Rare Events and Griffiths Phases in Topological Quantum Error Correction

This paper investigates how spatially or temporally non-uniform error rates—specifically those caused by rare, extended events like cosmic rays—affect quantum error correction, finding that while such events create a "Griffiths phase" with stretched-exponential failure rates in 1D repetition codes, they lead to a total loss of threshold in 2D toric codes.

The Gist

Imagine you are trying to build a massive, high-tech fortress made of millions of tiny, interconnected bricks. To keep the fortress standing, you have a team of "repair robots" (the Quantum Error Correction) constantly checking for cracks and fixing them.

Usually, scientists assume the weather is predictable: maybe a light drizzle every day, or a steady wind. But in the real world, things are messy. Sometimes, a massive, unexpected storm hits (like a cosmic ray striking a computer chip), or one specific section of the wall is built with slightly weaker materials (fabrication errors).

This paper investigates what happens when these "rare, big events" occur. The researchers found that the shape of the problem changes everything.

1. The 1D Repetition Code: The "Chain of Links"

Imagine your fortress is just a long, single chain of metal links. If a storm hits, it might weaken a few links in a row for a short time.

The researchers found that this creates a "Griffiths Phase." Think of this like a "Bad Luck Streak." Even if the storm isn't strong enough to snap the whole chain, you might get a series of weak links that makes the chain much more likely to fail than usual.

Instead of the chain failing suddenly and catastrophically, it fails in a "stretched" way—it’s not as strong as a perfect chain, but it’s not totally broken either. It’s like a rubber band that has been weathered; it still works, but it’s much more prone to snapping unexpectedly. The good news: You can still protect the information, it’s just much harder than you thought.

2. The 2D Toric Code: The "Sheet of Paper"

Now, imagine your fortress is a giant, flat sheet of paper. This is much more complex and powerful than the chain.

When a "rare event" (the storm) hits this sheet, it doesn't just hit a line; it hits a whole patch or a "plane" of the paper at once. The researchers discovered something scary here: The 2D code has no "Bad Luck Streak" phase—it only has "Total Collapse."

Because the error happens across a whole area, the "repair robots" get overwhelmed instantly. It’s like trying to patch a leaking boat, but instead of a small hole, a giant wave hits the entire hull at once. There is no middle ground where the boat is "slightly leaky"; once the storm hits a certain strength, the whole boat sinks.

The Big Takeaway

The paper tells us that geometry matters.

• If your errors happen in lines (like a single faulty wire), you can survive, but you have to prepare for "bad luck streaks" that make your system much weaker.
• If your errors happen in patches (like a cosmic ray hitting a whole area of a chip), your most advanced quantum computers could face "catastrophic failure" where they stop working entirely, regardless of how good the rest of the system is.

The Lesson for Scientists: If we want to build a real quantum computer, we can't just focus on making the "average" error rate low. We have to find ways to stop these "big storms" from happening, or build "shields" specifically designed to handle massive, sudden bursts of noise.

Technical Summary

Technical Summary: Rare Events and Griffiths Phases in Topological Quantum Error Correction

Authors: Adithya Sriram, Nicholas O’Dea, Yaodong Li, Tibor Rakovszky, Vedika Khemani (Stanford University)
Date: March 17, 2025

---

1. The Problem: Heterogeneous Noise and Spatio-Temporal Correlations

Standard Quantum Error Correction (QEC) theory typically assumes spatio-temporally uniform error rates (i.e., every qubit has the same probability of error at every time step). However, real-world quantum hardware is subject to heterogeneity caused by imperfect fabrication or stochastic, high-energy events like cosmic rays.

These events create "error bursts"—periods where error rates are temporarily and globally elevated across the entire code patch. The authors investigate how these extended spatio-temporal correlations qualitatively alter the performance of topological codes, specifically asking if they can destroy the error threshold or change the scaling of logical failure rates.

2. Methodology: Statistical Mechanics Mapping

The researchers utilize the mapping between QEC decoding and disordered statistical mechanics models. They study two representative cases:

• 1D Repetition Code: Mapped to a 2D Random Bond Ising Model (RBIM).
• 2D Toric Code: Mapped to a 3D Random Plaquette Ising Gauge Theory (RPGT).

The Noise Model:
They introduce a Bernoulli distribution for bit-flip error rates $p_{bf}(t)$:

• $p_{bulk}$: The baseline error rate (below the code's threshold).
• $p_{rare}$: An elevated error rate during "rare events."
• $\gamma$: The probability of a rare event occurring.

The study focuses on the Largest Rare Region (LRR). In a system of size $L$ with $L$ measurement rounds, the longest sequence of consecutive rare events scales as $L_\perp \propto \log(L)$. The authors hypothesize that these LRRs dominate the logical failure probability ($P_{fail}$).

3. Key Contributions and Results

The paper reveals a fundamental divergence in how different topological codes respond to correlated rare events, rooted in the dimensionality of the rare regions.

A. The 1D Repetition Code (Linear Rare Regions)

• Discovery of a Griffiths Phase: For the repetition code, rare regions are "linear" (1D in space, extending in time). The authors find that even if $p_{rare}$ is above the bulk threshold, the code can remain decodable.
• Scaling Behavior: They identify a new Griffiths phase where the logical failure rate does not decay exponentially (as in the conventional phase) but follows a stretched exponential decay: $P_{fail} \sim \exp(-L^{1-z})$, where $z$ is a continuously varying dynamical exponent.
• Phase Diagram:
  1. Ordered Phase: $P_{fail}$ decays exponentially.
  2. Griffiths Phase: $P_{fail}$ decays as a stretched exponential (decodable but significantly degraded).
  3. Non-decodable Phase: $P_{fail}$ becomes a constant (threshold reached).

B. The 2D Toric Code (Planar Rare Regions)

• Catastrophic Failure: Unlike the repetition code, the toric code has no decodable Griffiths phase. Because the rare regions are "planar" (2D in space, extending in time), they can undergo a phase transition independently of the bulk.
• Loss of Threshold: As soon as $p_{rare}$ exceeds the 3D bulk threshold ($p_{3D}^0$), the code loses its threshold entirely. Even as the bulk error rate $p_{bulk} \to 0$, the logical error rate does not vanish in the thermodynamic limit.
• Mechanism: The planar rare regions act as "weak links" that allow topological defects (flux tubes) to proliferate, making the code undecodable.

4. Significance and Implications

Theoretical Significance:

• The work provides a rigorous connection between disorder physics (specifically Griffiths effects and the McCoy-Wu model) and quantum error correction.
• It demonstrates that the dimensionality of the noise correlation is a decisive factor in determining the robustness of a topological code.

Experimental Significance:

• Hardware Design: For architectures using the toric code (or surface codes), the results imply that suppressing extended sequences of repeated rare events (e.g., through improved shielding or faster reset times) is not just an optimization but a requirement for scalability.
• Explaining the "Noise Floor": The paper provides a theoretical framework to explain experimental observations (such as those by Google) where a logical noise floor is detected but its specific source remains unidentified. It suggests that "error bursts" might be the culprit.
• Predictive Power: The scaling functions derived (specifically for the toric code) can be used to predict how error correction performance will degrade as code distance increases in the presence of non-ideal noise.