⚛️ quantum physics

Fusion-assisted decoding of non-Abelian topological order

This paper demonstrates that incorporating refined syndrome measurements resolving non-Abelian anyon fusion outcomes significantly enhances maximum likelihood decoder performance, enabling the recovery of logical qudits in non-Abelian topological codes like the $D_4$ quantum double model below a nonzero noise threshold.

Original authors: Rohith Sajith, Yabo Li, Zijian Song, Tsung-Cheng Lu, Carolyn Zhang

Published 2026-10-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Rohith Sajith, Yabo Li, Zijian Song, Tsung-Cheng Lu, Carolyn Zhang

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

In the quest to build a quantum computer, scientists face a fundamental problem: the information stored in these machines is incredibly fragile. Unlike a classical computer bit, which is simply a 0 or a 1, a quantum bit can exist in a delicate superposition of both states at once. This sensitivity makes it prone to errors caused by even the slightest environmental noise. To combat this, researchers use a strategy called topological quantum memory. Instead of storing data in a single particle, they encode it in the global shape of a system, much like how a knot in a rope holds its form even if the rope is wiggled locally. In this framework, errors appear as tiny, localized disturbances called anyons. If these disturbances are left alone, they can drift and eventually tangle the knot, destroying the information. The goal of a decoder is to act as a guide, identifying where these disturbances are and figuring out the most likely path they took, so the system can be corrected before the knot is ruined.

For a long time, scientists had a reliable method for decoding these systems when the disturbances behaved in a simple, predictable way. However, a new class of quantum systems has emerged that offers powerful advantages for computing, such as the ability to perform complex logical operations directly. These systems rely on "non-Abelian" anyons, which behave in a far more mysterious manner. Unlike their simpler cousins, non-Abelian anyons do not simply disappear when they meet; they can merge to create entirely new types of particles. This creates a significant hurdle for error correction: because the outcome of two anyons meeting is not always certain, standard decoding methods struggle to know which correction to apply. If the decoder guesses wrong, it might accidentally destroy the very information it is trying to save.

A team of researchers has now developed a refined approach to solving this problem, demonstrating that by gathering more detailed information about how these particles interact, they can significantly improve the reliability of the system. The team focused on a specific mathematical model known as the quantum double, which serves as a testbed for these complex quantum states. They discovered that the key to unlocking better performance lies in measuring not just where the disturbances are, but also how they fuse together. In their simulations, they tested three different levels of information gathering. The first level was a basic check to see if a disturbance was present at a specific location. The second level added the ability to identify exactly what kind of disturbance was there. The third and most detailed level went a step further, measuring the specific result when two disturbances were brought together.

The results of their simulations were striking. By adding the fusion measurements, the team found they could tolerate a much higher rate of noise before the system failed. In their specific model, the basic method could only handle a noise level of about 11 percent. When they added the ability to identify the specific type of disturbance, the limit rose to roughly 16 percent. But when they included the fusion measurements, the system could withstand noise up to about 21 percent. This represents a substantial improvement, suggesting that the extra effort to measure how particles merge pays off in a much more robust computer.

However, the researchers also uncovered a subtle and important limitation. While these fusion measurements are incredibly helpful for protecting a specific core piece of information, they can be dangerous for the rest of the data if the noise is too high. Because the path taken to measure the fusion can interact with the global structure of the system, an incorrect measurement can inadvertently reveal or alter the hidden information stored in the rest of the memory. The team found that as long as the noise remains below a certain threshold, the fusion measurements safely protect the core data without disturbing the rest. But if the noise gets too strong, the probability of making a wrong guess about how the particles are paired increases, and the measurement itself begins to scramble the information it was meant to protect.

This work provides a clear roadmap for how to build more resilient quantum memories. It shows that by carefully designing what we measure—specifically by including the outcomes of particle fusions—we can push the limits of how much noise a quantum computer can survive. The study confirms that for a specific type of quantum memory, it is possible to recover the entire logical state in a single round of measurements, provided the noise is kept below a critical point. While the full recovery of all data requires careful handling to avoid accidental disturbances, the ability to protect the most critical part of the information with such high efficiency is a major step forward. The researchers have effectively shown that the complexity of non-Abelian anyons, once seen as a barrier to error correction, can be turned into an asset if we know exactly how to look at them.

Technical Summary: Fusion-assisted decoding of non-Abelian topological order

Problem Statement
Maximum likelihood (ML) decoders are well-established for Abelian topological orders (e.g., the toric code), where they saturate information-theoretic thresholds by comparing the probabilities of different homology classes of errors. However, extending this framework to non-Abelian topological orders presents two fundamental challenges:

Non-deterministic Annihilation: Non-Abelian anyons fuse non-invertibly (e.g., $a \times \bar{a} = 1 + \sum c$ ), meaning local anyon positions do not uniquely determine the collective fusion state or the error history. Consequently, local syndromes alone cannot always distinguish between error histories that differ by logical operators.
Logical Operator Classification: Logical operators in non-Abelian orders do not form simple homology equivalence classes. A logical operator corresponding to a winding string combined with its conjugate may not yield a trivial operator but rather a coherence between other logical states. Furthermore, standard local syndrome measurements do not diagonalize the decohered density matrix as they do in Abelian cases.

This work addresses the decoding of quantum memories based on non-Abelian quantum double models (specifically the $D_4$ quantum double) under charge noise. The authors investigate whether refining syndrome measurements to include anyon fusion outcomes can systematically improve decoding thresholds and enable single-shot recovery.

Methodology
The authors construct explicit ML decoders for the $D_4$ quantum double model on a cylinder geometry under a charge decoherence channel that proliferates 2-dimensional ( $E$ ) non-Abelian anyons. They analyze three progressively refined sets of syndrome measurements:

$S^{(1)}$ (Coarse): Records only the presence or absence of the non-Abelian $E$ charge at each vertex.
$S^{(2)}$ (Onsite): Records the complete local charge at each vertex, distinguishing the $E$ charge from the four 1-dimensional Abelian charges ( $A_{xz}$ ).
$S^{(3)}$ (Fusion): Records the onsite charges plus the fusion outcomes of disjoint pairs of $E$ anyons (determined via a minimum-weight perfect matching, MWPM).

The authors map the calculation of syndrome probabilities for these sets to classical statistical-mechanics models (specifically variants of the random-bond Ising model and loop models). They utilize tensor network contractions to evaluate these partition functions exactly for finite system sizes. By performing finite-size scaling on the ML failure probability, they determine the decoding thresholds ( $p_c$ ) for each syndrome set.

A critical component of the methodology involves analyzing the impact of fusion measurements on logical information. The authors demonstrate that while fusion measurements along specific paths can disturb logical information outside the "central qubit" (by revealing non-central conjugacy classes), they preserve the central qubit subspace. They also construct a "heralded" decoder protocol to argue for a lower bound on the single-shot decodability of the full logical qudit.

Key Contributions and Results

Refined Syndromes Improve Thresholds: The authors demonstrate that resolving fusion outcomes significantly increases the decoding threshold. For the $D_4$ model under $E$ -charge noise, the thresholds are:
- $p_c^{(1)} \approx 0.111$ (Onsite $E$ presence only)
- $p_c^{(2)} \approx 0.165$ (Complete onsite charges)
- $p_c^{(3)} \approx 0.214$ (Onsite charges + Pair fusion outcomes)
  This confirms the inequality $p_c^{(1)} \leq p_c^{(2)} \leq p_c^{(3)}$ , showing that non-local fusion measurements provide substantial gains over local measurements alone.
Central Qubit Recovery: The authors identify a "central logical qudit" (associated with the center of the group $Z(D_4)$ ) which behaves similarly to Abelian topological orders. For this subspace, the non-Abelian fusion of $E$ anyons yields only Abelian outcomes, allowing the problem to be reduced to comparing classical probabilities of two homology classes. The high threshold of $p_c^{(3)} \approx 0.214$ applies to the recovery of this central qubit.
Fusion Measurements and Logical Disturbance: The paper provides a rigorous analysis showing that fusion measurements along homologically distinct paths can distinguish between non-central logical states (conjugacy classes $[r], [s], [rs]$) while leaving the central qubit ( $[1], [r^2]$ ) invariant. This implies that using fusion measurements to improve the central qubit threshold comes at the cost of inevitably losing the rest of the logical qudit if the noise is above the MWPM threshold.
Lower Bound for Full Qudit Recovery: The authors argue that the full logical qudit can be recovered in a single shot below a threshold $p_{full} \geq p_{MWPM} \approx 0.103$ . This is achieved by using a "heralded" approach where the pairing of anyons is dictated by the observed Abelian charges. However, the paper notes that this value represents a lower bound (derived from the MWPM threshold on a square lattice) and that the actual threshold for the full qudit is unknown. Furthermore, while single-shot recovery is theoretically possible, the authors suggest that in practice, a multi-step procedure is likely preferred to minimize measurement weight and avoid costly coherent transport.
Intrinsic Threshold Estimate: Using Rényi-2 coherent information calculations, the authors estimate the intrinsic information-theoretic threshold for the decohered state to be approximately $p_c \approx 0.46$ . They note that while their ML decoders improve with refined syndromes, they likely do not reach this intrinsic limit due to the inability of syndrome measurements to fully diagonalize the non-Abelian decohered density matrix.

Significance
The paper establishes a general framework for designing ML-type decoders for non-Abelian topological orders by incorporating fusion measurements. It demonstrates that for non-Abelian systems, the trade-off between measurement locality and decoding performance is critical: non-local fusion measurements are necessary to achieve high thresholds for specific logical subspaces (the central qubit). The work highlights a unique feature of non-Abelian topological order where the act of measuring fusion outcomes can inadvertently reveal logical information about non-central sectors, a phenomenon absent in Abelian codes. The results provide a concrete pathway for improving fault tolerance in quantum memories based on non-Abelian anyons, specifically showing that single-shot decoding is feasible for the central qubit of the $D_4$ quantum double with a non-zero threshold.

More like this