pith. sign in

arxiv: 2605.01258 · v1 · submitted 2026-05-02 · 🪐 quant-ph

Toward the Goldilocks blind compression of quantum states

Hyunho Cha , Chae-Yeun Park , Jungwoo Lee This is my paper

Pith reviewed 2026-05-09 15:05 UTC · model grok-4.3

classification 🪐 quant-ph
keywords quantum autoencodersblind compressionpure quantum statesfidelityancilla qubitsCPTP mapsGoldilocks architecture
0
0 comments X

The pith

A quantum autoencoder with exactly k encoder ancillas and n decoder ancillas achieves optimal fidelity for any distribution of pure n-qubit states.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Quantum autoencoders compress quantum data into a low-dimensional latent state while allowing reconstruction with high fidelity. The paper proves that for any distribution of pure n-qubit states, a specific architecture using k encoder ancillas and n decoder ancillas matches the highest average fidelity possible with any encoder-decoder pair that is completely positive and trace preserving. This number of encoder ancillas is shown to be necessary in some cases, while isometric decoders work exactly for several families but not universally, though the practical difference remains small.

Core claim

For every distribution of pure n-qubit states, there exists a QAE with exactly k encoder ancillas and n decoder ancillas that achieves the optimal fidelity over all CPTP encoder--decoder pairs. The encoder-side statement is sharp in that source families exist for which every optimal scheme necessarily uses at least k encoder ancillas. Isometric decoders attain optimality for several analytically tractable families, yet an explicit counterexample shows they are not always sufficient, although numerical experiments indicate the performance gap is negligible.

What carries the argument

The Goldilocks QAE architecture with exactly k encoder ancillas and n decoder ancillas, which attains the information-theoretic optimum for blind single-copy compression under average infidelity.

Load-bearing premise

The inputs are pure n-qubit states and optimality is measured by average infidelity over an unknown distribution under single-copy blind compression.

What would settle it

A distribution of pure n-qubit states for which no circuit with k encoder ancillas and n decoder ancillas reaches the optimal average fidelity of the best CPTP pair, or for which a general decoder improves fidelity substantially over an isometric one.

Figures

Figures reproduced from arXiv: 2605.01258 by Chae-Yeun Park, Hyunho Cha, Jungwoo Lee.

Figure 1
Figure 1. Figure 1: General structure of a QAE that compresses an view at source ↗
Figure 2
Figure 2. Figure 2: Infidelity curves for an isometric decoder ( view at source ↗
Figure 3
Figure 3. Figure 3: Test infidelity curves for source µ1,0.1, where n ∈ {2, 3, 4} and k ∈ {1, . . . , n − 1} (with 2-sigma confidence intervals). Across all settings, the performance differences among (n, k, 0, n − k)-QAE, (n, k, 0, n)-QAE, and (n, k, k, n)-QAE are clearly visible. The converged values for (2, 1, 0, 1)-QAE and (2, 1, 1, 2)-QAE are consistent with the first-order bounds derived in Appendix E.2. 0 100 200 300 4… view at source ↗
Figure 4
Figure 4. Figure 4: Test infidelity curves for MNIST source µ2, where n ∈ {2, 3, 4} and k ∈ {1, . . . , n−1} (with 2-sigma confidence intervals). Across all settings, the performance difference between (n, k, k, n−k)- QAE and (n, k, k, n − k + 1)-QAE is negligible relative to the gap between (n, k, k, n − k)-QAE and (n, k, 0, n − k)-QAE. 10 view at source ↗
Figure 5
Figure 5. Figure 5: Logical dependency graph of the results. Dashed nodes are used as cited black boxes. view at source ↗
read the original abstract

Quantum autoencoders (QAEs) are learning architectures that compress quantum data into a low-dimensional latent state while preserving the information needed for reconstruction. We study blind single-copy compression of quantum states through a $k$-qubit bottleneck and investigate the minimal circuit width required to attain the information-theoretic optimum under average infidelity. Between the conventional architecture, which is narrow but nonuniversal, and fully general \emph{completely positive and trace preserving} (CPTP) realizations, which are universal but overparameterized, we identify a \emph{Goldilocks} regime. We prove that for every distribution of pure $n$-qubit states, there exists a QAE with exactly $k$ encoder ancillas and $n$ decoder ancillas that achieves the optimal fidelity over all CPTP encoder--decoder pairs. The encoder-side statement is sharp in that we construct source families for which every optimal scheme necessarily uses at least $k$ encoder ancillas, thereby determining the universal encoder threshold exactly. On the decoder side, we show that isometric decoders are exactly optimal for several analytically tractable source families, but we also exhibit an explicit counterexample demonstrating that decoder isometry is not universally sufficient. Nevertheless, numerical experiments indicate that the performance gap is practically negligible.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper establishes that for any ensemble of pure n-qubit states, there exists a quantum autoencoder realizing the information-theoretically optimal average fidelity (under blind single-copy compression) using precisely k encoder ancillas and n decoder ancillas. The encoder ancilla count is shown to be tight via explicit source families requiring at least k ancillas in any optimal scheme. On the decoder side, the authors prove that isometric decoders achieve optimality for several tractable families but supply a counterexample where non-isometric decoders are strictly better; numerical experiments indicate the fidelity gap remains negligible in practice.

Significance. If the existence claims hold, the work precisely characterizes the minimal ancilla resources needed to attain the CPTP optimum in blind quantum compression, identifying a resource-efficient 'Goldilocks' regime between narrow non-universal circuits and fully general overparameterized maps. The combination of existence proofs, tightness constructions, and supporting numerics provides a concrete advance in understanding the circuit-width requirements for optimal quantum data compression.

major comments (2)
  1. [Proof of the main existence theorem for QAEs] The central existence result for the decoder (that n decoder ancillas suffice to realize an optimal CPTP map from the k-qubit latent space) requires explicit justification. Standard Stinespring theory bounds the required ancilla dimension by the Kraus rank, which can reach 2^{k+n} for maps from k to n qubits; the manuscript must show why an optimal decoder can always be chosen with Kraus rank at most 2^n (independent of k). This is load-bearing for the 'exactly n decoder ancillas' claim.
  2. [Section exhibiting the isometric-decoder counterexample] The counterexample demonstrating that isometric decoders are not universally optimal should be accompanied by an explicit calculation of the fidelity gap and the Kraus rank of the optimal decoder in that case, to confirm that the n-ancilla bound is tight but still sufficient.
minor comments (2)
  1. [Introduction and Figure 1] Clarify the precise definition of 'encoder ancillas' and 'decoder ancillas' in the circuit diagrams and the first paragraph of the introduction, including whether the encoder discards qubits or uses them as part of the latent space.
  2. [Numerical results] The numerical experiments section would benefit from reporting the exact ensemble sizes, optimization hyperparameters, and error bars on the reported fidelity gaps to allow reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive suggestions. We address each major comment below and will revise the manuscript accordingly to improve clarity and completeness.

read point-by-point responses

  1. Referee: [Proof of the main existence theorem for QAEs] The central existence result for the decoder (that n decoder ancillas suffice to realize an optimal CPTP map from the k-qubit latent space) requires explicit justification. Standard Stinespring theory bounds the required ancilla dimension by the Kraus rank, which can reach 2^{k+n} for maps from k to n qubits; the manuscript must show why an optimal decoder can always be chosen with Kraus rank at most 2^n (independent of k). This is load-bearing for the 'exactly n decoder ancillas' claim.

    Authors: We agree that the decoder ancilla bound requires more explicit justification in the proof. The existence argument in Theorem 3 constructs an optimal decoder via a Stinespring isometry from the k-qubit latent space into an n-qubit output tensored with an n-qubit environment; the average-fidelity objective admits an optimal channel whose minimal Kraus rank is at most 2^n, independent of k, because the relevant ensemble of compressed states lies in a subspace that permits this reduced dilation. We will expand the proof in the revised manuscript with a dedicated paragraph deriving this Kraus-rank bound from the optimization and citing the appropriate Stinespring properties. revision: yes

  2. Referee: [Section exhibiting the isometric-decoder counterexample] The counterexample demonstrating that isometric decoders are not universally optimal should be accompanied by an explicit calculation of the fidelity gap and the Kraus rank of the optimal decoder in that case, to confirm that the n-ancilla bound is tight but still sufficient.

    Authors: We accept this recommendation. The counterexample in Section IV will be augmented in the revision with the explicit numerical fidelity gap between the optimal CPTP decoder and the best isometric decoder, together with the Kraus rank of the optimal decoder, thereby confirming both that isometry is not always sufficient and that the n-ancilla construction remains adequate. revision: yes

Circularity Check

0 steps flagged

No significant circularity; claims rest on independent existence proofs and constructions.

full rationale

The paper advances an existence theorem for QAEs with fixed ancilla counts that achieve optimal average fidelity over CPTP pairs, supported by explicit constructions for the encoder sharpness and counterexamples for decoder isometry. These steps are mathematical proofs rather than reductions to fitted parameters, self-definitions, or load-bearing self-citations. The derivation chain does not collapse any claimed result to its own inputs by construction; the results are self-contained against external benchmarks of quantum information theory.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard quantum information axioms and the definition of optimality over CPTP maps; no free parameters or new entities are introduced.

axioms (2)
  • domain assumption Quantum states under consideration are pure n-qubit states
    The proofs and sharpness statements apply specifically to distributions of pure states.
  • standard math Optimality is defined with respect to average infidelity over all CPTP encoder-decoder pairs
    The information-theoretic benchmark is the minimum infidelity achievable by any completely positive trace-preserving map.

pith-pipeline@v0.9.0 · 5523 in / 1320 out tokens · 70000 ms · 2026-05-09T15:05:15.745868+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

71 extracted references · 71 canonical work pages

  1. [1]

    Physical Review A , volume=

    Quantum variational autoencoder utilizing regularized mixed-state latent representations , author=. Physical Review A , volume=. 2025 , publisher=

    work page 2025

  2. [2]

    Linear Algebra and its Applications , volume=

    Completely positive linear maps on complex matrices , author=. Linear Algebra and its Applications , volume=. 1975 , publisher=

    work page 1975

  3. [3]

    Some Open Problems in Quantum Information Theory

    Some open problems in quantum information theory , author=. arXiv preprint arXiv:0708.1902 , year=

    work page Pith review arXiv 1902

  4. [4]

    Linear Algebra and its Applications , volume=

    On the extreme points of quantum channels , author=. Linear Algebra and its Applications , volume=. 2016 , publisher=

    work page 2016

  5. [5]

    Physical Review Research , volume=

    Group-covariant extreme and quasiextreme channels , author=. Physical Review Research , volume=. 2022 , publisher=

    work page 2022

  6. [6]

    Physical Review A , volume=

    General fidelity limit for quantum channels , author=. Physical Review A , volume=. 1996 , publisher=

    work page 1996

  7. [7]

    Annales Henri Poincar

    Ancilla dimension in quantum channel discrimination , author=. Annales Henri Poincar. 2017 , organization=

    work page 2017

  8. [8]

    Quantum Info

    Nayak, Ashwin and Sen, Pranab , title =. Quantum Info. Comput. , pages =. 2007 , publisher =

    work page 2007

  9. [9]

    2010 , publisher=

    Quantum computation and quantum information , author=. 2010 , publisher=

    work page 2010

  10. [10]

    International Journal of Quantum Information , volume=

    On the existence of physical transformations between sets of quantum states , author=. International Journal of Quantum Information , volume=. 2004 , publisher=

    work page 2004

  11. [11]

    Simulating physics with computers

    Feynman, Richard P. Simulating physics with computers. Int. J. Theor. Phys

    work page

  12. [12]

    Science , volume=

    Universal quantum simulators , author=. Science , volume=. 1996 , publisher=

    work page 1996

  13. [13]

    Quantum computing in the

    Preskill, John , journal=. Quantum computing in the. 2018 , publisher=

    work page 2018

  14. [14]

    2022 , publisher=

    Noisy intermediate-scale quantum algorithms , author=. 2022 , publisher=

    work page 2022

  15. [15]

    The Bell System Technical Journal , volume=

    A mathematical theory of communication , author=. The Bell System Technical Journal , volume=. 1948 , publisher=

    work page 1948

  16. [16]

    Physical Review A , volume=

    Quantum coding , author=. Physical Review A , volume=. 1995 , publisher=

    work page 1995

  17. [17]

    Journal of Modern Optics , volume=

    A new proof of the quantum noiseless coding theorem , author=. Journal of Modern Optics , volume=. 1994 , publisher=

    work page 1994

  18. [18]

    Physical Review Letters , volume=

    Universal quantum information compression , author=. Physical Review Letters , volume=. 1998 , publisher=

    work page 1998

  19. [19]

    Physical Review A , volume=

    Quantum universal variable-length source coding , author=. Physical Review A , volume=. 2002 , publisher=

    work page 2002

  20. [20]

    Physical Review A , volume=

    Limits for compression of quantum information carried by ensembles of mixed states , author=. Physical Review A , volume=. 1998 , publisher=

    work page 1998

  21. [21]

    Journal of Mathematical Physics , volume=

    Trading quantum for classical resources in quantum data compression , author=. Journal of Mathematical Physics , volume=. 2002 , publisher=

    work page 2002

  22. [22]

    Physical Review Letters , volume=

    Experimental demonstration of quantum source coding , author=. Physical Review Letters , volume=. 2003 , publisher=

    work page 2003

  23. [23]

    Quantum , volume=

    On the Entanglement Cost of One-Shot Compression , author=. Quantum , volume=. 2020 , publisher=

    work page 2020

  24. [24]

    Science , volume=

    Reducing the dimensionality of data with neural networks , author=. Science , volume=. 2006 , publisher=

    work page 2006

  25. [25]

    Auto-encoding variational

    Kingma, Diederik P and Welling, Max , journal=. Auto-encoding variational

    work page

  26. [26]

    International Conference on Machine Learning , pages=

    Stochastic backpropagation and approximate inference in deep generative models , author=. International Conference on Machine Learning , pages=. 2014 , organization=

    work page 2014

  27. [27]

    Nature , volume=

    Quantum machine learning , author=. Nature , volume=. 2017 , publisher=

    work page 2017

  28. [28]

    Nature Reviews Physics , volume=

    Variational quantum algorithms , author=. Nature Reviews Physics , volume=. 2021 , publisher=

    work page 2021

  29. [29]

    Nature Communications , volume=

    Barren plateaus in quantum neural network training landscapes , author=. Nature Communications , volume=. 2018 , publisher=

    work page 2018

  30. [30]

    Quantum Science and Technology , volume=

    Quantum autoencoders for efficient compression of quantum data , author=. Quantum Science and Technology , volume=. 2017 , publisher=

    work page 2017

  31. [31]

    Quantum Science and Technology , volume=

    Quantum autoencoders via quantum adders with genetic algorithms , author=. Quantum Science and Technology , volume=. 2019 , publisher=

    work page 2019

  32. [32]

    Physical Review A , volume=

    Realization of a quantum autoencoder for lossless compression of quantum data , author=. Physical Review A , volume=. 2020 , publisher=

    work page 2020

  33. [33]

    Physical Review Applied , volume=

    Noise-assisted quantum autoencoder , author=. Physical Review Applied , volume=. 2021 , publisher=

    work page 2021

  34. [34]

    Positive functions on

    Stinespring, W Forrest , journal=. Positive functions on. 1955 , publisher=

    work page 1955

  35. [35]

    Annals of Physics , volume=

    General state changes in quantum theory , author=. Annals of Physics , volume=. 1971 , publisher=

    work page 1971

  36. [36]

    Physical Review D , volume=

    Anomaly detection in high-energy physics using a quantum autoencoder , author=. Physical Review D , volume=. 2022 , publisher=

    work page 2022

  37. [37]

    Machine Learning: Science and Technology , volume=

    Quantum autoencoders with enhanced data encoding , author=. Machine Learning: Science and Technology , volume=. 2021 , publisher=

    work page 2021

  38. [38]

    Physical Review A , volume=

    Quantum circuit autoencoder , author=. Physical Review A , volume=. 2024 , publisher=

    work page 2024

  39. [39]

    Quantum Science and Technology , volume=

    Clustering and enhanced classification using a hybrid quantum autoencoder , author=. Quantum Science and Technology , volume=. 2022 , publisher=

    work page 2022

  40. [40]

    npj Quantum Information , volume=

    Quantum autoencoders using mixed reference states , author=. npj Quantum Information , volume=. 2024 , publisher=

    work page 2024

  41. [41]

    Proceedings of the IEEE , volume=

    Gradient-based learning applied to document recognition , author=. Proceedings of the IEEE , volume=. 1998 , publisher=

    work page 1998

  42. [42]

    An algorithm for the machine calculation of complex

    Cooley, James W and Tukey, John W , journal=. An algorithm for the machine calculation of complex. 1965 , publisher=

    work page 1965

  43. [43]

    Nature , volume=

    Probing many-body dynamics on a 51-atom quantum simulator , author=. Nature , volume=. 2017 , publisher=

    work page 2017

  44. [44]

    2017 , publisher=

    Quantum sensing , author=. 2017 , publisher=

    work page 2017

  45. [45]

    Physical Review Letters , volume=

    Compressibility of quantum mixed-state signals , author=. Physical Review Letters , volume=. 2001 , publisher=

    work page 2001

  46. [46]

    arXiv preprint quant-ph/0104001 , year=

    Teleportation cost and hybrid compression of quantum signals , author=. arXiv preprint quant-ph/0104001 , year=

    work page arXiv

  47. [47]

    Physical Review A , volume=

    Visible compression of commuting mixed states , author=. Physical Review A , volume=. 2001 , publisher=

    work page 2001

  48. [48]

    Compression of sources of probability distributions and density operators

    Compression of sources of probability distributions and density operators , author=. arXiv preprint quant-ph/0208131 , year=

    work page Pith review arXiv

  49. [49]

    IEEE Transactions on Information Theory , volume=

    General mixed-state quantum data compression with and without entanglement assistance , author=. IEEE Transactions on Information Theory , volume=. 2022 , publisher=

    work page 2022

  50. [50]

    Physical Review A , volume=

    Quantum rate-distortion coding , author=. Physical Review A , volume=. 2000 , publisher=

    work page 2000

  51. [51]

    Quantum rate distortion, reverse

    Datta, Nilanjana and Hsieh, Min-Hsiu and Wilde, Mark M , journal=. Quantum rate distortion, reverse. 2012 , publisher=

    work page 2012

  52. [52]

    IEEE Transactions on Information Theory , volume=

    Quantum rate-distortion coding with auxiliary resources , author=. IEEE Transactions on Information Theory , volume=. 2013 , publisher=

    work page 2013

  53. [53]

    IEEE Transactions on Information Theory , volume=

    One-shot lossy quantum data compression , author=. IEEE Transactions on Information Theory , volume=. 2013 , publisher=

    work page 2013

  54. [54]

    IEEE Transactions on Information Theory , volume=

    Quantum rate-distortion coding of relevant information , author=. IEEE Transactions on Information Theory , volume=. 2018 , publisher=

    work page 2018

  55. [55]

    Proceedings of the 25th International Conference on Machine Learning , pages=

    Extracting and composing robust features with denoising autoencoders , author=. Proceedings of the 25th International Conference on Machine Learning , pages=

    work page

  56. [56]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Representation learning: A review and new perspectives , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2013 , publisher=

    work page 2013

  57. [57]

    Zhao, Shengjia and Song, Jiaming and Ermon, Stefano , journal=. Info

    work page

  58. [58]

    Contemporary Physics , volume=

    An introduction to quantum machine learning , author=. Contemporary Physics , volume=. 2015 , publisher=

    work page 2015

  59. [59]

    Quantum Science and Technology , volume=

    Parameterized quantum circuits as machine learning models , author=. Quantum Science and Technology , volume=. 2019 , publisher=

    work page 2019

  60. [60]

    Advanced Quantum Technologies , volume=

    Expressibility and entangling capability of parameterized quantum circuits for hybrid quantum-classical algorithms , author=. Advanced Quantum Technologies , volume=. 2019 , publisher=

    work page 2019

  61. [61]

    PRX Quantum , volume=

    Connecting ansatz expressibility to gradient magnitudes and barren plateaus , author=. PRX Quantum , volume=. 2022 , publisher=

    work page 2022

  62. [62]

    Quantum Science and Technology , volume=

    Quantum variational autoencoder , author=. Quantum Science and Technology , volume=. 2019 , publisher=

    work page 2019

  63. [63]

    Advanced Quantum Technologies , volume=

    Experimental implementation of a quantum autoencoder via quantum adders , author=. Advanced Quantum Technologies , volume=. 2019 , publisher=

    work page 2019

  64. [64]

    Physical Review Letters , volume=

    Experimental realization of a quantum autoencoder: The compression of qutrits via machine learning , author=. Physical Review Letters , volume=. 2019 , publisher=

    work page 2019

  65. [65]

    Automatica , volume=

    On compression rate of quantum autoencoders: Control design, numerical and experimental realization , author=. Automatica , volume=. 2023 , publisher=

    work page 2023

  66. [66]

    Physical Review Letters , volume=

    Quantum autoencoders to denoise quantum data , author=. Physical Review Letters , volume=. 2020 , publisher=

    work page 2020

  67. [67]

    Reports on Mathematical Physics , volume=

    Linear transformations which preserve trace and positive semidefiniteness of operators , author=. Reports on Mathematical Physics , volume=. 1972 , publisher=

    work page 1972

  68. [68]

    Physical Review A—Atomic, Molecular, and Optical Physics , volume=

    Channel-state duality , author=. Physical Review A—Atomic, Molecular, and Optical Physics , volume=. 2013 , publisher=

    work page 2013

  69. [69]

    Physical Review A—Atomic, Molecular, and Optical Physics , volume=

    Theoretical framework for quantum networks , author=. Physical Review A—Atomic, Molecular, and Optical Physics , volume=. 2009 , publisher=

    work page 2009

  70. [70]

    Annals of Mathematics , volume=

    Der Massbegriff in der Theorie der kontinuierlichen Gruppen , author=. Annals of Mathematics , volume=. 1933 , publisher=

    work page 1933

  71. [71]

    Physical Review Letters , volume=

    Adaptive quantum state tomography improves accuracy quadratically , author=. Physical Review Letters , volume=. 2013 , publisher=

    work page 2013