arxiv: 2604.05627 · v1 · submitted 2026-04-07 · 🪐 quant-ph

Recognition: 2 theorem links

· Lean Theorem

Loss-aware state space geometry for quantum variational algorithms

Ankit Gill , Kunal Pal

Authors on Pith no claims yet

Pith reviewed 2026-05-10 19:39 UTC · model grok-4.3

classification 🪐 quant-ph

keywords variational quantum algorithmsnatural gradient descentloss-aware optimizationquantum state geometryFubini-Study metricFisher information metricconformal variants

0 comments

The pith

Embedding the loss hypersurface into state space geometry produces loss-aware natural gradient updates that rescale step sizes while preserving descent direction.

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

The paper develops an optimization principle for variational problems that incorporates the geometry of the space of possible outcomes, not just the parameters themselves. It achieves this either by embedding the loss surface into a base statistical manifold or by constructing the relevant metric directly from overlaps between nearby quantum states. A sympathetic reader would care because standard natural gradient methods treat all directions equally with respect to outcomes, which can result in inefficient step lengths when optimizing expectation values. If the construction holds, the updates automatically adjust their magnitude to the local loss scale without changing the underlying direction of descent. The authors test the resulting methods on variational quantum circuits and a neural network task, observing that standard natural gradient is robust on average while the new conformal variants can improve best-case performance in favorable cases.

Core claim

The intrinsic geometry of the space of outcomes is accounted for either through an ambient construction that embeds the loss hypersurface into a base statistical manifold equipped with the Fisher information metric or Fubini-Study tensor, or through a direct first-principle construction based on overlaps of nearby states on the projective Hilbert space. This approach, together with a family of conformal variants, produces loss-aware natural gradient updates that rescale the effective step size while preserving the descent direction.

What carries the argument

Loss-aware natural gradient update, obtained by embedding the loss hypersurface into the statistical manifold with its Fisher or Fubini-Study geometry, or by direct overlap construction on the projective Hilbert space.

If this is right

The effective step size is rescaled according to the local geometry of the loss surface.
The direction of each update remains identical to that of the conventional natural gradient.
A family of conformal variants supplies additional forms of the same loss-aware rescaling.
Benchmarks on variational quantum circuits and neural networks indicate possible improvements in best-case convergence under suitable conditions.
Standard natural gradient continues to show the highest average robustness across the tested examples.

Where Pith is reading between the lines

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

The same embedding idea could be applied to classical probability-based optimization tasks where outcome spacing matters.
Hybrid schemes that combine loss-aware rescaling with adaptive learning-rate schedules might reduce the need for manual tuning.
Whether the conformal family remains stable when the number of variational parameters grows large is not yet settled by the presented benchmarks.
Direct comparison against other geometry-aware methods such as quantum natural gradient with different metrics would clarify relative strengths.

Load-bearing premise

That embedding the loss hypersurface or using state overlaps will generate stable, useful rescalings of the step size without introducing instabilities or extra hyperparameters.

What would settle it

A set of repeated optimization runs on standard variational quantum eigensolver circuits where the proposed loss-aware updates produce worse average final loss values or more frequent divergence than standard natural gradient would falsify the practical benefit.

Figures

Figures reproduced from arXiv: 2604.05627 by Ankit Gill, Kunal Pal.

**Figure 2.** Figure 2: FIG. 2: Learning landscapes of the optimization dynamics for [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗

**Figure 3.** Figure 3: FIG. 3: Learning landscapes of the optimization dynamics for [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗

**Figure 4.** Figure 4: FIG. 4: (a) Convergence of the median relative energy error as a function of iteration (Above), with dashed lines indicating the [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗

**Figure 5.** Figure 5: FIG. 5: Win rate ( [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗

**Figure 6.** Figure 6: FIG. 6: Training loss (left) as a function of optimisation steps. Curves show the mean over the top-15 runs for each optimiser. [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗

**Figure 7.** Figure 7: FIG. 7: Variation of the Ricci scalar curvatures with [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗

**Figure 9.** Figure 9: FIG. 9: Variation of the Ricci scalar curvatures with [PITH_FULL_IMAGE:figures/full_fig_p023_9.png] view at source ↗

read the original abstract

The natural gradient descent optimisation technique is an efficient optimising protocol for broad classes of classical and quantum systems that takes the underlying geometry of the parameter manifold into account by means of using either the Fisher information metric of the classical probability distribution function or the Fubini-Study tensor of the associated parametrised quantum states in the consequent update rules. Even though the natural gradient descent procedure utilises the geometry of the space of probability or states, it is, however, insensitive to the measure of parametrised distance on the space of possible outcomes when the corresponding optimising problem is considered for the expectation value of a classical or quantum observable with respect to the probability distribution or the quantum state. In this work, we introduce a generic optimising principle, where the intrinsic geometry of the space of outcomes has been taken into account suitably, either by using an ambient space construction with a base statistical manifold with the usual Fisher information metric (or the Fubini-Study tensor), where the loss hypersurface is embedded to, or by means of a first-principle construction from the overlap of nearby quantum states on the projective Hilbert space. This construction as well as a family of conformal variants yields a form of loss-aware natural gradient updates that rescale the effective step size while preserving the descent direction. We benchmark the resulting optimisers on variational quantum circuit examples and on a classical neural network task, finding that, while the standard natural gradient remains the most robust on average, the proposed conformal schemes can improve best-case convergence in favourable regimes.

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.

Desk Editor's Note private letter to a colleague

This paper gives a clean geometric way to fold loss-surface structure into natural gradient for VQAs through conformal rescalings, but the gains stay modest and regime-specific.

read the letter

The core move here is to adjust natural gradient so the optimizer feels the geometry of the loss values themselves, not just the underlying state manifold. They do this in two ways: embed the loss hypersurface into the usual statistical manifold with its Fisher or Fubini-Study metric, or build a metric directly from overlaps of nearby states on projective space. Both routes produce conformal factors that stretch or shrink the step length while leaving the descent direction unchanged in the tangent space. That is a standard Riemannian fact, and the paper treats it straightforwardly without hidden assumptions or extra fitted parameters. The construction is first-principles rather than ad-hoc, which is the main novelty relative to earlier natural-gradient work on variational circuits. The benchmarks on quantum circuits and a classical neural net are reported honestly: the new schemes can improve best-case convergence in favorable regimes, yet standard natural gradient remains more robust on average. That level of candor is useful. The soft spots are proportionate. Because the direction is preserved, this is essentially a position-dependent step-size rule rather than a wholesale change in the optimizer. The advantage appearing only in best-case runs suggests the method may not deliver consistent wins across random initializations or harder landscapes, and practitioners would still need to decide which conformal variant to use. If the full paper's quantitative results lack error bars or broad statistical tests, the empirical case stays preliminary. This is aimed at people already using or extending natural gradient for variational quantum algorithms and similar geometric optimization tasks. The thinking is clear and the geometry is handled correctly, so it deserves a serious referee even if the practical payoff looks narrow.

Referee Report

1 major / 2 minor

Summary. The manuscript proposes a loss-aware extension of natural gradient descent for variational quantum algorithms and classical optimization tasks. By either embedding the loss hypersurface into a base manifold equipped with the Fisher information metric (or Fubini-Study tensor) or constructing a metric directly from overlaps of nearby states on projective Hilbert space, the authors derive updates that incorporate outcome-space geometry. These constructions, together with a family of conformal rescalings, produce loss-aware natural gradients that rescale the effective step size while preserving the descent direction in the tangent space. Numerical benchmarks on variational quantum circuits and a classical neural network are reported, with the finding that standard natural gradient remains most robust on average while the conformal variants can improve best-case convergence in favorable regimes.

Significance. If the central geometric construction holds, the work supplies a first-principles, parameter-free route to making natural-gradient methods sensitive to the loss landscape without altering the optimization direction or introducing new hyperparameters. This is a direct consequence of the conformal invariance of the descent direction under positive scalar rescaling of the metric, a standard fact of Riemannian geometry. The inclusion of both quantum and classical benchmarks, even if qualitative, adds practical relevance for variational quantum algorithms.

major comments (1)

[Numerical experiments section] Numerical experiments section: The abstract and main text state that benchmarks were performed and that conformal schemes can improve best-case convergence, yet no quantitative results, error bars, number of independent runs, or data-exclusion criteria are supplied. This absence prevents verification of the magnitude and statistical reliability of the claimed improvement over standard natural gradient.

minor comments (2)

[Abstract] The abstract would be clearer if it briefly indicated the specific VQC ansatze and classical NN architecture used in the benchmarks.
[Main text] Notation for the conformal factor and the overlap-based metric should be introduced with an explicit equation number in the main text to aid readability.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and for recommending minor revision. We address the single major comment below.

read point-by-point responses

Referee: [Numerical experiments section] Numerical experiments section: The abstract and main text state that benchmarks were performed and that conformal schemes can improve best-case convergence, yet no quantitative results, error bars, number of independent runs, or data-exclusion criteria are supplied. This absence prevents verification of the magnitude and statistical reliability of the claimed improvement over standard natural gradient.

Authors: We agree that additional quantitative information would strengthen the presentation of the benchmarks. The manuscript currently emphasizes qualitative observations of convergence behavior, consistent with the referee's summary. In the revised version we will expand the numerical experiments section to include error bars derived from multiple independent runs, the precise number of runs performed for each variational quantum circuit and neural-network task, and any data-exclusion or averaging criteria employed. These additions will permit direct assessment of the statistical reliability of the reported best-case improvements under the conformal schemes. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper derives loss-aware natural gradient updates via an ambient embedding of the loss hypersurface into the statistical manifold (with Fisher/Fubini-Study metric) or a direct overlap-based construction on projective Hilbert space. These produce a conformal rescaling of the metric, which by standard Riemannian geometry multiplies the natural-gradient vector by a position-dependent positive scalar while preserving its tangent-space direction. The central claim is therefore a direct geometric consequence of the stated constructions and does not reduce to any fitted parameter, self-citation, or renamed empirical pattern. Benchmarks are presented only as empirical checks, not as part of the derivation. No load-bearing step collapses to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard differential-geometric constructions (Fisher metric, Fubini-Study tensor) plus a new embedding or overlap-based definition of loss-aware metric; no free parameters or invented entities are described in the abstract.

axioms (2)

standard math The Fubini-Study metric or Fisher information metric correctly captures the geometry of the parameter manifold for quantum states or probability distributions.
Invoked when defining the base manifold into which the loss hypersurface is embedded.
domain assumption Nearby quantum states on the projective Hilbert space have an overlap that can be used to define a loss-aware metric.
Stated as the first-principle construction alternative to ambient embedding.

pith-pipeline@v0.9.0 · 5565 in / 1399 out tokens · 23674 ms · 2026-05-10T19:39:34.705177+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

g_LA_ij(θ) = g_FIM_ij(θ) + ∂_i L(θ) ∂_j L(θ); conformal class g_CLA = Ω² g_LA with Ω² = (1 + g^{FS} ∂L ∂L)^{-γ} etc.; effective step η_CLA-3-NG_eff = (1 + σ)^γ η_LA-NG_eff
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Biorthogonal LA-AFS tensor FS_LA(α)_ij and non-metric ±α-connections Γ_LA-1(α), Γ_LA-2(-α) from overlap expansion

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

105 extracted references · 53 canonical work pages · 2 internal anchors

[1]

Mind the gaps: The fraught road to quantum advantage

J. Eisert and J. Preskill, arXiv preprint arXiv:2510.19928 (2025)

work page arXiv 2025
[2]

Peruzzo, J

A. Peruzzo, J. McClean, P. Shadbolt, M.-H. Yung, X.-Q. Zhou, P. J. Love, A. Aspuru-Guzik, and J. L. O brien, Nature communications5, 4213 (2014)

2014
[3]

Tilly, H

J. Tilly, H. Chen, S. Cao, D. Picozzi, K. Setia, Y . Li, E. Grant, L. Wossnig, I. Rungger, G. H. Booth, et al., Physics Reports 986, 1 (2022)

2022
[4]

J. R. McClean, J. Romero, R. Babbush, and A. Aspuru-Guzik, New Journal of Physics18, 023023 (2016). 23 FIM LA RCLA1 RCLA2 RCLA3 0 1 2 3 4 5 -1.3 -1.2 -1.1 -1.0 -0.9 -0.8 -0.7 -0.6 κ R Ricci Scalar Curvature (γ= 0.7, θ²= - 0.2) FIG. 9: Variation of the Ricci scalar curvatures with the width of the Gaussian kernelκ, withγ≤1 for the five families of metrics ...

2016
[5]

Variational quantum algorithms for nonlinear problems

M. Lubasch, J. Joo, P. Moinier, M. Kiffner, and D. Jaksch, Phys. Rev. A101, 010301 (2020), URLhttps://link.aps. org/doi/10.1103/PhysRevA.101.010301

work page doi:10.1103/physreva.101.010301 2020
[6]

Joseph, T

A. Joseph, T. White, V . Chandra, and M. McGuigan, arXiv preprint arXiv:2202.09906 (2022)

work page arXiv 2022
[7]

Fontana, M

E. Fontana, M. S. Rudolph, R. Duncan, I. Rungger, and C. Cˆırstoiu, npj Quantum Inf.11, 84 (2025), 2306.05400

work page arXiv 2025
[8]

Y . Shao, F. Wei, S. Cheng, and Z. Liu, Phys. Rev. Lett.133, 120603 (2024), URLhttps://link.aps.org/doi/10.1103/ PhysRevLett.133.120603

2024
[9]

A. M. Romero, J. Engel, H. L. Tang, and S. E. Economou, Phys. Rev. C105, 064317 (2022), 2203.01619

work page arXiv 2022
[10]

Carrasco-Codina, E

M. Carrasco-Codina, E. Costa, A. M. Romero, J. Men ´endez, and A. Rios, Phys. Rev. C113, 024332 (2026), 2507.13819

work page arXiv 2026
[11]

P. P. Popov, M. Meth, M. Lewenstein, P. Hauke, M. Ringbauer, E. Zohar, and V . Kasper, Phys. Rev. Res.6, 013202 (2024), 2307.15173

work page arXiv 2024
[12]

Consiglio, W

M. Consiglio, W. J. Chetcuti, C. Bravo-Prieto, S. Ramos-Calderer, A. Minguzzi, J. I. Latorre, L. Amico, and T. J. G. Apollaro, J. Phys. A55, 265301 (2022), 2106.15552

work page arXiv 2022
[13]

Cimini, M

V . Cimini, M. Valeri, S. Piacentini, F. Ceccarelli, G. Corrielli, R. Osellame, N. Spagnolo, and F. Sciarrino, npj Quantum Inf. 10, 26 (2024), 2308.02643

work page arXiv 2024
[14]

Barkai and S

J. Wu and T. H. Hsieh, Phys. Rev. Lett.123, 220502 (2019), URLhttps://link.aps.org/doi/10.1103/PhysRevLett. 123.220502

work page doi:10.1103/physrevlett 2019
[15]

Selisko, M

J. Selisko, M. Amsler, T. Hammerschmidt, R. Drautz, and T. Eckl, Quantum Sci. Technol.9, 015026 (2024), 2208.07621

work page arXiv 2024
[16]

Chawla, Shweta, K

P. Chawla, Shweta, K. R. Swain, T. Patel, R. Bala, D. Shetty, K. Sugisaki, S. B. Mandal, J. Riu, J. Nogu ´e, et al., Phys. Rev. A111, 022817 (2025), URLhttps://link.aps.org/doi/10.1103/PhysRevA.111.022817

work page doi:10.1103/physreva.111.022817 2025
[17]

J. Y . Araz, R. G. Jha, F. Ringer, and B. Sambasivam (2024), 2406.15545

work page arXiv 2024
[18]

Medina, A

I. Medina, A. Drinko, G. I. Correr, P. C. Azado, and D. O. Soares-Pinto, Phys. Rev. A110, 012443 (2024), 2310.07617

work page arXiv 2024
[19]

A. M. Alvertis, A. Khan, T. Iadecola, P. P. Orth, and N. Tubman, Quantum9, 1748 (2025), 2408.00836

work page arXiv 2025
[20]

J. Gidi, B. Candia, A. Mu ˜noz-Moller, A. Rojas, L. Pereira, M. Mu ˜noz, L. Zambrano, and A. Delgado, Physical Review A 108, 032409 (2023)

2023
[21]

S. E. Smart and P. Narang, Physical Review A110, 052430 (2024)

2024
[22]

Miura, Quantum Information Processing25, 34 (2026)

R. Miura, Quantum Information Processing25, 34 (2026)

2026
[23]

Rogerson and A

D. Rogerson and A. Roy, arXiv preprint arXiv:2408.12583 (2024)

work page arXiv 2024
[24]

Malvetti, C

E. Malvetti, C. Arenz, G. Dirr, and T. Schulte-Herbr ¨uggen, arXiv preprint arXiv:2405.12039 (2024)

work page arXiv 2024
[25]

V ogl, The European Physical Journal Plus140, 848 (2025)

M. V ogl, The European Physical Journal Plus140, 848 (2025)

2025
[26]

Wiedmann, D

M. Wiedmann, D. Burgarth, G. Dirr, T. Schulte-Herbr ¨uggen, E. Malvetti, and C. Arenz, arXiv preprint arXiv:2509.05295 (2025)

work page arXiv 2025
[27]

Li and S.-X

Z.-L. Li and S.-X. Zhang, Phys. Rev. Res.8, 013266 (2026), 2508.06358

work page arXiv 2026
[28]

A. Joch, G. S. Uhrig, and B. Fauseweh, Quantum Sci. Technol.10, 035032 (2025), 2501.17533

work page arXiv 2025
[29]

A. K. Patra, V . D. Ghevade, R. Bhat, R. Maitra, et al., arXiv preprint arXiv:2512.01605 (2025)

work page internal anchor Pith review Pith/arXiv arXiv 2025
[30]

Choudhury, S

A. Choudhury, S. Halder, R. Maitra, and D. Ghosh, arXiv preprint arXiv:2510.15678 (2025)

work page arXiv 2025
[31]

Clemente, A

G. Clemente, A. Crippa, K. Jansen, S. Ram´ırez-Uribe, A. E. Renter´ıa-Olivo, G. Rodrigo, G. F. R. Sborlini, and L. Vale Silva, Phys. Rev. D108, 096035 (2023), URLhttps://link.aps.org/doi/10.1103/PhysRevD.108.096035

work page doi:10.1103/physrevd.108.096035 2023
[32]

B ¨arligea, B

A. B ¨arligea, B. Poggel, and J. M. Lorenz, Phys. Rev. A112, 032407 (2025), URLhttps://link.aps.org/doi/10. 1103/rgyh-8xw8

2025
[33]

K. M. Sherbert, H. Amer, S. E. Economou, E. Barnes, and N. J. Mayhall, Phys. Rev. Appl.23, 024036 (2025), URL https://link.aps.org/doi/10.1103/PhysRevApplied.23.024036

work page doi:10.1103/physrevapplied.23.024036 2025
[34]

Q. Wang, R. D’Cunha, A. Mitra, S. K. Gray, M. Otten, and L. Gagliardi, J. Phys. Chem. A129, 7999 (2025), 2501.13371

work page arXiv 2025
[35]

K. N. Okada, K. Osaki, K. Mitarai, and K. Fujii, Phys. Rev. Res.5, 043217 (2023), URLhttps://link.aps.org/doi/ 10.1103/PhysRevResearch.5.043217

work page doi:10.1103/physrevresearch.5.043217 2023
[36]

M. S. Jattana, F. Jin, H. De Raedt, and K. Michielsen, Phys. Rev. Appl.19, 024047 (2023), URLhttps://link.aps.org/ doi/10.1103/PhysRevApplied.19.024047

work page doi:10.1103/physrevapplied.19.024047 2023
[37]

T. J. Sewell, N. Bao, and S. P. Jordan, Phys. Rev. A107, 042620 (2023), URLhttps://link.aps.org/doi/10.1103/ PhysRevA.107.042620. 24

2023
[38]

Kim, K.-M

B. Kim, K.-M. Hu, M.-H. Sohn, Y . Kim, Y .-S. Kim, S.-W. Lee, and H.-T. Lim, Science Advances10, eado3472 (2024)

2024
[39]

D. Lee, B. Bilash, J. Lee, H.-T. Lim, Y . Kim, S.-W. Lee, and Y .-S. Kim, npj Quantum Information (2026)

2026
[40]

Y . Sato, H. C. Watanabe, R. Raymond, R. Kondo, K. Wada, K. Endo, M. Sugawara, and N. Yamamoto, Phys. Rev. A108, 022429 (2023), URLhttps://link.aps.org/doi/10.1103/PhysRevA.108.022429

work page doi:10.1103/physreva.108.022429 2023
[41]

Nakayama, K

A. Nakayama, K. Mitarai, L. Placidi, T. Sugimoto, and K. Fujii, Phys. Rev. Res.7, 033048 (2025), URLhttps://link. aps.org/doi/10.1103/c43x-9866

work page doi:10.1103/c43x-9866 2025
[42]

da Silva Fonseca, C

M. da Silva Fonseca, C. Moraes Porto, N. Armando Cabrera Carpio, G. de Souza Tavares de Morais, N. Henrique Morgon, R. Alfonso Nome, and C. Jorge Villas-Boas, Braz. J. Phys.56, 8 (2026), 2505.04768

work page arXiv 2026
[43]

Callison and N

A. Callison and N. Chancellor, Phys. Rev. A106, 010101 (2022), 2207.06850

work page arXiv 2022
[44]

B. C. B. Symons, D. Galvin, E. Sahin, V . Alexandrov, and S. Mensa, J. Phys. A56, 453001 (2023), 2305.07323

work page arXiv 2023
[45]

T. H. Kyaw, M. B. Soley, B. Allen, P. Bergold, C. Sun, V . S. Batista, and A. Aspuru-Guzik, Quantum Sci. Technol.9, 01LT01 (2024), 2208.10470

work page arXiv 2024
[46]

Cerezo, A

M. Cerezo et al., Nature Rev. Phys.3, 625 (2021), 2012.09265

work page arXiv 2021
[48]

Watanabe, K

R. Watanabe, K. Fujii, and H. Ueda, Phys. Rev. Res.6, 023009 (2024), URLhttps://link.aps.org/doi/10.1103/ PhysRevResearch.6.023009

2024
[49]

Amari, Neural computation10, 251 (1998)

S.-I. Amari, Neural computation10, 251 (1998)

1998
[50]

K. N. Quinn, M. C. Abbott, M. K. Transtrum, B. B. Machta, and J. P. Sethna, Reports on Progress in Physics86, 035901 (2023)

2023
[51]

Nagaoka,Methods of information geometry, vol

S.-i Amari, and H. Nagaoka,Methods of information geometry, vol. 191 (American Mathematical Soc., 2000)

2000
[52]

Amari, Advances in neural information processing systems9(1996)

S.-i. Amari, Advances in neural information processing systems9(1996)

1996
[53]

Rattray, D

M. Rattray, D. Saad, and S.-i. Amari, Physical review letters81, 5461 (1998)

1998
[54]

Amari, R

S.-i. Amari, R. Karakida, and M. Oizumi, inThe 22nd International Conference on Artificial Intelligence and Statistics (PMLR, 2019), pp. 694–702

2019
[55]

J. Liu, Y . Tang, and P. Zhang, Physical Review E111, 025304 (2025)

2025
[56]

Patel and M

D. Patel and M. M. Wilde, Physical Review A112, 052421 (2025)

2025
[57]

Miyahara, Quantum Machine Intelligence7, 98 (2025)

H. Miyahara, Quantum Machine Intelligence7, 98 (2025)

2025
[58]

Quantum natural gradient,

J. Stokes, J. Izaac, N. Killoran, and G. Carleo, Quantum4, 269 (2020), 1909.02108

work page arXiv 2020
[59]

Yamamoto, arXiv e-prints arXiv:1909.05074 (2019), 1909.05074

N. Yamamoto, arXiv e-prints arXiv:1909.05074 (2019), 1909.05074

work page arXiv 1909
[60]

Kolotouros and P

I. Kolotouros and P. Wallden, Quantum8, 1503 (2024)

2024
[61]

C. Shi, V . Dunjko, and H. Wang, Quantum Science and Technology11, 015060 (2026)

2026
[62]

Dell’Anna, R

F. Dell’Anna, R. G´omez-Lurbe, A. P´erez, and E. Ercolessi, Physical Review A112, 022612 (2025)

2025
[63]

A. Roy, S. Erramilli, and R. M. Konik, Physical Review Research6, 043083 (2024)

2024
[64]

Minervini, D

M. Minervini, D. Patel, and M. M. Wilde, Physical Review A112, 022424 (2025)

2025
[65]

Koczor and S

B. Koczor and S. C. Benjamin, Phys. Rev. A106, 062416 (2022), URLhttps://link.aps.org/doi/10.1103/ PhysRevA.106.062416

2022
[66]

Haug and M

T. Haug and M. Kim, Physical Review A106, 052611 (2022)

2022
[67]

T. R. Harvey, arXiv preprint arXiv:2509.03594 (2025)

work page arXiv 2025
[68]

Eguchi, Hiroshima Mathematical Journal22, 631 (1992)

S. Eguchi, Hiroshima Mathematical Journal22, 631 (1992)

1992
[69]

Kullback and R

S. Kullback and R. A. Leibler, The annals of mathematical statistics22, 79 (1951)

1951
[70]

Amari, The Annals of Statistics pp

S.-I. Amari, The Annals of Statistics pp. 357–385 (1982)

1982
[71]

Zhu and R

H. Zhu and R. Rohwer, Neural Processing Letters2, 28 (1995)

1995
[72]

Provost and G

J. Provost and G. Vallee, Communications in Mathematical Physics76, 289 (1980)

1980
[73]

Facchi, R

P. Facchi, R. Kulkarni, V . Man’Ko, G. Marmo, E. Sudarshan, and F. Ventriglia, Physics Letters A374, 4801 (2010)

2010
[74]

D. C. Brody and L. P. Hughston, Journal of Geometry and Physics38, 19 (2001), quant-ph/9906086

work page arXiv 2001
[75]

Geometrical formulation of quantum mechanics,

A. Ashtekar and T. A. Schilling, arXiv e-prints gr-qc/9706069 (1997), gr-qc/9706069

work page internal anchor Pith review arXiv 1997
[76]

T. W. Kibble, Communications in Mathematical Physics65, 189 (1979)

1979
[77]

S. L. Braunstein and C. M. Caves, Phys. Rev. Lett.72, 3439 (1994), URLhttps://link.aps.org/doi/10.1103/ PhysRevLett.72.3439

1994
[78]

T. R. Field and L. P. Hughston, Journal of Mathematical Physics40, 2568 (1999)

1999
[79]

Anandan, Foundations of Physics21, 1265 (1991)

J. Anandan, Foundations of Physics21, 1265 (1991)

1991
[80]

du Buisson and H

P. Zanardi and N. Paunkovi ´c, Phys. Rev. E74, 031123 (2006), URLhttps://link.aps.org/doi/10.1103/PhysRevE. 74.031123

work page doi:10.1103/physreve 2006
[81]

Zanardi, P

P. Zanardi, P. Giorda, and M. Cozzini, Physical Review Letters99, 100603 (2007)

2007

Showing first 80 references.