes or hardware-specific.
- Direct Hardware Validation: A prospective experiment on ibm strasbourg using a specific hexagonal syndrome circuit to test Z3 anti-bunching predictions.
Statistical Analysis:
- Calculation of the Fano factor ( $F = \text{Var}(n)/\langle n \rangle$ ) to distinguish between Poissonian (random) and sub-Poissonian (regulated/anti-bunched) statistics.
- Analysis of burst scaling (linear vs. quadratic) and spatial/temporal correlations.
The Regime Classifier Decoder:
- A new decoding algorithm introduced to distinguish between Binary Errors (B31) and Ternary Transitions (T75).
- Mechanism: Before applying a correction, the classifier scores flagged syndrome nodes based on five structural features:
  1. Isolation: Neighbors are quiet (anti-bunched pattern).
  2. Boundary Status: Reduced coordination (fewer neighbors).
  3. Density Contrast: Strong self-signal, weak neighbor signal.
  4. Chirality: Non-zero chirality nodes.
  5. Temporal Consistency: Persistence across time steps.
- Action: Nodes classified as Ternary (T75) are abstained from (left uncorrected); Binary nodes (B31) are corrected normally.
Mixed Error Model: A simulation calibrated to IBM hardware statistics, generating a mix of random binary errors and structured ternary transitions (where $f = 1 - F$ ).

3. Key Contributions

A. Discovery of Sub-Poissonian Syndrome Statistics

The paper provides robust evidence that IBM Eagle r3 hardware exhibits sub-Poissonian statistics ( $F = 0.856 \pm 0.03$ ), indicating that error events are not independent but anti-bunched.

Key Finding: The Fano factor is distance-independent (ANOVA $p=0.79$ ), proving the effect originates at the physical error level, not from decoder artifacts.
Burst Scaling: Syndrome bursts scale linearly with code distance ( $R^2 = 0.9999$ ), suggesting errors are geometrically constrained to boundaries rather than the bulk volume.

B. Cross-Platform Falsification (The "Willow" Control)

To rule out generic surface-code artifacts, the authors analyzed Google's Willow processor (grid topology, no P-gate asymmetry).

Result: Willow exhibits super-Poissonian statistics ( $F = 2.42$ ), super-linear burst scaling, and positive spatial correlation.
Significance: This confirms the sub-Poissonian signal is specific to the IBM heavy-hex architecture and its specific gate asymmetries (P-gate), not a universal property of QEC.

C. The Regime Classifier & Selective Abstention

The paper introduces a decoder that improves performance by doing less.

Mechanism: By identifying and abstaining from correcting ternary transitions, the decoder avoids "miscorrection" (applying a gate to a valid state).
Performance: On a mixed error model calibrated to IBM data, the classifier reduces Logical Error Rates (LER) by 7–19% at static detection depth ( $\tau=1$ ).
Falsification: When applied to Google Willow data, the classifier showed zero improvement, confirming the benefit is specific to hardware with cooperative ternary structure.

D. Theoretical Connection to Physics

The paper links the observed Fano factor to fundamental physics constants within the "Merkabit" framework:

The Fano factor at code distance $d=7$ relates to the Strong Coupling Constant ( $\alpha_s$ ).
The authors propose the identity $\alpha_s \approx F/7 \approx 5/42$ , suggesting the sub-Poissonian error correlations are a manifestation of the strong force's coupling structure on the hexagonal lattice.

4. Results

Metric	IBM Eagle r3 (Heavy-Hex)	Google Willow (Grid)
Fano Factor ( $F$ )	0.856 (Sub-Poissonian)	2.42 (Super-Poissonian)
Distance Dependence	None ( $p=0.79$ )	Strong ( $p \approx 0$ )
Burst Scaling	Linear ( $R^2=0.9999$ )	Super-linear ( $R^2=0.9999$ )
Spatial Correlation	Anti-bunched (Negative)	Bunched (Positive)
Classifier Effect	+7% to +19% LER reduction	~0% (No effect)

Abstention Rates: The classifier correctly identifies and abstains from correcting 75–98% of ternary transitions.
Miscorrection Avoidance: The improvement is driven entirely by stopping incorrect corrections, not by fixing errors better.
Gate Sensitivity: Direct hardware validation showed that using "wrong-direction" CNOTs (increasing circuit depth) destroyed the sub-Poissonian signal ( $F$ rose to 1.207), confirming the signal relies on native connectivity and gate coherence.

5. Significance and Implications

Paradigm Shift in QEC: The paper inverts the standard logic of error correction. It demonstrates that less correction can yield better performance if the hardware possesses cooperative error structures that binary decoders misinterpret as noise.
Hardware-Aware Decoding: It establishes that decoders must be "regime-aware." For hardware exhibiting sub-Poissonian statistics (like IBM Eagle), a classification step is required to distinguish valid ternary states from binary errors.
No Hardware Overhead: The proposed solution requires no hardware changes, only a modification to the software decoder to include a classification/abstention step.
Theoretical Unification: The work suggests a deep connection between quantum architecture (Eisenstein lattices), error statistics (Fano factor), and fundamental physics (Strong Coupling Constant), proposing that "noise" may actually be a signal in an unmeasured basis (ternary degree of freedom).
Predictive Power: The framework successfully predicted the absence of the effect on Google Willow before the data was analyzed, validating the hypothesis that the effect is driven by specific circuit asymmetries (P-gates) rather than just topology.

In conclusion, the paper argues that standard QEC actively destroys quantum information on IBM hardware by correcting valid ternary states. By recognizing and preserving these states through selective abstention, logical error rates can be significantly reduced.