Pith. sign in

REVIEW 2 major objections 2 minor 26 references

Only complete primal-dual warm starts reliably accelerate ACOPF solves with IPOPT; partial initializations often increase time or cause failures.

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 · grok-4.3

2026-06-27 15:58 UTC pith:LUJ3L5FW

load-bearing objection Only the complete oracle primal-dual restart reliably speeds up IPOPT ACOPF solves on these cases; most partial combinations slow things down or fail. the 2 major comments →

arxiv 2606.08984 v1 pith:LUJ3L5FW submitted 2026-06-08 math.OC cs.SYeess.SY

Not All Warm Starts Help: Benchmarking Primal-Dual Initializations for ACOPF Algorithms

Babak Taheri , Daniel K. Molzahn This is my paper
classification math.OC cs.SYeess.SY
keywords ACOPFwarm startsprimal-dual initializationIPOPToptimal power flowinterior-point methodsconvergence reliabilityPGLib
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.

The paper benchmarks initialization strategies for AC optimal power flow solved by the interior-point method IPOPT across 19 PGLib-OPF networks ranging from 5 to 30,000 buses. It tests every non-empty subset of the primal blocks together with matching dual multipliers under oracle conditions, plus several practical DC-seeded variants. Only the full primal-plus-dual restart produces consistent convergence on all baseline-convergent cases and delivers a 47.6 percent median solve-time reduction. Twelve of the fourteen partial oracle configurations produce negative median speedups, and several fail repeatedly on larger instances. The results further show that full coverage of bound multipliers drives robustness far more than partial dual information, and that the net benefit of DC seeding disappears once presolve cost is counted.

Core claim

Among the oracle primal-plus-dual configurations, only the complete restart reliably converges on every baseline-convergent case, reaching a 47.6% median solve-time speedup. Twelve of the 14 partial O-PD combinations have negative median speedups, and several fail repeatedly on larger networks. Decomposing the dual into constraint and bound multipliers shows that coverage, not the presence of duals per se, governs robustness: the full bound-multiplier vector reaches 90.7% convergence and a +26.8% median speedup, whereas block-matched coverage drops to 70.4% and -31.1%.

What carries the argument

Oracle primal-plus-dual (O-PD) configurations that test all 15 non-empty subsets of the primal blocks {P_g, Q_g, V_m, V_a} together with their matching dual multipliers, further decomposed into constraint versus bound multipliers.

Load-bearing premise

The 19 PGLib-OPF instances and the specific choice of IPOPT are representative enough for the observed patterns on partial versus full coverage to generalize to other solvers, real-world networks, and practical non-oracle predictors.

What would settle it

Repeating the full set of oracle and DC-seeded benchmarks on the same 19 instances but with a different interior-point solver such as KNITRO and finding that several partial primal-dual combinations then match or exceed the complete restart would falsify the necessity of full coverage.

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

If this is right

  • Only full coverage of both primal variables and dual multipliers produces reliable convergence and positive median speedup.
  • Full bound-multiplier coverage alone reaches 90.7 percent convergence and +26.8 percent median speedup.
  • Any block-matched partial dual coverage reduces convergence to 70.4 percent and produces -31.1 percent median speedup.
  • Learned warm-start methods should predict the full primal vector first; if only partial coverage is feasible, prioritize voltage variables and avoid inconsistent dual predictions unless the primal estimate is nearly complete.
  • The apparent benefit of DC seeding disappears once the DCOPF presolve time is included in the total (p = 0.4171).

Where Pith is reading between the lines

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

  • The same coverage requirement is likely to appear when the same initialization experiments are run with other interior-point solvers or on networks outside the PGLib collection.
  • Practical learned predictors will need loss terms that penalize incomplete or inconsistent dual coverage even when the primal estimate is only moderately accurate.
  • Extending the benchmark to networks larger than 30,000 buses would test whether the failure rate of partial restarts grows with problem size.
  • The ordering recommendation (full primal first, voltages next) supplies a concrete target for training data collection in warm-start research.

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

2 major / 2 minor

Summary. The manuscript benchmarks primal-dual warm-start strategies for AC optimal power flow (ACOPF) solved with IPOPT on 19 PGLib-OPF instances ranging from 5 to 30,000 buses. It evaluates all 15 non-empty subsets of the primal blocks {P_g, Q_g, V_m, V_a} under oracle conditions, decomposes duals into constraint and bound multipliers, and compares three DC-seeded practical initializations. Key findings are that only the complete oracle primal-dual restart reliably converges on all baseline-convergent cases (47.6% median speedup), twelve of fourteen partial O-PD combinations yield negative median speedups with repeated failures on larger networks, full bound-multiplier coverage reaches 90.7% convergence while block-matched partial coverage drops to 70.4%, and DC seeding benefits lose statistical significance once presolve cost is included (p=0.4171). The paper derives ordering recommendations for learned warm-start predictors.

Significance. If the reported patterns hold, the work supplies actionable, empirically tested guidance on which warm-start components matter for interior-point ACOPF solvers, especially the distinction between dual presence and full coverage. The exhaustive subset enumeration, median-based speedups, convergence statistics, and hypothesis testing on standard instances constitute a reproducible empirical contribution that can directly inform the design of learned initializers. Strengths include the oracle-vs-practical separation and the explicit focus on bound vs. constraint multipliers.

major comments (2)
  1. [abstract/results] Abstract and results (benchmark tables): the central claim that only the complete O-PD restart converges on every baseline-convergent case while 12/14 partial combinations produce negative medians rests on a single IPM solver (IPOPT) and the 19 PGLib instances; because interior-point codes differ in multiplier initialization and update rules, the observed ordering of primal-block importance may not transfer, and no cross-solver verification is reported.
  2. [conclusion] Conclusion (target ordering for learned methods): the recommendation to 'predict the full primal vector first; if only partial coverage is possible, prioritize voltage variables' is extrapolated from oracle experiments; without additional trials using noisy or incomplete predictors, it is unclear whether the relative importance of voltage blocks versus others persists under realistic prediction error.
minor comments (2)
  1. [abstract] Abstract: the p-value for DC seeding (p=0.4171) is reported without stating the exact statistical test or whether multiple-comparison correction was applied.
  2. [methods] The manuscript would benefit from explicit mention of how default multiplier values are chosen for uncovered bounds in the partial-coverage experiments.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and positive evaluation of the manuscript. We address each major comment below, acknowledging limitations where appropriate and indicating planned revisions.

read point-by-point responses

  1. Referee: [abstract/results] Abstract and results (benchmark tables): the central claim that only the complete O-PD restart converges on every baseline-convergent case while 12/14 partial combinations produce negative medians rests on a single IPM solver (IPOPT) and the 19 PGLib instances; because interior-point codes differ in multiplier initialization and update rules, the observed ordering of primal-block importance may not transfer, and no cross-solver verification is reported.

    Authors: We acknowledge that all experiments use IPOPT on the PGLib-OPF set and that other interior-point implementations may initialize or update multipliers differently, potentially altering the observed block-importance ordering. IPOPT remains the dominant solver for large-scale ACOPF, which motivated the choice, but we agree the findings are solver-specific. We will revise the abstract and results section to state explicitly that the reported patterns and ordering apply to IPOPT and that cross-solver verification lies outside the present scope. revision: yes

  2. Referee: [conclusion] Conclusion (target ordering for learned methods): the recommendation to 'predict the full primal vector first; if only partial coverage is possible, prioritize voltage variables' is extrapolated from oracle experiments; without additional trials using noisy or incomplete predictors, it is unclear whether the relative importance of voltage blocks versus others persists under realistic prediction error.

    Authors: The oracle setting isolates the intrinsic value of each initialization component by removing prediction noise, thereby establishing an idealized performance target for learned methods. We agree that the persistence of the voltage-priority ordering under realistic error remains untested. We will revise the conclusion to clarify that the ordering is derived under oracle conditions and that its robustness to predictor noise requires separate empirical study. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical benchmarking with direct measurements

full rationale

The paper performs systematic numerical experiments comparing 15 primal-block subsets and DC-seeded variants on 19 fixed PGLib-OPF instances solved with IPOPT. All reported speedups, convergence rates, and statistical tests (e.g., p=0.4171) are direct outputs of these runs against baseline solves; no equations, fitted parameters, or predictions are derived from prior results within the paper. No self-citations are invoked as load-bearing mathematical premises, and the target ordering for learned methods is presented as an empirical recommendation rather than a derived theorem. The work is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

This is an empirical benchmarking study; no mathematical derivations, fitted constants, or new postulated entities are introduced. All claims rest on direct numerical comparisons against baseline solves on standard test cases.

axioms (1)
  • domain assumption IPOPT interior-point method converges reliably on the tested ACOPF instances when started from the chosen baselines.
    The paper measures speedups and failures relative to these baseline solves.

pith-pipeline@v0.9.1-grok · 5892 in / 1405 out tokens · 16019 ms · 2026-06-27T15:58:22.112490+00:00 · methodology

0 comments
read the original abstract

Warm starts are widely used to accelerate AC optimal power flow (ACOPF) solves, but the impact of different initialization strategies has received limited systematic study, particularly for the primal-dual interior-point methods that dominate large-scale ACOPF algorithms. This paper benchmarks initialization strategies for ACOPF solved with the interior-point solver IPOPT on 19 PGLib-OPF instances (5 to 30,000 buses), testing all 15 non-empty subsets of the primal blocks $\{P_g, Q_g, V_m, V_a\}$ under oracle conditions and three DC-seeded combinations in a practical setting. The experiments show that most partial primal-plus-dual restarts increase solve time or reduce convergence reliability. Among the oracle primal-plus-dual (O-PD) configurations, only the complete restart reliably converges on every baseline-convergent case, reaching a $47.6\%$ median solve-time speedup. Twelve of the 14 partial O-PD combinations have negative median speedups, and several fail repeatedly on larger networks. Decomposing the dual into constraint and bound multipliers shows that \emph{coverage}, not the presence of duals per se, governs robustness: the full bound-multiplier vector reaches 90.7\% convergence and a $+26.8$\% median speedup, whereas block-matched coverage (oracle multipliers on some bounds, defaults on the rest) drops to 70.4\% and $-31.1$\%. Practical DC seeding sometimes helps the AC solve, but the benefit is no longer statistically significant once the DCOPF presolve cost is included in the end-to-end comparison ($p = 0.4171$). For learned warm-start methods, the results support the following target ordering: predict the full primal vector first; if only partial coverage is possible, prioritize voltage variables; and avoid partial or inconsistent dual predictions unless the primal estimate is nearly complete.

Figures

Figures reproduced from arXiv: 2606.08984 by Babak Taheri, Daniel K. Molzahn.

Figure 1
Figure 1. Figure 1: Median AC solve-time speedup (%) across all 15 non-empty [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Median AC solve-time speedup (%) across all 15 initialization-block subsets for the five dual modes: O-PO, O-PD, O-CD, O-BD, and O-PD-AB. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Why interior-point initialization is delicate. For a variable with bounds [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Left: Baseline AC solve time versus case-wise best O-PD and case-wise best DC-seeded AC solve. Right: Baseline total time versus case-wise best [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Case network size (nb buses, log scale) versus runtime (log scale) for the baseline, best oracle AC primal-plus-dual restart, best DC-seeded AC solve, and best DC end-to-end workflow. 1 2 3 4 5 6 7 8 Performance ratio τ 0% 20% 40% 60% 80% 100% Fraction of cases Performance profiles of representative static initialization policies baseline default oracle primal only Qg Vm oracle primal dual Vm oracle constr… view at source ↗
Figure 6
Figure 6. Figure 6: Performance profiles [26] of representative static initialization policies [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Per-case AC solve-time speedup ordered by ascending O-PD speedup. [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Box plots of per-case AC solve-time speedup for the case-wise best [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 11
Figure 11. Figure 11: Box plots of end-to-end speedup (%) relative to baseline total time [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: AC solve-time speedup (%) versus network size for the case-wise [PITH_FULL_IMAGE:figures/full_fig_p014_12.png] view at source ↗

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

26 extracted references · 3 canonical work pages · 1 internal anchor

  1. [1]

    Nocedal and S

    J. Nocedal and S. J. Wright,Numerical Optimization, 2nd ed., ser. Springer Series in Operations Research and Financial Engineering. New York, NY: Springer, 2006

    2006

  2. [2]

    On the implementation of an interior- point filter line-search algorithm for large-scale nonlinear programming,

    A. W ¨achter and L. T. Biegler, “On the implementation of an interior- point filter line-search algorithm for large-scale nonlinear programming,” Mathematical Programming, vol. 106, no. 1, pp. 25–57, May 2006

    2006

  3. [3]

    Initialisation, decoupling, hot start, and warm start in direct nonlinear interior point algorithm for optimal power flows,

    Y . C. Wu and A. S. Debs, “Initialisation, decoupling, hot start, and warm start in direct nonlinear interior point algorithm for optimal power flows,” IEE Proceedings-Generation, Transmission and Distribution, vol. 148, no. 1, pp. 67–75, Jan 2001

    2001

  4. [4]

    DeepOPF: A feasibility- optimized deep neural network approach for AC optimal power flow problems,

    X. Pan, M. Chen, T. Zhao, and S. H. Low, “DeepOPF: A feasibility- optimized deep neural network approach for AC optimal power flow problems,”IEEE Systems Journal, vol. 17, no. 1, pp. 673–683, Mar 2023

    2023

  5. [5]

    Learning optimal solutions for extremely fast AC optimal power flow,

    A. S. Zamzam and K. Baker, “Learning optimal solutions for extremely fast AC optimal power flow,” inIEEE SmartGridComm, Tempe, AZ, USA, Nov 2020

    2020

  6. [6]

    Learning warm-start points for AC optimal power flow,

    K. Baker, “Learning warm-start points for AC optimal power flow,” in29th IEEE International Workshop on Machine Learning for Signal Processing (MLSP), Pittsburgh, PA, USA, Oct 2019, pp. 1–6

    2019

  7. [7]

    Fast and explainable warm-start point learning for AC optimal power flow using decision tree,

    Y . Cao, H. Zhao, G. Liang, J. Zhao, H. Liao, and C. Yang, “Fast and explainable warm-start point learning for AC optimal power flow using decision tree,”International Journal of Electrical Power & Energy Systems, vol. 153, p. 109369, Nov 2023

    2023

  8. [8]

    MAT- POWER: Steady-state operations, planning, and analysis tools for power systems research and education,

    R. D. Zimmerman, C. E. Murillo-S ´anchez, and R. J. Thomas, “MAT- POWER: Steady-state operations, planning, and analysis tools for power systems research and education,”IEEE Transactions on Power Systems, vol. 26, no. 1, pp. 12–19, Feb 2011

    2011

  9. [9]

    The power grid library for benchmarking AC optimal power flow algorithms.arXiv preprint arXiv:1908.02788, 2019

    S. Babaeinejadsarookolaee, A. Birchfield, R. D. Christie, C. Coffrin, C. L. DeMarco, R. Diao, M. Ferris, S. Fliscounakis, S. Greene, R. Huang, C. Josz, R. Korab, B. C. Lesieutre, J. Maeght, D. Molzahn, T. J. Overbye, P. Panciatici, B. Park, J. Snodgrass, and R. D. Zimmerman, “The power grid library for benchmarking AC optimal power flow algorithms,” 2019,...

    work page arXiv 2019

  10. [10]

    Reoptimization with the primal-dual interior point method,

    J. Gondzio and A. Grothey, “Reoptimization with the primal-dual interior point method,”SIAM Journal on Optimization, vol. 13, no. 3, pp. 842–864, 2003

    2003

  11. [11]

    On warm starts for interior methods,

    A. Forsgren, “On warm starts for interior methods,” inSystem Modeling and Optimization: Proceedings of the 22nd IFIP TC7 Conference, ser. IFIP International Federation for Information Processing, F. Ceragioli, A. Dontchev, H. Furuta, K. Marti, and L. Pandolfi, Eds., vol. 199. Boston, MA: Springer, 2006, pp. 51–66

    2006

  12. [12]

    A linear-programming approxima- tion of AC power flows,

    C. Coffrin and P. Van Hentenryck, “A linear-programming approxima- tion of AC power flows,”INFORMS Journal on Computing, vol. 26, no. 4, pp. 718–734, 2014

    2014

  13. [13]

    DC power flow revisited,

    B. Stott, J. Jardim, and O. Alsac ¸, “DC power flow revisited,”IEEE Transactions on Power Systems, vol. 24, no. 3, pp. 1290–1300, Aug 2009

    2009

  14. [14]

    WARP: A Benchmark for Primal-Dual Warm-Starting of Interior-Point Solvers

    D. Suri, H. Hilmarsson, and S. Bose, “W ARP: A benchmark for primal-dual warm-starting of interior-point solvers,” 2026. [Online]. Available: https://arxiv.org/abs/2605.05728

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  15. [15]

    Dual conic proxies for AC optimal power flow,

    G. Qiu, M. Tanneau, and P. Van Hentenryck, “Dual conic proxies for AC optimal power flow,”Electric Power Systems Research, vol. 236, p. 110661, 2024

    2024

  16. [16]

    Dual conic proxy for semidefinite relaxation of AC optimal power flow,

    ——, “Dual conic proxy for semidefinite relaxation of AC optimal power flow,” 2025. [Online]. Available: https://arxiv.org/abs/2502.06978

    work page arXiv 2025

  17. [17]

    Individual comparisons by ranking methods,

    F. Wilcoxon, “Individual comparisons by ranking methods,”Biometrics Bulletin, vol. 1, no. 6, pp. 80–83, Dec 1945

    1945

  18. [18]

    A simple sequentially rejective multiple test procedure,

    S. Holm, “A simple sequentially rejective multiple test procedure,” Scandinavian Journal of Statistics, vol. 6, no. 2, pp. 65–70, 1979

    1979

  19. [19]

    Estimates of location based on rank tests,

    J. L. Hodges and E. L. Lehmann, “Estimates of location based on rank tests,”The Annals of Mathematical Statistics, vol. 34, no. 2, pp. 598– 611, Jun 1963

    1963

  20. [20]

    SciPy 1.0: Fundamental algorithms for scientific computing in Python,

    P. Virtanenet al., “SciPy 1.0: Fundamental algorithms for scientific computing in Python,”Nature Methods, vol. 17, pp. 261–272, 2020

    2020

  21. [21]

    A fully asynchronous multifrontal solver using distributed dynamic scheduling,

    P. R. Amestoy, I. S. Duff, J.-Y . L’Excellent, and J. Koster, “A fully asynchronous multifrontal solver using distributed dynamic scheduling,” SIAM Journal on Matrix Analysis and Applications, vol. 23, no. 1, pp. 15–41, 2001

    2001

  22. [22]

    W. E. Hartet al.,Pyomo: Optimization Modeling in Python, 2nd ed. Cham, Switzerland: Springer, 2017

    2017

  23. [23]

    Parallelizing the dual revised simplex method,

    Q. Huangfu and J. A. J. Hall, “Parallelizing the dual revised simplex method,”Mathematical Programming Computation, vol. 10, no. 1, pp. 119–142, 2018

    2018

  24. [24]

    IPOPT options documentation,

    COIN-OR Project, “IPOPT options documentation,” https://coin-or. github.io/Ipopt/OPTIONS.html, accessed: 2026-03-21

    2026

  25. [25]

    A. V . Fiacco,Introduction to Sensitivity and Stability Analysis in Nonlinear Programming, ser. Mathematics in Science and Engineering. New York: Academic Press, 1983, vol. 165

    1983

  26. [26]

    Benchmarking optimization software with performance profiles,

    E. D. Dolan and J. J. Mor ´e, “Benchmarking optimization software with performance profiles,”Mathematical Programming, vol. 91, no. 2, pp. 201–213, Jan 2002. 10 TABLE V COMPLETE WARM-START RESULTS MATRIX. EACH SUCCESSFUL ORACLE CELL REPORTSACSOLVE TIME[S] /SPEEDUP[%] / IPOPTITERATIONS. DCROWS FOLLOW THE SAME CONVENTION.F=SOLVER FAILURE. COLUMN HEADERS:5=P...

    2002