2022
DOI: 10.1613/jair.1.13666
|View full text |Cite
|
Sign up to set email alerts
|

Better Decision Heuristics in CDCL through Local Search and Target Phases

Abstract: On practical applications, state-of-the-art SAT solvers dominantly use the conflict-driven clause learning (CDCL) paradigm. An alternative for satisfiable instances is local search solvers, which is more successful on random and hard combinatorial instances. Although there have been attempts to combine these methods in one framework, a tight integration which improves the state of the art on a broad set of application instances has been missing. We present a combination of techniques that achieves such an impr… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

0
5
0

Year Published

2022
2022
2024
2024

Publication Types

Select...
6
2

Relationship

1
7

Authors

Journals

citations
Cited by 11 publications
(5 citation statements)
references
References 40 publications
0
5
0
Order By: Relevance
“…Most techniques (including the two most important, vivification and probing) either fit into our new PCDCL base calculus or do not require any change (like random walk [10] that is conjectured to be the reason for the major performance improvement in 2020). One major technique that we cannot currently express is variable elimination, because models are changed and need to be fixed.…”
Section: Discussionmentioning
confidence: 99%
See 1 more Smart Citation
“…Most techniques (including the two most important, vivification and probing) either fit into our new PCDCL base calculus or do not require any change (like random walk [10] that is conjectured to be the reason for the major performance improvement in 2020). One major technique that we cannot currently express is variable elimination, because models are changed and need to be fixed.…”
Section: Discussionmentioning
confidence: 99%
“…In order to study the performance we have run 3 different IsaSAT versions: the original SML solver (using MLton with the LLVM backend), the first port of the IsaSAT solver, and the current version with inprocessing and various other improvements on heuristics that do not require any change on our PCDCL calculus, notably rephasing and target phases [10] (but no local search) and the alternation between aggressive restarts (heuristically seems better for UNSAT) and few restarts (seems better for SAT) following the ideas of Chanseok Oh [24]. We run all the benchmarks from the SAT Competition 2022 on an Intel Xeon E5-2620 v4 CPU at 2.10 GHz (with turbo-mode disabled) with a memory limit of 7 GB and a timeout of 5000 s. For comparison, we have included versat [23] and CreuSAT [25].…”
Section: Performancementioning
confidence: 99%
“…More recently, the solvers CaDiCaL and Kissat call a local search solver to produce a promising assignment used for target phases [BF20]. A deeper cooperation between CDCL and local search is proposed in [CZ21b;Cai+22] to improve target phases, rephasing and branching. Finally, other solving paradigms can be used to solve SAT although this is not common as satisfiability is a powerful formalism with highly efficient dedicated solvers.…”
Section: Other Methodsmentioning
confidence: 99%
“…A different approach, first proposed in Mrs. Beaver [26] and subsequently improved and extended in a sequence of publications [28,27], can be viewed as a non-traditional stochastic local search (SLS) which-somewhat unintuitively yet very successfully-uses a complete SAT solver to guide search space traversal from one solution to another, ideally better one. Integrating local search with complete solvers has been shown to increase efficiency in other contexts as well, such as in SAT solvers [8] and finite-domain CP solvers [17].…”
Section: Introductionmentioning
confidence: 99%