betaFeedback

WorksPhysics and Astronomy

Entropic quantum error correction for independent and correlated amplitude damping errors

Abstract

The idea of using quantum entropy for quantum error correction (QEC) has existed for a long time. However, its computational complexity has limited its practical use: even a [[4,1]][[4,1]] code for standard independent amplitude damping errors involves a 256×256256\times 256 Hermitian matrix, so the approach scales poorly as the codeword length increases. For correlated errors, the situation is even worse. In this paper, we show that, by using the insights developed in our previous paper and exploiting certain structures of quantum conditional mutual information, we can solve these problems exactly by hand, using only 2×22\times 2 unitary matrices. By examining the entropy, we identify the leakage channels and the amount of leakage exactly, then define the recovery map accordingly. The rules defining our recovery map are simple, intuitive, and computationally inexpensive: for independent errors, it is based on 16 explicitly given orthonormal vectors and 6 instructions; for correlated errors, it is based on 6 explicitly given orthonormal vectors and 6 instructions. Our method sacrifices little in terms of performance: for the widely used entanglement fidelity, our simple QEC recovery map achieves a sharp result; for the worst-case fidelity, the best result is about 9%9\% better than ours; for correlated errors, there are no comparable results for short codewords. Among the known techniques, our recovery map is the first non-global one. The calculations use elementary linear algebra and calculus, and is presented in great detail.

Introduction

Developing scalable quantum error correction (QEC) is a major bottleneck in achieving practical fault-tolerant quantum computing; see, for example, [8].

In this paper, we study entropy-based QEC for both independent and correlated amplitude damping errors. Our goal is to develop a non-global recovery map based on quantum entropy with little loss in performance, in the hope of improving scalability.

We begin with a brief review of existing results.

The Knill–Laflamme condition [17] assumes that errors map the code subspace into nearly orthogonal error subspaces, allowing exact recovery. Thus, it does not work well for amplitude damping errors.

The study of approximate quantum error correction began with [19].

For independent amplitude damping errors and short codewords, Fletcher, Shor, and collaborators used semidefinite programming in [6, 7], while Biswas and Mandayam used the Petz map in [3, 2]. However, these techniques are global because they must account for every error, which makes scaling to long codewords difficult. For example, the standard semidefinite program cannot be extended to a distance-7 code; see, for example, [18].

The challenge is greater for correlated errors because the complexity of global recovery increases dramatically. Pauli twirling is used to approximate these errors by standard Pauli errors. However, this approach requires very long codewords to obtain meaningful results because Pauli twirling cannot eliminate correlations between errors; see [4]. It can also overestimate the logical error rate; see [15].

We now discuss entropic quantum error correction, which is the main technique used in this paper.

The entropy involved is quantum conditional mutual information (QCMI). For a tripartite quantum density matrix ρABC\rho_{ABC} describing subsystems AA, BB, and CC, the QCMI is defined as

I(A:C∣B)=Tr⁡ρABC(t)log⁡2ρABC(t)−Tr⁡ρAB(t)log⁡2ρAB(t)+Tr⁡ρB(t)log⁡2ρB(t)−Tr⁡ρBC(t)log⁡2ρBC(t)(1)\begin{aligned} I(A:C\mid B) ={}& \operatorname{Tr}\rho_{ABC}(t)\log_{2}\rho_{ABC}(t) - \operatorname{Tr}\rho_{AB}(t)\log_{2}\rho_{AB}(t) \\ &+ \operatorname{Tr}\rho_{B}(t)\log_{2}\rho_{B}(t) - \operatorname{Tr}\rho_{BC}(t)\log_{2}\rho_{BC}(t) \tag*{(1)} \end{aligned}

where ρABC\rho_{ABC} is the density matrix of the whole system, and ρAB(t)\rho_{AB}(t), ρBC(t)\rho_{BC}(t), and ρB(t)\rho_{B}(t) are the reduced density matrices defined by

ρAB(t):=Tr⁡CρABC(t),ρBC(t)=Tr⁡AρABC(t),ρB(t):=Tr⁡ACρABC(t).\rho_{AB}(t) := \operatorname{Tr}_{C}\rho_{ABC}(t), \qquad\rho_{BC}(t) = \operatorname{Tr}_{A}\rho_{ABC}(t), \qquad\rho_{B}(t) := \operatorname{Tr}_{AC}\rho_{ABC}(t).

Here, Tr⁡D\operatorname{Tr}_{D} denotes the partial trace over subsystem DD.

In 1973, Lieb and Ruskai proved in [20] that

I(A:C∣B)≥0.(2)I(A:C|B) \ge0. \tag*{(2)}

Petz proved in [24, 25] that perfect recovery is possible if and only if I(A:C∣B)=0I(A:C|B)=0.

When I(A:C∣B)≥0I(A:C|B) \geq0, one needs to find a recovery map for QEC. A local recovery map is preferable because it is scalable and is therefore highly sought after. This is the focus of the present paper and subsequent papers.

The idea of using QCMI for quantum error correction dates back to at least 1996. Schumacher and Nielsen began to develop ideas for QEC based on QCMI; see [26, 27, 28] and the more recent results in [13, 5, 16]. This approach connects recoverability with an information-theoretic measure of correlations. However, the problem is difficult because it involves large Choi matrices. For example, for four physical qubits subject to amplitude damping errors, the matrix is 256×256256 \times256.

In the recent papers [13, 5, 16, 14], recovery maps are constructed using entropies. However, the gate count scales exponentially with the code length, partly because the operators involved require manipulating the entire density matrix. This makes implementation difficult; for example, implementing the inverse of a large density matrix requires a deep circuit.

New ideas for recovery maps were developed in [21, 2]. However, these techniques are still global, and only independent errors were considered.

We now present our results.

In the present paper, we use the four-qubit Leung code [19] and follow the line of reasoning in [26, 27, 28]. The system has three parts: the reference RR, the logical state system SS, and the environment EE. The QCMI in (1.1) then takes the form

I(R:E∣S):=Tr⁡ρRSE(t)log⁡2ρRSE(t)−Tr⁡ρRS(t)log⁡2ρRS(t)+Tr⁡ρS(t)log⁡2ρS(t)−Tr⁡ρSE(t)log⁡2ρSE(t).(3)\begin{aligned} I(R:E\mid S) &: = \operatorname{Tr}\rho_{RSE}(t)\log_{2}\rho_{RSE}(t)-\operatorname{Tr}\rho_{RS}(t)\log_{2}\rho_{RS}(t) \\ &\quad+\operatorname{Tr}\rho_{S}(t)\log_{2}\rho_{S}(t)-\operatorname{Tr}\rho_{SE}(t)\log_{2}\rho_{SE}(t). \tag*{(3)} \end{aligned}

Here, the reference system is not subject to any external forces, SS is the codeword system, and EE is the environment. Even though the codeword length is only 4, the codeword system and the environment involve a 256×256256 \times256 Choi matrix, which is difficult to process.

In the present paper, we solve the problem by hand in a reasonable amount of time, using only 2×22 \times2 unitary matrices. Building on the insights developed in our previous paper [11] and exploiting the structure of the entropy, we let the entropy dictate most of the recovery rules, with the help of three technical lemmas: Lemmas A.1, B.1, and C.1. By examining the list of entropy leakages, we identify exactly the leakage channels, the amount of leakage, and the best way to correct the errors. We then define the recovery map accordingly.

We also extend our technique to correlated errors and identify the structure of their correlations.

Our QEC technique is non-global because the entropy provides a list of leakage errors ordered by the amount of leakage, and we construct the recovery map to correct the largest leakages and ignore the rest. This differs from the existing techniques in [6, 13, 5, 16, 21, 2].

Here we can not claim that our recovery map is local because the codeword considered here is too short.

The problem of constructing local recovery maps for various error models will be addressed in future work, including models of ZZ crosstalk [23, 30, 12]. As discussed earlier, an efficient local recovery map is desirable because it holds the promise of overcoming scalability limitations on the path to practical fault-tolerant quantum computation.

Next, we discuss the performance of our recovery maps.

Our instructions for implementing the recovery maps are simple and intuitive. They are given in (2.26)–(2.31) for independent errors and in (3.28)–(3.29) for correlated errors. For independent errors, the recovery map is based on 16 explicitly given orthonormal vectors and 6 instructions, listed in (2.22)–(2.24) and (2.26)–(2.31), respectively. For correlated errors, it is based on 6 explicitly given orthonormal vectors in (65) and 6 instructions in (68) and (69).

To quantify performance, we use the widely adopted entanglement fidelity, denoted by FEntF_{\mathrm{Ent}} in (4) and (5) below; see [22]. Our simple recovery map achieves a sharp result for this metric, matching the result obtained in [6] using a method designed to optimize entanglement fidelity.

To take more metrics into consideration, the following results were obtained in [2], where the parameter γ\gamma corresponds to our tt:

FEnt=1−1.25γ2+O(γ3),Fmin⁡2=1−1.15γ2+O(γ3)(4)\begin{aligned} F_{\mathrm{Ent}} &= 1 - 1.25\gamma^{2} + O(\gamma^{3}), \\ F_{\min}^{2} &= 1 - 1.15\gamma^{2} + O(\gamma^{3}) \tag*{(4)} \end{aligned}

where FEntF_{\mathrm{Ent}} is the entanglement fidelity and Fmin⁡2F_{\min}^{2} is the worst-case fidelity. For independent errors, our Theorems 2.2 and 2.3 below establish the following results for these two metrics:

FEnt=Fmin⁡2=1−1.25t2+O(t3).(5)F_{\mathrm{Ent}} = F_{\min}^{2} = 1 - 1.25t^{2} + O(t^{3}). \tag*{(5)}

Thus, our method achieves the same sharp result for entanglement fidelity, while our result for worst-case fidelity is about 9% worse. The performance of other recovery maps is reported in [2]. As discussed earlier, our method and the associated computations are much simpler, with little loss in performance. Moreover, our technique is non-global.

For correlated errors, the limitations of existing techniques require very long codewords to obtain meaningful results. Since our codeword is short, there are no comparable results available.

The paper is organized as follows. In Sections 2 and 3, we formulate the main results for independent and correlated errors. In Section 4, we develop a convenient expression for quantum conditional mutual information. In Sections 6–13, we prove the results stated in Sections 2 and 3. In Appendices A, B, and C, we provide technical results.

Main results for independent amplitude damping errors

We begin by defining the quantum system, which consists of three subsystems: the reference system, denoted by RR; the logical state system, denoted by SS; and the radiation field, denoted by EE.

Initially, the system is in the pure state

∣Vini⟩:=12∣R1⟩∣0L⟩∣Ω⟩+12∣R2⟩∣1L⟩∣Ω⟩,(6)\lvert V_{\mathrm{ini}}\rangle:= \frac{1}{\sqrt{2}}\lvert R_1\rangle\lvert0_L\rangle\lvert\Omega\rangle+ \frac{1}{\sqrt{2}}\lvert R_2\rangle\lvert1_L\rangle\lvert\Omega\rangle, \tag*{(6)}

where ∣R1⟩\lvert R_1\rangle and ∣R2⟩\lvert R_2\rangle are reference states that are not subject to any external forces, while ∣0L⟩\lvert0_L\rangle and ∣1L⟩\lvert1_L\rangle are the logical states. Here, we consider the Leung code [19], whose logical states ∣0L⟩\lvert0_L\rangle and ∣1L⟩\lvert1_L\rangle are defined in terms of four physical qubits:

∣0L⟩:=12(∣0000⟩+∣1111⟩),∣1L⟩:=12(∣0011⟩+∣1100⟩).(7)\begin{aligned} \lvert0_L\rangle&:= \frac{1}{\sqrt{2}}\left(\lvert0000\rangle+ \lvert1111\rangle\right),\\ \lvert1_L\rangle&:= \frac{1}{\sqrt{2}}\left(\lvert0011\rangle+ \lvert1100\rangle\right). \tag*{(7)} \end{aligned}

For each physical qubit, the ground state is ∣0⟩\lvert0\rangle and the excited state is ∣1⟩\lvert1\rangle. These states are represented by the two-dimensional vectors

∣0⟩:=(10);and∣1⟩:=(01)(8)\lvert0\rangle:= \begin{pmatrix}1\\0\end{pmatrix}; \quad\text{and} \quad\lvert1\rangle:= \begin{pmatrix}0\\1\end{pmatrix} \tag*{(8)}

The state ∣Ω⟩\lvert\Omega\rangle is the vacuum state of the radiation field.

For the quantum model governing the evolution of the whole system, we use the standard Hamiltonian together with ideas from [10, 9]. We assume that the speed of light is c=∞c = \infty. This assumption is reflected in the definition of the operator PkP_k in (9) and in its property stated below. The sole purpose of this assumption is to make the present paper fully mathematically rigorous without discussing the Born–Oppenheimer approximation. We emphasize that the Kraus operators derived here are the same as those in the existing literature, so this assumption does not affect the results. Moreover, the assumption c=∞c = \infty is consistent with current laboratory technology.

The Hamiltonian for the whole system is

H=IdR⊗HSE≔IdR⊗∑k=14(IdS⊗Pk+Hk)(9)H = I d_{R} \otimes H_{SE} \coloneqq I d_{R} \otimes\sum_{k=1}^{4} \left(I d_{S} \otimes P_{k} + H_{k}\right) \tag*{(9)}

where HSEH_{SE} is the Hamiltonian acting on the logical state system and the environment. Each term IdS⊗Pk+HkI d_{S} \otimes P_{k} + H_{k} acts only on the kk-th qubit and its environment. We assume that the actions on different qubits are identical and independent. Here,

Hk≔σk+⊗ak∗(0)+σk−⊗ak(0)H_{k} \coloneqq\sigma_{k}^{+} \otimes a_{k}^{*}(0) + \sigma_{k}^{-} \otimes a_{k}(0)

and

σk+≔[0100],σk−≔[0010]\sigma_{k}^{+} \coloneqq\begin{bmatrix} 0 & 1 \\ 0 & 0 \end{bmatrix}, \qquad\sigma_{k}^{-} \coloneqq\begin{bmatrix} 0 & 0 \\ 1 & 0 \end{bmatrix}

are the raising and lowering operators on the kk-th qubit. The operator PkP_{k} is the Hamiltonian of the radiation field and satisfies

eitPkak#(s)e−itPk=ak#(t+s)(10)e^{itP_{k}}a_{k}^{\#}(s)e^{-itP_{k}} = a_{k}^{\#}(t+s) \tag*{(10)}

The creation and annihilation operators ak∗(t)a_{k}^{*}(t) and ak(t)a_{k}(t) satisfy

References

  1. [1]H. Barnum and E. Knill. Reversing quantum dynamics with near-optimal quantum and classical fidelity. Journal of Mathematical Physics, 43(5):2097–2106, May 2002.DOI
  2. [2]D. Biswas and P. Mandayam. Universal syndrome-based recovery for noise-adapted quantum error correction. arXiv preprint arXiv:2510.08719, 2025.arxiv.org/abs/2510.08719
  3. [3]D. Biswas, G. M. Vaidya, and P. Mandayam. Noise-adapted recovery circuits for quantum error correction. Phys. Rev. Res., 6:043034, Oct 2024.DOI
  4. [4]A. Brillant, R. N. Rajmohan, P. Groszkowski, A. Seif, J. Koch, and A. Clerk. Noise correlations as a resource in pauli-twirled circuits. arXiv preprint arXiv:2603.12054, 2026.arxiv.org/abs/2603.12054
  5. [5]O. Fawzi and R. Renner. Quantum conditional mutual information and approximate Markov chains. Comm. Math. Phys., 340(2):575–611, 2015.
  6. [6]A. S. Fletcher, P. W. Shor, and M. Z. Win. Optimum quantum error recovery using semidefinite programming. Phys. Rev. A, 75:012338, Jan 2007.DOI
  7. [7]A. S. Fletcher, P. W. Shor, and M. Z. Win. Channel-adapted quantum error correction for the amplitude damping channel. IEEE Transactions on Information Theory, 54(12):5705–5718, 2008.arxiv.org/abs/0710.1052
  8. [8]A. G. Fowler, M. Mariantoni, J. M. Martinis, and A. N. Cleland. Surface codes: Towards practical large-scale quantum computation. Phys. Rev. A, 86:032324, Sep 2012.DOI
  9. [9]J. Fröhlich, Z. Gang, and A. Pizzo. A tentative completion of quantum mechanics. In Physics and the Nature of Reality: Essays in Memory of Detlef Dürr, pages 151–165. Springer, 2024.
  10. [10]J. Fröhlich, Z. Gang, and A. Pizzo. A theory of quantum jumps: J. fröhlich, z. gang, a. pizzo. Communications in Mathematical Physics, 406(8):195, 2025.
  11. [11]Z. Gang. Exact characterizations for quantum conditional mutual information and some other entropies. arXiv preprint arXiv:2603.14650, 2026.
  12. [12]S. Gicev, B. Harper, H. Kang, M. Usman, and M. Sevior. Crosstalk in contemporary quantum devices. arXiv preprint arXiv:2605.26528, 2026.arxiv.org/abs/2605.26528
  13. [13]A. Gilyén, S. Lloyd, I. Marvian, Y. Quek, and M. M. Wilde. Quantum algorithm for petz recovery channels and pretty good measurements. Phys. Rev. Lett., 128:220502, Jun 2022.DOI
  14. [14]M. Junge, R. Renner, D. Sutter, M. M. Wilde, and A. Winter. Universal recovery maps and approximate sufficiency of quantum relative entropy. Ann. Henri Poincaré, 19(10):2955–2978, 2018.
  15. [15]A. Katabarwa and M. R. Geller. Logical error rate in the pauli twirling approximation. Scientific reports, 5(1):14670, 2015.DOI
  16. [16]K. Kato and F. G. Brandão. Quantum approximate markov chains are thermal: K. kato, fgsl brandão. Communications in Mathematical Physics, 370(1):117–149, 2019.
  17. [17]E. Knill, R. Laflamme, and L. Viola. Theory of quantum error correction for general noise. Phys. Rev. Lett., 84:2525–2528, Mar 2000.pubmed.ncbi.nlm.nih.gov/11018926
  18. [18]R. L. Kosut and D. A. Lidar. Quantum error correction via convex optimization. Quantum Information Processing, 8(5):443–459, 2009.
  19. [19]D. W. Leung, M. A. Nielsen, I. L. Chuang, and Y. Yamamoto. Approximate quantum error correction can lead to better codes. Phys. Rev. A, 56:2567–2573, Oct 1997.
  20. [20]E. H. Lieb and M. B. Ruskai. Proof of the strong subadditivity of quantum-mechanical entropy. J. Mathematical Phys., 14:1938–1941, 1973. With an appendix by B. Simon.
  21. [21]H. K. Ng and P. Mandayam. Simple approach to approximate quantum error correction based on the transpose channel. Physical Review A, 81(6), June 2010.
  22. [22]M. A. Nielsen. The entanglement fidelity and quantum error correction. arXiv preprint quant-ph/9606012, 1996.DOI
  23. [23]A. Paetznick, B. W. Reichardt, M. P. da Silva, C. Ryan-Anderson, D. Aasen, J. M. Bello-Rivas, J. P. Campora III, R. Chao, A. Chernoguzov, W. van Dam, J. M. Dreiling, C. Foltz, F. Frachon, J. P. Gaebler, T. M. Gatterman, L. Grans-Samuelsson, D. Gresh, D. Hayes, N. Hewitt, C. Holliman, C. V. Horst, J. Johansen, D. Lucchetti, Y. Matsuoka, M. Mills, S. A. Moses, B. Neyenhuis, A. Paz, J. Pino, P. Siegfried, A. Sundaram, D. Tom, Z. Wang, S. J. Wernli, M. Zanner, R. P. Stutz, and K. M. Svore. Improved quantum processor logical error rates via correction and detection. Nature, 654:349–355, Jun 2026.
  24. [24]D. Petz. Sufficient subalgebras and the relative entropy of states of a von Neumann algebra. Comm. Math. Phys., 105(1):123–131, 1986.
  25. [25]D. Petz. Sufficiency of channels over von Neumann algebras. Quart. J. Math. Oxford Ser. (2), 39(153):97–108, 1988.DOI
  26. [26]B. Schumacher. Sending entanglement through noisy quantum channels. Phys. Rev. A, 54:2614–2628, Oct 1996.pubmed.ncbi.nlm.nih.gov/9913770
  27. [27]B. Schumacher and M. A. Nielsen. Quantum data processing and error correction. Phys. Rev. A, 54:2629–2635, Oct 1996.DOI
  28. [28]B. Schumacher and M. D. Westmoreland. Approximate quantum error correction. Quantum Information Processing, 1(1):5–12, 2002.DOI
  29. [29]G. Zheng, W. He, G. Lee, and L. Jiang. Near-optimal performance of quantum error correction codes. Physical Review Letters, 132(25), June 2024.DOI
  30. [30]Z. Zhou, A. Ji, and Y. Ding. Surface code error correction with crosstalk noise. In 2025 IEEE International Conference on Quantum Computing and Engineering (QCE), volume 1, pages 627–636. IEEE, 2025.arxiv.org/abs/2503.04642

Paper details

vlnc:W_vs752kqb
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents