betaFeedback

WorksPhysics and Astronomy

Shadow bootstrapping: Nearly minimal-setting pure-state tomography with shallow Clifford measurements

Abstract

We present a pure-state tomography algorithm that uses a nearly minimal number of measurement settings while achieving optimal sample complexity with nonadaptive single-copy shallow Clifford measurements. For an arbitrary nn-qubit pure state ∣ψ⟩∈Cd|ψ\rangle \in \mathbb{C}^{d}, where d=2nd=2^n, the algorithm uses O(d/ε)O(d/ε) copies and O~(1)\widetilde{O}(1) measurement settings to output a state vector ∣ψ^⟩|\hatψ\rangle satisfying 1−∣⟨ψ∣ψ^⟩∣2≤ε1-|\langleψ|\hatψ\rangle|^2\leε with high probability. Each measurement is implemented by a Clifford circuit of depth O(log⁡n)O(\log n). Reusing each measurement setting allows the data to be stored compactly as per-basis outcome histograms. Given these histograms, classical post-processing runs in O~(d)\widetilde{O}(d) time. Our key technique, shadow bootstrapping, uses the current estimate to reduce the variance of subsequent residual estimation, enabling progressively more accurate reconstruction with only a constant number of new measurement settings per stage.

Introduction

Quantum state tomography (QST) aims to obtain a classical description of an unknown quantum state. A major focus of QST research has been to minimize the number of copies required for accurate reconstruction. For an nn-qubit pure state ∣ψ⟩|\psi\rangle in dimension d=2nd = 2^n, the optimal sample complexity is Θ(d/ϵ)\Theta(d/\epsilon) for infidelity 1−∣⟨ψ^∣ψ⟩∣2≤ϵ1 - |\langle\hat{\psi}|\psi\rangle|^2 \le\epsilon and constant failure probability. This scaling can already be achieved using independent single-copy measurements [16, 29 SSW25, 17, 14].

Pure-state tomography also provides a way to learn mixed states. Recent reductions based on random purification channels [TWZ25] produce copies of the same purification of a mixed state. Learning this purification yields a classical description from which an estimate of the original mixed state can be recovered [30]. The collective operations across copies are confined to the purification step; the remaining reconstruction uses pure-state tomography.

Another important resource for QST is the number of distinct measurement settings. Each setting specifies a unitary circuit followed by computational-basis readout, and the same circuit can be applied repeatedly to fresh copies. Introducing a new setting can require circuit compilation and changes to the control program [11]. Even with optimal sample complexity and shallow circuits, using many distinct measurement settings can add substantial experimental overhead. This motivates reusing a smaller collection of measurement settings [7 ZL23, HW23, KVG+25].

Reusing the same settings also allows measurement outcomes to be stored more compactly. A raw outcome list records every observed bit string b∈{0,1}nb \in\{0, 1\}^n separately, along with the corresponding measurement basis UU. Repeated measurements in each basis can instead be summarized by a histogram, with observed bit strings as keys and their occurrence counts as values. Each histogram contains at most dd entries, and each count uses only logarithmically many bits in the number of shots. For example, the outcomes of O(d/ϵ)O(d/\epsilon) shots in a single basis can be stored in O~(d)\widetilde{O}(d) memory. These histograms can be accumulated directly during the experiment without storing a raw outcome list. Working directly with this smaller input can reduce the cost of classical reconstruction as well as its memory requirements.

Learning from few measurement bases has been studied in several settings. A constant number of bases can suffice for pure-state identification and reconstruction from exact outcome distributions [4 GCℒ+15, CHIK+16, Vor13, 24, 1]. Some of these works bound the reconstruction error when the outcome probabilities are known only approximately, but they do not provide explicit sample complexity bounds in terms of dimension and target infidelity. More recently, finite-sample protocols using randomly chosen bases have reduced the dimension dependence of the number of measurement settings, but this number still grows polynomially with inverse accuracy [2, 8].

Here, we aim to improve this accuracy dependence by taking more shots per measurement basis, while keeping both the total number of samples and the classical runtime small. Our analysis uses shadow estimators [18] constructed from randomized measurements [EFH+23], with multiple shots taken in each sampled basis. For these estimators, taking more shots reduces the noise in the empirical outcome bit string frequencies. However, the estimator can still fluctuate with the random choice of bases, even if the outcome probabilities in each sampled basis are known exactly. Establishing accurate reconstruction with few settings therefore requires controlling both the finite-shot noise and these fluctuations.

On the computational side, compact histogram input does not automatically lead to efficient classical post-processing. Explicitly forming a shadow estimator as a dense d×dd \times d matrix would already require d2d^2 entries, losing the computational benefit of the compact input. To achieve nearly linear reconstruction time, we must exploit the measurement structure to recover a length-dd state vector directly from the histograms without forming such a matrix.

In this work, we give a sample-optimal pure-state tomography algorithm using shallow measurements, a nearly minimal number of measurement settings, and nearly linear classical post-processing time from outcome histograms. To our knowledge, this is the first algorithm to achieve these guarantees simultaneously.

Theorem 1.1 (Informal, see Theorem 7.3). For any nn-qubit pure state ∣ψ⟩∈Cd|\psi\rangle\in\mathbb{C}^{d}, where d=2nd = 2^{n}, and any ϵ∈(0,1)\epsilon\in(0,1), there is a tomography algorithm using nonadaptive single-copy Clifford measurements of depth O(log⁡n)O(\log n) that outputs a unit vector ∣ψ^⟩|\hat{\psi}\rangle satisfying 1−∣⟨ψ^∣ψ⟩∣2≤ϵ1 - |\langle\hat{\psi}|\psi\rangle|^{2} \le\epsilon with probability at least 1−δ1-\delta. Its measurement and computational resources satisfy

N=O~(1),T=O(dlog⁡(1/δ)/ϵ),Timehist=O~(d),N = \widetilde{O}(1), \qquad T = O\left(d\log(1/\delta)/\epsilon\right), \qquad\mathrm{Time}_{\mathrm{hist}} = \widetilde{O}(d),

where NN is the number of sampled measurement settings, TT is the total number of copies, and Timehist\mathrm{Time}_{\mathrm{hist}} is the classical post-processing time given per-setting outcome histograms. Here, O~(⋅)\widetilde{O}(\cdot) suppresses polylogarithmic factors in dd, 1/ϵ1/\epsilon, and 1/δ1/\delta.

Related Work

Tomography from few measurement bases. Compressed-sensing approaches exploit the low-rank structure of density matrices to reconstruct quantum states from an incomplete set of observable expectation values [15, 25, 11, 22]. For Pauli observables, O~(rd)\widetilde{O}(rd) expectation values suffice to recover a rank-rr state [11]. The number of measurement bases is a related but different resource, since each basis gives access to a full distribution over dd outcomes. A complementary line of work focused directly on the number of measurement bases, showing that a constant number suffices to distinguish pure states [4, 12, 3 Vor13, 24, 1]. These works do not, however, derive explicit bounds on the number of copies required as a function of dd and the target infidelity ϵ\epsilon.

The Hamiltonian-updates approach of [2] reduces both the number of measurement settings and classical computational costs, with rigorous finite-sample guarantees. For pure-state infidelity, the bounds summarized in Table 1 require O~(1/ϵ)\widetilde{O}(1/\epsilon) settings and O~(d/ϵ2)\widetilde{O}(d/\epsilon^{2}) samples. The number of settings thus depends only polylogarithmically on dimension, but the required number of samples has suboptimal dependence on ϵ\epsilon, and the classical runtime bound is superlinear in dd. Their analyzed global Clifford and unitary 4-design implementations also do not establish an O(log⁡n)O(\log n) measurement-depth bound.

ReferenceTotal copies (TT)Settings (NN)Circuit depthClassical time
[2]O~(d/ϵ2)\widetilde{O}(d/\epsilon^{2}), δ=1/3\delta= 1/3O~(1/ϵ)\widetilde{O}(1/\epsilon)O(log⁡d)O(\log d)O~(d2/ϵ5/2)\widetilde{O}(d^{2}/\epsilon^{5/2})
[17]O(d/ϵ)O(d/\epsilon), δ=d−100\delta= d^{-100}O(d/ϵ)O(d/\epsilon)poly⁡(log⁡d,log⁡(1/ϵ))\operatorname{poly}(\log d,\log(1/\epsilon))poly⁡(d,1/ϵ)\operatorname{poly}(d,1/\epsilon)
[30]O((d+log⁡(1/δ))/ϵ)O((d+\log(1/\delta))/\epsilon)O((d+log⁡(1/δ))/ϵ)O((d+\log(1/\delta))/\epsilon)poly⁡(d)\operatorname{poly}(d)poly⁡(d,1/ϵ)\operatorname{poly}(d,1/\epsilon)
[8]O~(dlog⁡(1/δ)/ϵ)\widetilde{O}(d\log(1/\delta)/\epsilon)O~(1/ϵ)\widetilde{O}(1/\epsilon)O(log⁡log⁡d)O(\log\log d)poly⁡(d)\operatorname{poly}(d)
[GGH^{+}26]O~(dlog⁡(1/δ)/ϵ)\widetilde{O}(d\log(1/\delta)/\epsilon)O~(d/ϵ)\widetilde{O}(d/\epsilon)O(1)O(1)O~(d/ϵ)\widetilde{O}(d/\epsilon)
This workO(dlog⁡(1/δ)/ϵ)O(d\log(1/\delta)/\epsilon)O~(1)\widetilde{O}(1)O(log⁡log⁡d)O(\log\log d)O~(d)∗\widetilde{O}(d)^{*}

Table 1. Resource upper bounds for nn-qubit pure-state tomography with infidelity ϵ\epsilon and failure probability at most δ\delta, where d=2nd = 2^{n}. Here O~\widetilde{O} suppresses polylogarithmic factors in dd, 1/ϵ1/\epsilon, and 1/δ1/\delta. Settings denote upper bounds on the number of distinct measurement bases. The classical runtime marked with an asterisk (*) assumes per-basis outcome histograms as input.

More recently, [8] obtained few-setting tomography using Clifford measurements of depth O(log⁡n)O(\log n). Their pure-state specialization uses O~(d/ϵ)\widetilde{O}(d/\epsilon) samples and O~(1/ϵ)\widetilde{O}(1/\epsilon) settings. The sample complexity is nearly optimal, but the number of settings still depends polynomially on inverse accuracy, and a nearly linear runtime guarantee for the full reconstruction is not established. Our algorithm reduces the accuracy dependence of the number of settings to logarithmic while achieving optimal sample complexity and nearly linear reconstruction time from histograms.

Sample-optimal pure-state tomography. Independent single-copy measurements suffice for sample-optimal pure-state tomography [17, 30, 14]. They average estimators of the form (d+1)U∣b⟩⟨b∣U−I(d+1)U\lvert b\rangle\langle b\rvert U-I and recover a pure-state estimate from a top eigenvector of the resulting matrix. [17] transfer high matrix-moment bounds from Haar-random measurements to efficiently implementable approximate designs and apply Markov’s inequality. This gives O(d/ϵ)O(d/\epsilon) copy complexity with failure probability at most d−100d^{-100}. [30] bound the moments of scalar quadratic forms and combine Bernstein’s inequality with a covering-net argument. Their analysis achieves O((d+log⁡(1/δ))/ϵ)O((d+\log(1/\delta))/\epsilon) copies, separating the confidence overhead from the dimension dependence. Both protocols independently sample a fresh measurement basis per copy, so their upper bounds on the number of sampled measurement bases have a linear dependence on dd. For each measurement, [17] use poly⁡(n)\operatorname{poly}(n) gates, whereas [30] use poly⁡(d)\operatorname{poly}(d) gates. Neither construction establishes an O(log⁡n)O(\log n) measurement-depth bound. In classical post-processing, both reconstruct the state from a top eigenvector of a d×dd \times d empirical matrix, with poly⁡(d)\operatorname{poly}(d)-time guarantees but no nearly linear reconstruction bound O~(d)\widetilde{O}(d).

Nearly time-optimal pure-state tomography. [13] gave an algorithm using nonadaptive Pauli measurements with sample complexity O~(d/ϵ)\widetilde{O}(d/\epsilon). With a raw list of measurement outcomes as input, their algorithm runs in time O~(d/ϵ)\widetilde{O}(d/\epsilon) nearly linear in the list length. At each level of their reconstruction tree, they use Frobenius distance estimation to determine the coefficients for combining the child estimates. Their distance estimator balances the number of sampled Pauli observables against the repetitions per observable across multiple scales to achieve nearly optimal sample complexity. [19] extended this line of work to recovering the leading eigenvector of a mixed state using nonadaptive Pauli measurements, achieving sample complexity and runtime nearly linear in dd for fixed accuracy and spectral gap. However, for pure-state tomography, both works give an upper bound of O~(d/ϵ)\widetilde{O}(d/\epsilon) on the number of distinct measurement bases. Using shallow Clifford measurements, our algorithm requires only O~(1)\widetilde{O}(1) measurement settings while achieving the optimal O(d/ϵ)O(d/\epsilon) sample complexity.

Multi-shot shadow estimation and common randomized measurements. Multi-shot shadow estimation reuses each randomized measurement circuit for multiple shots and analyzes the resulting correlations and variance [HW23, ZL23]. Repeated shots suppress sampling noise, but fluctuations due to the random choice of measurement basis remain. Common randomized measurements (CRM) [VRS+24] reduce these fluctuations by comparing experimental measurements with predictions from a reference state represented, for example, as a matrix product state (MPS) or a full classical description. Using the same random bases correlates the measured and predicted contributions, reducing variance when the reference is sufficiently close to the target state. [VRS+24] bound the variance of estimators for linear and multcopy observables in terms of the reference accuracy, and numerically demonstrate variance reduction when a reference state close to ρ\rho is available. in advance. In contrast, our algorithm progressively improves an estimate of an arbitrary unknown pure state through successive stages, using only tomography data and requiring no prior classical approximation of the target state.

Small measurement ensembles for classical shadows. Previous studies construct small collections of measurement bases for classical shadow estimation [ZLZ24, YCZ26, WC24]. Our focus is how many bases are sampled and used in an experiment to achieve a target accuracy while maintaining optimal sample complexity. A small measurement collection alone does not guarantee this finite-sample efficiency. Conversely, our protocol samples and reuses only a few bases from a much larger Clifford ensemble, without requiring the ensemble itself to be small.

Technical Overview

We first explain how shadow bootstrapping halves the target infidelity at each stage using a constant number of new measurement settings. This refinement requires an initial estimate with constant overlap with the target state. We then show how to obtain such an estimate using only O(d)O(d) copies and few shallow measurement settings, so that initialization preserves the optimal dependence on dd in the total sample complexity. Finally, we explain how both procedures can be implemented efficiently from outcome histograms.

Shadow Bootstrapping

The shadow channel M\mathcal{M} maps a state ρ\rho to the mean of the projectors U†∣b⟩⟨b∣UU^{\dagger}|b\rangle\langle b|U obtained from randomized measurements [18]. Applying its inverse to each projector gives an unbiased shadow estimator of ρ\rho [Eq. (3)]. Using these estimators, shadow bootstrapping refines an initial estimate with constant overlap with the target state through successive corrections. We partition the measurement data into independent batches, indexed by t=1,…,Jt = 1, \ldots, J. At stage tt, we use the estimate ∣σt−1⟩|\sigma_{t-1}\rangle from the previous stage as a reference to construct ∣σt⟩|\sigma_t\rangle. The purpose of this reference is to reduce fluctuations due to the random choice of measurement bases, which cannot be removed merely by taking more shots per basis.

The key idea can be understood by analogy with classical variance reduction. Suppose we want to estimate E[X]\mathbb{E}[X] and have a prediction YY whose mean is known. Define Z:=X−Y+E[Y]Z := X - Y + \mathbb{E}[Y]. Then E[Z]=E[X]\mathbb{E}[Z] = \mathbb{E}[X] and

Var⁡(Z)=Var⁡(X−Y)≤E[(X−Y)2].\operatorname{Var}(Z) = \operatorname{Var}(X-Y) \le\mathbb{E}[(X-Y)^2].

Thus, when YY closely predicts XX on the same random input, ZZ estimates the same expectation E[X]\mathbb{E}[X] with small variance.

In our setting, ∣σt−1⟩|\sigma_{t-1}\rangle predicts the outcome probabilities ∣⟨z∣U∣σt−1⟩∣2|\langle z|U|\sigma_{t-1}\rangle|^2 in each measurement basis UU. We subtract these predictions from the outcome frequencies observed in the experiment and use the differences to estimate the correction ψ−σt−1\psi-\sigma_{t-1}. Adding back σt−1\sigma_{t-1} gives an unbiased estimator of ψ\psi. Because the observed and predicted contributions use the same basis, their difference becomes less sensitive to the random choice of measurement basis as σt−1\sigma_{t-1} approaches ψ\psi.

Let NtN_t denote the number of independently sampled measurement bases at stage tt, and StS_t the number of shots per basis. Our analysis gives

E[1−∣⟨ψ∣σt⟩∣2]≤CNt(E[1−∣⟨ψ∣σt−1⟩∣2]+dSt)\mathbb{E}[1-|\langle\psi|\sigma_t\rangle|^2] \le\frac{C}{N_t}\left(\mathbb{E}[1-|\langle\psi|\sigma_{t-1}\rangle|^2]+\frac{d}{S_t}\right)

for a universal constant CC. Without iterative refinement, we would take S=Θ(d)S = \Theta(d) shots per basis and N=Θ(1/ϵ)N = \Theta(1/\epsilon) measurement bases to achieve the target accuracy ϵ\epsilon. By updating the reference after each stage, we can instead choose Nt=Θ(1)N_t = \Theta(1) and St=Θ(d 2t)S_t = \Theta(d\,2^t), with constants chosen so that the upper bound on expected infidelity is halved at each stage. After O(log⁡(1/ϵ))O(\log(1/\epsilon)) stages, the estimate therefore achieves infidelity at most ϵ\epsilon with constant failure probability. All measurement bases and shot counts are fixed in advance; only the reference used in classical post-processing changes between stages.

Beyond pure states, we extend shadow bootstrapping to low-rank tomography with nonadaptive single-copy measurements, achieving nearly optimal bounds on both the number of measurement settings and the sample complexity (Appendix § 11). We expect the same variance-reduction strategy to apply more broadly to channel learning and observable estimation.

Initialization

The refinement procedure requires an initial pure-state estimate ∣σ0⟩|\sigma_0\rangle with constant overlap with the unknown state ∣ψ⟩|\psi\rangle. Initialization therefore need not achieve optimal dependence on accuracy. It must still use few shallow measurement settings and admit efficient processing of outcome histograms.

References

  1. [1]Charles H. Baldwin, Ivan H. Deutsch, and Amir Kalev. Strictly-complete measurements for bounded-rank quantum-state tomography. Physical Review A, 93(5):052105, 2016. [pp. 3, 4]DOI
  2. [2]Fernando GSL Brandão, Richard Kueng, and Daniel Stilck França. Fast and robust quantum state tomography from few basis measurements. arXiv preprint arXiv:2009.08216, 2020. [pp. 3, 4, 35]DOI
  3. [3]Claudio Carmeli, Teiko Heinosaari, Michael Kech, Jussi Schultz, and Alessandro Toigo. Stable pure state quantum tomography from five orthonormal bases. Europhysics Letters, 115(3):30001, 2016. [pp. 3, 4]DOI
  4. [4]Claudio Carmeli, Teiko Heinosaari, Jussi Schultz, and Alessandro Toigo. How many orthonormal bases are needed to distinguish all pure quantum states? The European Physical Journal D, 69(7):179, 2015. [pp. 3, 4]DOI
  5. [5]Gyungmin Cho and Dohun Kim. Sample-optimal single-copy quantum state tomography via shallow depth measurements. arXiv preprint arXiv:2509.12703, 2025. [pp. 9, 10]arxiv.org/abs/2509.12703
  6. [6]Gyungmin Cho and Dohun Kim. Shallow randomized measurement in noisy quantum devices. arXiv preprint arXiv:2504.15698, 2025. [pp. 9, 10]arxiv.org/abs/2504.15698
  7. [7]Zhuo Chen, Guoding Liu, and Xiongfeng Ma. Optimizing circuit reusing and its application in randomized benchmarking. Quantum, 9:1606, 2025. [p. 3]DOI
  8. [8]Gyungmin Cho, Changhun Oh, and Dohun Kim. Quantum state learning beyond approximate unitary designs. arXiv preprint arXiv:2609.39994, 2026. [pp. 3, 4, 5, 10, 33, 34, 35]arxiv.org/abs/2609.39994
  9. [10]Fino and Algazi. Unified matrix treatment of the Fast Walsh-Hadamard Transform. IEEE Transactions on Computers, C-25(11):1142–1146, 1976. [p. 10]DOI
  10. [11]Steven T Flammia, David Gross, Yi-Kai Liu, and Jens Eisert. Quantum tomography via compressed sensing: error bounds, sample complexity and efficient estimators. New Journal of Physics, 14(9):095022, 2012. [pp. 3, 4]
  11. [12]D Goyeneche, G Cañas, S Etcheverry, ES Gómez, GB Xavier, G Lima, and A Delgado. Five measurement bases determine pure quantum states on any dimension. Physical review letters, 115(9):090401, 2015. [pp. 3, 4]pubmed.ncbi.nlm.nih.gov/26371631
  12. [13]Sabee Grewal, Meghal Gupta, William He, Aniruddha Sen, and Mihir Singhal. Nearly time-optimal pure state tomography with Pauli measurements. arXiv preprint arXiv:2601.04444, 2026. [pp. 4, 5, 7, 16, 18]arxiv.org/abs/2601.04444
  13. [14]Madalin Guță, Jonas Kahn, Richard Kueng, and Joel A Tropp. Fast state tomography with optimal error bounds. Journal of Physics A: Mathematical and Theoretical, 53(20):204001, 2020. [pp. 3, 5]arxiv.org/abs/1809.11162
  14. [15]David Gross, Yi-Kai Liu, Steven T Flammia, Stephen Becker, and Jens Eisert. Quantum state tomography via compressed sensing. Physical review letters, 105(15):150401, 2010. [p. 4]DOI
  15. [16]Jeongwan Haah, Aram W Harrow, Zhengfeng Ji, Xiaodi Wu, and Nengkun Yu. Sample-optimal tomography of quantum states. In Proceedings of the forty-eighth annual ACM symposium on Theory of Computing, pages 913–925, 2016. [p. 3]
  16. [17]Jeongwan Haah, Robin Kothari, Ryan O’Donnell, and Ewin Tang. Query-optimal estimation of unitary channels in diamond distance. In 2023 IEEE 64th Annual Symposium on Foundations of Computer Science (FOCS), pages 363–390. IEEE, 2023. [pp. 3, 4, 5, 28]DOI
  17. [18]Hsin-Yuan Huang, Richard Kueng, and John Preskill. Predicting many properties of a quantum system from very few measurements. Nature Physics, 16(10):1050–1057, 2020. [pp. 3, 6, 9, 10, 34]DOI
  18. [19]Dutch Hansen and Jerry Li. Quantum 1-PCA with Pauli measurements in nearly linear time. arXiv preprint arXiv:2610.06808, 2026. [p. 5]arxiv.org/abs/2610.06808
  19. [20]Teiko HeinosaarI, Luca Mazzarella, and Michael M. Wolf. Quantum tomography under prior information. Communications in Mathematical Physics, 318:355–374, 2013. [p. 35]
  20. [22]Richard Kueng, Holger Rauhut, and Ulrich Terstiege. Low rank matrix recovery from rank one measurements. Applied and Computational Harmonic Analysis, 42(1):88–116, 2017. [p. 4]
  21. [24]Michael Kech and Michael M Wolf. Constrained quantum tomography of semi-algebraic sets with applications to low-rank matrix recovery. Information and Inference: A Journal of the IMA, 6(2):171–195, 2017. [pp. 3, 4]
  22. [25]Yi-Kai Liu. Universal low-rank matrix recovery from Pauli measurements. Advances in Neural Information Processing Systems, 24, 2011. [p. 4]arxiv.org/abs/1103.2816
  23. [27]Angus Lowe and Ashwin Nayak. Lower bounds for learning quantum states with single-copy measurements. ACM Transactions on Computation Theory, 17(1), March 2025. [p. 35]
  24. [28]Dmitri Maslov. Linear depth stabilizer and quantum Fourier transformation circuits with no auxiliary qubits in finite-neighbor quantum architectures. Physical Review A—Atomic, Molecular, and Optical Physics, 76(5):052310, 2007. [p. 23]
  25. [29]Ryan O’Donnell and John Wright. Efficient quantum tomography. In Proceedings of the 48th Annual ACM Symposium on Theory of Computing, pages 899–912, 2016. [p. 3]arxiv.org/abs/1508.01907
  26. [30]Angelos Pelecanos, Jack Spilecki, Ewin Tang, and John Wright. Mixed state tomography reduces to pure state tomography. arXiv preprint arXiv:2511.15806, 2025. [pp. 3, 4, 5]arxiv.org/abs/2511.15806
  27. [31]Thomas Schuster, Jonas Haferkamp, and Hsin-Yuan Huang. Random unitaries in extremely low depth. Science, 389(6755):92–96, 2025. [p. 9]DOI

Paper details

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

views · 3

downloads · 0

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

  • Physics and Astronomy

GPT-5.6, GPT-6

View statement
Looking for related papers…
Contents