Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Quantum Natural Gradient

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read The paper defines quantum natural gradient descent on the Fubini-Study metric, shows its block-diagonal form costs one measurement per commuting layer, and demonstrates faster convergence on 7-, 9-, and 11-qubit circuits.

desk verdict The paper that made quantum natural gradient a standard variational-circuit tool: the theoretical core is sound and the block-diagonal QGT estimator is a real contribution, but the empirical 'dramatically fewer iterations' claim rests on single-run curves and needs statistical support. read the letter →

arxiv 1909.02108 v3 pith:2AJYCZNR submitted 2019-09-04 quant-ph cs.LGstat.ML

James Stokes , Josh Izaac , Nathan Killoran , Giuseppe Carleo This is my paper · ORCID
classification quant-phcs.LGstat.ML MSC 81P68 PACS 03.67.-a
keywords quantumnaturalgradientFubini-Studymetricgeometrictensorvariationalcircuitsparameter-shiftruleimaginary-timeevolutionoptimizationdescent
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Variational quantum circuits are trained by nudging parameters downhill, but the usual Euclidean notion of "downhill" depends on how the circuit happens to be parameterized. This paper argues that the natural geometry lives on the space of pure quantum states, and that the correct descent direction is given by the Fubini-Study metric, the real part of the quantum geometric tensor. The resulting quantum natural gradient update is invariant under reparameterization and, in the classical limit, reduces to the Fisher information matrix. The paper also shows that for circuits whose parametrized gates commute within each layer, the block-diagonal metric can be estimated with a single measurement setting per layer, and demonstrates on 7-, 9-, and 11-qubit circuits that this preconditioner reaches the ground-state objective in far fewer iterations than vanilla gradient descent or Adam.

What carries the argument

The load-bearing object is the quantum geometric tensor, $G_{ij}=\langle\partial_i\psi|\partial_j\psi\rangle-\langle\partial_i\psi|\psi\rangle\langle\psi|\partial_j\psi\rangle$, whose real part is the Fubini-Study metric on the space of pure states. The optimizer uses the pseudo-inverse $g^+$ of this metric as a preconditioner, solving $g(\theta_t)(\theta_{t+1}-\theta_t)=-\eta\nabla L(\theta_t)$ instead of materializing the inverse. For layered circuits the metric is approximated block-diagonally: whenever all parametrized gates in a layer commute, the generators $K_i$ satisfy $[K_i,K_j]=0$, the product $K_iK_j$ is Hermitian, and each block reduces to a covariance $\langle\psi_l|K_iK_j|\psi_l\rangle-\langle\psi_l|K_i|\psi_l\rangle\langle\psi_l|K_j|\psi_l\rangle$ that can be read off from one measurement in the shared eigenbasis of the layer's generators.

What would settle it

Simulate the exact update $\theta_{t+1}=\theta_t-\eta g^+(\theta_t)\nabla L(\theta_t)$ for the same physical two-qubit state written in two different parameter coordinate systems; reparametrization invariance predicts identical trajectories on the state manifold, so any divergence in the state-space path would falsify the claim.

Watch

Extended reading notes

Core claim

The central claim is that the dynamics $\theta_{t+1}=\theta_t-\eta g^+(\theta_t)\nabla L(\theta_t)$, with $g_{ij}(\theta)=\operatorname{Re}[G_{ij}(\theta)]$, is the reparameterization-invariant steepest descent on the pure-state manifold. Here $G_{ij}(\theta)=\langle\partial_i\psi_\theta|\partial_j\psi_\theta\rangle-\langle\partial_i\psi_\theta|\psi_\theta\rangle\langle\psi_\theta|\partial_j\psi_\theta\rangle$ is the quantum geometric tensor. For a circuit factored into layers whose parametrized gates commute, the $l$-th block of the metric becomes $G^{(l)}_{ij}=\langle\psi_l|K_iK_j|\psi_l\rangle-\langle\psi_l|K_i|\psi_l\rangle\langle\psi_l|K_j|\psi_l\rangle$, which is Hermitian and therefore equals its own real part; this block can be evaluated with a single commuting measurement per layer, and a diagonal variant with even less overhead. Numerical experiments on 7-, 9-, and 11-qubit circuits with five layers, and on 9-qubit circuits with three to six layers, show the quantum natural gradient reaching the target in far fewer iterations than vanilla gradient descent or Adam, with the advantage growing with qubit number and persisting with depth.

Load-bearing premise

The efficient single-measurement estimator assumes that every parametrized gate inside a circuit layer commutes with the others in that layer; if a layer mixes non-commuting gates, the block-diagonal expression is no longer the Fubini-Study metric and the one-setting scheme does not apply.

Editorial extensions

If this is right

  • On the tested circuits, quantum natural gradient reaches the minimum in a small number of iterations, while vanilla gradient descent fails to find it on the 7-, 9-, and 11-qubit benchmarks; Adam finds it but needs more iterations.
  • Each natural-gradient step costs $2d+L$ quantum evaluations: $2d$ for parameter-shift gradients and $L$ for the metric blocks, one per layer.
  • The diagonal approximation to the metric gives convergence close to the block-diagonal version with greatly reduced classical processing, bringing wall time down to roughly the level of vanilla gradient descent.
  • Because the metric does not depend on the observable being minimized, the preconditioner reflects state-space geometry rather than loss curvature, making it complementary to historical-averaging methods such as Adam.
  • In the zero-step-size limit, the quantum natural gradient dynamics coincides with imaginary-time evolution restricted to the variational subspace.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A direct untested transfer would be to QAOA-style MaxCut circuits, whose layers contain commuting $ZZ$ interactions but whose cost landscape is combinatorial rather than the two-local observable used here; the iteration-count advantage is not guaranteed to carry over.
  • Because the metric blocks are independent of the loss Hamiltonian, the same $L$ measurements per layer could be cached and reused when one circuit family is trained on several objectives, amortizing the per-step cost across tasks.
  • For circuits with non-commuting gates inside a layer, a natural extension is to split each layer into maximal commuting subsets and estimate a metric block per subset; this would preserve the geometric interpretation at the price of more than one measurement setting per layer.
  • Combining quantum natural gradient with Adam-style historical averaging is explicitly left open by the paper, and the supplementary results with a natural-gradient-modified Adam suggest such hybrids may converge even faster.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces a quantum analogue of natural gradient descent for variational quantum circuits. The update replaces the Euclidean metric in the gradient-descent proximal problem with the real part of the Quantum Geometric Tensor (the Fubini-Study metric pulled back to the parameter manifold), giving a reparametrization-invariant update (Eqs. 10-13). The authors derive a block-diagonal approximation of the metric for layered circuits, show that each block can be expressed in terms of expectation values of products of generators (Eq. 27), and provide a measurement scheme requiring a single state preparation per layer when the generators within a layer commute (Secs. 2.5-2.6). They also establish an equivalence with imaginary-time evolution (Sec. A.2) and report numerical comparisons on a ladder ansatz with 7-11 qubits (Figs. 1-4).

Significance. If the results hold, the paper makes a valuable contribution: it provides a principled, parameterization-invariant geometry for variational quantum optimization, an explicit and implementable way to estimate the metric on quantum hardware (with open-source PennyLane implementation), and a proof-of-concept that geometric preconditioning can reduce the number of optimization iterations. The derivation of the metric blocks is clean and internally consistent, and the imaginary-time equivalence is a genuine structural insight. The main weakness is that the empirical evidence for "dramatically fewer iterations" rests on single runs with fixed hyperparameters, so the practical advantage is not yet statistically established.

major comments (3)
  1. [Section 3, Figures 1-4] The paper's central practical claim—that QNG reaches the minimum in dramatically fewer iterations than vanilla GD and Adam—is supported only by single-run cost curves. Each panel shows a single trajectory from one random initialization (the same initialization as Ref. [24]) with no error bars, no multiple seeds, and no summary statistics; Figs. 3-4 in Sec. A.5 have the same limitation for wall time and number of quantum evaluations. Because single trajectories can be unrepresentative, the numerical advantage is not established. I request multiple random initializations with median and quantile bands and, for the shot-noise runs, repeated evaluations to characterize variance.
  2. [Figures 1-4 and Sec. A.5] The comparison against Adam uses a single learning rate, 0.01, for every optimizer. Adam's default suggested rate in Ref. [18] is 0.001, and optimizer performance is strongly sensitive to this hyperparameter. The observed iteration advantage may therefore reflect an unfavorable baseline setting rather than a property of the geometry. Please report a small learning-rate sweep or sensitivity curves for each optimizer and for both the block and diagonal QNG variants, and state the chosen rates explicitly in the captions.
  3. [Section 2.5, Eqs. (19)-(28)] The block-diagonal metric is equal to the Fubini-Study metric block, and the single-setting measurement is valid, only when all parameterized generators within each layer commute, so that K_i K_j is Hermitian. While the paper states this as a defining property of a layer, the abstract and introduction promise an efficient algorithm for parametrized quantum circuits without this qualification. Please state the scope in the abstract and introduction, and indicate which standard ansatz families (e.g., QAOA, hardware-efficient circuits with one rotation per qubit per layer, and circuits with multi-qubit parametric entanglers) satisfy or violate the condition; for families violating it, the presented scheme does not apply and the approximation error relative to the full QNG update is not quantified.
minor comments (5)
  1. [Section 3, Fig. 1] The text states that the block-diagonal approximation is omitted for n = 11, but the figure legend and caption still list the block-diagonal QNG curve; please reconcile the legend or the text.
  2. [Eq. (39)] The notation "arccos2" should read "arccos^2"; the same notational issue should be checked wherever the squared arccos appears.
  3. [Sec. 2.1] The statement that CP^{N-1} possesses "a unique unitarily-invariant metric tensor" should be qualified as unique up to an overall scale; otherwise it is technically imprecise.
  4. [Sec. 2.4] The interval conventions in Eqs. (15)-(17) are terse; a short example or an explanatory sentence for (l:L] and [1:l) would improve readability.
  5. [Secs. 2.5-2.6] The notation drops the layer index from the generator K, which is convenient but occasionally confusing in Eqs. (25)-(27) when comparing different layers; consider retaining a layer subscript.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the QNG update, block-diagonal metric, and imaginary-time equivalence are derived from the Fubini-Study geometry and explicit commuting-layer assumptions, not from the objective or fitted parameters.

full rationale

The derivation chain is self-contained. Section 2.1 defines the Fubini-Study metric from the fidelity distance on CP^{N-1} (Eqs. 5-7), and Appendix A.1 derives Re[G] as the infinitesimal squared line element from the Taylor expansion of |<psi_theta, psi_{theta+dtheta}>|^2, so the metric is not defined in terms of the loss L. The update rule in Eq. (13) is obtained by solving the first-order optimality condition of the metric-weighted steepest-descent problem in Eq. (10); the reparametrization invariance is a property of the Riemannian metric, not a fitted input. The block-diagonal QGT in Eqs. (19)-(28) is a direct calculation for layers whose parametric gates commute, an assumption stated explicitly as the defining property of a layer in Sections 2.4-2.5, and the single-setting measurement follows from the commuting observables in Eq. (32); there is no parameter fit hidden in the measurement scheme. The imaginary-time equivalence in Appendix A.2 is proved from the same equations (Eqs. 46-53), not assumed as a premise. The numerical comparisons use the external McClean et al. circuit with a fixed learning rate of 0.01 for all optimizers and no fitted constants; the lack of multiple seeds or error bars is an empirical robustness limitation, not a circularity. Self-citations (PennyLane, parameter-shift rule, overparameterization discussion) are tooling or context and do not carry the central argument. No step reduces, by construction or by self-citation, to its own output.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central algorithm uses no fitted constants: the metric is derived from pure-state geometry, and the numerical advantage is an empirical observation. The listed axioms are standard math or explicit assumptions about the circuit architecture; the commuting-layers assumption is the most restrictive.

free parameters (2)
  • Learning rate eta = 0.01
    Used in all numerical experiments for all optimizers. It is chosen by hand, not fitted, but the convergence curves depend on it.
  • Measurement shots per expectation value = 8192 (and 1024 in supplementary)
    Sets the noise level in cost and metric estimates. Chosen by hand; no error bars are attached to the resulting curves.
assumptions (5)
  • standard math The space of pure quantum states is CP^(N-1) with the unique unitarily invariant Fubini-Study metric.
    Invoked in Section 2.1 to justify using Re[G] as the natural metric on the variational family.
  • domain assumption The objective is the expectation value of a Hermitian operator on pure states prepared by a unitary circuit.
    Eq. (9) defines L(theta)=1/2 <psi_theta|H|psi_theta>; the algorithm and experiments assume this form.
  • domain assumption All parametrized gates in a layer commute, so [K_i,K_j]=0.
    Section 2.5 uses this to derive Eq. (27) and the single-measurement scheme.
  • standard math Standard calculus identities hold: normalization gives Re<psi|partial_i psi>=0 and Taylor expansions to second order are valid.
    Used in Supplementary A.1 for the Fubini-Study metric line element.
  • domain assumption The parameter-shift rule gives unbiased analytic gradients on quantum hardware.
    Used in Section 3 for gradient estimates; standard for Pauli rotation gates.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum Natural Gradient." pith.science (2026). https://pith.science/paper/2AJYCZNR

@misc{pith2026190902108,
  author       = {Pith},
  title        = {Pith review of: Quantum Natural Gradient},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2AJYCZNR}},
  note         = {Machine review of arXiv:1909.02108}
}
read the original abstract

A quantum generalization of Natural Gradient Descent is presented as part of a general-purpose optimization framework for variational quantum circuits. The optimization dynamics is interpreted as moving in the steepest descent direction with respect to the Quantum Information Geometry, corresponding to the real part of the Quantum Geometric Tensor (QGT), also known as the Fubini-Study metric tensor. An efficient algorithm is presented for computing a block-diagonal approximation to the Fubini-Study metric tensor for parametrized quantum circuits, which may be of independent interest.

Figures

Figures reproduced from arXiv: 1909.02108 by the authors.

Figure 1
Figure 1. The cost function value for n = 7, 9, 11 qubits and l = 5 layers as a function of training iteration for four different optimization dynamics. 8192 shots (sam￾ples) are used per required expectation value during op￾timization. 0 00 00 "% )% ! $%"$ *0 *0 *0 [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. The cost function value for n = 9 qubits and l = 3, 4, 5, 6 layers as a function of training iteration for four different optimization dynamics. 8192 shots (sam￾ples) are used per required expectation value during op￾timization. Furthermore, since every operator in Sl com￾mutes, this implies that the number of state preparations is reduced from the naive counting |Sl | = n(n + 1)/2 to just a single measurement. 3 Nu… view at source ↗
Figure 3
Figure 3. The cost function value for n = 9 qubits and l = 3, 4, 5, 6 layers as a function of training iteration (top), wall time (middle), and number of quantum evaluations (bottom) for various optimization techniques; vanilla gradient descent (blue), Adam (green), Adam modified to use the natural gradient (orange), Nelder-Mead (purple), COBYLA (cyan), the Quantum Natural Gradient (block-diagonal approximation) (red), and th… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: The cost function value for n = 9 qubits, l = 6 layers as a function of training iteration (top), wall time (middle), and number of quantum evaluations (bottom) for various optimization techniques; vanilla gradient descent (blue), Adam (green), Adam modified to use the…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Spectral Geometry and Bosonic-Bloch Probes: Explorations in Quantum Learning

    quant-ph 2026-06 unverdicted novelty 5.0 of 10

    Training reorganizes output similarity graphs in quantum networks, increasing spectral dimension by 0.23, with bosonic interference correlations and Bloch drift enabling high-ROC-AUC anomaly detection via a proposed s...

Reference graph

Works this paper leans on

36 extracted references · 9 canonical work pages · cited by 1 Pith paper

  1. [36]

    Theory of variational quantum simulation.Quantum, 3:191, 2019

    Xiao Yuan, Suguru Endo, Qi Zhao, Ying Li, and Simon C Benjamin. Theory of variational quantum simulation.Quantum, 3:191, 2019. DOI: 10.22331/q-2019-10-07-191. Accepted in Quantum 2020-05-08, click title to verify. Published under CC-BY 4.0. 15

    doi:10.22331/q-2019-10-07-191 2019

  2. [24]

    Barren plateaus in quantum neural network training landscapes.Nature communications, 9(1): 4812, 2018

    Jarrod R McClean, Sergio Boixo, Vadim N Smelyanskiy, Ryan Babbush, and Hartmut Neven. Barren plateaus in quantum neural network training landscapes.Nature communications, 9(1): 4812, 2018. DOI: 10.1038/s41467-018-07090-4

    doi:10.1038/s41467-018-07090-4 2018

  3. [18]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    arXiv 2014

  4. [1]

    Natural gradient works efficiently in learning

    Shun-Ichi Amari. Natural gradient works efficiently in learning. Neural Computation, 10(2): 251–276, 1998. DOI: 10.1162/089976698300017746

    doi:10.1162/089976698300017746 1998

  5. [2]

    Pennylane: Automatic differentiation of hybrid quantum-classical computations.arXiv preprint arXiv:1811.04968, 2018

    VilleBergholm, JoshIzaac, MariaSchuld, ChristianGogolin, M.SohaibAlam, ShahnawazAhmed, Juan Miguel Arrazola, Carsten Blank, Alain Delgado, Soran Jahangiri, Keri McKiernan, Jo- hannes Jakob Meyer, Zeyue Niu, Antal Szàva, and Nathan Killoran. Pennylane: Automatic differentiation of hybrid quantum-classical computations.arXiv preprint arXiv:1811.04968, 2018

    arXiv 2018

  6. [3]

    Geometric speed limit of accessible many-body state preparation

    Marin Bukov, Dries Sels, and Anatoli Polkovnikov. Geometric speed limit of accessible many-body state preparation. Physical Review X, 9(1):011034, 2019. DOI: 10.1103/PhysRevX.9.011034

    work page doi:10.1103/physrevx.9.011034 2019

  7. [4]

    Localization and glassy dynamicsofmany-bodyquantumsystems

    Giuseppe Carleo, Federico Becca, Marco Schiró, and Michele Fabrizio. Localization and glassy dynamicsofmany-bodyquantumsystems. Scientific reports, 2:243, 2012. DOI:10.1038/srep00243

    work page doi:10.1038/srep00243 2012

  8. [5]

    Light-cone effect and supersonic correlations in one-and two-dimensional bosonic superfluids

    Giuseppe Carleo, Federico Becca, Laurent Sanchez-Palencia, Sandro Sorella, and Michele Fab- rizio. Light-cone effect and supersonic correlations in one-and two-dimensional bosonic superfluids. Physical Review A, 89(3):031602, 2014. DOI: 10.1103/PhysRevA.89.031602

    work page doi:10.1103/physreva.89.031602 2014

Show all 36 references
  1. [6]

    Demonstration of adiabatic variational quantum computing with a superconducting quantum coprocessor.arXiv preprint arXiv:1905.03150, 2019

    Ming-Cheng Chen, Ming Gong, Xiao-Si Xu, Xiao Yuan, Jian-Wen Wang, Can Wang, Chong Ying, Jin Lin, Yu Xu, Yulin Wu, et al. Demonstration of adiabatic variational quantum computing with a superconducting quantum coprocessor.arXiv preprint arXiv:1905.03150, 2019

    1905 arXiv

  2. [7]

    Efficient quantum measurement of pauli operators

    Ophelia Crawford, Barnaby van Straaten, Daochen Wang, Thomas Parks, Earl Campbell, and Stephen Brierley. Efficient quantum measurement of pauli operators. arXiv preprint arXiv:1908.06942, 2019

    1908 arXiv

  3. [8]

    The expressive power of parame- terized quantum circuits.arXiv preprint arXiv:1810.11922, 2018

    Yuxuan Du, Min-Hsiu Hsieh, Tongliang Liu, and Dacheng Tao. The expressive power of parame- terized quantum circuits.arXiv preprint arXiv:1810.11922, 2018

    2018 arXiv

  4. [9]

    Classification with quantum neural networks on near term processors

    Edward Farhi and Hartmut Neven. Classification with quantum neural networks on near term processors. arXiv preprint arXiv:1802.06002, 2018

    2018 arXiv

  5. [10]

    A quantum approximate optimization algorithm

    Edward Farhi, Jeffrey Goldstone, and Sam Gutmann. A quantum approximate optimization algorithm. arXiv preprint arXiv:1411.4028, 2014

    2014 arXiv

  6. [11]

    Minimizing state preparations in variational quantum eigensolver by partitioning into commuting families.arXiv preprint arXiv:1907.13623, 2019

    Pranav Gokhale, Olivia Angiuli, Yongshan Ding, Kaiwen Gui, Teague Tomesh, Martin Suchara, Margaret Martonosi, and Frederic T Chong. Minimizing state preparations in variational quantum eigensolver by partitioning into commuting families.arXiv preprint arXiv:1907.13623, 2019

    1907 arXiv

  7. [12]

    Practical optimization for hybrid quantum- classical algorithms

    Gian Giacomo Guerreschi and Mikhail Smelyanskiy. Practical optimization for hybrid quantum- classical algorithms. arXiv preprint arXiv:1701.01450, 2017

    2017 arXiv

  8. [13]

    Low-depth gradient measurements can improve convergence in variational hybrid quantum-classical algorithms.arXiv preprint arXiv:1901.05374, 2019

    Aram Harrow and John Napp. Low-depth gradient measurements can improve convergence in variational hybrid quantum-classical algorithms.arXiv preprint arXiv:1901.05374, 2019

    1901 arXiv

  9. [14]

    Towards quantum machine learning with tensor networks.Quantum Science and Technology, 4:024001, 2018

    William James Huggins, Piyush Patil, Bradley Mitchell, K Birgitta Whaley, and Miles Stouden- mire. Towards quantum machine learning with tensor networks.Quantum Science and Technology, 4:024001, 2018. DOI: 10.1088/2058-9565/aaea94

    2018 doi

  10. [15]

    Three factors influencing minima in sgd.arXiv preprint arXiv:1711.04623, 2017

    Stanislaw Jastrzebski, Zachary Kenton, Devansh Arpit, Nicolas Ballas, Asja Fischer, Yoshua Ben- gio, and Amos Storkey. Three factors influencing minima in sgd.arXiv preprint arXiv:1711.04623, 2017

    2017 arXiv

  11. [16]

    Quantum compilation and circuit optimisation via energy dissipation

    Tyson Jones and Simon C Benjamin. Quantum compilation and circuit optimisation via energy dissipation. arXiv preprint arXiv:1811.03147, 2018

    2018 arXiv

  12. [17]

    Variational quantum algorithms for discovering hamiltonian spectra.Physical Review A, 99(6):062304, 2019

    Tyson Jones, Suguru Endo, Sam McArdle, Xiao Yuan, and Simon C Benjamin. Variational quantum algorithms for discovering hamiltonian spectra.Physical Review A, 99(6):062304, 2019. DOI: 10.1103/PhysRevA.99.062304

    2019 doi

  13. [19]

    Geometry and non- adiabatic response in quantum and classical systems

    Michael Kolodrubetz, Dries Sels, Pankaj Mehta, and Anatoli Polkovnikov. Geometry and non- adiabatic response in quantum and classical systems. Physics Reports, 697:1–87, 2017. DOI: 10.1016/j.physrep.2017.07.001

    2017 doi

  14. [20]

    Springer, 1981

    PH Kramer and Marcos Saraceno.Geometry of the time-dependent variational principle in quan- tum mechanics. Springer, 1981. DOI: 10.1007/3-540-10271-X_317. Accepted in Quantum 2020-05-08, click title to verify. Published under CC-BY 4.0. 14

    1981 doi

  15. [21]

    Efficient variational quantum simulator incorporating active error minimization

    Ying Li and Simon C Benjamin. Efficient variational quantum simulator incorporating active error minimization. Physical Review X, 7(2):021050, 2017. DOI: 10.1103/PhysRevX.7.021050

    2017 doi

  16. [22]

    Fisher-rao metric, ge- ometry, and complexity of neural networks

    Tengyuan Liang, Tomaso Poggio, Alexander Rakhlin, and James Stokes. Fisher-rao metric, ge- ometry, and complexity of neural networks. InThe 22nd International Conference on Artificial Intelligence and Statistics, pages 888–896, 2019

    2019

  17. [23]

    Varia- tional ansatz-based quantum simulation of imaginary time evolution.npj Quantum Information, 5(1):1–6, 2019

    Sam McArdle, Tyson Jones, Suguru Endo, Ying Li, Simon C Benjamin, and Xiao Yuan. Varia- tional ansatz-based quantum simulation of imaginary time evolution.npj Quantum Information, 5(1):1–6, 2019. DOI: 10.1038/s41534-019-0187-2

    2019 doi

  18. [25]

    Quantumcircuitlearning

    KosukeMitarai, MakotoNegoro, MasahiroKitagawa, andKeisukeFujii. Quantumcircuitlearning. Physical Review A, 98(3):032309, 2018. DOI: 10.1103/PhysRevA.98.032309

    2018 doi

  19. [26]

    Path-SGD: Path-normalized optimization in deep neural networks

    Behnam Neyshabur, Ruslan R Salakhutdinov, and Nati Srebro. Path-SGD: Path-normalized optimization in deep neural networks. In Advances in Neural Information Processing Systems, pages 2422–2430, 2015

    2015

  20. [27]

    A variational eigenvalue solver on a photonic quantum processor

    Alberto Peruzzo, Jarrod McClean, Peter Shadbolt, Man-Hong Yung, Xiao-Qi Zhou, Peter J Love, Alán Aspuru-Guzik, and Jeremy L O’Brien. A variational eigenvalue solver on a photonic quantum processor. Nature Communications, 5:4213, 2014. DOI: 10.1038/ncomms5213

    2014 doi

  21. [28]

    Information-geometry of quantum states

    Dénes Petz. Information-geometry of quantum states. InQuantum Probability Communications: Volume X, pages 135–157. World Scientific, 1998. DOI: 10.1142/9789812816054_0006

    1998 doi

  22. [29]

    Quantum computing in the NISQ era and beyond.Quantum, 2:79, 2018

    John Preskill. Quantum computing in the NISQ era and beyond.Quantum, 2:79, 2018. DOI: 10.22331/q-2018-08-06-79

    2018 doi

  23. [30]

    Circuit-centric quantum classi- fiers.arXiv preprint arXiv:1804.00633, 2018

    Maria Schuld, Alex Bocharov, Krysta Svore, and Nathan Wiebe. Circuit-centric quantum classi- fiers.arXiv preprint arXiv:1804.00633, 2018. DOI: 10.1103/PhysRevA.101.032308

    2018 arXiv

  24. [31]

    Evaluat- ing analytic gradients on quantum hardware

    Maria Schuld, Ville Bergholm, Christian Gogolin, Josh Izaac, and Nathan Killoran. Evaluat- ing analytic gradients on quantum hardware. Physical Review A, 99(3):032331, 2019. DOI: 10.1103/PhysRevA.99.032331

    2019 doi

  25. [32]

    Weak binding between two aromatic rings: Feeling the van der waals attraction by quantum monte carlo methods.The Journal of Chemical Physics, 127(1):014105, 2007

    Sandro Sorella, Michele Casula, and Dario Rocca. Weak binding between two aromatic rings: Feeling the van der waals attraction by quantum monte carlo methods.The Journal of Chemical Physics, 127(1):014105, 2007. DOI: 10.1063/1.2746035

    2007 doi

  26. [33]

    Multivariate stochastic approximation using a simultaneous perturbation gradient approximation

    James C Spall et al. Multivariate stochastic approximation using a simultaneous perturbation gradient approximation. IEEE Transactions on Automatic Control, 37(3):332–341, 1992. DOI: 10.1109/9.119632

    1992 doi

  27. [34]

    Geometric phases in physics.Geometric Phases In Physics

    F Wilczek and A Shapere. Geometric phases in physics.Geometric Phases In Physics. Series: Ad- vanced Series in Mathematical Physics, ISBN: 978-9971-5-0621-6. WORLD SCIENTIFIC, Edited by F Wilczek and A Shapere, vol. 5, 5, 1989. DOI: 10.1142/0613

    1989 doi

  28. [35]

    PennyLane source code

    Xanadu Quantum Technologies. PennyLane source code. https://github.com/XanaduAI/ pennylane, 2019. [Online; accessed 3-Mar-2020]

    2019

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.