Pith. sign in

REVIEW 4 major objections 6 minor 2 cited by

Quantum Optimization via Gradient-Based Hamiltonian Descent

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Gradient-based Quantum Hamiltonian Descent claims a quadratic-in-time convergence rate and order-of-magnitude gains over QHD and classical baselines.

desk verdict Genuine continuous-time extension of QHD with a real Lyapunov proof, but the headline performance claim rests on discrete simulations outside the theorem hypotheses. read the letter →

arxiv 2505.14670 v1 pith:HDHAG7F4 submitted 2025-05-20 quant-ph cs.LGmath.OC

Jiaqi Leng , Bin Shi This is my paper
classification quant-phcs.LGmath.OC MSC 68Q1281P6890C25
keywords gradient-basedQHDquantumHamiltoniandescentcontinuousoptimizationLyapunovfunctiontunnelingfirst-ordersimulationconvergencerate
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

This paper proposes a quantum algorithm for continuous optimization that feeds the gradient of the objective into the Hamiltonian of a Schrödinger evolution, extending Quantum Hamiltonian Descent (QHD) from a zeroth-order method to a first-order one. The main theoretical claim is that the continuous-time dynamics converge to a global minimizer at rate $O(t^{-2})$ for convex objectives (Theorem 1) and drive the expected squared gradient norm to zero at $O(t^{-2})$ under a condition on the squared-gradient landscape (Theorem 4). A product-formula discretization yields Algorithm 1 with query complexity linear in dimension, and numerical tests on convex and non-convex 2D functions show faster convergence and higher success probability than QHD, accelerated gradient descent, and momentum SGD, by an order of magnitude in some cases. The theoretical guarantees are for the continuous-time dynamics; the paper explicitly leaves a complete convergence analysis of the discretized algorithm to future work.

What carries the argument

The load-bearing object is the gradient-augmented time-dependent Hamiltonian $\hat H(t) = \tfrac12\sum_{j=1}^d (t^{-3/2}\hat p_j + \alpha t^{3/2}\hat v_j)^2 + \tfrac{\beta}{2}t^3\|\nabla f\|^2 + (t^3+\gamma t^2)f$, obtained by canonical quantization of a Lagrangian in which the velocity term is coupled to $\nabla f$. Convergence is carried by two operator-valued Lyapunov functions, one for function values and one for gradient norms; the non-increasingness of their expectation values follows from the commutation relations in Lemma 3, which convert the time derivative of the Lyapunov function into non-positive convexity terms. The discretization uses a three-way operator splitting into kinetic, gradient-anticommutator, and multiplicative potential parts, with the gradient-anticommutator term simulated by quantum singular value transformation.

What would settle it

Simulate Algorithm 1 on a smooth convex objective under Theorem 1's hypotheses ($\alpha>0$, $\beta=0$, $\gamma\ge 3\alpha$) and fit $\log \mathbb{E}[f(X_k)]$ against $\log t_k$; the central claim predicts a slope of $-2$, so observing a significantly different slope would show the discrete algorithm does not inherit the continuous-time rate.

Watch

Extended reading notes

Core claim

The paper's central discovery is that inserting gradient information directly into the quantum Hamiltonian permits a Lyapunov-function proof of convergence. With the operator $A_j = t^{-3/2}\hat p_j + \alpha t^{3/2}\hat v_j$ in $\hat H(t) = \tfrac12\sum_j A_j^2 + \tfrac{\beta}{2}t^3\|\nabla f\|^2 + (t^3+\gamma t^2)f$, the expected function value satisfies $\mathbb{E}[f(X_t)] \le (K_0+D_0)/(t^2+\omega t)$ for convex $f$ when $\beta=0$ and $\gamma\ge\max(3\alpha,0)$, and the expected squared gradient norm is bounded by $2(K_0+D_0')/(\beta t^2)$ when $\beta>0$ and condition (17) holds. The proof constructs operator-valued Lyapunov functions whose expectation values are non-increasing under the Schrödinger flow, using operator commutation relations to cancel all growing terms. The same construction also yields an efficient discrete implementation: with quantum access to $f$ and $\nabla f$, each iteration costs $\tilde O(d)$ queries, matching the per-iteration cost of classical accelerated methods.

Load-bearing premise

The paper's practical claims rest on the assumption that the discrete product-formula algorithm inherits the convergence proved for the continuous-time Schrödinger dynamics, an inheritance the paper explicitly leaves unproved.

Editorial extensions

If this is right

  • For convex objectives, continuous-time gradient-based QHD provably achieves $\mathbb{E}[f(X_t)] \le O(t^{-2})$, the same asymptotic convergence class as accelerated gradient methods.
  • When $\beta>0$ and condition (17) holds, the expected squared gradient norm also decays as $O(t^{-2})$, giving a stationary-point guarantee rather than only a function-value guarantee.
  • The discrete Algorithm 1 is implementable with $O(K)$ queries to a function-value oracle and $\tilde O(\alpha d h K L)$ queries to a gradient oracle, so the per-iteration cost is linear in dimension.
  • On the tested non-convex 2D problems (Styblinski-Tang, Michalewicz, Cube-Wave, and Rastrigin), gradient-based QHD attains lower final objective values and higher success probability than QHD, accelerated gradient descent, and momentum SGD.
  • The numerical results suggest that the inclusion of gradient information also improves the chance of identifying the global minimum, not merely the local convergence rate.

Reading between the lines

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

  • The theorems are stated for parameters $\alpha>0$ (Theorem 1 requires $1/\alpha \ge T_0 > 0$) and $\beta>0$ in Theorem 4, while the experiments use $\alpha=-0.1$ or $-0.05$ and $\beta=0$; a reader should treat the numerical speedup as evidence about the discretized dynamics, not as covered by the proven rates.
  • If the product-formula discretization can be shown to inherit the continuous Lyapunov decay, then the $\tilde O(d)$ per-iteration cost would become a true end-to-end complexity statement, making gradient-based QHD directly comparable to classical accelerated methods in query count.
  • Because Theorem 4 only needs the squared-gradient magnitude $G(x)=\|\nabla f(x)\|^2$ to be convex, the method has a plausible route to saddle-point escape beyond the convex case; testing it on objectives with convex $\|\nabla f\|^2$ but nonconvex $f$ would isolate this mechanism.
  • The gradient coupling term plays a role parallel to the Hessian-velocity correction in high-resolution classical ODEs, suggesting that any advantage observed here may transfer to stochastic or constrained variants of quantum Hamiltonian descent; this connection is left implicit in the paper.
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

4 major / 6 minor

Summary. The paper introduces gradient-based QHD, a modification of Quantum Hamiltonian Descent that incorporates gradient information into the time-dependent Hamiltonian. For the continuous-time Schrödinger dynamics, the authors prove O(t^-2) decay of the expected function value (Theorem 1, under β=0, γ≥max(3α,0), and convexity) and O(t^-2) decay of the expected squared gradient norm (Theorem 4, under β>0, condition (17), and convexity-related assumptions), using Lyapunov function arguments. They then propose a product-formula discretization (Algorithm 1), give a gate-complexity analysis (Theorem 6), and report numerical experiments on 2D convex and non-convex test problems in which the proposed method converges faster and finds lower objective values than QHD, NAG, and SGDM. The abstract claims that gradient-based QHD outperforms existing quantum and classical methods by at least an order of magnitude.

Significance. If the continuous-time theorems are correct, the paper offers a useful new Lyapunov-based convergence analysis for a gradient-enhanced quantum optimization Hamiltonian, and the specific Lyapunov constructions are a genuine technical contribution that could inform future work on quantum Hamiltonian methods. The algorithm's per-iteration cost is linear in dimension, and the code is provided, which supports reproducibility. However, the headline practical claim of order-of-magnitude improvement is supported only by simulations of a discretized algorithm whose convergence is explicitly left unproven and whose parameters lie outside the regime covered by Theorems 1 and 4. The theoretical results are thus not yet connected to the numerical evidence, and the paper's central practical assertion is not substantiated.

major comments (4)
  1. [§6.2, §6.3 vs. Theorem 1/4] The experimental parameter settings violate the hypotheses of the theorems. In §6.2 the authors set α=-0.1, β=0, and t0=1, while Theorem 1 requires 1/α ≥ T0 > 0, which forces α>0. In §6.3 they set α=-0.05, β=0, and t0=0, again contradicting α>0 of Theorem 1 and the β>0 requirement of Theorem 4. Consequently, the observed convergence in the experiments cannot be attributed to the proved continuous-time dynamics, and the abstract's order-of-magnitude claim is left without theoretical support.
  2. [§5.1, Algorithm 1] The convergence theorems apply to the exact continuous-time Schrödinger evolution (15), not to the product-formula discretization used in the numerical experiments. The paper explicitly states that a complete understanding of the discrete-time algorithm's convergence is left for future study, and the Trotter error bound given there implies a step size h∼t_k^{-3/2} to control per-step error. The fixed step sizes h=0.005–0.2 over t_k up to about 5 do not satisfy this requirement, so the reported numerical convergence could be an artifact of the specific integrator rather than a property of the proposed Hamiltonian dynamics.
  3. [Theorem 4 vs. Lemma 5] There is a mismatch between the hypotheses of Theorem 4 and those of Lemma 5, which is cited as the proof of the theorem. Theorem 4 assumes γ≥max(3α,0) and β>0, while Lemma 5 assumes γ>0 and α≥max(β,0). The theorem's hypotheses do not imply α≥β; for example α=0.1 and β=5 satisfy the theorem's stated conditions with a sufficiently small T0 but violate Lemma 5. Thus the Lyapunov inequality F'(t)≤0 is not established under the theorem's stated assumptions, and Theorem 4 is unproven as written. The authors should either add α≥max(β,0) to Theorem 4 or prove Lemma 5 under weaker assumptions.
  4. [Remark 2 (after Theorem 4)] The remark claims that convexity of G(x)=||∇f(x)||^2 is a sufficient condition for the identity G(x)−∇G(x)^T x ≤ 0. This is false: for G(x)=e^x, one has G(x)−∇G(x)^T x = e^x(1−x), which is positive for x<1. The theorem itself remains valid if condition (17) is assumed, but the remark should be corrected or replaced with a genuinely sufficient condition.
minor comments (6)
  1. [Abstract and §6] The abstract's claim of outperforming all compared methods by at least an order of magnitude is too strong given that the experiments cover only a few 2D test functions and the parameters are outside the proven regime; this claim should be qualified to the specific numerical settings.
  2. [§3.2] There are typos: 'Feymann's path integral' should be 'Feynman's path integral', and 'Hamtiltonian' in §3 should be 'Hamiltonian'.
  3. [Figure 1 caption] The caption reads 'PPPk denotes the success probability at iteration k'; this should be 'P_k denotes the success probability at iteration k'.
  4. [§4.1] The notation ⟨f⟩_t is used in the paragraph after Theorem 1 without definition; it should be defined as ⟨Ψ(t)|f|Ψ(t)⟩ for clarity.
  5. [§6.2] The phrase 'while the (classical) QHD exhibits a slower convergence rate' is confusing, as QHD is a quantum algorithm; presumably 'classical' refers to the classical baselines, so the wording should be corrected.
  6. [§5.1] The statement 'we treat the step size h as an independent parameter in the complexity analysis' is not fully justified: Theorem 6's complexity contains an explicit factor h, and if accuracy requires h∼t^{-3/2}, the end-to-end complexity would scale differently. This should be discussed explicitly.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the Lyapunov convergence proofs are derived in-paper from explicit operator identities, and no fitted parameter or self-cited uniqueness claim is used to force the stated rates.

full rationale

The paper's central theoretical claims are Theorems 1 and 4, which give O(t^-2) rates for continuous-time Schrödinger evolution under Hamiltonian (5). These are proved in Appendix C via a self-contained Lyapunov construction: the derivative E'(t) = <∂_t Ô + i[H,Ô]> is computed from explicit commutation relations (Lemma 3), and the non-positivity follows from the stated assumptions (convexity and ω≥0 for Theorem 1; condition (17) and β>0 for Theorem 4). No constant is fitted to data to obtain the rate; α, β, γ, and T0 are free parameters appearing in stated inequalities, and the bound constants are initial-state expectations. Prior work by the authors (Leng et al. 2023a; Shi et al. 2022) motivates the Hamiltonian form but is not used as a black-box theorem; the quantization and the Lyapunov calculation are performed in this paper. The numerical 'order of magnitude' claim is an empirical assertion, not derived from the theorems, and the paper itself flags the main gap: 'A complete understanding of the convergence of the discrete-time algorithm, however, is left for future study' (Section 5.1). The experiments also use α = -0.1 or -0.05 with β = 0, outside the hypotheses of Theorems 1 and 4, so the numerical behavior is not a theorem-forced prediction; this is a support gap/correctness risk, not circularity. No uniqueness theorem is imported from the authors, no ansatz is smuggled in via citation, and no known result is simply renamed. Hence no circular step is identifiable from the paper's own equations.

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

The central theoretical results require convexity or condition (17), smoothness of f, and standard canonical quantization. The algorithm further assumes quantum oracles for f and grad f, plus product-formula discretization with no proven error bound. The hyperparameters alpha, beta, gamma, h, and T0 are hand-chosen, and the reported numerical values fall outside the proven parameter regime; no parameters were fitted to data in the derivation.

free parameters (5)
  • alpha (gradient coupling in Hamiltonian (5)) = -0.1 in convex experiment; -0.05 in nonconvex experiments
    Hand-chosen. Theorem 1's condition 1/alpha >= T0 > 0 requires alpha > 0, but all experiments use negative alpha, so the theory does not cover the tested regime.
  • beta (coefficient of t^3 ||grad f||^2 term) = 0 in all experiments
    Set to zero in all experiments. Theorem 4 requires beta > 0, so the gradient-norm convergence theorem does not apply to the simulated configurations.
  • gamma (coefficient of t^2 f term) = 5 in all experiments
    Hand-chosen. It satisfies gamma >= max(3alpha,0) for the negative alpha values, but no sensitivity analysis is reported.
  • step size h = 0.2 for convex; 0.01, 0.02, 0.005 for nonconvex problems
    Fixed step size treated as an independent parameter; Section 5.1 states that discrete-time convergence is left for future study.
  • initial evolution time T0 (t0) = 1 for convex; 0 for nonconvex
    Theorem statements assume T0 > 0; nonconvex simulations start at t0 = 0, outside the theorem hypotheses.
assumptions (5)
  • domain assumption The objective f is convex and continuously differentiable (Theorems 1 and 4).
    Used in the Lyapunov proofs to guarantee f(x)-x^T grad f(x) <= 0 and f(x) >= 0 (Appendix C.2, Eq. 35).
  • domain assumption For Theorem 4, f satisfies G(x)-x^T grad G(x) <= 0 with G=||grad f||^2.
    Stated as condition (17) and used to make the I2(t) term non-positive in Lemma 5, Appendix C.3.
  • standard math The classical Hamiltonian (7) can be quantized by the canonical mapping p_j -> -i d/dx_j with anti-commutator ordering for non-commuting terms.
    Canonical quantization procedure in Section 3.2; the anti-commutator keeps the gradient term Hermitian.
  • domain assumption Quantum oracles for f and grad f can be implemented with O(1) queries and H_{k,2} can be block-encoded with norm O(alpha d N L).
    Assumed in Lemma 9 and Theorem 6, Appendix D; the complexity statement relies on QSVT block-encoding constructions.
  • ad hoc to paper The exact continuous-time Schrodinger solution is a valid model for the discretized Algorithm 1.
    This is the load-bearing unproven premise: all numerical claims use Algorithm 1, but no discrete-time convergence bound is provided (Section 5.1).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum Optimization via Gradient-Based Hamiltonian Descent." pith.science (2026). https://pith.science/paper/HDHAG7F4

@misc{pith2026250514670,
  author       = {Pith},
  title        = {Pith review of: Quantum Optimization via Gradient-Based Hamiltonian Descent},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HDHAG7F4}},
  note         = {Machine review of arXiv:2505.14670}
}
read the original abstract

With rapid advancements in machine learning, first-order algorithms have emerged as the backbone of modern optimization techniques, owing to their computational efficiency and low memory requirements. Recently, the connection between accelerated gradient methods and damped heavy-ball motion, particularly within the framework of Hamiltonian dynamics, has inspired the development of innovative quantum algorithms for continuous optimization. One such algorithm, Quantum Hamiltonian Descent (QHD), leverages quantum tunneling to escape saddle points and local minima, facilitating the discovery of global solutions in complex optimization landscapes. However, QHD faces several challenges, including slower convergence rates compared to classical gradient methods and limited robustness in highly non-convex problems due to the non-local nature of quantum states. Furthermore, the original QHD formulation primarily relies on function value information, which limits its effectiveness. Inspired by insights from high-resolution differential equations that have elucidated the acceleration mechanisms in classical methods, we propose an enhancement to QHD by incorporating gradient information, leading to what we call gradient-based QHD. Gradient-based QHD achieves faster convergence and significantly increases the likelihood of identifying global solutions. Numerical simulations on challenging problem instances demonstrate that gradient-based QHD outperforms existing quantum and classical methods by at least an order of magnitude.

Figures

Figures reproduced from arXiv: 2505.14670 by the authors.

Figure 1
Figure 1. Numerical comparison of successful probability across iterations for both QHD and gradient-based QHD applied to the Styblinski-Tang function. P k denotes the success probability at iteration k. where the first derivative X˙ represents velocity in classical mechanics. By transforming this equation into its canonical form, we obtain:    X˙ = V, V˙ = − 3 t V˙ − ∇f(X). This canonical form establishes the foundation f… view at source ↗
Figure 2
Figure 2. Numerical performance comparison of various algo￾rithms on the Styblinski-Tang function. Furthermore, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Numerical comparison of various optimization algo￾rithms on the convex objective function (19), including function values and success probability. 6.3. Non-convex optimization More details on these test problems are available in Ap￾pendix E.2. We now turn our attention to the numerical comparison of gradient-based QHD against three baseline algorithms, including SGDM, NAG, and QHD, in non-convex optimiza￾tion settin… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Numerical comparison of various optimization algo￾rithms on the Michalewicz function, including function values and success probability [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Numerical comparison of various optimization algo￾rithms on the Cube-Wave function, including function values and success probability. and yields notably lower terminal objective function values. In the Cube-Wave function, for instance, the final objective value obtain…

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Stochastic Quantum Hamiltonian Descent

    quant-ph 2025-07 conditional novelty 6.0 of 10

    SQHD is a gate-based quantum algorithm that approximates a Lindblad dynamics blending Hamiltonian descent with stochastic component noise, giving an order-2 weak approximation and an O(1/t + eta sigma*) convergence bo...

  2. Heisenberg-Limited Quantum Hamiltonian Learning via Randomly Spread Product-States

    quant-ph 2025-07 conditional novelty 5.0 of 10

    Randomly spread product states and random Pauli measurements activate all spectral gaps of a Hamiltonian, giving a finite-time quadratic Fisher-information window and enabling simultaneous, beyond-Standard-Quantum-Lim...

Reference graph

Works this paper leans on

41 extracted references · 29 canonical work pages · cited by 2 Pith papers

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    and Sra, S

    Ahn, K. and Sra, S. From nesterov’s estimate sequence to riemannian acceleration. In Conference on Learning Theory, pp.\ 84--118. PMLR, 2020

    work page 2020

  3. [3]

    Time-dependent unbounded hamiltonian simulation with vector norm scaling

    An, D., Fang, D., and Lin, L. Time-dependent unbounded hamiltonian simulation with vector norm scaling. Quantum, 5: 0 459, 2021

    work page 2021

  4. [4]

    Time-dependent hamiltonian simulation of highly oscillatory dynamics and superconvergence for schr \"o dinger equation

    An, D., Fang, D., and Lin, L. Time-dependent hamiltonian simulation of highly oscillatory dynamics and superconvergence for schr \"o dinger equation. Quantum, 6: 0 690, 2022

    work page 2022

  5. [5]

    A quantum central path algorithm for linear optimization

    Augustino, B., Leng, J., Nannicini, G., Terlaky, T., and Wu, X. A quantum central path algorithm for linear optimization. arXiv preprint arXiv:2311.03977, 2023

    arXiv 2023

  6. [6]

    W., Childs, A

    Berry, D. W., Childs, A. M., Su, Y., Wang, X., and Wiebe, N. Time-dependent hamiltonian simulation with L ^ 1 -norm scaling. Quantum, 4: 0 254, 2020

    work page 2020

  7. [7]

    I., and Wilson, A

    Betancourt, M., Jordan, M. I., and Wilson, A. C. On symplectic optimization. arXiv preprint arXiv:1802.03653, 2018

    arXiv 2018

  8. [8]

    Quantum langevin dynamics for optimization

    Chen, Z., Lu, Y., Wang, H., Liu, Y., and Li, T. Quantum langevin dynamics for optimization. arXiv preprint arXiv:2311.15587, 2023

    arXiv 2023

Show all 41 references
  1. [9]

    Quantum hamiltonian descent for graph partition

    Cheng, J., Zhou, R., Gan, Y., Qian, C., and Liu, J. Quantum hamiltonian descent for graph partition. arXiv preprint arXiv:2411.14696, 2024

    2024

  2. [10]

    M., Su, Y., Tran, M

    Childs, A. M., Su, Y., Tran, M. C., Wiebe, N., and Zhu, S. Theory of trotter error with commutator scaling. Physical Review X, 11 0 (1): 0 011020, 2021

    2021

  3. [11]

    M., Leng, J., Li, T., Liu, J.-P., and Zhang, C

    Childs, A. M., Leng, J., Li, T., Liu, J.-P., and Zhang, C. Quantum simulation of real-space dynamics. Quantum, 6: 0 860, 2022

    2022

  4. [12]

    Optimizing quantum optimization algorithms via faster quantum gradient computation

    Gily \'e n, A., Arunachalam, S., and Wiebe, N. Optimizing quantum optimization algorithms via faster quantum gradient computation. In Proceedings of the Thirtieth Annual ACM-SIAM Symposium on Discrete Algorithms, pp.\ 1425--1444. SIAM, 2019 a

    2019

  5. [13]

    H., and Wiebe, N

    Gily \'e n, A., Su, Y., Low, G. H., and Wiebe, N. Quantum singular value transformation and beyond: exponential improvements for quantum matrix arithmetics. In Proceedings of the 51st Annual ACM SIGACT Symposium on Theory of Computing, pp.\ 193--204, 2019 b

    2019

  6. [14]

    and Boyd, S

    Giselsson, P. and Boyd, S. Monotonicity and restart in fast gradient methods. In 53rd IEEE Conference on Decision and Control, pp.\ 5058--5063. IEEE, 2014

    2014

  7. [15]

    Riemannian accelerated gradient methods via extrapolation

    Han, A., Mishra, B., Jawanpuria, P., and Gao, J. Riemannian accelerated gradient methods via extrapolation. In International Conference on Artificial Intelligence and Statistics, pp.\ 1554--1585. PMLR, 2023

    2023

  8. [16]

    Jordan, S. P. Fast quantum algorithm for numerical gradient estimation. Physical review letters, 95 0 (5): 0 050501, 2005

    2005

  9. [17]

    and Prakash, A

    Kerenidis, I. and Prakash, A. Quantum gradient descent for linear systems and least squares. Physical Review A, 101 0 (2): 0 022316, 2020

    2020

  10. [18]

    and Tao, M

    Kong, L. and Tao, M. Quantitative convergences of lie group momentum optimizers. arXiv preprint arXiv:2405.20390, 2024

    2024 arXiv

  11. [19]

    Krichene, W., Bayen, A., and Bartlett, P. L. Accelerated mirror descent in continuous and discrete time. Advances in neural information processing systems, 28, 2015

    2015

  12. [20]

    Qhdopt: A software for nonlinear optimization with quantum hamiltonian descent

    Kushnir, S., Leng, J., Peng, Y., Fan, L., and Wu, X. Qhdopt: A software for nonlinear optimization with quantum hamiltonian descent. INFORMS Journal on Computing, 2024

    2024

  13. [21]

    Quantum H amiltonian D escent

    Leng, J., Hickman, E., Li, J., and Wu, X. Quantum H amiltonian D escent. arXiv preprint arXiv:2303.01471, 2023 a

    2023 arXiv

  14. [22]

    A quantum-classical performance separation in nonconvex optimization

    Leng, J., Zheng, Y., and Wu, X. A quantum-classical performance separation in nonconvex optimization. arXiv preprint arXiv:2311.00811, 2023 b

    2023 arXiv

  15. [23]

    Expanding hardware-efficiently manipulable hilbert space via hamiltonian embedding

    Leng, J., Li, J., Peng, Y., and Wu, X. Expanding hardware-efficiently manipulable hilbert space via hamiltonian embedding. arXiv preprint arXiv:2401.08550, 2024

    2024

  16. [24]

    Analysis and design of optimization algorithms via integral quadratic constraints

    Lessard, L., Recht, B., and Packard, A. Analysis and design of optimization algorithms via integral quadratic constraints. SIAM Journal on Optimization, 26 0 (1): 0 57--95, 2016

    2016

  17. [25]

    On efficient quantum block encoding of pseudo-differential operators

    Li, H., Ni, H., and Ying, L. On efficient quantum block encoding of pseudo-differential operators. Quantum, 7: 0 1031, 2023

    2023

  18. [26]

    On efficient optimal transport: An analysis of greedy and accelerated mirror descent algorithms

    Lin, T., Ho, N., and Jordan, M. On efficient optimal transport: An analysis of greedy and accelerated mirror descent algorithms. In International Conference on Machine Learning, pp.\ 3982--3991. PMLR, 2019

    2019

  19. [27]

    Towards provably efficient quantum algorithms for large-scale machine-learning models

    Liu, J., Liu, M., Liu, J.-P., Ye, Z., Wang, Y., Alexeev, Y., Eisert, J., and Jiang, L. Towards provably efficient quantum algorithms for large-scale machine-learning models. Nature Communications, 15 0 (1): 0 434, 2024

    2024

  20. [28]

    J., and Li, T

    Liu, Y., Su, W. J., and Li, T. On quantum speedups for nonconvex optimization via quantum tunneling walks. Quantum, 7: 0 1030, 2023

    2023

  21. [29]

    N., and Fujii, K

    Mizuta, K., Ikeda, T. N., and Fujii, K. Explicit error bounds with commutator scaling for time-dependent product and multi-product formulas. arXiv preprint arXiv:2410.14243, 2024

    2024

  22. [30]

    A method for solving the convex programming problem with convergence rate o (1/k2)

    Nesterov, Y. A method for solving the convex programming problem with convergence rate o (1/k2). In Dokl akad nauk Sssr, volume 269, pp.\ 543, 1983

    1983

  23. [31]

    and Candes, E

    O’donoghue, B. and Candes, E. Adaptive restart for accelerated gradient schemes. Foundations of computational mathematics, 15: 0 715--732, 2015

    2015

  24. [32]

    Quantum gradient descent and newton’s method for constrained polynomial optimization

    Rebentrost, P., Schuld, M., Wossnig, L., Petruccione, F., and Lloyd, S. Quantum gradient descent and newton’s method for constrained polynomial optimization. New Journal of Physics, 21 0 (7): 0 073023, 2019

    2019

  25. [33]

    On the hyperparameters in stochastic gradient descent with momentum

    Shi, B. On the hyperparameters in stochastic gradient descent with momentum. Journal of Machine Learning Research, 25 0 (236): 0 1--40, 2024

    2024

  26. [34]

    S., Jordan, M

    Shi, B., Du, S. S., Jordan, M. I., and Su, W. J. Understanding the acceleration phenomenon via high-resolution differential equations. Mathematical Programming, pp.\ 1--70, 2022

    2022

  27. [35]

    Shi, B., Su, W., and Jordan, M. I. On learning rates and schr \"o dinger operators. Journal of Machine Learning Research, 24 0 (379): 0 1--53, 2023

    2023

  28. [36]

    Siegel, J. W. Accelerated optimization with orthogonality constraints. arXiv preprint arXiv:1903.05204, 2019

    1903 arXiv

  29. [37]

    Su, W., Boyd, S., and Candes, E. J. A differential equation for modeling nesterov's accelerated gradient method: Theory and insights. Journal of Machine Learning Research, 17 0 (153): 0 1--43, 2016

    2016

  30. [38]

    C., and Jordan, M

    Wibisono, A., Wilson, A. C., and Jordan, M. I. A variational perspective on accelerated methods in optimization. proceedings of the National Academy of Sciences, 113 0 (47): 0 E7351--E7358, 2016

    2016

  31. [39]

    C., Recht, B., and Jordan, M

    Wilson, A. C., Recht, B., and Jordan, M. I. A lyapunov analysis of accelerated methods in optimization. Journal of Machine Learning Research, 22 0 (113): 0 1--34, 2021

    2021

  32. [40]

    An improved sample complexity lower bound for (fidelity) quantum state tomography

    Yuen, H. An improved sample complexity lower bound for (fidelity) quantum state tomography. Quantum, 7: 0 890, 2023

    2023

  33. [41]

    Quantum algorithms for escaping from saddle points

    Zhang, C., Leng, J., and Li, T. Quantum algorithms for escaping from saddle points. Quantum, 5: 0 529, 2021

    2021

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.