Pith. sign in

REVIEW 3 major objections 3 minor 48 references

The paper establishes that finite-measurement quantum reservoir memory is governed by a delay-space quantum Fisher information matrix, and that engineered Clifford routing can make that matrix exactly diagonal, substantially improving finit

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 11:24 UTC pith:XGU5RV2O

load-bearing objection The analytical core is genuinely new and largely sound; the finite-shot benchmark claim needs extra validation before it can carry the weight the paper puts on it. the 3 major comments →

arxiv 2607.29219 v1 pith:XGU5RV2O submitted 2026-07-31 quant-ph

Fisher-Orthogonal Memory in Quantum Reservoir Computing

Ce Wang , Xingze Qiu This is my paper
classification quant-ph
keywords quantum reservoir computingquantum Fisher informationfinite-shot measurementdelay-space QFIMClifford routing orbitPauli stringsecho state propertynonlinear memory tasks
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Quantum reservoir computing stores past inputs in a driven many-body state, but with finite measurement shots, two different input histories are indistinguishable if they push the state in nearly the same direction. The paper formalizes this as a local multiparameter estimation problem via a delay-space quantum Fisher information matrix, whose off-diagonal entries measure interference between delays. It then constructs reservoirs whose delayed response operators are distinct Pauli strings routed along a Clifford orbit, making the delay-space Fisher matrix exactly diagonal and analytically programmable. Under finite-shot local Pauli readout, these reservoirs retain sharp memory windows and substantially outperform optimized random spin reservoirs for both linear delay reconstruction and nonlinear product-delay tasks. The nonlinear advantage is traced to second-order response channels that inherit the same Pauli-routing structure.

Core claim

The central claim is that quantum reservoir memory can be engineered against the finite-measurement bottleneck by making different delay coordinates Fisher-orthogonal, meaning each historical input perturbs the reservoir state along a statistically independent direction. The paper introduces the delay-space quantum Fisher information matrix and proves that for a reservoir built from a Clifford routing orbit plus a Pauli write-store block, this matrix is strictly diagonal: the tangent operator at delay k is a specific Pauli string, and its magnitude is controlled by how many damping hits that string has accumulated. Since distinct Pauli strings are Fisher-orthogonal at the identity steady sta

What carries the argument

The central object is a reservoir channel built from an input rotation, a Pauli write-store block W_r(Q,G), and a Clifford routing generator V_K that cyclically permutes Pauli strings, P_{j+1}=V_K P_j V_K^†. The routing is chosen so that the orbit is maximal: K=2^N+1 steps and every nonidentity Pauli orbit receives a damping hit, which guarantees the echo state property. Because distinct Pauli strings are orthogonal in the Fisher metric at the identity steady state, the delay-space QFIM is exactly diagonal; the binary damping word η (which routed strings anticommute with the damping string G) sets the fading profile, H^(K)=π^2(1−r^2) r^{2n_{k−1}} δ_{kl}. The write-store anticommutation {Q,G}

Load-bearing premise

The load-bearing assumption is that the Fisher-orthogonality proven for infinitesimal perturbations around the maximally mixed state continues to hold at the finite input amplitudes used in the numerical tasks.

What would settle it

Compute the exact delay-space QFIM (or the observed classical Fisher matrix of the local Pauli readout) for the Clifford reservoirs at the finite input amplitude δw=0.25 used in the benchmarks. If off-diagonal entries become comparable to the diagonal ones at that operating point, or if the advantage over the optimized random spin reservoir vanishes when δw is taken toward zero, the claim that Fisher-orthogonal routing is the cause of the improvement would be refuted.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the Fisher-orthogonal construction holds, finite-shot quantum reservoir computers can be designed analytically rather than by random search: the memory window and fading profile are read off from the damping word.
  • The diagonal QFIM means each delayed input can be estimated without interference from other delays, so a linear readout can in principle saturate per-delay estimation precision.
  • The construction guarantees the echo state property for any nonidentity damping string, so the reservoir forgets old inputs at a controlled rate with no fine-tuning.
  • Second-order (product) responses follow the same Pauli route, so nonlinear memory tasks inherit the same orthogonal structure and remain accessible to linear Pauli readout.
  • The observed improvement over optimized random spin reservoirs under finite-shot local Pauli readout, if correct, provides a practical route to measurement-efficient quantum reservoir computing on near-term devices.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper, the same design principle may apply to other readout settings such as collective or adaptive measurements, where Fisher orthogonality could yield even larger gains because the off-diagonal readout noise is already suppressed.
  • The algebraic run-length limit on transparent memory segments suggests a possible speed limit on how long a memory window can be made undamped in this class of reservoirs; testing non-Clifford routing would clarify whether that limit is fundamental or an artifact of the Clifford structure.
  • A direct experimental test would be to measure the classical Fisher matrix of the local Pauli outcomes for a Clifford reservoir at finite input amplitude: if off-diagonal entries stay small, the design principle is validated operationally, and if not, the advantage is tied to the operating point.
  • The broader idea of making delayed inputs statistically independent in a reservoir may transfer to classical reservoir computing or neuromorphic hardware, where finite-sample identifiability is also the practical bottleneck.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 3 minor

Summary. The paper introduces a delay-space quantum Fisher information matrix (QFIM) for quantum reservoir computing and uses it as a design target. For a single-qubit reservoir it identifies a one-parameter Fisher-cycle family and proves, via a Gill–Massar-based argument in the Supplemental Material, that this family achieves the global optimum for three-delay weighted estimation. It then generalizes the construction to multi-qubit reservoirs using Clifford routing orbits and Singer-cycle Pauli algebra, deriving a strictly diagonal QFIM (Eq. S4.36) with programmable damping, an ESP theorem, and second-order response channels for product delays. The theoretical construction is benchmarked in numerical simulations against random transverse-field Ising reservoirs for linear and product-delay tasks under exact and finite-shot local Pauli readout, reporting that the Clifford reservoirs retain sharp memory windows and substantially improve over the Ising baseline.

Significance. If the central claim holds, the paper provides a genuinely useful design principle: reservoir memory can be engineered so that different delays correspond to Fisher-orthogonal directions, reducing finite-measurement interference. The strengths are the analytical single-qubit optimality proof (S3.2), the closed-form diagonal QFIM for Singer cycles (S4.4), the algebraic ESP condition (S4.5), and the explicit second-order response theory for product tasks (S4.8). These are substantive and mostly checkable. The main risk is that the headline numerical advantage over Ising is demonstrated only at finite amplitude and without statistical error bars, while the theoretical guarantee is derived at the infinitesimal, maximally-mixed operating point.

major comments (3)
  1. [S4.4 / S5.3 (Eq. S4.36 vs δw=0.25)] The strict diagonality of H^{(K)} is derived at ρ*=I/d for infinitesimal δs_k. The task benchmarks use δw=0.25, so s_t∈[0.25,0.75]; at these amplitudes the write-store channel is not unital, the state is not maximally mixed, and the exact SLD metric need not make distinct Pauli strings orthogonal. The paper lists 'infinitesimal input perturbations' as an idealization in S5.3 but then attributes the Fig. 4 finite-shot improvement to the Fisher-orthogonal design without verifying the transfer. Please add a numerical check of the exact QFIM (or Fisher overlap matrix) at the operating amplitude, or benchmark at smaller δw and show the advantage is governed by the linear-response structure.
  2. [S5.3 / Fig. 4] The finite-shot panels (b),(d) report pointwise optimized NRMSE without error bars or repeated seeds. The Ising baseline is selected online over 1000 accepted reservoirs across delays and tasks, so multiple-comparisons bias is possible. To support 'substantially improve', report mean ± standard error over independent realizations (e.g., different random input sequences and shot samples) and state the number of repeats.
  3. [S5.3 finite-shot noise model] Clifford finite-shot features are drawn from the full multinomial eight-outcome distribution, whereas the Ising finite-shot features are drawn from a Gaussian approximation to the multinomial using exact covariance. If the Gaussian approximation is inaccurate for small counts or nonlinear features, the comparison is biased. Use the same sampling protocol for both reservoirs, or include a numerical validation that the Gaussian approximation reproduces the multinomial statistics for the Ising case.
minor comments (3)
  1. [Eq. (4) and S1.4] The truncation length L in Eq. (4) is a Taylor-expansion cutoff, while the benchmarks use the full recurrent history. Clarify this distinction in the main text to avoid implying that the benchmarks are limited to L delays.
  2. [S5.3] The text says 'thousands of choices' for the Ising baseline, but the reported library is 1000 accepted reservoirs. Align the wording with the actual simulation.
  3. [Main text after Eq. (14)] Consider qualifying 'strictly diagonal' as 'strictly diagonal at the linear-response fixed point' to avoid overstatement, given the finite-amplitude benchmarks.

Circularity Check

0 steps flagged

No significant circularity: the diagonal QFIM is a derived consequence of the Clifford/Singer construction, not a fitted target, and the finite-shot task benchmarks are independent held-out tests.

full rationale

The paper's central derivation is self-contained rather than circular. The delay-space QFIM is defined from standard SLD Fisher information, and the claimed diagonal form H^{(K)}_{\eta_K}(r)_{kl} = \pi^2(1-r^2) r^{2n_{k-1}(\eta_K)} \delta_{kl} (Eq. S4.36) is derived from the explicit Pauli-orbit construction: the write-store block injects perturbation along Q, routing maps it to distinct Pauli strings P_k, and the maximally-mixed fixed point makes the QFIM proportional to the Hilbert-Schmidt inner product, so distinct Pauli strings are automatically Fisher-orthogonal. This is a mathematical consequence of the ansatz, not a fit to the benchmarks. The single-qubit Gill-Massar analysis is likewise a lower-bound proof with saturation conditions, not an extraction of the answer from the data. The numerical benchmarks use finite-amplitude inputs (\delta w=0.25), a full recurrent history, local Pauli readout, and held-out test errors; the Clifford parameters (Q,G,r,B) are chosen on validation data and the reported errors are on test data, so the comparison is not a fitted quantity being relabeled as a prediction. The few self-citations (e.g., refs. [24,35]) are background references and are not load-bearing. The paper also explicitly lists its linear-response idealizations and then separately tests finite-amplitude behavior, which is an acknowledged limitation or risk of transfer, not a circular step. No equation is shown to reduce to its own input, no fitted parameter is renamed as a prediction, and no load-bearing claim rests solely on a self-citation. Therefore the appropriate score is 0.

Axiom & Free-Parameter Ledger

3 free parameters · 6 axioms · 0 invented entities

No new physical entities are introduced. The main hand-set degrees of freedom are the fading parameter r, the finite-amplitude scale δw, and the Ising baseline sampling ranges. The central construction rests on standard quantum estimation theory plus the assumption that Pauli-string orthogonality at the maximally-mixed fixed point remains the right design criterion away from the linear regime.

free parameters (3)
  • r (fading parameter) = optimized per task/delay; r_opt ≈ 0.565 for uniform GM cost
    Controls the write-store damping trade-off in W_r. It is a design parameter tuned on validation data (grid 0.08:0.06:0.98) and analytically in the GM cost; it is not fixed by independent data.
  • δw (input amplitude) = 0.25
    Sets the finite-amplitude operating point for task benchmarks (S5.3). This value is chosen by hand and determines the nonlinearity strength and the effective shot-noise resource R=N_shot(δw)^2; results may depend on it.
  • Ising baseline sampling ranges = J∈[-1,1], h_x,h_z∈[-4,4], τ∈[0.05,10]
    Define the pool of random Ising reservoirs against which Clifford reservoirs are compared. The 'optimized baseline' claim depends on these ranges being sufficiently broad and representative.
axioms (6)
  • standard math SLD quantum Fisher information and Gill-Massar bound are valid for single-copy local estimation
    Used throughout (S2) to define H and C_GM; standard textbook results.
  • standard math QFI data-processing inequality under CPTP maps
    Used in S3 Step 2 to bound write strength h_w against contraction spectrum.
  • domain assumption Input encoding |ψ_s> has QFI π^2 at s0=1/2
    The sinusoidal encoding Eq. (1) is the chosen input map; its QFI is π^2. The proof of the write-retention bound assumes this specific encoding and a pure input qubit.
  • domain assumption Reservoir fixed point is maximally mixed ρ*=I/d, so QFIM reduces to Hilbert-Schmidt inner product
    The Clifford write-store channel is unital, giving ρ*=I/d; the Fisher-orthogonality of distinct Pauli strings and Eq. (S4.36) rely on this. Non-unital reservoirs would not inherit the diagonal structure.
  • ad hoc to paper Finite-amplitude benchmarks inherit the linear-response Fisher-orthogonality design
    The diagonal QFIM is proven only for infinitesimal perturbations around s0 (Eq. S4.36); the task benchmarks use δw=0.25, where the exact QFIM is not diagonal. The paper asserts rather than proves that the finite-shot advantage transfers.
  • domain assumption The random Ising scan is a representative optimized non-Clifford baseline
    The comparison in Fig. 4 relies on 1000 accepted Ising samples with selected ranges; the pointwise validation-best envelope is treated as the optimized baseline. A different sampling distribution or more candidates could change the margin.

pith-pipeline@v1.3.0-daily-deepseek · 24957 in / 19388 out tokens · 179818 ms · 2026-08-03T11:24:39.581894+00:00 · methodology

0 comments
read the original abstract

Quantum reservoir computing processes temporal information through driven many-body dynamics, but its performance is ultimately limited by how accurately past inputs can be extracted from finite measurements. Here we formulate this limitation as a local multiparameter estimation problem and introduce a delay-space quantum Fisher information matrix to quantify the distinguishability of memory traces. This perspective identifies Fisher-orthogonal memory as a measurement-efficient design principle: different delays should perturb the reservoir state along statistically independent directions. We first analyze the single-qubit limit using the Gill--Massar bound, revealing an optimal write-store-routing trade-off. Guided by this structure, we construct solvable multi-qubit reservoirs based on Clifford routing orbits and Singer-cycle Pauli algebra. The resulting dynamics yield diagonal, analytically programmable Fisher memory matrices with controlled dissipation profiles. Under finite-shot local Pauli readout, these reservoirs retain sharp memory windows and substantially improve over optimized random Ising reservoirs for both linear delay reconstruction and nonlinear product-delay tasks. The nonlinear advantage is traced to second-order response channels that inherit the same Pauli-routing structure. Our results provide an analytically controlled route toward measurement-efficient quantum reservoir computing.

Figures

Figures reproduced from arXiv: 2607.29219 by Ce Wang, Xingze Qiu.

Figure 1
Figure 1. Figure 1: FIG. 1. Single-qubit GM cost for the weighted three-delay [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: for the K = 5 example). As the Pauli string Pj propagates along the CRO driven by VK, its dynamical envelope is fully determined by the damping string G: at each step, Pj picks up a fading factor r if it anticommutes with G (a hit), and passes through undamped otherwise. This sequence of hits directly defines the binary damp￾ing pattern ηK(Q, G) (ηK,j = 1 for a hit and 0 otherwise) visually encoded in [PI… view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Fisher-cost landscape of Clifford damping patterns. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Task-specific prediction performance under a sin [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

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

48 extracted references · 8 linked inside Pith

  1. [1]

    The rotationV cyc generates the three-axis routeZ7→ X7→Y7→Z, withH Had the Hadamard gate andS= diag(1,i)

    The block form ofW r is written in the input-qubit basis, whilethePaulioperatorsactonthereservoirqubit. The rotationV cyc generates the three-axis routeZ7→ X7→Y7→Z, withH Had the Hadamard gate andS= diag(1,i). The induced quantum channel factorizes the write- store step from the subsequent routing dynamics as Φs,r(ρ) =V cycΨs,r(ρ)V † cyc. At the operating...

  2. [2]

    Fujii and K

    K. Fujii and K. Nakajima, Physical Review Applied8, 024030 (2017)

    2017

  3. [3]

    Ghosh, A

    S. Ghosh, A. Opala, M. Matuszewski, T. Paterek, and T. C. H. Liew, npj Quantum Information5, 35 (2019)

    2019

  4. [4]

    Mujal, R

    P. Mujal, R. Martínez-Peña, J. Nokkala, J. García-Beni, G. L. Giorgi, M. C. Soriano, and R. Zambrini, Advanced Quantum Technologies4, 2100027 (2021)

    2021

  5. [5]

    Nakajima, K

    K. Nakajima, K. Fujii, M. Negoro, K. Mitarai, and M. Kitagawa, Physical Review Applied11, 034021 (2019)

    2019

  6. [6]

    Chen and H

    J. Chen and H. I. Nurdin, Quantum Information Process- ing18, 198 (2019)

    2019

  7. [7]

    Nokkala, R

    J. Nokkala, R. Martínez-Peña, G. L. Giorgi, V. Parigi, M. C. Soriano, and R. Zambrini, Communications Physics 4, 53 (2021)

    2021

  8. [8]

    Martínez-Peña and J.-P

    R. Martínez-Peña and J.-P. Ortega, Physical Review E 107, 035306 (2023)

    2023

  9. [9]

    Sannia, R

    A. Sannia, R. Martínez-Peña, M. C. Soriano, G. L. Giorgi, and R. Zambrini, Quantum8, 1291 (2024)

    2024

  10. [10]

    Kubota, Y

    T. Kubota, Y. Suzuki, S. Kobayashi, Q. H. Tran, N. Ya- mamoto, and K. Nakajima, Physical Review Research5, 023057 (2023)

    2023

  11. [11]

    Martínez-Peña and J.-P

    R. Martínez-Peña and J.-P. Ortega, Physical Review E 111, 065306 (2025)

    2025

  12. [12]

    Schütte, N

    N.-E. Schütte, N. Götting, H. Müntinga, M. List, D. Brunner, and C. Gies, Expressivity of quantum reser- voir computers (2025), arXiv:2501.15528 [quant-ph]

    Pith/arXiv arXiv 2025

  13. [13]

    Čindrak, L

    S. Čindrak, L. Giebeler, N. Götting, C. Gies, and K. Lüdge, Memory-nonlinearity trade-off across quantum reservoir computing frameworks (2026), arXiv:2603.21371 [quant-ph]

    arXiv 2026

  14. [14]

    Negoro, K

    M. Negoro, K. Mitarai, K. Fujii, K. Nakajima, and M. Kitagawa, Machine learning with controllable quan- tum dynamics of a nuclear spin ensemble in a solid (2018), arXiv:1806.10910 [quant-ph]

    Pith/arXiv arXiv 2018

  15. [15]

    Suzuki, Q

    Y. Suzuki, Q. Gao, K. C. Pradel, K. Yasuoka, and N. Ya- mamoto, Scientific Reports12, 1353 (2022)

    2022

  16. [16]

    Martínez-Peña, J

    R. Martínez-Peña, J. Nokkala, G. L. Giorgi, R. Zam- brini, and M. C. Soriano, Cognitive Computation15, 1440 (2023)

    2023

  17. [17]

    Y. Hou, J. Hua, Z. Wu, W. Xia, Y. Chen, X. Li, Z. Li, X. Peng, and J. Du, Physical Review Letters136, 120602 (2026)

    2026

  18. [18]

    Mujal, R

    P. Mujal, R. Martínez-Peña, G. L. Giorgi, M. C. Soriano, and R. Zambrini, npj Quantum Information9, 16 (2023)

    2023

  19. [19]

    Yasuda, Y

    T. Yasuda, Y. Suzuki, T. Kubota, K. Nakajima, Q. Gao, W. Zhang, S. Shimono, H. I. Nurdin, and N. Yamamoto, Quantum reservoir computing with re- peated measurements on superconducting devices (2023), arXiv:2310.06706 [quant-ph]

    Pith/arXiv arXiv 2023

  20. [20]

    Kobayashi, K

    K. Kobayashi, K. Fujii, and N. Yamamoto, PRX Quan- tum5, 040325 (2024)

    2024

  21. [21]

    Monomi, W

    T. Monomi, W. Setoyama, and Y. Hasegawa, Feedback- enhanced quantum reservoir computing with weak mea- 6 surements (2025), arXiv:2503.17939 [quant-ph]

    Pith/arXiv arXiv 2025

  22. [22]

    C. Zhu, P. J. Ehlers, H. I. Nurdin, and D. Soh, npj Quan- tum Information11, 195 (2025)

    2025

  23. [23]

    Franceschetto, M

    G. Franceschetto, M. Płodzień, M. Lewenstein, A. Acín, and P. Mujal, Physical Review X16, 021002 (2026)

    2026

  24. [24]

    J. Chen, H. I. Nurdin, and N. Yamamoto, Physical Re- view Applied14, 024065 (2020)

    2020

  25. [25]

    W. Xia, J. Zou, X. Qiu, F. Chen, B. Zhu, C. Li, D.-L. Deng, and X. Li, Science Bulletin68, 2321 (2023)

    2023

  26. [26]

    E. L. Connerty, E. N. Evans, G. Angelatos, and V. Narayanan, Communications Physics 10.1038/s42005- 026-02652-1 (2026)

    doi:10.1038/s42005- 2026

  27. [27]

    Paparelle, J

    I. Paparelle, J. Henaff, J. García-Beni, E. Gillet, D. Mon- tesinos, G. L. Giorgi, M. C. Soriano, R. Zambrini, and V. Parigi, Nature Photonics20, 413 (2026)

    2026

  28. [28]

    Y. Kora, H. Zadeh-Haghighi, T. C. Stewart, K. Heshami, and C. Simon, Physical Review A110, 042416 (2024)

    2024

  29. [29]

    Karimi, H

    A. Karimi, H. Zadeh-Haghighi, Y. Kora, and C. Si- mon, The role of entanglement in quantum reservoir com- puting with coupled kerr nonlinear oscillators (2025), arXiv:2508.11175 [quant-ph]

    Pith/arXiv arXiv 2025

  30. [30]

    Askari, Y

    S. Askari, Y. Kora, and C. Simon, Spin-network quantum reservoir computing with distributed inputs: The role of entanglement (2025), arXiv:2511.04900 [quant-ph]

    arXiv 2025

  31. [31]

    Nahum, S

    A. Nahum, S. Vijay, and J. Haah, Physical Review X8, 021014 (2018)

    2018

  32. [32]

    Xiong, Z

    W. Xiong, Z. Holmes, A. Angrisani, Y. Suzuki, T. Chotibut, and S. Thanasilp, Role of scrambling and noise in temporal information processing with quantum systems (2025), arXiv:2505.10080 [quant-ph]

    Pith/arXiv arXiv 2025

  33. [33]

    Kobayashi and Y

    K. Kobayashi and Y. Motome, Physical Review Letters 136, 040602 (2026)

    2026

  34. [34]

    Sannia, G

    A. Sannia, G. L. Giorgi, and R. Zambrini, Exponential concentration and symmetries in quantum reservoir com- puting (2025), arXiv:2505.10062 [quant-ph]

    Pith/arXiv arXiv 2025

  35. [35]

    M. N. Ivaki, M. Karjula, and T. Ala-Nissila, Physical Review A113, L060401 (2026)

    2026

  36. [36]

    W. Xia, S. Cao, X. Qiu, and X. Li, Quantum magic and non-commutativity as computational resources in quan- tum reservoir computing (2026), arXiv:2607.12035 [quant- ph]

    Pith/arXiv arXiv 2026

  37. [37]

    C. W. Helstrom,Quantum Detection and Estimation Theory(Academic Press, New York, 1976)

    1976

  38. [38]

    S. L. Braunstein and C. M. Caves, Physical Review Let- ters72, 3439 (1994)

    1994

  39. [39]

    M. G. A. Paris, International Journal of Quantum Infor- mation7, 125 (2009)

    2009

  40. [40]

    Szczykulska, T

    M. Szczykulska, T. Baumgratz, and A. Datta, Advances in Physics: X1, 621 (2016)

    2016

  41. [41]

    S. Ragy, M. Jarzyna, and R. Demkowicz-Dobrzański, Physical Review A94, 052108 (2016)

    2016

  42. [42]

    See Supplemental Material for derivations of the delay- space QFIM and Gill–Massar cost, a rigorous proof of the single-qubit Fisher-cycle optimum, the Clifford/Singer construction and its echo-state properties, second-order response theory, and numerical protocols

  43. [43]

    R. D. Gill and S. Massar, Physical Review A61, 042312 (2000)

    2000

  44. [44]

    Zhu, Physical Review Letters116, 040501 (2016)

    H. Zhu, Physical Review Letters116, 040501 (2016)

    2016

  45. [45]

    O. Kern, K. S. Ranade, and U. Seyfarth, Journal of Physics A: Mathematical and Theoretical43, 275305 (2010)

    2010

  46. [46]

    Hahto and J

    M. Hahto and J. Nokkala, New Journal of Physics27, 094510 (2025)

    2025

  47. [47]

    Liu, Q.-X

    D.-S. Liu, Q.-X. Jie, C.-L. Zou, X.-F. Ren, and G.-C. Guo, Physical Review A113, 042401 (2026). 7 Supplemental Material for Fisher-Orthogonal Memory in Quantum Reservoir Computing S1. QUANTUM FISHER MEMORY MATRIX We first collect the local estimation framework used in the main text. Each classical inputs∈[0,1]is encoded into a fresh input qubit, |ψs⟩= cos...

    2026

  48. [48]

    Dividing Eq

    Hence (∇px)T H−1(∇px)≤Tr(ρE 2 x)−p 2 x.(S2.5) Since0≤E x ≤I, we haveE 2 x ≤λ max(Ex)Ex ≤Tr(E x)Ex, and henceTr(ρE2 x)≤Tr(E x)px. Dividing Eq. (S2.5) by px and summing over the outcomes therefore gives the complete bound in one step, Tr H−1I(M) = X x (∇px)T H−1(∇px) px ≤ X x Tr(ρE2 x) px −p x ≤ X x [Tr(Ex)−p x] =d sys −1.(S2.6) The last form is a compact w...

    2002