Pith. sign in

REVIEW 3 major objections 4 minor 33 references

The paper shows that treating all difference constraints x−y≤d with one global propagator, built on incremental shortest-path algorithms, can replace separate per-constraint propagators and substantially speed up finite-domain constraint so

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 ↗

A single global propagator for difference constraints, equipped with clause-learning explanations, speeds up constraint solvers on scheduling and temporal problems.

T0 review reviewed 2026-08-01 challenge →

load-bearing objection Solid, well-engineered global difference propagator with real empirical gains, but the domain-propagation theorem doesn't cover the actual encoding, so the theory is oversold. the 3 major comments →

arxiv 2607.20022 v1 pith:LI4EFKXO submitted 2026-07-22 cs.AI

Global Difference Constraint Propagation for Constraint Programming

Lucas Kletzander , Jip J. Dekker , Andreas Schutt , Peter J. Stuckey This is my paper
classification cs.AI
keywords difference constraintsglobal propagationconstraint programmingbounds consistencylazy clause generationshortest pathshalf-reificationscheduling
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 reading

The paper is trying to establish that difference constraints—simple inequalities of the form x−y≤d—should be propagated globally rather than one constraint at a time. It builds a single propagator that treats the whole set as a weighted graph, uses shortest-path reasoning to detect unsatisfiable cycles and derive new bounds, and produces clause explanations so it can work inside a lazy clause generation solver. Under stated assumptions, this global propagator is a domain propagator for the entire conjunction of difference constraints. The paper reports that this global treatment eliminates pathological slow propagation behavior and gives large speedups on scheduling benchmarks.

Core claim

The central claim is that all difference constraints in a model can and should be handled by one global propagator. The propagator views the constraints as a weighted directed graph, maintains a valid potential function, and uses incremental algorithms to add constraints and bound changes, detect negative cycles, and extract all implied bounds. It also derives clause explanations for each propagation, which is essential for use in a lazy clause generation solver. Under the assumptions that domains are range domains and no Boolean variable appears twice in the constraint set, this global propagator is a domain propagator for the conjunction, meaning it removes exactly the values that cannot p

What carries the argument

The constraint graph G_C, with an edge x→y of weight d for each constraint x−y≤d, together with the shortest-path theorem stating that C entails x−y≤d exactly when the shortest path from x to y has weight at most d. The propagator maintains a potential function that keeps reduced edge costs non-negative, and uses incremental satisfiability and implication algorithms plus dedicated bound-propagation algorithms (IncLB and IncUB) to process changes efficiently. Explanations are generated from predecessor pointers recorded during shortest-path computations, including explanations of negative cycles and Boolean propagations based on bounds.

Load-bearing premise

The domain-propagation guarantee rests on the assumption that no Boolean variable appears in more than one difference constraint, bounds constraint, or added half-reified constraint; when that assumption fails, as it does in the solver's own encodings of reified constraints, the propagator remains sound but is not proven to be a domain propagator.

What would settle it

Construct a constraint set with range domains and two constraints sharing the same Boolean variable, for example b ⇒ x−y ≤ 1 and b ⇒ y−z ≤ 1, with b fixed true and x,y,z given ranges. Run the global propagator and check whether every value remaining in each variable's domain can be extended to a full solution of the conjunction. If any remaining value has no support, the claimed domain-propagation completeness fails under exactly the conditions the theorem excludes.

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

If this is right

  • A single propagation pass can reach a fixpoint that would otherwise require many separate propagator invocations; worst-case chains such as the O(n³) example collapse to O(n log n + m) per propagation.
  • Half-reified difference constraints that are disentailed by the whole constraint set are detected even when no individual constraint implies the disentailment.
  • Every propagation can be explained in clausal form, so the global propagator can be embedded in a lazy clause generation solver without sacrificing clause learning.
  • An initial simplification phase using all-pairs shortest paths, redundant-edge removal, and node unification accounts for most of the practical speedup, especially on unsatisfiable scheduling instances.
  • On scheduling benchmarks, the global propagator increases the number of optimal and unsatisfiable instances solved and reduces runtime compared to separate difference-constraint propagation.

Where Pith is reading between the lines

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

  • Editorially, the domain-propagation guarantee is narrower than the implemented solver: the paper's own encoding of reified constraints introduces Boolean variables that appear in two implied constraints, violating the stated assumption. The empirical results therefore support usefulness, not the full completeness theorem, in those configurations.
  • Editorially, the finding that the incremental implication check is usually better disabled suggests a broader design principle: in a lazy clause generation solver, deferring implication checks until conflict analysis can be more efficient than eagerly checking every consequence.
  • Editorially, the global propagator opens a loop that the paper leaves as future work: other global propagators, such as disjunctive scheduling propagators, could infer new difference constraints and feed them back into the difference-logic graph, yielding even stronger propagation.
  • Editorially, the separation of bounds propagation and Boolean propagation priorities indicates that treating these as independent scheduling decisions within one global propagator can outperform a monolithic design.
Share X Bluesky LinkedIn Reddit HN

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 / 4 minor

Summary. The paper proposes a global propagator for difference constraints x−y≤d and their half-reified variants b⇒x−y≤d in a constraint programming / lazy clause generation solver. It adapts incremental shortest-path algorithms (IncSat/IncImp) for satisfiability and implication, adds IncLB/IncUB for simultaneous bounds propagation, and contributes explanation schemes together with a full integration into the Rust-based LCG solver Huub, including preprocessing and simplification. The headline claims are that, under a no-duplicate-Boolean assumption, the global propagator is a domain propagator for the constraint set, and that treating difference constraints globally can substantially outperform treating them as individual propagators. Experiments compare Huub with and without the propagator against Chuffed and CP-SAT on MiniZinc Challenge benchmarks and on ProdCons and RCPSP/max scheduling benchmarks.

Significance. The practical contribution is real and valuable: this appears to be the first general-purpose LCG solver integration of a global difference-logic propagator with explanations, and the paper provides a systematic evaluation of explanation types, priorities, and constraint-acquisition levels. The complexity improvements over naive propagation (O(n log n + m) instead of O(nm) for bounds updates) and the simplification stage are useful advances. The paper also ships reproducible experiments and supplementary material. However, the theoretical 'domain propagator' claim is not established for the implemented system, and the main MiniZinc comparison is partly in-sample. These issues weaken but do not invalidate the core engineering contribution.

major comments (3)
  1. [§4.1 (final paragraph) and §5.1] The paper states, without proof, that under the assumption that no Boolean variable appears twice in the difference constraints, bounds constraints, and added half-reified constraints, the global propagator is a domain propagator. Section 5.1 defines the implemented encoding: a reified constraint b⇔x−y≤d is decomposed into b⇒x−y≤d and ¬b⇒y−x≤−d−1, so the same Boolean labels two implied edges. Levels 1–3 as implemented therefore operate outside the stated assumption. This is not cosmetic. For C={b⇒x−y≤0, b⇒y−x≤−1}, the only satisfying assignments have b=0; a domain propagator for C must set b=0 from any domain in which both implications occur. The described propagator, however, does not activate either implication until b is assigned true and does not combine the two half-reified constraints to derive b=0. Thus the domain-propagator guarantee is false for the implemented solver. Even in t
  2. [§6.3.1 and Table 1] The final configuration is selected by scoring all 162 configurations on the MiniZinc Challenge benchmark, and Table 1 then reports the improvement of Huub(gp+s) over Huub(sp) on that same benchmark. This is an in-sample model-selection result: the headline MiniZinc gain may overstate the general-purpose advantage because the configuration was chosen to maximize points on these exact instances. The issue does not affect the ProdCons/RCPSPmax comparisons, where the configuration is selected on small subsets and evaluated on full sets, nor the worst-case experiment in §6.1. The authors should provide a held-out evaluation, a nested validation, or at least clearly state that the MiniZinc comparison is exploratory rather than confirmatory.
  3. [§4.2, Algorithms 3 and IncUB] Theorems 1–2 provide the underlying theory, but the actual propagation algorithms IncLB/IncUB are presented informally with an example and no correctness proof. No theorem states that IncLB computes exactly the new lower bounds derivable from the current difference constraints and bounds, or that its pruning condition ('does not explore all variables') preserves completeness. The claimed O(n log n + m) bound also assumes a priority queue with decrease-key, while §5.1 states that the implementation uses a simple queue with stale duplicate entries; a separate argument is needed for the implemented complexity. Since these algorithms are the central engine of the propagator, this gap should be closed in a revision.
minor comments (4)
  1. [§4.2, Example 7] Typo: 'Dijkstra' is written as 'Dijktra'. Also, the treatment of the condition in line 12 of Algorithm 3 and the conversion of δ values into bounds in lines 18–20 should be rephrased for clarity, since the notation alternates between δ→v0(x), −δ, and minD without a precise definition of the returned bounds.
  2. [§5.1] The mapping for inequalities and reified equalities introduces auxiliary Boolean variables and multiple difference constraints, but the paper does not explicitly connect this with the no-duplicate-Boolean assumption in §4.1 until the reader discovers the mismatch. A forward reference or a short remark in §5.1 would help.
  3. [§6.4, Table 1] The column header 'Satisfied #to nr time' is ambiguous: it mixes the number of instances with a solution but timeout with the number of satisfaction instances solved and their runtime. Please split these columns or clarify the header.
  4. [Throughout] Minor typos: 'invokation' in §6.1, 'Bellman Ford' vs 'Bellman-Ford' in Algorithm 4, and inconsistent hyphenation of 'half-reified'. These are purely presentation issues.

Circularity Check

0 steps flagged

No significant circularity: the propagator algorithms derive from established shortest-path theory and are evaluated empirically; the noted domain-propagator caveat is a scope/correctness issue, not a circular derivation.

full rationale

The paper's central derivation maps difference constraints to shortest-path graphs (Definition 1, Theorem 1) and builds propagation on incremental Bellman-Ford/Dijkstra algorithms from Cotton and Maler [6] and Ramalingam et al. [11]. These are established external results, not re-statements of the paper's conclusions. The claimed domain-propagator property in Section 4.1 is stated under an explicit assumption that no Boolean variable appears twice; Section 5.1's reification encoding (b⇔x−y≤d as b⇒x−y≤d and ¬b⇒y−x≤−d−1) can indeed duplicate Booleans, meaning the implemented configuration may fall outside the proven regime. This is a limitation in the scope of the theoretical guarantee, not circularity: the theorem is not obtained by assuming its own conclusion, and the gap is about the encoding satisfying the theorem's hypotheses. The experimental Sections 6.2–6.4 compare configurations on benchmarks and do not disguise tuned parameters as predictions. Self-citations such as [1] and [24] are used to situate the work and to describe prior implementations; they are not load-bearing for the soundness or completeness argument, which is based on standard shortest-path correctness. No fitted constant, renamed known result, or definitional equivalence between input and output was found. Therefore, the appropriate circularity score is 0.

Axiom & Free-Parameter Ledger

1 free parameters · 4 axioms · 0 invented entities

The paper adds no physical or logical entities. It relies on known shortest-path machinery plus two domain assumptions; the no-repeated-Boolean assumption is violated by the implementation's own encodings for reified/equality/inequality constraints. The only tuned quantities are solver configuration switches.

free parameters (1)
  • Difference-logic configuration switches (level, priorities, IncImp, explanation type) = Level 1, bound priority 0, Boolean priority 4, IncImp off, Lifted explanations
    Selected as the best of 162 combinations on the MiniZinc challenge benchmark (Section 6.3.1) and then used for the headline comparison in Table 1 on the same benchmark; the empirical gains are therefore partly tuning-dependent.
axioms (4)
  • standard math Shortest-path theory for difference constraints (Theorem 1, Corollary 1), Dijkstra, Bellman-Ford, Johnson's algorithm, and valid potential functions.
    Used throughout Sections 3–5 as the algorithmic foundation; taken from the cited literature rather than proved in this paper.
  • domain assumption Range domains only.
    Section 2 restricts to range domains, and all correctness statements, including the domain-propagator claim, are conditional on this.
  • domain assumption No Boolean variable appears twice in the set of difference constraints, bounds constraints, and half-reified constraints.
    Explicit in the final paragraph of Section 4.1; this is the key assumption for the domain-propagator completeness guarantee and is violated by the Section 5.1 encodings for reified/equality/inequality constraints.
  • domain assumption Bounds propagation reaches the same fixpoint regardless of the order in which propagators run.
    Invoked in the proof of Theorem 2(a): "Since any order of propagating constraints leads to the same result, D' is a false domain." This is standard for monotone propagators but is asserted rather than proved in the text.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Global Difference Constraint Propagation for Constraint Programming." pith.science (2026). https://pith.science/paper/LI4EFKXO

@misc{pith2026260720022,
  author       = {Pith},
  title        = {Pith review of: Global Difference Constraint Propagation for Constraint Programming},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LI4EFKXO}},
  note         = {Machine review of arXiv:2607.20022}
}
Share X Bluesky LinkedIn Reddit HN
abstract

Difference constraints of the form $x - y \leq d$ are well studied, with efficient algorithms for satisfaction and implication, because of their connection to shortest paths. Finite domain propagation algorithms, however, typically do not make use of these algorithms, and treat each difference constraint as a separate propagator. Propagation does guarantee completeness of solving, but can be needlessly slow. In this paper we describe how to build a (bounds consistent) global propagator for difference constraints that treats them all simultaneously. SAT modulo theory solvers have included theory solvers for difference constraints for some time. While a theory solver for difference constraints gives the basis of a global difference constraint propagator, we show how the requirements on the propagator are quite different. Crucially, we show how to explain propagations by a global difference constraint propagator, in order to use it within a lazy clause generation solver. We give experiments showing that treating difference constraints globally can substantially improve on the standard propagation approach.

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

33 extracted references · 6 canonical work pages

  1. [1]

    In: Antoy, S., Albert, E

    Feydy, T., Schutt, A., Stuckey, P.J.: Global difference constraint propagation for finite domain solvers. In: Antoy, S., Albert, E. (eds.) Proceedings of the 10th In- ternational ACM SIGPLAN Conference on Principles and Practice of Declarative Programming. PPDP ’08, pp. 226–235. Association for Computing Machinery, New York, NY, USA (2008). https://doi.or...

    arXiv 2008

  2. [2]

    (ed.): Handbook of Theoretical Computer Science: Volume A: Algorithms and Complexity

    Leeuwen, J. (ed.): Handbook of Theoretical Computer Science: Volume A: Algorithms and Complexity. Elsevier and MIT Press, Cambridge, MA, USA (1990)

    1990

  3. [3]

    In: Lee, J.H.M

    Feydy, T., Somogyi, Z., Stuckey, P.J.: Half-reification and flattening. In: Lee, J.H.M. (ed.) Proceedings of the 17th International Conference on Principles and 32 Practice of Constraint Programming. LNCS, vol. 6876, pp. 286–301. Springer, Berlin, Heidelberg (2011)

    2011

  4. [4]

    In: Baader, F., Voronkov, A

    Nieuwenhuis, R., Oliveras, A., Tinelli, C.: Abstract DPLL and abstract DPLL modulo theories. In: Baader, F., Voronkov, A. (eds.) Logic for Programming, Ar- tificial Intelligence, and Reasoning, 11th International Conference, LPAR 2004, Montevideo, Uruguay, March 14-18, 2005, Proceedings. Lecture Notes in Com- puter Science, vol. 3452, pp. 36–50. Springer,...

    work page doi:10.1007/978-3-540-32275-7 2004

  5. [5]

    JACM53(6), 937–977 (2006)

    Niewenhuis, R., Oliveras, A., Tinelli, C.: Solving SAT and SAT Modulo Theo- ries: from an abstract Davis-Putnam-Logemann-Loveland procedure to DPLL(T). JACM53(6), 937–977 (2006)

    2006

  6. [6]

    In: Theory and Applications of Satisfiability Testing - SAT 2006

    Cotton, S., Maler, O.: Fast and Flexible Difference Constraint Propagation for DPLL(T). In: Theory and Applications of Satisfiability Testing - SAT 2006. LNCS, vol. 4121, pp. 170–183. Springer, Berlin, Heidelberg (2006)

    2006

  7. [7]

    Constraints14(3), 357–391 (2009)

    Ohrimenko, O., Stuckey, P.J., Codish, M.: Propagation via lazy clause generation. Constraints14(3), 357–391 (2009)

    2009

  8. [8]

    In: Proceedings of the European Symposium on Algorithms, pp

    Cherkassky, B.V., Goldberg, A.V.: Negative-cycle detection algorithms. In: Proceedings of the European Symposium on Algorithms, pp. 349–363 (2006)

    2006

  9. [9]

    JACM 28(4), 769–779 (1981)

    Shostak, R.: Deciding linear inequalities by computing loop residues. JACM 28(4), 769–779 (1981)

    1981

  10. [10]

    Artificial Intelli- gence49(1), 61–95 (1991) https://doi.org/10.1016/0004-3702(91)90006-6

    Dechter, R., Meiri, I., Pearl, J.: Temporal constraint networks. Artificial Intelli- gence49(1), 61–95 (1991) https://doi.org/10.1016/0004-3702(91)90006-6

    work page doi:10.1016/0004-3702(91)90006-6 1991

  11. [11]

    Algorithmica23, 261–275 (1999)

    Ramalingan, G., Song, J., Joskowicz, L., Miller, R.E.: Solving systems of difference constraints incrementally. Algorithmica23, 261–275 (1999)

    1999

  12. [12]

    In: Logic for Programming, Artificial Intelligence, and Reasoning

    Wang, C., Ivanˇ ci´ c, F., Ganai, M., Gupta, A.: Deciding Separation Logic Formulae by SAT and Incremental Negative Cycle Elimination. In: Logic for Programming, Artificial Intelligence, and Reasoning. LNCS, vol. 3835, pp. 322–336. Springer, Berlin, Heidelberg (2005)

    2005

  13. [13]

    https://doi.org/10.5281/ zenodo.15591852

    Dekker, J.J., Stuckey, P.J., Zhong, A.Z.: Huub (2026). https://doi.org/10.5281/ zenodo.15591852 . https://huub.solutions

    2026

  14. [14]

    In: Garcia de la Banda, M

    Dekker, J.J., Ingatiev, A., Stuckey, P.J., Zhong, A.Z.: Towards modern and modular SAT for LCG. In: Garcia de la Banda, M. (ed.) Proceedings of the 30th International Conference on Principles and Practice of Constraint Pro- gramming. LIPICS, pp. 42–14212. Schloss Dagstuhl, Dagstuhl, Germany (2025). https://doi.org/10.4230/LIPIcs.CP.2025.42 33

    work page doi:10.4230/lipics.cp.2025.42 2025

  15. [15]

    Fazekas, K., Niemetz, A., Preiner, M., Kirchweger, M., Szeider, S., Biere, A.: IPASIR-UP: user propagators for CDCL. In: SAT. LIPIcs, vol. 271, pp. 8–1813. Schloss Dagstuhl - Leibniz-Zentrum f¨ ur Informatik, Dagstuhl, Germany (2023)

    2023

  16. [16]

    MIT press, Cambridge, MA, USA (2022)

    Cormen, T.H., Leiserson, C.E., Rivest, R.L., Stein, C.: Introduction to Algo- rithms. MIT press, Cambridge, MA, USA (2022)

    2022

  17. [17]

    In: Bessiere, C

    Lagerkvist, M., Schulte, C.: Advisors for incremental propagation. In: Bessiere, C. (ed.) Proceedings of the 13th International Conference on Principles and Prac- tice of Constraint Programming. LNCS, vol. 4741, pp. 409–422. Springer, Berlin, Heidelberg (2007)

    2007

  18. [18]

    Proceedings of the AAAI Conference on Artificial Intelligence38(8), 8004–8012 (2024) https://doi.org/10.1609/aaai

    Fichte, J.K., Geibinger, T., Hecher, M., Schl¨ ogel, M.: Parallel empirical eval- uations: Resilience despite concurrency. Proceedings of the AAAI Conference on Artificial Intelligence38(8), 8004–8012 (2024) https://doi.org/10.1609/aaai. v38i8.28638

    doi:10.1609/aaai 2024

  19. [19]

    https://github.com/chuffed/chuffed

    Chu, G., Stuckey, P.J., Schutt, A., Ehlers, T., Gange, G., Francis, K.: Chuffed, a lazy clause generation solver (2026). https://github.com/chuffed/chuffed

    2026

  20. [20]

    v9.15 (2026)

    Perron, L., Didier, F.: CP-SAT. v9.15 (2026). https://developers.google.com/ optimization/cp/cp solver/

    2026

  21. [21]

    AI magazine35(2), 55–60 (2014)

    Stuckey, P.J., Feydy, T., Schutt, A., Tack, G., Fischer, J.: The MiniZinc challenge 2008–2013. AI magazine35(2), 55–60 (2014)

    2008

  22. [22]

    https: //doi.org/10.5281/zenodo.21037348

    Kletzander, L., Dekker, J.J., Schutt, A., Stuckey, P.J.: Data Appendix for Global Difference Constraint Propagation for Constraint Programming (2026). https: //doi.org/10.5281/zenodo.21037348

    work page doi:10.5281/zenodo.21037348 2026

  23. [23]

    Constraints15(3), 307–316 (2010) https://doi.org/10.1007/s10601-010-9093-0

    Stuckey, P.J., Becket, R., Fischer, J.: Philosophy of the MiniZinc challenge. Constraints15(3), 307–316 (2010) https://doi.org/10.1007/s10601-010-9093-0

    doi:10.1007/s10601-010-9093-0 2010

  24. [24]

    In: Rue- her, M

    Schutt, A., Stuckey, P.J.: Explaining producer/consumer constraints. In: Rue- her, M. (ed.) Principles and Practice of Constraint Programming, pp. 438–454. Springer, Cham (2016)

    2016

  25. [25]

    Journal of scheduling16(3), 273–289 (2013)

    Schutt, A., Feydy, T., Stuckey, P.J., Wallace, M.G.: Solving RCPSP/max by lazy clause generation. Journal of scheduling16(3), 273–289 (2013)

    2013

  26. [26]

    Stuckey, Banda, M.G., Maher, M., Marriott, K., Slaney, J., Somogyi, Z., Wallace, M., Walsh, T.: The G12 project: Mapping solver independent models to efficient solutions

    P.J. Stuckey, Banda, M.G., Maher, M., Marriott, K., Slaney, J., Somogyi, Z., Wallace, M., Walsh, T.: The G12 project: Mapping solver independent models to efficient solutions. In: Beek, P.V. (ed.) Proceedings of the 11th International Conference on Principles and Practice of Constraint Programming. LNCS, pp. 13–16. Springer, Berlin, Heidelberg (2005)

    2005

  27. [27]

    In: Gent, I.P

    Feydy, T., Stuckey, P.J.: Lazy clause generation reengineered. In: Gent, I.P. (ed.) 34 Principles and Practice of Constraint Programming - CP 2009, 15th International Conference, CP 2009, Lisbon, Portugal, September 20-24, 2009, Proceedings. Lecture Notes in Computer Science, vol. 5732, pp. 352–366. Springer, Berlin, Heidelberg (2009). https://doi.org/10....

    work page doi:10.1007/978-3-642-04244-7 2009

  28. [28]

    PhD thesis, the University of Melbourne (2011)

    Chu, G.: Improving combinatorial optimization. PhD thesis, the University of Melbourne (2011)

    2011

  29. [29]

    In: Banda, M.G

    Hebrard, E.: Disjunctive Scheduling in Tempo. In: Banda, M.G. (ed.) 31st Inter- national Conference on Principles and Practice of Constraint Programming (CP 2025). Leibniz International Proceedings in Informatics (LIPIcs), vol. 340, pp. 13– 11322. Schloss Dagstuhl – Leibniz-Zentrum f¨ ur Informatik, Dagstuhl, Germany (2025). https://doi.org/10.4230/LIPIcs...

    doi:10.4230/lipics.cp.2025.13 2025

  30. [30]

    In: Gal, K., Now´ e, A., Nalepa, G.J., Fairstein, R., R˘ adulescu, R

    Bit-Monnot, A.: Enhancing hybrid CP-SAT search for disjunctive scheduling. In: Gal, K., Now´ e, A., Nalepa, G.J., Fairstein, R., R˘ adulescu, R. (eds.) 26th Euro- pean Conference on Artificial Intelligence (ECAI 2023)), pp. 255–262. IOS Press, Amsterdam, The Netherlands (2023). https://doi.org/10.3233/F AIA230278

    work page doi:10.3233/f 2023

  31. [31]

    In: Proceedings of the 3rd International Workshop on Temporal Representation and Reasoning (TIME 1996), pp

    Cesta, A., Oddi, A.: Gaining efficiency and flexibility in the simple tempo- ral problem. In: Proceedings of the 3rd International Workshop on Temporal Representation and Reasoning (TIME 1996), pp. 45–50 (1996)

    1996

  32. [32]

    Constraints23(2), 210–250 (2018) https://doi.org/10.1007/ s10601-018-9281-x

    Laborie, P., Rogerie, J., Shaw, P., Vil ´ ım, P.: IBM ILOG CP optimizer for scheduling. Constraints23(2), 210–250 (2018) https://doi.org/10.1007/ s10601-018-9281-x

    2018

  33. [33]

    Cherkassky, B.V., Goldberg, A.V., Radzik, T.: Shortest paths algorithms: Theory and experimental evaluation. Mathematical programming73(2), 129–174 (1996) 35 A Configuration Experiments onProdCons This section shows the graphs for the configuration experiments on the smallProdCons instance set. All configurations outperform the baseline which requires mor...

    1996

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.