Pith. sign in

REVIEW 3 major objections 6 minor 39 references

Adding Hessian-driven damping to inertial primal-dual dynamics makes the primal-dual gap decay as O(1/t^{α-1}) for strongly convex saddle point problems, even when the strong convexity parameters are unknown.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 00:26 UTC pith:CIPOXQXA

load-bearing objection A solid extension of inertial primal-dual dynamics with Hessian damping; the adaptive strongly-convex rate holds, but Lemma 10 has a curable typo and global existence is assumed. the 3 major comments →

arxiv 2607.26235 v2 pith:CIPOXQXA submitted 2026-07-28 math.OC

Inertial Primal Dual Dynamics with Hessian-driven Damping for Saddle Point Problems

Zepeng Wang , Juan Peypouquet This is my paper
classification math.OC MSC 37N4090C25
keywords saddle point problemsinertial primal dual dynamicsHessian-driven dampingstrong convexityconvergence ratesaugmented LagrangianLyapunov energycontinuous-time optimization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper proposes two inertial primal-dual dynamical systems for bilinearly coupled saddle point problems, each modified with Hessian-driven damping terms. For convex-concave problems, the trajectories approach a saddle point with the primal-dual gap decaying as O(1/t²). The central new result is that when f and g are strongly convex—even if the strong convexity parameters are unknown—the same system achieves a faster asymptotic rate O(1/t^{α-1}) for any α≥3, provided the damping parameter r is at least max(β_f,β_g). If the parameters are known, a separate heavy-ball style system with Hessian damping gives exponential convergence, and a third system extends the approach to affinely constrained optimization via the augmented Lagrangian. A sympathetic reader cares because this suggests Hessian-driven damping can reconcile smoother trajectories with parameter-free acceleration in continuous-time optimization.

Core claim

The paper's central claim is that including Hessian-driven damping terms β_f ∇²f(x)ẋ and β_g ∇²g(y)ẏ in the inertial primal-dual system (PD-AVD-H) yields an asymptotic primal-dual gap rate O(1/t^{α-1}) for strongly convex–strongly concave saddle point problems, without any knowledge of the strong convexity parameters μ_f, μ_g. This is established in Theorem 12(iii) under α≥3, θ=1/(α-1), r≥max(β_f,β_g), and t≥T, where T depends on the possibly unknown μ's. The proof uses a Lyapunov energy whose derivative becomes non-positive after t≥T, with the Hessian terms contributing negative quadratic terms that accelerate the decay. The paper also proves O(1/t²) for the convex case, exponential decay w

What carries the argument

The central object is the energy functional W(t), defined in (9) for the convex/adaptive analysis and in (14) for the known-parameter case. W(t) combines a kinetic term ||tṡ+(α-1)(s-s*)||², a primal-dual gap term t(γt+r)F(s), and cross terms involving ⟨∇f(x),x-x*⟩-(f(x)-f(x*)) (and the analogous g-term), each weighted by the Hessian damping coefficients β_f and β_g. Lemma 5 shows that with θ=1/(α-1) the bilinear coupling terms cancel exactly, and the derivative Ẇ becomes a sum of negative semidefinite terms; the Hessian damping appears as -β_f t²⟨∇²f(x)ẋ,ẋ⟩ and -β_g t²⟨∇²g(y)ẏ,ẏ⟩. Under strong convexity these are bounded below by -β_f μ_f t²‖ẋ‖² and -β_g μ_g t²‖ẏ‖², which in Lemma 10 convert

Load-bearing premise

The theorems assume the existence of a global twice-differentiable trajectory (x,y):[t₀,∞)→X×Y, but no existence, uniqueness, or continuation result is proved; in infinite-dimensional Hilbert spaces, C² without Lipschitz gradients is not enough to guarantee such a trajectory exists.

What would settle it

Numerically integrate (PD-AVD-H) for a strongly convex–strongly concave quadratic, e.g., f(x)=μ_x‖x‖²/2, g(y)=μ_y‖y‖²/2, A=I, with α=4, any β>0, θ=1/3, and check whether the primal-dual gap L(x(t),y*)-L(x*,y(t)) decays as t^{-3} for large t; if the observed exponent is strictly less than 3, Theorem 12(iii) fails. Alternatively, verify algebraically that the energy W(t) from (9) satisfies the differential inequality tẆ+(α-3)W≤0 after the time T defined in the theorem—a mistake in Lemma 10 would invalidate the entire strong-convexity analysis.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If correct, (PD-AVD-H) is an inertial primal-dual dynamical system whose strongly convex convergence rate 1/t^{α-1} does not require knowledge of the strong convexity parameters—the acceleration is inherent to the flow.
  • The same Lyapunov framework yields O(1/t²) in the convex case and exponential decay in the known-parameter case, so a single family of systems covers all three regimes with a unified energy analysis.
  • Hessian-driven damping does not slow convergence; it reduces trajectory oscillations (demonstrated numerically in Section 4.4) while preserving or improving the rates, showing that smoothing and fast convergence are compatible.
  • For affinely constrained problems, the augmented-Lagrangian system with Hessian damping achieves O(1/t²) for both the gap and the constraint violation, extending earlier primal-dual dynamics to the Hessian-damped setting.
  • The rates are asymptotic and hold only after a time T that depends on the strong convexity parameters; larger α gives faster decay but also a potentially later onset of the accelerated regime.

Where Pith is reading between the lines

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

  • If the O(1/t^{α-1}) rate is genuine, a discretization of (PD-AVD-H) might inherit this adaptivity without restart or line-search, though the paper provides no discrete algorithm and the Hessian terms would need numerical approximation in practice.
  • The parameter-free claim is only asymptotic: for problems with very small μ_f or μ_g, the time T before acceleration appears may be enormous, so the improved rate could be unobservable in practice for ill-conditioned problems.
  • Because the system requires C² functions, the analysis does not cover nonsmooth or piecewise-smooth convex functions; a natural extension would be a differential-inclusion formulation or a first-order reduction to see whether the same rates persist.
  • The energy proof relies on the exact cancellation at θ=1/(α-1); small perturbations of θ, as explored in other primal-dual dynamics, may destroy the monotonicity of W, suggesting that the rate could degrade discontinuously with θ.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The manuscript proposes three inertial primal-dual second-order ODE systems with Hessian-driven damping: two for bilinearly coupled saddle point problems (PD-AVD-H and PD-HBF-H) and one for affinely constrained convex optimization problems. For the convex-concave case the paper proves trajectory boundedness, O(1/t) velocity decay (O(1/t^2) in squared norm), and an O(1/t^2) decay of the primal-dual gap. When f and g are strongly convex with unknown strong-convexity parameters, it claims the adaptive rate O(1/t^{alpha-1}) for the gap with alpha>=3, under the condition r>=max(beta_f,beta_g) and theta=1/(alpha-1). When the strong-convexity parameters are known, a linearly convergent system is proposed. For affinely constrained problems, O(1/t^2) rates are proved for the augmented Lagrangian gap, the feasibility measure, and the objective value. The proofs are Lyapunov-based and largely self-contained.

Significance. If correct, the main contribution is Theorem 12(iii): Hessian-driven damping yields an asymptotic O(1/t^{alpha-1}) primal-dual gap rate under strong convexity without knowledge of mu_f and mu_g, improving on the convex O(1/t^2) rate. This is a nontrivial and interesting extension of the inertial primal-dual literature. The manuscript provides explicit Lyapunov functions, parameter thresholds, and rates, and the proofs are mostly self-contained. The affinely constrained application complements existing results [27,28], and the numerical example illustrates the expected stabilization effect. The main caveats are an algebraic error in Lemma 10 and the absence of an existence/continuation statement for the trajectories.

major comments (3)
  1. [Lemma 10, Eq. (12)] The coefficient multiplying beta_f and beta_g in the Bregman brackets is incorrect. Combining Lemma 5 with inequality (12) gives the bracket gamma t + r - (alpha-2) beta_f - 2(alpha-1)(alpha-3)/(mu_f t), not gamma t + r - (alpha-4) beta_f - ...; the same holds for beta_g. For alpha=5, beta=1, gamma t + r = 6, mu=1, t=5, the printed bracket is positive while the correct bracket is negative, reversing the sign of that contribution. Therefore the proof of Lemma 10 as written does not establish the inequality needed for Theorem 12. The rate is likely recoverable because the explicit T in Theorem 12 is already large enough when r>=beta_f,beta_g, but the statement and proof must be corrected.
  2. [Sections 2-4, Theorems 7/12/15/25] Every theorem begins 'Let (x,y) be a solution of the system' and no existence or continuation result is supplied. Under the stated C^2 assumptions on Hilbert spaces, local existence is standard, and the Lyapunov estimates give the a priori bounds needed for global continuation, but this is never stated. Please add a proposition or remark on local existence and global continuation (or explicitly assume that a global solution exists). Without this, the convergence rates are conditional on a hypothesis that is not guaranteed by the stated assumptions.
  3. [Theorem 25, proof] The proof asserts that, 'proceeding as in the proof of Lemma 18', the energy E_{\ell_\tau} satisfies the differential inequality with L_\rho(x,\ell_\tau)-L_\rho(x*,\ell_\tau) in place of F(x). This is not immediate: E_\ell uses the shifted equilibrium s*_\ell and L_\rho(x,\ell)-L_\rho(x*,\ell) contains the extra term <ell-y*, Ax-b>, whose contribution must be tracked through the dynamics. This calculation is the core of Theorem 25 and should be included explicitly or stated as a separate lemma.
minor comments (6)
  1. [Lemma 5, proof] In the expression for \dot W_1 there is a typo: 'beta_g t <\nabla^2 g(y) \dot x, y-y*>' should read '\dot y'.
  2. [Theorem 15, proof] The proof lists a part (iv) ('the result follows by applying smoothness') that is not present in the theorem statement. Either add the statement or remove the item.
  3. [Section 4] The label (PD-AVD-H) is reused for the augmented-Lagrangian system, although it is a different system from the one in Section 2. Rename to avoid confusion, e.g., (PD-AL-H).
  4. [Figure 1] The axis labels appear garbled ('10□2', '10□1', etc.); the formatting should be fixed.
  5. [Theorem 25] The parameter xi=1/theta is not defined in the theorem statement; define it explicitly or refer to Lemma 18 where it is set.
  6. [Abstract] The abstract says 'two inertial primal dual dynamical systems' but the paper proposes a third system in Section 4; adjust the wording.

Circularity Check

0 steps flagged

No significant circularity; the convergence proofs are self-contained Lyapunov estimates with no fitted inputs or load-bearing self-citations.

full rationale

The paper's results are derived from the proposed dynamics via explicit energy functions and differential inequalities. No parameter is fitted to data and then renamed a prediction; the rates O(1/t^2), O(1/t^{alpha-1}), and O(e^{-alpha t/2}) follow from Lyapunov inequalities whose coefficients (alpha, beta_f, beta_g, gamma, r, theta) are free and explicitly stated. The self-citations [14] and [21] are contextual background references, not used to justify the central convergence theorems. The unproved existence of a global twice-differentiable trajectory is a regularity/existence assumption, not a circular step. The skeptical concern about Lemma 10's coefficient (alpha-4) versus the algebraically derived (alpha-2) is a possible correctness issue in the proof as printed, but it does not make the argument circular, since the desired bound is not assumed as an input.

Axiom & Free-Parameter Ledger

7 free parameters · 4 axioms · 0 invented entities

The central claims rest on standard convex analysis plus the unproved existence of global solutions. The parameters of the dynamics are free, not fitted. No new physical or mathematical entities are introduced.

free parameters (7)
  • α
    Damping parameter; α≥3; determines the O(1/t^{α−1}) rate.
  • β_f
    Hessian damping coefficient on x; needs r≥β_f; enables strong-convexity adaptivity.
  • β_g
    Hessian damping coefficient on y; needs r≥β_g; enables strong-convexity adaptivity.
  • γ
    Gradient scaling coefficient; appears as the 1/(γ t²) rate constant.
  • r
    Extra damping coefficient in (γ+r/t); condition r≥max(β_f,β_g).
  • θ
    Coupling time-scale; fixed as 1/(α−1) in §2, 2/α in §3, and in [1/(α−1),1/2] in §4.
  • ρ
    Augmented Lagrangian penalty parameter in §4; ρ>0.
axioms (4)
  • domain assumption Existence of a global C² solution (x,y):[t0,∞)→X×Y for the dynamics
    All theorems are conditional on this; not proved.
  • domain assumption f and g are convex and twice continuously differentiable on real Hilbert spaces; A is bounded linear
    Standing assumptions of the paper.
  • domain assumption f,g strongly convex with moduli μ_f,μg>0
    For Theorems 12 and 15.
  • standard math Standard convex analysis: for convex C² f, ⟨∇²f(x)v,v⟩≥0; for μ-strongly convex f, ⟨∇f(x),x−x*⟩ − (f(x)−f(x*)) ≥ (μ/2)||x−x*||²
    Used in Lemmas 5, 10, 14 and Theorem 15.

pith-pipeline@v1.3.0-alltime-deepseek · 17233 in / 40943 out tokens · 310955 ms · 2026-08-01T00:26:50.081322+00:00 · methodology

0 comments
read the original abstract

Featuring Hessian-driven damping, two inertial primal dual dynamical systems are proposed for solving smooth saddle point problems with bilinear coupling. For convex-concave functions, we establish a convergence rate $\mathcal{O}\left( \frac{1}{t^2} \right)$ for the primal dual gap; for strongly convex-strongly concave functions, we obtain an asymptotic rate $\mathcal{O}\left( \frac{1}{t^{\alpha-1}} \right)$ ($\alpha\ge 3$ is the damping parameter) without knowledge of the strong convexity parameters, and an accelerated linear convergence rate when the strong convexity parameters are known. As an application of the proposed inertial systems, we also consider the affinely constrained convex optimization problem, and develop an inertial system with Hessian-driven damping, which complements existing results.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

39 extracted references · 7 linked inside Pith

  1. [1]

    Soviet Mathematics Doklady27(2), 372–376 (1983)

    Nesterov, Y.: A method for solving the convex programming problem with convergence rateO 1 k2 . Soviet Mathematics Doklady27(2), 372–376 (1983)

    1983

  2. [2]

    SIAM Journal on Imaging Sciences2(1), 183–202 (2009)

    Beck, A., Teboulle, M.: A fast iterative shrinkage-thresholding algorithm for linear inverse problems. SIAM Journal on Imaging Sciences2(1), 183–202 (2009)

    2009

  3. [3]

    Mathematical Programming159, 81–107 (2016)

    Kim, D., Fessler, J.A.: Optimized first-order methods for smooth convex mini- mization. Mathematical Programming159, 81–107 (2016)

    2016

  4. [4]

    Applied Mathematics & Optimization88(77) (2023)

    Park, C., Park, J., Ryu, E.K.: Factor- √ 2 acceleration of accelerated gradient methods. Applied Mathematics & Optimization88(77) (2023)

    2023

  5. [5]

    Journal of Machine Learning Research17(153), 1–43 (2016)

    Su, W., Boyd, S., Cand` es, E.J.: A differential equation for modeling Nesterov’s accelerated gradient method: Theory and insights. Journal of Machine Learning Research17(153), 1–43 (2016)

    2016

  6. [6]

    Mathematical Programming168, 123–175 (2018)

    Attouch, H., Chbani, Z., Peypouquet, J., Redont, P.: Fast convergence of iner- tial dynamics and algorithms with asymptotic vanishing viscosity. Mathematical Programming168, 123–175 (2018)

    2018

  7. [7]

    arXiv:2510.23513 (2025)

    Jang, U., Ryu, E.K.: Point convergence of Nesterov’s accelerated gradient method: An AI-assisted proof. arXiv:2510.23513 (2025)

    arXiv 2025

  8. [8]

    Turkish Journal of Mathematics41(3), 681–685 (2017)

    May, R.: Asymptotic for a second-order evolution equation with convex potential and vanishing damping term. Turkish Journal of Mathematics41(3), 681–685 (2017)

    2017

  9. [9]

    SIAM Journal on Optimization26(3), 1824–1834 (2016)

    Attouch, H., Peypouquet, J.: The rate of convergence of Nesterov’s acceler- ated forward-backward method is actually faster than 1/k 2. SIAM Journal on Optimization26(3), 1824–1834 (2016)

    2016

  10. [10]

    Journal of Differential Equations 20 261(10), 5734–5783 (2016)

    Attouch, H., Peypouquet, J., Redont, P.: Fast convex optimization via iner- tial dynamics with Hessian driven damping. Journal of Differential Equations 20 261(10), 5734–5783 (2016)

    2016

  11. [11]

    Journal de Math´ ematiques Pures et Appliqu´ ees81(8), 747–779 (2002)

    Alvarez, F., Attouch, H., Bolte, J., Redont, P.: A second-order gradient-like dissipative dynamical system with Hessian-driven damping.: Application to opti- mization and mechanics. Journal de Math´ ematiques Pures et Appliqu´ ees81(8), 747–779 (2002)

    2002

  12. [12]

    Mathematical Programming 195, 79–148 (2022)

    Shi, B., Du, S.S., Jordan, M.I., Su, W.: Understanding the acceleration phe- nomenon via high-resolution differential equations. Mathematical Programming 195, 79–148 (2022)

    2022

  13. [13]

    SIAM Journal on Optimization34(2), 2150–2168 (2024)

    Li, B., Shi, B., Yuan, Y.: Linear convergence of forward-backward accelerated algorithms without knowledge of the modulus of strong convexity. SIAM Journal on Optimization34(2), 2150–2168 (2024)

    2024

  14. [14]

    arXiv:2506.21730 (2025)

    Wang, Z., Peypouquet, J.: Fast convex optimization via inertial systems with asymptotically vanishing viscosity and Hessian-driven damping. arXiv:2506.21730 (2025)

    Pith/arXiv arXiv 2025

  15. [15]

    USSR computational mathematics and mathematical physics4(5), 791–803 (1964)

    Polyak, B.T.: Some methods of speeding up the convergence of iteration meth- ods. USSR computational mathematics and mathematical physics4(5), 791–803 (1964)

    1964

  16. [16]

    the continuous dynamical system: global exploration of the local minima of a real-valued function by asymptotic analysis of a dissipative dynamical system

    Attouch, H., Goudou, X., Redont, P.: The heavy ball with friction method, I. the continuous dynamical system: global exploration of the local minima of a real-valued function by asymptotic analysis of a dissipative dynamical system. Communications in Contemporary Mathematics2(1), 1–34 (2000)

    2000

  17. [17]

    arXiv:1903.05671 (2019)

    Siegel, J.W.: Accelerated first-order methods: Differential equations and Lya- punov functions. arXiv:1903.05671 (2019)

    Pith/arXiv arXiv 1903

  18. [18]

    Mathematical Programming195, 735–781 (2022)

    Luo, H., Chen, L.: From differential equation solvers to accelerated first-order methods for convex optimization. Mathematical Programming195, 735–781 (2022)

    2022

  19. [19]

    SIAM Journal on Optimization 32(3), 1817–1842 (2022)

    Aujol, J.-F., Dossal, C., Rondepierre, A.: Convergence rates of the heavy ball method for quasi-strongly convex optimization. SIAM Journal on Optimization 32(3), 1817–1842 (2022)

    2022

  20. [20]

    Mathematical Programming 193, 113–155 (2022)

    Attouch, H., Chbani, Z., Fadili, J., Riahi, H.: First-order optimization algorithms via inertial systems with Hessian driven damping. Mathematical Programming 193, 113–155 (2022)

    2022

  21. [21]

    arXiv:2502.16953 (2025)

    Wang, Z., Peypouquet, J.: Accelerated gradient methods via inertial systems with Hessian-driven damping. arXiv:2502.16953 (2025)

    Pith/arXiv arXiv 2025

  22. [22]

    21 Springer, New York (2004)

    Nesterov, Y.: Introductory Lectures on Convex Optimization: A Basic Course. 21 Springer, New York (2004)

    2004

  23. [23]

    IF AC-PapersOnLine53(2), 7362–7367 (2020)

    Zeng, X., Dou, L., Chen, J.: Accelerated first-order continuous-time algorithm for solving convex-concave bilinear saddle point problem. IF AC-PapersOnLine53(2), 7362–7367 (2020)

    2020

  24. [24]

    arXiv:2606.18724 (2026)

    He, X., Fang, Y.-P.: Fast primal-dual methods for convex-concave bilinear saddle point problems: continuous-time dynamics and discrete algorithms. arXiv:2606.18724 (2026)

    Pith/arXiv arXiv 2026

  25. [25]

    Applied Mathematics & Optimization 89(30) (2024)

    He, X., Hu, R., Fang, Y.: A second order primal–dual dynamical system for a con- vex–concave bilinear saddle point problem. Applied Mathematics & Optimization 89(30) (2024)

    2024

  26. [26]

    Computational Optimization and Applications90, 151–192 (2025)

    Ding, K., Fliege, J., Vuong, P.T.: Fast convergence of the primal-dual dynamical system and corresponding algorithms for a nonsmooth bilinearly coupled sad- dle point problem. Computational Optimization and Applications90, 151–192 (2025)

    2025

  27. [27]

    IEEE Transactions on Automatic Control68(3), 1760–1767 (2023)

    Zeng, X., Lei, J., Chen, J.: Dynamical primal-dual Nesterov accelerated method and its application to network optimization. IEEE Transactions on Automatic Control68(3), 1760–1767 (2023)

    2023

  28. [28]

    Journal of Differential Equations303, 369–406 (2021)

    Bot ¸, R.I., Nguyen, D.-K.: Improved convergence rates and trajectory conver- gence for primal-dual dynamical systems with vanishing damping. Journal of Differential Equations303, 369–406 (2021)

    2021

  29. [29]

    arXiv:2605.18236 (2026)

    He, X., Huang, N.-J., Xiao, Y.-B., Fang, Y.-P.: Trajectory convergence ando(t −2) rates for Nesterov accelerated primal-dual dynamics without Lipschitz gradient assumption. arXiv:2605.18236 (2026)

    Pith/arXiv arXiv 2026

  30. [30]

    Applied Mathematics & Optimization 88(27) (2023)

    Hulett, D.A., Nguyen, D.-K.: Time rescaling of a primal-dual dynamical system with asymptotically vanishing damping. Applied Mathematics & Optimization 88(27) (2023)

    2023

  31. [31]

    SIAM Journal on Control and Optimization59(5), 3278–3301 (2021)

    He, X., Hu, R., Fang, Y.P.: Convergence rates of inertial primal-dual dynamical methods for separable convex optimization problems. SIAM Journal on Control and Optimization59(5), 3278–3301 (2021)

    2021

  32. [32]

    second-order primal

    He, X., Hu, R., Fang, Y.-P.: “second-order primal” + “first-order dual” dynamical systems with time scaling for linear equality constrained convex optimization problems. IEEE Transactions on Automatic Control67(8), 4377–4383 (2022)

    2022

  33. [33]

    Automatica146, 110547 (2022) 22

    He, X., Hu, R., Fang, Y.-P.: Fast primal–dual algorithm via dynamical system for a linearly constrained convex optimization problem. Automatica146, 110547 (2022) 22

    2022

  34. [34]

    Journal of Optimization Theory and Applications193, 704–736 (2022)

    Attouch, H., Chbani, Z., Fadili, J., Riahi, H.: Fast convergence of dynamical ADMM via time scaling of damped inertial dynamics. Journal of Optimization Theory and Applications193, 704–736 (2022)

    2022

  35. [35]

    Mathematical Programming200, 147–197 (2023)

    Bot ¸, R.I., Csetnek, E.R., Nguyen, D.-K.: Fast augmented Lagrangian method in the convex regime with convergence guarantees for the iterates. Mathematical Programming200, 147–197 (2023)

    2023

  36. [36]

    arXiv:2605.19467 (2026)

    He, X., Huang, N.-J., Xiao, Y.-B., Fang, Y.-P.: Convergence of iterates and improved rates for accelerated augmented Lagrangian methods for linearly constrained convex optimization. arXiv:2605.19467 (2026)

    Pith/arXiv arXiv 2026

  37. [37]

    Optimization74(2), 365–390 (2025)

    He, X., Tian, F., Li, A.-q., Fang, Y.-P.: Convergence rates of mixed primal-dual dynamical systems with Hessian driven damping. Optimization74(2), 365–390 (2025)

    2025

  38. [38]

    Foundations of Computational Mathematics25, 163–222 (2025)

    Bot ¸, R.I., Csetnek, E.R., Nguyen, D.-K.: Fast optimistic gradient descent ascent (OGDA) method in continuous and discrete time. Foundations of Computational Mathematics25, 163–222 (2025)

    2025

  39. [39]

    arXiv:2509.08258 (2025) 23

    He, X., Fang, Y.-P.: Nesterov acceleration for strongly convex-strongly con- cave bilinear saddle point problems: discrete and continuous-time approaches. arXiv:2509.08258 (2025) 23

    Pith/arXiv arXiv 2025