On November 16, 2026, two peer-reviewed papers published in the journal Nature Physics reported results that surpass the break-even point in autonomous quantum error correction (AQEC), which does not rely on measurement and feedback. Conventional active quantum error correction requires repeatedly measuring signs of errors and applying corrective operations based on decisions made by a classical computer, which has led to challenges such as control latency and hardware complexity. Using different superconducting circuits, the two studies succeeded in extending the lifetime of a logical qubit beyond that of the best physical qubit in the same device, using autonomous correction mechanisms built into the physical system itself.
From "Measure and Fix" to "Preserve Through Dissipation": How Autonomous Quantum Error Correction Works
In conventional active quantum error correction, ancilla qubits are used to repeatedly measure signs of errors (syndromes), and a classical computer instructs corrective operations in real time based on the results. However, repeating this measurement and feedback cycle introduces delays and new errors, and also places a heavy burden on the control hardware. Two studies offer an alternative solution to this problem: a paper (DOI: 10.1038/s41567-026-03493-z) by a team including Zhongchu Ni, Ling Hu, and Dapeng Yu, and a paper (DOI: 10.1038/s41567-026-03528-5) by a team including Lida Sun, Yifang Xu, and Luyan Sun.
Autonomous quantum error correction (AQEC) is an approach that removes repeated syndrome measurement and classical feedback from the error correction process. In both studies, microwaves are applied continuously or periodically from outside to coherently drive the quantum system. Combined with an energy dissipation pathway designed in advance, this automatically expels the excess entropy produced by errors from the system. Rather than a controller measuring and judging each error as it occurs, the dynamics of the physical system itself return the quantum state to its original code space.
Note, however, that measurement is unnecessary only while error correction is running; measurement has not been eliminated from every stage of the experiment. Both teams performed final state readout and quantum process tomography to evaluate the lifetime of the protected quantum state. Quantum process tomography is a technique for examining how accurately a quantum operation was carried out.
Key to understanding these results is the criterion known as "break-even." It means that the coherence lifetime of an error-corrected logical qubit exceeds the lifetime of the best physical qubit available in the same device. It is not an absolute lifetime record beating the physical qubits of every quantum device, but a criterion for judging whether error correction produced a net gain within the same device.
Among earlier active quantum error correction work, an experiment by Sivak and colleagues (DOI: 10.1038/s41586-023-05782-6), published in Nature on March 22, 2023, used measurement and feedback to report a coherence gain of relative to the best physical qubit in the same device. However, the publicly available materials we could confirm do not give the absolute lifetime of each of the compared systems. The two bosonic AQEC studies discussed here, meanwhile, were released as preprints as arXiv:2509.22191 on September 26, 2025, and arXiv:2509.26042 on September 30 of the same year, and were later published in Nature Physics after peer review. Both studies aimed to demonstrate that an autonomous scheme using no measurement or feedback can also cross this break-even point.
Autonomously Correcting Single-Photon Loss and Extending Qubit Lifetime by 18%
In the first study, the team of Ni, Hu, Cai and colleagues built a device combining a three-dimensional superconducting microwave cavity made of high-purity aluminum, a transmon serving as an ancilla qubit, and a lossy stripline resonator equipped with a Purcell filter. For the logical qubit, they adopted the lowest-order binomial code.
\[ |0_L\rangle=\frac{|0\rangle+|4\rangle}{\sqrt{2}},\qquad |1_L\rangle=|2\rangle \]
This code is designed to correct single-photon loss, the loss of one photon, which is the dominant source of error in superconducting resonators.
In the device, the entropy of the error caused by single-photon loss is transferred to the ancilla transmon through coherent operations. The entropy is then expelled to the outside through the stripline resonator, restoring the logical qubit to its original code space. This sequence of operations requires no syndrome measurement or classical feedback, and each AQEC cycle was executed in about 150 μs.
In the experiment, the fitted lifetime of the binomial logical qubit protected by AQEC reached about 2.0 ms. This is 1.18 times, or an 18% extension, compared with the best physical qubit in the same device: an uncorrected qubit using the cavity's Fock states and . It also recorded a lifetime 3.3 times that of the binomial code without error correction and 15.0 times that of the transmon qubit in the device. The absolute lifetimes of these comparison targets are not given in the publicly available materials we could confirm.
The lifetime was determined by fitting the time evolution of the quantum process fidelity to the following exponential model.
\[ F_\chi(t)=A\exp\left(-\frac{t}{\tau}\right)+0.25 \]
Here, is the characteristic lifetime obtained from the fit, and the constant term of 0.25 corresponds to the process fidelity when the qubit state is completely depolarized.
The team also ran experiments with another bosonic code, called the code, and obtained an AQEC lifetime of about 1.75 ms. This is about 2.2 times that of the unprotected code and about 2% longer than the lifetime of the best physical qubit in the same device. For this comparison as well, the absolute lifetimes of the unprotected code and the physical qubit are not given in the publicly available materials we could confirm.
In the code experiment, about 12,000 single-shot measurements with error correction and about 32,000 without were used to calculate standard errors by the bootstrap method. The uncertainty in the lifetimes was evaluated as the standard error of the fit. This memory experiment demonstrates protection against single-photon loss within a single device, so the type of error addressed is limited.
Break-Even Passed by a Different Method, Extending Qubit Lifetime by 4%
In the second study, the team of Lida Sun, Yifang Xu, Yilong Zhou and colleagues implemented their own AQEC protocol derived from quantum channel simulation. To encode the logical qubit, they used the lowest-order binomial code in a long-lived bosonic mode, as Ni and colleagues did. The recovery mechanism for expelling the error entropy, however, is different.
In this experiment, the entropy produced by single-photon loss is first coherently mapped onto an ancilla qubit, Y1. It is then transferred by a swap operation to a remote ancilla qubit, Y2, and a driven reset called a Raman reset is applied to Y2, dissipating the entropy. This allows errors to be corrected without repeated measurement or classical feedback. Each cycle includes a 220 μs free evolution interval, and the team fitted the decay of process fidelity over 20 rounds of correction cycles.
As a result, the coherence time of the AQEC-protected logical qubit was 1,627 μs. This exceeds the 1,565 μs of the best physical qubit in the same device, a qubit using the Fock states and of resonator S1, giving a lifetime ratio of 1.04, or a 4% extension. The lifetime of the binomial code without error correction was 629 μs, so AQEC extended the coherence time by about 2.6 times.
The process fidelity after one round also shows the effect of AQEC. With AQEC it remained at 89.1%, whereas without AQEC it fell to 75.5%. The process fidelity of encoding and decoding alone, without free evolution, was 98.5%.
The team also examined in detail the loss factors that govern error correction performance. The Raman reset drive deliberately shortened the effective lifetime of ancilla qubit Y2 from 110 μs to 4 μs, and the single-photon lifetime of resonator S2 from 1,170 μs to under 5 μs. This allowed the entropy to be expelled quickly to the outside.
In addition, a calibrated error model predicted a per-round fidelity of 87.2%, in good agreement with the measured value of 87.0%. This analysis showed that decoherence occurring in the ancilla qubit during the recovery operation is the main factor limiting performance. The paper also suggests that this framework could in principle be applied to arbitrary recovery maps, but what the experiment verified is recovery from single-photon loss for one binomial code.
Comparing the results of the two studies gives the following.
| 研究 | 回復の実装方式 | 符号と対象エラー | 保護時のコヒーレンス時間 | 同一デバイス内の最良物理基準 | その他の測定結果 |
|---|---|---|---|---|---|
| Ni、Hu、Cai氏ら(DOI: 10.1038/s41567-026-03493-z) | 補助トランズモンへ転送し、パーセルフィルター付き損失共振器で散逸。サイクル約150 μs | 最低次二項符号、単一光子損失 | 約2.0 ms(物理基準の1.18倍) | 共振器の無訂正Fock量子ビット (絶対値は公表資料で確認できず) | 無保護二項符号の3.3倍、トランズモンの15.0倍。 符号では約1.75 ms(物理基準を約2%超過) |
| Lida Sun、Yifang Xu、Yilong Zhou氏ら(DOI: 10.1038/s41567-026-03528-5) | Y1へ写像後、Y2へスワップし、ラマンリセットで散逸。自由発展220 μs、20ラウンド | 最低次二項符号、単一光子損失 | 1,627 μs(物理基準の1.04倍) | 共振器S1の無訂正Fock量子ビット (1,565 μs) | 無保護二項符号629 μsの約2.6倍。1ラウンド後の忠実度89.1%(無保護75.5%) |
Applications to Quantum Sensing: What the Two Experiments Show and What Remains
Ni and colleagues also tested whether AQEC could improve the sensitivity of quantum metrology (sensing). The experiment used Ramsey interferometry with a superposition of the cavity's Fock states and , measuring a tiny frequency shift while repeatedly running AQEC.
Performance was evaluated using a "normalized Fisher information" that counts all the time required for initialization, interference, characterization, readout, and AQEC in the total single-shot measurement time. The maximum Fisher information of the AQEC-protected measurement protocol reached 1.4 kHz, a gain of 3.3 dB over the unprotected version of the same sensing protocol and 6.3 dB over a reference system using a superposition of the uncorrected and states.
These, however, are maximum gains, not values obtained uniformly at every interference time. The absolute Fisher information values of the unprotected version and the reference system used for comparison are not given in the publicly available materials we could confirm. The error bars on the Fisher information were calculated by propagating the uncertainty of the fitting parameters. These results are a laboratory demonstration of frequency estimation, and they do not confirm improved performance in a specific measuring instrument or in sensors deployed in actual operation.
Research on autonomous quantum error correction is also progressing beyond superconducting circuits. In another preprint using ion traps (arXiv:2504.16746), released on April 23, 2025, dephasing was autonomously corrected in an environment where low-frequency noise had been deliberately added, and a logical qubit lifetime of ms and a physical qubit lifetime of ms were reported. However, because both the quantum platform used and the error model targeted differ, it cannot be compared on equal terms with the two superconducting resonator experiments here.
Using different superconducting circuits and recovery schemes, the two Nature Physics papers independently demonstrated that autonomous quantum error correction through engineered dissipation can pass the break-even point. Because the circuit configurations differ, however, neither directly reproduced the other's experimental results. The lifetime improvements also remain modest, at 18% and 4%, with decoherence in the ancilla system and other factors limiting performance.
As for application to quantum computation, these results are at a stage centered on protecting the memory of a logical qubit. To develop this into large-scale fault-tolerant quantum computing, it will be necessary to further reduce errors arising in the ancilla system, and to demonstrate logical gate operations protected by error correction as well as operations among multiple logical qubits.
