Pith. sign in

REVIEW 5 minor 37 references

Generalized Nonlinear Imaginary-Time Evolution

T0 review · 0 major / 5 minor · reviewed 2026-07-13 · grok-4.5

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

Pith's one-line read Nonlinear imaginary-time evolution optimizes general pure-state costs and converges locally exponentially.

desk verdict Clean geometric generalization of ITE to arbitrary pure-state costs, with a correct local exponential-rate theorem and exact multi-state QNGD equivalence; useful and ready for referees. read the letter →

arxiv 2607.09599 v1 pith:LPRTI43D submitted 2026-07-10 quant-ph

Chenyu Shi , Hao Wang , Jin-Fu Chen This is my paper · ORCID
classification quant-ph
keywords imaginary-timeevolutionFubini-StudymetricquantumnaturalgradientvariationalalgorithmsstatepreparationFisherinformationvarianceminimization
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

Standard imaginary-time evolution finds ground states by flowing downhill in energy under the Fubini-Study geometry of pure states. This paper generalizes that flow to any smooth cost defined on pure states (or several pure states at once). The resulting nonlinear imaginary-time evolution (NITE) is the Fubini-Study gradient flow of the chosen cost; its effective generator is state-dependent, so the differential equation is nonlinear. A hardware-efficient variational version of NITE is exactly quantum natural gradient descent on the same cost. The authors prove local exponential convergence to non-degenerate minima, apply the method to variance minimization, quantum Fisher-information maximization, and penalty-based excited-state preparation, and show that it converges faster and more reliably than ordinary gradient descent.

What carries the argument

Nonlinear Imaginary-Time Evolution (NITE): the ordinary differential equation d|ϕ>/dt = −(H(ϕ)−⟨H(ϕ)⟩)|ϕ>, where the state-dependent Hermitian generator H(ϕ) is the Hilbert-Schmidt gradient of the cost evaluated at the pure-state density matrix; multi-state NITE couples several such equations.

What would settle it

On any of the three example tasks, replace the Fubini-Study metric by the Euclidean metric (ordinary gradient descent) or deliberately choose an initial state whose trajectory leaves the positive-Hessian neighborhood; if the observed convergence rate remains exponential and identical to the NITE rate, the local-exponential claim fails.

Watch

Extended reading notes

Core claim

Any twice-differentiable cost on pure quantum states (or on a tuple of them) defines a nonlinear imaginary-time evolution given by the Fubini-Study gradient flow. That continuous flow decreases the cost monotonically and converges locally exponentially to any non-degenerate local minimum; its first-order variational discretization is precisely quantum natural gradient descent for that cost.

Load-bearing premise

Near the target minimum the Fubini-Study Hessian must stay strictly positive-definite inside a whole geodesically convex neighborhood, and every trajectory that starts nearby must remain inside that neighborhood.

Editorial extensions

If this is right

  • Any pure-state variational subroutine whose cost is not a simple expectation value can replace ordinary gradient steps by NITE/QNGD steps and inherit local exponential convergence.
  • Multi-state NITE supplies a single coupled flow that simultaneously prepares an orthonormal low-energy subspace rather than one eigenstate at a time.
  • The same construction applies immediately to matrix-product states, neural quantum states, or other variational manifolds once the Fubini-Study metric on that manifold is available.
  • When the cost is a nonlinear composite of expectation values (least-squares or cross-entropy loss), NITE becomes a geometry-aware training rule for variational quantum classifiers.

Reading between the lines

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

  • Because the exponential rate is controlled by the smallest positive eigenvalue of the Fubini-Study Hessian, one can diagnose and regularize near-degenerate directions before training begins.
  • The multi-state metric sum that appears in the variational update is formally identical to the metric of a product-state manifold, suggesting a direct lift to multi-parameter quantum sensing or multi-objective variational algorithms.
  • If the free-energy functional of a purified thermal state is used as the cost, the same NITE construction yields a geometry-aware imaginary-time flow for finite-temperature variational manifolds.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

0 major / 5 minor

Summary. The manuscript generalizes normalized imaginary-time evolution from energy expectation values to arbitrary C^{2} cost functions on pure states (and products of pure states). The resulting Nonlinear Imaginary-Time Evolution (NITE) is the Fubini–Study gradient flow of the cost; its generator is the state-dependent Hermitian operator obtained from the Hilbert–Schmidt gradient of the cost (Eqs. (5)–(6) and multi-state analogues (8)–(9)). The authors prove local exponential convergence to non-degenerate local minima under a positive-definiteness assumption on the Fubini–Study Hessian (Theorem 1 and the manifold version Theorem 2 in the supplement), recover the classical spectral-gap rate of ordinary ITE as a special case (Remark 1), and show that the variational discretization of NITE is exactly quantum natural gradient descent (single-state and multi-state forms, supplement §§III–IV). Numerical illustrations on variance minimization, quantum Fisher information maximization, and penalty-based simultaneous ground/excited-state preparation demonstrate faster and more robust convergence than Euclidean gradient descent, both with hardware-efficient ansätze and with full-space parameterization.

Significance. If the claims hold, the work supplies a clean geometric unification of imaginary-time evolution and quantum natural gradient for objectives that are not simple expectation values—precisely the setting of many modern variational subroutines (variance minimization, QFI maximization, penalty methods). The local-convergence theorem is a standard but carefully specialized Riemannian Polyak–Łojasiewicz argument that recovers the known spectral-gap rate, and the exact equivalence of the variational form to QNGD is useful for near-term implementations. The multi-state extension and the explicit applications give a practical route to low-energy subspace preparation and metrology-oriented state preparation. Strengths include the transparent continuous-time derivation, the reduction proofs, the local-rate guarantee with an explicit Hessian hypothesis, and the side-by-side numerics (means ± std over 10 seeds, full-space checks that isolate ansatz expressivity).

minor comments (5)
  1. Fig. 2 caption and main-text discussion of the multi-state experiment should state the concrete numerical values of the penalty coefficients λ0, λ1, λp (or the general λk, λij) used in the runs; they are free parameters of the cost (15)–(16) and affect the landscape.
  2. In the main text after Eq. (5), a one-sentence pointer that the non-uniqueness of H(ϕ) is resolved by the Hilbert–Schmidt definition (and that this choice recovers ordinary ITE) would help readers who do not immediately consult the supplement.
  3. Supplement Fig. S4 reports an empirical rate comparison with 2λ*; a brief remark in the main text that the observed slopes are consistent with the local-rate prediction of Theorem 1 would strengthen the link between theory and numerics.
  4. A few typographical inconsistencies appear (e.g., “EffcientSU2”, “withrespecttothe”, “andpo-tential”); a light copy-edit pass would remove them.
  5. The outlook mentions finite-temperature free-energy functionals and quantum-machine-learning losses; a short clarifying sentence that the same Hilbert–Schmidt construction applies once the cost is written as a function of the purified state (or of the density operator) would make the claimed generality more concrete.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; NITE, its QNGD equivalence, and local exponential rate are derived self-containedly from the Fubini–Study gradient flow.

full rationale

The paper defines NITE as the Fubini–Study gradient flow of a general C^{2} cost E on pure states (or product of pure states), constructs the associated state-dependent Hermitian generator via the Hilbert–Schmidt gradient of the density-operator form of E, and recovers ordinary ITE when E is an expectation value. The variational McLachlan projection of that flow onto a parameterized manifold is shown by direct least-squares algebra (supplement §§III–IV) to be exactly the quantum natural gradient update (single-state or multi-state sum of metrics). Local exponential convergence (Theorem 1) follows from the standard Riemannian Polyak–Łojasiewicz inequality that holds once the Fubini–Study Hessian is positive-definite on a geodesically convex neighborhood; the classical spectral-gap rate of ITE is recovered as the special case (Remark 1). No parameter is fitted and then re-used as a prediction, no uniqueness theorem is imported from overlapping authors, and the sole self-citation ([29]) is an incidental remark that the multi-state metric sum is a special case of WA-QNGD, not a load-bearing premise. All steps are therefore independent of their own conclusions.

Assumptions & free parameters 3 free parameters · 3 assumptions · 1 invented entities

The central mathematical claim rests only on standard Riemannian geometry of complex projective space plus the definition of the Hilbert-Schmidt gradient; the free parameters appear only in the numerical illustrations and do not enter the theorems.

free parameters (3)
  • penalty coefficients λ0, λ1, λp (or λk, λij)
    Chosen by hand so that the global minimum of the multi-state cost coincides with the lowest K eigenstates; values are not derived from first principles.
  • learning rate / discrete time step η
    Appears in the variational QNGD discretization; must be chosen small enough for stability but is not fixed by theory.
  • ansatz depth (number of EfficientSU2 layers)
    Hyper-parameter of the hardware-efficient circuit used in the main-text numerics; full-space controls later remove it.
assumptions (3)
  • standard math The pure-state manifold is the complex projective space CP^{d-1} equipped with the Fubini-Study metric.
    Standard geometric setting for pure quantum states; used from the first paragraph of the geometric interpretation.
  • domain assumption The cost E is twice continuously differentiable (C^{2}) and the target critical point is a non-degenerate local minimum of the Fubini-Study Hessian.
    Explicit hypothesis of Theorem 1; without it the local exponential rate is not claimed.
  • standard math The Hilbert-Schmidt gradient of a phase-invariant cost yields a Hermitian operator H(ϕ) satisfying H(ϕ)|ϕ⟩ = |g(ϕ)⟩.
    Follows from the Riesz representation theorem on the real vector space of Hermitian matrices (supplement §I).
invented entities (1)
  • Nonlinear Imaginary-Time Evolution (NITE) and its multi-state extension independent evidence
    purpose: Defines the continuous-time flow for arbitrary pure-state costs and for joint multi-state costs.
    The central new dynamical system introduced by the paper; independent evidence is the reduction to ordinary ITE when the cost is an expectation value and the match to QNGD.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generalized Nonlinear Imaginary-Time Evolution." pith.science (2026). https://pith.science/paper/LPRTI43D

@misc{pith2026260709599,
  author       = {Pith},
  title        = {Pith review of: Generalized Nonlinear Imaginary-Time Evolution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LPRTI43D}},
  note         = {Machine review of arXiv:2607.09599}
}
read the original abstract

Imaginary-time evolution (ITE) is a powerful method for ground-state preparation of a given Hamiltonian. The normalized ITE can be viewed as a gradient flow of the energy expectation value with respect to the Fubini--Study metric. In this work, we propose a generalized nonlinear imaginary-time evolution (NITE) for more general quantum state-preparation tasks. We further present a hardware-efficient variational implementation of NITE and reveal its connection to quantum natural gradient descent. NITE is applied to several subroutine tasks, including variance minimization in variational eigensolvers, probe-state preparation in variational quantum sensing, and excited-state preparation using penalty terms. We prove that NITE achieves a local exponential convergence rate under reasonable assumptions. Our results show that NITE outperforms standard gradient descent and can serve as an efficient optimization method for variational tasks beyond ground-state preparation.

Figures

Figures reproduced from arXiv: 2607.09599 by the authors.

Figure 1
Figure 1. FIG. 1. Illustrative diagram comparing ITE and NITE. Stan [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Comparison of NITE and standard gradient descent [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references

  1. [1]

    Mario Motta, Chong Sun, Adrian T. K. Tan, Matthew J. O’Rourke, Erika Ye, Austin J. Minnich, Fernando G. S. L. Brandão, and Garnet Kin-Lic Chan. Determin- ing eigenstates and thermal states on a quantum com- puter using quantum imaginary time evolution.Nature Physics, 16(2):205–210, February 2020

    2020

  2. [2]

    Benjamin, and Xiao Yuan

    Sam McArdle, Tyson Jones, Suguru Endo, Ying Li, Si- mon C. Benjamin, and Xiao Yuan. Variational ansatz- based quantum simulation of imaginary time evolution. npj Quantum Information, 5(1):75, September 2019

    2019

  3. [3]

    Quantum Natural Gradient.Quantum, 4:269, May 2020

    JamesStokes, JoshIzaac, NathanKilloran, andGiuseppe Carleo. Quantum Natural Gradient.Quantum, 4:269, May 2020

    2020

  4. [4]

    Implementation of quantum imaginary-time evolution method on NISQ devices by introducing nonlocal approx- imation.npj Quantum Information, 7(1):85, June 2021

    Hirofumi Nishi, Taichi Kosugi, and Yu-ichiro Matsushita. Implementation of quantum imaginary-time evolution method on NISQ devices by introducing nonlocal approx- imation.npj Quantum Information, 7(1):85, June 2021

    2021

  5. [5]

    Benjamin

    Xiao Yuan, Suguru Endo, Qi Zhao, Ying Li, and Si- mon C. Benjamin. Theory of variational quantum simu- lation.Quantum, 3:191, October 2019

    2019

  6. [6]

    McMahon, Mahum Pervez, and Christian 6 Arenz

    Nathan A. McMahon, Mahum Pervez, and Christian 6 Arenz. Equatingquantumimaginarytimeevolution, Rie- mannian gradient flows, and stochastic implementations. Physical Review Research, 8(2):023024, April 2026

    2026

  7. [7]

    Orth, and Yong-Xin Yao

    Niladri Gomes, Anirban Mukherjee, Feng Zhang, Thomas Iadecola, Cai-Zhuang Wang, Kai-Ming Ho, Pe- ter P. Orth, and Yong-Xin Yao. Adaptive Varia- tional Quantum Imaginary Time Evolution Approach for Ground State Preparation.Advanced Quantum Tech- nologies, 4(12):2100114, December 2021

    2021

  8. [8]

    Quantum imaginary-time evolution al- gorithm for quantum field theories with continuous vari- ables.Physical Review A, 105(1):012412, January 2022

    Kübra Yeter-Aydeniz, Eleftherios Moschandreou, and George Siopsis. Quantum imaginary-time evolution al- gorithm for quantum field theories with continuous vari- ables.Physical Review A, 105(1):012412, January 2022

    2022

Show all 37 references
  1. [9]

    M. B. Hastings, D. Wecker, B. Bauer, and M. Troyer. Im- proving Quantum Algorithms for Quantum Chemistry

  2. [10]

    Love, Alán Aspuru- Guzik, and Jeremy L

    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 Com- munications, 5(1):4213, July 2014

    2014

  3. [11]

    Thetheoryofvariationalhybrid quantum-classical algorithms.New Journal of Physics, 18(2):023023, February 2016

    Jarrod R McClean, Jonathan Romero, Ryan Babbush, andAlánAspuru-Guzik. Thetheoryofvariationalhybrid quantum-classical algorithms.New Journal of Physics, 18(2):023023, February 2016

    2016

  4. [12]

    Booth, and Jonathan Tennyson

    Jules Tilly, Hongxiang Chen, Shuxiang Cao, Dario Pi- cozzi, Kanav Setia, Ying Li, Edward Grant, Leonard Wossnig, Ivan Rungger, George H. Booth, and Jonathan Tennyson. The Variational Quantum Eigensolver: A re- view of methods and best practices.Physics Reports, 986:1–128, Nove...

    2022

  5. [13]

    Variational quantum eigensolvers by variance min- imization.Chinese Physics B, 31(12):120301, November 2022

    Dan-Bo Zhang, Bin-Lin Chen, Zhan-Hao Yuan, and Tao Yin. Variational quantum eigensolvers by variance min- imization.Chinese Physics B, 31(12):120301, November 2022

    2022

  6. [14]

    Hobday, P

    I. Hobday, P. D. Stevenson, and J. Benstead. Variance minimization for nuclear structure on a quantum com- puter.Physical Review C, 111(6):064321, June 2025

    2025

  7. [15]

    Variational Quan- tum Eigensolver with Mutual Variance-Hamiltonian Op- timization.Chinese Physics Letters, 40(1):010303, Jan- uary 2023

    Bin-Lin Chen and Dan-Bo Zhang. Variational Quan- tum Eigensolver with Mutual Variance-Hamiltonian Op- timization.Chinese Physics Letters, 40(1):010303, Jan- uary 2023

    2023

  8. [16]

    A variational toolbox for quantum multi- parameter estimation.npj Quantum Information, 7(1):89, June 2021

    Johannes Jakob Meyer, Johannes Borregaard, and Jens Eisert. A variational toolbox for quantum multi- parameter estimation.npj Quantum Information, 7(1):89, June 2021

    2021

  9. [17]

    Beckey, M

    Jacob L. Beckey, M. Cerezo, Akira Sone, and Patrick J. Coles. Variational quantum algorithm for estimating the quantum Fisher information.Physical Review Research, 4(1):013083, February 2022

    2022

  10. [18]

    Benjamin MacLellan, Piotr Roztocki, Stefanie Czischek, and Roger G. Melko. End-to-end variational quantum sensing.npj Quantum Information, 10(1):118, November 2024

    2024

  11. [19]

    Variational Quantum Computation of Excited States

    Oscar Higgott, Daochen Wang, and Stephen Brierley. Variational Quantum Computation of Excited States. Quantum, 3:156, July 2019

    2019

  12. [20]

    Variationalquantumalgorithms for discovering Hamiltonian spectra.Physical Review A, 99(6):062304, June 2019

    Tyson Jones, Suguru Endo, Sam McArdle, Xiao Yuan, andSimonC.Benjamin. Variationalquantumalgorithms for discovering Hamiltonian spectra.Physical Review A, 99(6):062304, June 2019

    2019

  13. [21]

    Ignacio Cirac

    Tao Shi, Eugene Demler, and J. Ignacio Cirac. Varia- tional study of fermionic and bosonic systems with non- Gaussian states: Theory and applications.Annals of Physics, 390:245–302, March 2018

    2018

  14. [22]

    The Complex Gradient Operator and the CR-Calculus, 2009

    Ken Kreutz-Delgado. The Complex Gradient Operator and the CR-Calculus, 2009. Version Number: 1

    2009

  15. [23]

    A short tutorial on Wirtinger Cal- culus with applications in quantum information, 2023

    Kelvin Koor, Yixian Qiu, Leong Chuan Kwek, and Patrick Rebentrost. A short tutorial on Wirtinger Cal- culus with applications in quantum information, 2023. Version Number: 1

    2023

  16. [24]

    Supplemental Material

  17. [25]

    Convergence and ef- ficiency proof of quantum imaginary time evolution for bounded order systems, 2025

    Tobias Hartung and Karl Jansen. Convergence and ef- ficiency proof of quantum imaginary time evolution for bounded order systems, 2025. Version Number: 2

    2025

  18. [26]

    Benavides-Riveros, Prachi Sharma, and Fedor Šimkovic

    Carlos L. Benavides-Riveros, Prachi Sharma, and Fedor Šimkovic. Augmenting Imaginary-Time Evolution with Local Geometric Information, 2026. Version Number: 1

    2026

  19. [27]

    Benjamin

    Bálint Koczor and Simon C. Benjamin. Quantum natu- ral gradient generalized to noisy and nonunitary circuits. Physical Review A, 106(6):062416, December 2022

    2022

  20. [28]

    Avoiding local minima in variational quantum eigensolvers with the natural gradient optimizer.Physi- cal Review Research, 2(4):043246, November 2020

    David Wierichs, Christian Gogolin, and Michael Kasto- ryano. Avoiding local minima in variational quantum eigensolvers with the natural gradient optimizer.Physi- cal Review Research, 2(4):043246, November 2020

    2020

  21. [29]

    Weighted approximate quantum natural gradient for variational quantum eigensolver.Quantum Science and Technology, 11(1):015060, March 2026

    Chenyu Shi, Vedran Dunjko, and Hao Wang. Weighted approximate quantum natural gradient for variational quantum eigensolver.Quantum Science and Technology, 11(1):015060, March 2026

    2026

  22. [30]

    Pennylanequantum chemistry datasets.https://pennylane.ai/datasets/ h3-plus-molecule, 2023

    UtkarshAzadandStepanFomichev. Pennylanequantum chemistry datasets.https://pennylane.ai/datasets/ h3-plus-molecule, 2023

    2023

  23. [31]

    The density-matrix renormalization group in the age of matrix product states.Annals of Physics, 326(1):96–192, January 2011

    Ulrich Schollwöck. The density-matrix renormalization group in the age of matrix product states.Annals of Physics, 326(1):96–192, January 2011

    2011

  24. [32]

    Solving the quan- tum many-body problem with artificial neural networks

    Giuseppe Carleo and Matthias Troyer. Solving the quan- tum many-body problem with artificial neural networks. Science, 355(6325):602–606, February 2017

    2017

  25. [33]

    Ignacio Cirac

    Tao Shi, Eugene Demler, and J. Ignacio Cirac. Varia- tional Approach for Many-Body Systems at Finite Tem- perature.Physical Review Letters, 125(18):180602, Oc- tober 2020

    2020

  26. [34]

    Ignacio Cirac

    Sirui Lu, Giacomo Giudice, and J. Ignacio Cirac. Varia- tional neural and tensor network approximations of ther- mal states.Physical Review B, 111(7):075102, February 2025

    2025

  27. [35]

    Quan- tum circuit learning.Physical Review A, 98(3):032309, September 2018

    K.Mitarai, M.Negoro, M.Kitagawa, andK.Fujii. Quan- tum circuit learning.Physical Review A, 98(3):032309, September 2018

    2018

  28. [36]

    Svore, and Nathan Wiebe

    Maria Schuld, Alex Bocharov, Krysta M. Svore, and Nathan Wiebe. Circuit-centric quantum classifiers.Phys- ical Review A, 101(3):032308, March 2020

    2020

  29. [37]

    Bromley, Josh Izaac, Maria Schuld, and Nathan Killoran

    Andrea Mari, Thomas R. Bromley, Josh Izaac, Maria Schuld, and Nathan Killoran. Transfer learning in hy- brid classical-quantum neural networks.Quantum, 4:340, October 2020. 7 Supplemental Material: Generalized Nonlinear Imaginary-Time Evolution Chenyu Shi1,2, Hao Wang1,2,∗, an...

    2020

Pith tools

Reviewed July 13, 2026 · model on record in the stance chip above.