Pith. sign in

REVIEW 3 major objections 5 minor 44 references

Finite Horizon Optimization: Framework and Applications

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Choosing stepsizes for a fixed T-iteration budget accelerates the primal-dual LP method from O(κ) to O(√κ).

desk verdict Promising finite-horizon stepsize idea for primal-dual LP with a real bug in the headline theorem—still worth a serious referee. read the letter →

arxiv 2412.21068 v1 pith:LWJJ5WLW submitted 2024-12-30 math.OC

Yushun Zhang , Dmitry Rybin , Zhi-Quan Luo This is my paper · ORCID
classification math.OC MSC 90C0590C2290C47
keywords finitehorizonoptimizationprimal-dualmethodstepsizerulesemidefiniteprogrammingminimaxproblemChebyshevpolynomiallinearNetlibbenchmark
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 introduces finite horizon optimization, a framework for tuning an iterative algorithm's hyperparameters so that it performs well after exactly T steps, where T is a fixed and finite iteration budget, rather than as T grows to infinity. Applied to the primal-dual method for linear programming, the paper proposes a Finite Horizon stepsize rule that solves a convex semidefinite program with 4x4 matrix constraints; the SDP's optimality exploits a hidden convexity in the stepsize-selection minimax problem. The paper proves that, at the pre-fixed T-th iteration, this rule reaches $O(\sqrt{\kappa}\log(1/\epsilon))$ complexity, against the $\Omega(\kappa\log(1/\epsilon))$ lower bound for the optimal constant stepsize, for the unconstrained linear-system case. On more than 90 Netlib LP instances, the reported average is a 3.9x speedup over the optimal constant stepsize, saving about 75% of the wall-clock time to reach the same precision.

What carries the argument

The machinery is the block-diagonalization of the primal-dual update matrix $M=\begin{bmatrix}\beta A^\top A & A^\top\\ -A & 0\end{bmatrix}$ into $2\times 2$ blocks $B(\sigma)=\begin{bmatrix}\beta\sigma^2 & \sigma\\ -\sigma & 0\end{bmatrix}$ using the singular value decomposition of $A$. With the cumulative product $\Gamma=(I-\eta_T M)\cdots(I-\eta_1 M)$, the objective $\|\Gamma\|_{\mathrm{op}}$ becomes the spectral norm of a $T$-th-degree matrix polynomial in $B(\sigma)$, and the paper proves that controlling this norm is equivalent to a semidefinite program with $4\times 4$ matrix variables (30); after discretizing the singular-value interval $[\mu,L]$ into $n_{\mathrm{sample}}$ samples, the SDP (31) is solved to obtain the polynomial coefficients. The stepsizes are the reciprocals of the polynomial's roots, computed via a companion matrix. The proof of Theorem 1 then uses the eigen-decomposition of $M$ (Lemma 1), a block-diagonal eigenvalue computation of $\Gamma^\top\Gamma$ (Lemma 2), and the Chebyshev minimax polynomial to bound the resulting contraction by $(1-2/(1+\sqrt{\kappa}))^T$.

What would settle it

For a fixed matrix $A$ with known singular-value bounds, compute Algorithm 1's stepsizes and evaluate the actual worst-case contraction of $\prod_{t=1}^T(I-\eta_t M)$ over the interval $[\mu,L]$ with a much finer grid; if the contraction exceeds $\sqrt{2+4\gamma}(1-2/(1+\sqrt{\kappa}))^T$, then the discretized SDP did not solve the continuous minimax problem and the theorem's bound does not hold for the implemented stepsizes.

Watch

Extended reading notes

Core claim

The central discovery is that the minimax stepsize-selection problem for the primal-dual method—choose $\eta_1,\dots,\eta_T$ to minimize the worst-case contraction $\|(I-\eta_T M)\cdots(I-\eta_1 M)\|$ over LP constraint matrices with singular values in $[\mu,L]$—is exactly convex, despite the non-symmetric update matrix $M$ and the product form of the objective. The paper establishes this by block-diagonalizing $M$ via the SVD of $A$ into $2\times 2$ blocks $B(\sigma)$, rewriting the contraction as a $T$-th-order matrix polynomial, and then casting the spectral-norm control as a linear matrix inequality of size $4\times 4$. The resulting SDP (31) returns polynomial coefficients whose roots give the stepsizes. For the unconstrained case (32), where the LP reduces to solving $Ax=b$, Theorem 1 shows this stepsize rule attains $\mathrm{dist}(z_T,\mathcal{Z}^*)\le \sqrt{2+4\gamma}(1-2/(1+\sqrt{\kappa}))^T\mathrm{dist}(z_0,\mathcal{Z}^*)$, an $O(\sqrt{\kappa}\log(1/\epsilon))$ guarantee at the $T$-th iteration, versus the $\Omega(\kappa\log(1/\epsilon))$ lower bound for the optimal constant stepsize. The experiments then report the 3.9x speedup on Netlib, with the theoretical guarantee applying to the unprojected update while the experiments use the projected update with small measured projection ratios.

Load-bearing premise

The argument hinges on the discretized SDP that outputs the stepsizes being an exact solution of the continuous minimax problem; if that step is approximate, the stated $\sqrt{\kappa}$-rate guarantee at iteration $T$ does not follow.

Editorial extensions

If this is right

  • The same primal-dual algorithm, using only a different stepsize schedule, reduces the condition-number dependence at the T-th iteration from $O(\kappa)$ to $O(\sqrt{\kappa})$ for the unconstrained case.
  • No guarantee is claimed for iterations before T; the experimental gap curves show the error barely improves until the final step, then drops sharply, so the schedule is tuned for one target horizon and must be cycled for longer runs.
  • The SDP subproblem involves only 4x4 matrix inequalities and costs a fixed, negligible time (an average 11.6 seconds, under 3.1% of total runtime on Netlib), independent of the LP dimension.
  • On more than 90 Netlib LP instances, the rule saves on average 74% of iterations and 75% of wall-clock time to reach the same precision as the optimal constant stepsize.
  • If the projection onto the nonnegative orthant rarely activates (measured below 5% in the experiments), the no-projection analysis is a good proxy for the projected method's behavior.

Reading between the lines

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

  • Beyond the paper: the same 2x2 block-diagonalization trick should apply to any saddle-point algorithm whose update matrix is block-structured by a rank-revealing factorization of the constraint matrix, so the hidden-convexity route may extend to PDHG and other first-order LP solvers.
  • Beyond the paper: the sharp final-step increase in the stepsize schedule resembles the silver stepsize patterns in gradient descent; a direct comparison of the two schedules could clarify whether finite-horizon LP stepsize design and accelerated GD share the same underlying minimax principle.
  • Beyond the paper: the paper's own cyclical-repetition trick suggests an anytime variant: repeat the T-step schedule; testing whether the oracle complexity of the cyclic schedule is $O(\sqrt{\kappa}\log(1/\epsilon))$ with the same constant would give an asymptotic guarantee that the single-shot schedule lacks.
  • Beyond the paper: a proof that the projection ratio stays below some small threshold under finite-horizon stepsizes (not just empirical evidence below 5%) would upgrade the numerical speedup into the same theoretical guarantee as Theorem 1 for the projected method.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes a finite-horizon framework for tuning iterative algorithms under a fixed iteration budget T and instantiates it for the primal-dual method for linear programming (LP). For the standard-form LP, the primal update's projection is modeled as a random Bernoulli mask, the residue term in equation (24) is dropped, and the resulting worst-case stepsize design problem is reformulated as a semi-infinite SDP (30), then discretized as SDP (31). The polynomial coefficients obtained from the SDP are factored to produce the Finite Horizon stepsize rule (Algorithm 1). Theorem 1 claims an O(sqrt(kappa) log(1/epsilon)) rate at the pre-fixed T-th iteration for the unprojected problem (32), compared with an Omega(kappa log(1/epsilon)) lower bound for the optimal constant stepsize (Proposition 1). Experiments on the Netlib benchmark report an average 3.9x speedup and a 75% wall-clock time saving.

Significance. The finite-horizon viewpoint is timely, and the reported empirical speedup on a standard LP benchmark is potentially valuable. The paper provides code and uses a reproducible public benchmark. However, the central theoretical guarantee is not established for the algorithm as implemented, and the proof of Theorem 1 contains a concrete algebraic error in the Chebyshev bound. The framework and the empirical study are interesting enough to warrant a major revision, provided the theoretical claims are corrected and explicitly delimited.

major comments (3)
  1. [Section 4.2, Eqs. (45)-(48)] The passage from (47) to (48) silently drops the factor 2 in the standard Chebyshev lower bound C_T(x) >= (x + sqrt(x^2-1))^T / 2. For x = (kappa+1)/(kappa-1), the correct upper bound on 1/C_T(x) is 2 * ((sqrt(kappa)-1)/(sqrt(kappa)+1))^T, not ((sqrt(kappa)-1)/(sqrt(kappa)+1))^T. Consequently the right-hand side of Theorem 1 is smaller than the true minimax polynomial value for concrete parameters (for kappa=100, T=10, gamma=1/3, the stated RHS is about 0.250, while 1/C_10(101/99) is about 0.264). Since the operator norm of the error matrix is at least the maximum of |p(lambda)| over the eigenvalues of M, no real stepsize sequence can satisfy the stated bound. The asymptotic complexity claim survives if the constant is corrected to 2*sqrt(2+4*gamma), but the theorem as written is false.
  2. [Section 4.2, Step 4 and Algorithm 1] The proof uses the identity 'Algorithm 1 = min_eta max_sigma', but Algorithm 1 solves the discretized SDP (31) with nsample=200 grid points and an SCS solver run for a fixed number of iterations (sdp_iter, e.g., 100 or 20). No grid-refinement bound, optimality certificate, or exactness argument is provided, so the theorem does not apply to the polynomial coefficients actually returned by the solver. Additionally, the factorization step in Algorithm 1 assumes that the polynomial p(x) = 1 + a_1 x + ... + a_T x^T has real roots; this is neither guaranteed by the SDP formulation nor verified by the algorithm.
  3. [Theorem 1 vs. Section 5 experiments] Theorem 1 is stated for the unconstrained problem (32), where the update is the linear recurrence (33). The experiments in Section 5 run the projected primal-dual method (10) on standard-form LPs with the nonnegativity constraint x >= 0. The bridge between the two is the heuristic random-projection model in Section 2 and the numerical observation in Figure 6 that the projection ratio is small. This does not prove that the unprojected analysis transfers to the tested algorithm. The abstract's claim of a 'theoretical acceleration guarantee' for the method therefore overstates what is actually proved.
minor comments (5)
  1. [Throughout] There are numerous typographical errors that should be corrected: 'posts' should be 'poses', 'eigenvlaues' should be 'eigenvalues', 'thse' should be 'these', 'asymtopic' should be 'asymptotic', 'intances' should be 'instances', 'primarilly' should be 'primarily', and 'Finit Horizon' should be 'Finite Horizon'.
  2. [Algorithm 1 and Section 5.2] Algorithm 1 says to solve SDP (31) using Interior Point Methods, but Section 5.2 uses the SCS solver with sdp_iter=100 (and sdp_iter=20 for T=10). The solver and its accuracy settings should be stated consistently.
  3. [Table 1] The SIERRA row reports 33931 iterations, which is identical to the PILOT row; this is likely a copy-paste error and should be checked.
  4. [Section 4.2, Step 3] The proof uses the notation r_{i,1} and r_{i,2} without defining it; these are presumably the eigenvalues of B_i^2 from Lemma 2 and should be introduced explicitly.
  5. [Section 5.1] The global optimization method is called 'Stimulated Annealing'; the standard name is 'Simulated Annealing'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the SDP-optimized stepsize and the Chebyshev bound are independent inputs; the flagged issues are correctness/rigor gaps, not circular reductions.

full rationale

The paper's claimed derivation chain is not circular. The Finite Horizon stepsize rule is produced by an SDP (31) whose objective is the T-step worst-case error over singular values, and Theorem 1 then bounds this worst-case error by invoking standard external facts: the block diagonalization of M (Lemmas 1-2), the bound on the cross term, and the Chebyshev minimax polynomial (Proposition 2). The rate (1 - 2/(1+sqrt(kappa)))^T is not an input to the SDP; it is an upper bound supplied by a feasible polynomial p(lambda) with p(0)=1 after the stepsize optimization. The empirical 3.9x speedup on Netlib is a measured comparison, not a fitted parameter renamed as a prediction. Self-citations (Sun et al. 2020, Sun and Ye 2021, Shi et al. 2011) appear only as background on non-symmetric matrix analysis or as a WMMSE reference, and none carries the proof of Theorem 1 or the SDP equivalence. The main weaknesses are correctness/rigor gaps rather than circularity. Section 4.2 Step 4 asserts 'Algorithm 1 = min_{eta} max_{sigma}' although Algorithm 1 solves the discretized SDP (31) with nsample=200 and a fixed SCS iteration budget, with no optimality certificate or grid-refinement bound; the theorem applies to the unprojected problem (32) while the experiments run the projected method (10), bridged only by empirical projection-ratio evidence (Figure 6); and the Chebyshev chain from (47) to (48) silently drops a factor of 2, which affects the stated constant in Theorem 1. None of these reduces a prediction to its inputs by construction, so the circularity score is 0.

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

The paper introduces no new physical or mathematical entities such as particles or forces. The main unproven burdens are modeling choices: the random projection model, the discretization of the sigma interval, and the manual choice of beta and solver tolerances. These are assumption-like rather than entity-like.

free parameters (5)
  • beta (augmented Lagrangian coefficient) = 4/mu
    The paper sets beta = 4/mu in all experiments; the theory only requires beta >= 2/mu. The factor 4 is chosen by hand and affects both the SDP input and the theoretical constants.
  • nsample (number of discretization points for sigma) = 200
    The continuous interval [mu,L] is discretized into nsample evenly spaced points to form the finite SDP (31). No error bound is given for this discretization, and the resulting stepsize is not guaranteed to be feasible for un-sampled sigma values.
  • sdp_iter (SCS solver iterations) = 100 (20 for T=10)
    The SDP is solved approximately with a fixed iteration count, not to a certified optimum. This means the returned stepsizes are not verified minimizers of even the discretized SDP.
  • alpha (SCS parameter) = 1.5
    A solver parameter chosen by the authors; its effect on the solution quality is not analyzed.
  • Gaussian initialization mean = 5 (or 80 in toy example)
    The authors pick N(5,1) to keep the primal variable away from zero and reduce projection frequency. This is a free choice that influences whether the unprojected analysis approximately applies.
assumptions (5)
  • standard math Chebyshev polynomials give the optimal minimax polynomial on an interval with a fixed value at 0 (Proposition 2).
    Used in Step 4 of the proof of Theorem 1 to bound the polynomial growth of the product update.
  • standard math The update matrix M can be block-diagonalized into 2x2 blocks B(sigma) via the SVD of A, and the worst-case over z not in Z* reduces to the worst case over sigma in [mu,L].
    This is the core reduction in Section 2.2, relying on the singular value decomposition and the null-space structure of M. The calculations appear correct.
  • domain assumption The singular values of A lie in [mu, L] with both bounds known to the algorithm.
    The SDP (31) and the theoretical constants depend on mu and L as problem data. In practice these are estimated from A, and the paper does not discuss the effect of estimation error.
  • ad hoc to paper The projection matrices P_t in the primal-dual method can be modeled as i.i.d. Bernoulli random projections with small p_proj, and the residue term (b) in (24) can be neglected.
    Section 2.2 introduces the random projection model and discards the p_proj residue based on numerical observation rather than proof. This is load-bearing for connecting the SDP design to the projected algorithm used in experiments.
  • ad hoc to paper The SDP (31), with nsample discretization and fixed SCS iterations, produces polynomial coefficients whose polynomial p(x) has real roots that can be inverted to valid real stepsizes.
    Algorithm 1 factors p(x) and uses the reciprocals of its roots as stepsizes. A real-rootedness guarantee is not given, and the discretization is not proven to preserve the minimax property. If the roots are complex, the resulting stepsizes are invalid.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Finite Horizon Optimization: Framework and Applications." pith.science (2026). https://pith.science/paper/LWJJ5WLW

@misc{pith2026241221068,
  author       = {Pith},
  title        = {Pith review of: Finite Horizon Optimization: Framework and Applications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LWJJ5WLW}},
  note         = {Machine review of arXiv:2412.21068}
}
abstract

In modern engineering scenarios, there is often a strict upper bound on the number of algorithm iterations that can be performed within a given time limit. This raises the question of optimal algorithmic configuration for a fixed and finite iteration budget. In this work, we introduce the framework of finite horizon optimization, which focuses on optimizing the algorithm performance under a strict iteration budget $T$. We apply this framework to linear programming (LP) and propose Finite Horizon stepsize rule for the primal-dual method. The main challenge in the stepsize design is controlling the singular values of $T$ cumulative product of non-symmetric matrices, which appears to be a highly nonconvex problem, and there are very few helpful tools. Fortunately, in the special case of the primal-dual method, we find that the optimal stepsize design problem admits hidden convexity, and we propose a convex semidefinite programming (SDP) reformulation. This SDP only involves matrix constraints of size $4 \times 4$ and can be solved efficiently in negligible time. Theoretical acceleration guarantee is also provided at the pre-fixed $T$-th iteration, but with no asymptotic guarantee. On more than 90 real-world LP instances, Finite Horizon stepsize rule reaches an average 3.9$\times$ speed-up over the optimal constant stepsize, saving 75\% wall-clock time. Our numerical results reveal substantial room for improvement when we abandon asymptotic guarantees, and instead focus on the performance under finite horizon. We highlight that the benefits are not merely theoretical - they translate directly into computational speed-up on real-world problems.

Figures

Figures reproduced from arXiv: 2412.21068 by the authors.

Figure 1
Figure 1. (a): The growing mismatch between traditional optimization theory and modern applications such as auto-vehicle: classical theory focuses on T → ∞, while many applications can only afford small T. (b): The trajectories of Finite Horizon stepsize rule and the optimal constant stepsize on the augmented Lagrangian function of a simple linear programming (LP). The Finite Horizon stepsize rule for T = 2 reaches the optima… view at source ↗
Figure 2
Figure 2. Some practical applications with strict iteration budget for solvers. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Results on the toy LP instance (54). For Finite Horizon stepsize rule, we find that (a): the optimality gap will drop sharply at the T-th step; (b) there will be a sudden surge at the T-th step; (c): the iterates by Finite Horizon stepsize rules will take smaller steps in the sharp region, and then take large steps in the flat region. Problem (56) is a 1-dimensional (nonconvex) optimization problem. It can be solved… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The evolution spectral norm of three types of error matrices along iterations. We use the total [PITH_FULL_IMAGE:figures/full_fig_p023_4.png]
Figure 5
Figure 5. Figure 5: An example of Finite Horizon stepsize rule on real-world LP instance ( [PITH_FULL_IMAGE:figures/full_fig_p025_5.png]
Figure 6
Figure 6. Figure 6: The projection ratio (i.e., the proportion of entries in x that are forced to 0) along iterations is consis￾tently small (< 5%). In [PITH_FULL_IMAGE:figures/full_fig_p025_6.png]
Figure 7
Figure 7. Figure 7: (a): Finite Horizon stepsize rule for T = 10 on all the intances in Netlib. We observe that the stepsize at the final iteration (T = 10) always increases sharply. (b,c): Average iteration complexity and running time to on Netlib. Our stepsize rule achieves on average 3…
Figure 8
Figure 8. Figure 8: The curves of optimality gap and the detailed stepsize rules (for the first 2 steps) of the results in [PITH_FULL_IMAGE:figures/full_fig_p038_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 18 canonical work pages

  1. [9]

    Defazio, X

    A. Defazio, X. A. Yang, H. Mehta, K. Mishchenko, A. Khaled, and A. Cutkosky. The road less scheduled. arXiv preprint arXiv:2405.15682,

    arXiv

  2. [11]

    O. Fercoq. Monitoring the convergence speed of pdhg to find better primal and dual step sizes. arXiv preprint arXiv:2403.19202,

    work page Pith review arXiv

  3. [13]

    arXiv preprint arXiv:1706.02677,

    arXiv

  4. [14]

    Grimmer, K

    B. Grimmer, K. Shu, and A. L. Wang. Accelerated gradient descent via long steps. arXiv preprint arXiv:2309.09961,

    arXiv

  5. [15]

    Grimmer, K

    B. Grimmer, K. Shu, and A. L. Wang. Accelerated objective gap and gradient norm convergence for gradient descent via long steps. arXiv preprint arXiv:2403.14045, 2024a. B. Grimmer, K. Shu, and A. L. Wang. Composing optimized stepsize schedules for gradient descent. arXiv preprint arXiv:2410.16249, 2024b. B. Grimmer, K. Shu, and A. L. Wang. A strengthened ...

    arXiv

  6. [16]

    S. D. Gupta, R. M. Freund, X. A. Sun, and A. Taylor. Nonlinear conjugate gradient methods: worst-case convergence rates via computer-assisted analyses. arXiv preprint arXiv:2301.01530,

    work page Pith review arXiv

  7. [17]

    S. Hu, Y. Tu, X. Han, C. He, G. Cui, X. Long, Z. Zheng, Y. Fang, Y. Huang, W. Zhao, et al. Minicpm: Unveiling the potential of small language models with scalable training strategies. arXiv preprint arXiv:2404.06395,

    arXiv

  8. [18]

    Ibrahim, B

    A. Ibrahim, B. Thérien, K. Gupta, M. L. Richter, Q. Anthony, T. Lesort, E. Belilovsky, and I. Rish. Simple and scalable strategies to continually pre-train large language models. arXiv preprint arXiv:2403.08763,

    arXiv

Show all 44 references
  1. [19]

    U. Jang, S. D. Gupta, and E. K. Ryu. Computer-assisted design of accelerated composite optimization methods: Optista. arXiv preprint arXiv:2305.15704,

  2. [21]

    Kim and J

    D. Kim and J. A. Fessler. Another look at the fast iterative shrinkage/thresholding algorithm (fista). SIAM Journal on Optimization, 28(1):223–250, 2018a. D. Kim and J. A. Fessler. Generalizing the optimized gradient method for smooth convex minimization. SIAM Journal on Optim...

    1920

  3. [22]

    Kornowski and O

    G. Kornowski and O. Shamir. Open problem: Anytime convergence rate of gradient descent. arXiv preprint arXiv:2406.13888,

    arXiv

  4. [24]

    Li and R

    D. Li and R. Sun. On a faster r-linear convergence rate of the barzilai-borwein method. arXiv preprint arXiv:2101.00205,

    arXiv

  5. [26]

    H. Lu. First-order methods for linear programming. arXiv preprint arXiv:2403.14535,

    arXiv

  6. [27]

    Lu and J

    H. Lu and J. Yang. cupdlp. jl: A gpu implementation of restarted primal-dual hybrid gradient for linear programming in julia. arXiv preprint arXiv:2311.12180, 2023a. H. Lu and J. Yang. On a unified and simplified proof for the ergodic convergence rates of ppm, pdhg and admm. a...

    arXiv

  7. [31]

    Pedregosa

    F. Pedregosa. Residual polynomials and the chebyshev method. http://fa.bianp.net/blog/2020/ polyopt/,

    2020

  8. [34]

    L. N. Smith. Cyclical learning rates for training neural networks. In 2017 IEEE winter conference on applications of computer vision (WACV), pages 464–472. IEEE,

    2017

  9. [35]

    M. Sy. Optimization strategies for low-latency 5g nr ldpc decoding on general purpose processor. In 2023 International Conference on Control, Communication and Computing (ICCC), pages 1–6. IEEE,

    2023

  10. [36]

    B. Wang, S. Ma, J. Yang, and D. Zhou. Relaxed proximal point algorithm: Tight complexity bounds and acceleration without momentum. arXiv preprint arXiv:2410.08890,

    arXiv

  11. [37]

    H. Wang, M. Fazlyab, S. Chen, and V . M. Preciado. Robust convergence analysis of three-operator splitting. In 2019 57th Annual Allerton Conference on Communication, Control, and Computing (Allerton), pages 391–398. IEEE,

    2019

  12. [41]

    limiting error ratios

    Z. Xiong and R. M. Freund. Computational guarantees for restarted pdhg for lp based on" limiting error ratios" and lp sharpness. arXiv preprint arXiv:2312.14774,

  13. [42]

    Xiong and R

    Z. Xiong and R. M. Freund. The role of level-set geometry on the performance of pdhg for conic linear optimization. arXiv preprint arXiv:2406.01942,

    arXiv

  14. [43]

    Zhang and R

    Z. Zhang and R. Jiang. Accelerated gradient descent by concatenation of stepsize schedules. arXiv preprint arXiv:2410.12395,

  15. [44]

    Zhang, J

    Z. Zhang, J. D. Lee, S. S. Du, and Y. Chen. Anytime acceleration of gradient descent. arXiv preprint arXiv:2411.17668,

    arXiv

  16. [1916]

    L. Mones. A gentle introduction to optimal power flow. https://invenia.github.io/blog/2021/06/18/ opf-intro/,

    2021

  17. [1964]

    X. Qian, F. Altché, P . Bender, C. Stiller, and A. de La Fortelle. Optimal trajectory planning for autonomous driving integrating logical constraints: An miqp perspective. In 2016 IEEE 19th international conference on intelligent transportation systems (ITSC), pages 205–210. IEEE,

    2016

  18. [1984]

    Luner and B

    A. Luner and B. Grimmer. Performance estimation for smooth and strongly convex sets. arXiv preprint arXiv:2410.14811,

    arXiv

  19. [1988]

    Rotaru, F

    T. Rotaru, F. Glineur, and P . Patrinos. Exact worst-case convergence rates of gradient descent: a complete analysis for all constant stepsizes over nonconvex and convex functions. arXiv preprint arXiv:2406.17506,

  20. [1997]

    J. Wu, P . L. Bartlett, M. Telgarsky, and B. Yu. Large stepsize gradient descent for logistic loss: Non- monotonicity of the loss improves optimization efficiency. arXiv preprint arXiv:2402.15926,

    arXiv

  21. [2004]

    Y. Cai, J. Wu, S. Mei, M. Lindsey, and P . L. Bartlett. Large stepsize gradient descent for non-homogeneous two-layer networks: Margin improvement and fast optimization. arXiv preprint arXiv:2406.08654,

    arXiv

  22. [2006]

    Cyrus, B

    S. Cyrus, B. Hu, B. Van Scoy, and L. Lessard. A robust accelerated optimization algorithm for strongly convex functions. In 2018 Annual American Control Conference (ACC), pages 1376–1381. IEEE,

    2018

  23. [2008]

    Bok and J

    J. Bok and J. M. Altschuler. Accelerating proximal gradient descent via silver stepsizes. arXiv preprint arXiv:2412.05497,

    arXiv

  24. [2011]

    G. C. Chasparis and J. S. Shamma. Linear-programming-based multi-vehicle path planning with adversaries. In Proceedings of the 2005, American Control Conference, 2005., pages 1072–1077. IEEE,

    2005

  25. [2012]

    H. Fu, C. Li, X. Liu, J. Gao, A. Celikyilmaz, and L. Carin. Cyclical annealing schedule: A simple approach to mitigating kl vanishing. arXiv preprint arXiv:1903.10145,

    1903 arXiv

  26. [2013]

    Z. Xiong. Accessible theoretical complexity of the restarted primal-dual hybrid gradient method for linear programs with unique optima. arXiv preprint arXiv:2410.04043,

    arXiv

  27. [2015]

    O’Donoghue

    34 B. O’Donoghue. Operator splitting for a homogeneous embedding of the linear complementarity problem. SIAM Journal on Optimization, 31(3):1999–2023,

    1999

  28. [2016]

    B. Li, L. Yang, Y. Chen, S. Wang, Q. Chen, H. Mao, Y. Ma, A. Wang, T. Ding, J. Tang, et al. Pdhg-unrolled learning-to-optimize method for large-scale linear programming. arXiv preprint arXiv:2406.01908,

    arXiv

  29. [2017]

    De Klerk, F

    E. De Klerk, F. Glineur, and A. B. Taylor. Worst-case convergence analysis of inexact gradient and newton methods through semidefinite programming performance estimation. SIAM Journal on Optimization, 30(3): 2053–2082,

  30. [2018]

    J. M. Altschuler and P . A. Parrilo. Acceleration by stepsize hedging i: Multi-step descent and the silver stepsize schedule. arXiv preprint arXiv:2309.07879,

    arXiv

  31. [2019]

    36 K. Wen, Z. Li, J. Wang, D. Hall, P . Liang, and T. Ma. Understanding warmup-stable-decay learning rates: A river valley loss landscape perspective. arXiv preprint arXiv:2410.05192,

    arXiv

  32. [2020]

    Defazio, A

    A. Defazio, A. Cutkosky, H. Mehta, and K. Mishchenko. When, why and how much? adaptive learning rate scheduling by refinement. arXiv preprint arXiv:2310.07831,

    arXiv

  33. [2021]

    Loshchilov and F

    I. Loshchilov and F. Hutter. Sgdr: Stochastic gradient descent with warm restarts. arXiv preprint arXiv:1608.03983,

    arXiv

  34. [2022]

    Kiessling, A

    D. Kiessling, A. Zanelli, A. Nurkanovi´ c, J. Gillis, M. Diehl, M. Zeilinger, G. Pipeleers, and J. Swevers. A feasible sequential linear programming algorithm with application to time-optimal path planning problems. In 2022 IEEE 61st Conference on Decision and Control (CDC), p...

    2022

  35. [2023]

    Babaeinejadsarookolaee, A

    S. Babaeinejadsarookolaee, A. Birchfield, R. D. Christie, C. Coffrin, C. DeMarco, R. Diao, M. Ferris, S. Flis- counakis, S. Greene, R. Huang, et al. The power grid library for benchmarking ac optimal power flow algorithms. arXiv preprint arXiv:1908.02788,

    1908 arXiv

  36. [2024]

    Q. Deng, Q. Feng, W. Gao, D. Ge, B. Jiang, Y. Jiang, J. Liu, T. Liu, C. Xue, Y. Ye, et al. New developments of admm-based interior point methods for linear programming and conic programming. arXiv preprint arXiv:2209.01793,

    arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.