A SAT Solver + Computer Algebra Attack on the Minimum Kochen–Specker Problem

Authors: Zhengyu Li, Curtis Bright, Vijay Ganesh

IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Tables 2 and 3 contain our experimental results. As can be seen from Table 2, the sequential version of MATHCHECK outperforms the sequential SAT-only (Maple SAT) and CAS-only (nauty) tools.
Researcher Affiliation Academia Zhengyu Li1 , Curtis Bright2 , Vijay Ganesh1 1School of Computer Science, Georgia Institute of Technology, United States 2School of Computer Science, University of Windsor, Canada brian.li@gatech.edu, cbright@uwindsor.ca, vganesh45@gatech.edu
Pseudocode No The paper describes methods and algorithms in text but does not include any structured pseudocode or algorithm blocks (e.g., labeled 'Pseudocode' or 'Algorithm').
Open Source Code Yes 1Code at https://github.com/Brian Li009/MathCheck
Open Datasets No The paper addresses a mathematical search problem (finding the minimum Kochen Specker vector system) using SAT solvers and CAS, rather than empirical training on a labeled dataset.
Dataset Splits No The paper describes a mathematical search and verification process, not a machine learning methodology that uses training, validation, and test dataset splits.
Hardware Specification Yes Results in Table 2 were conducted on a cluster of Intel E52683 v4 Broadwell @ 2.1GHz CPUs, each with access to 4 Gi B of RAM and running 64-bit Cent OS Linux 7. Results in Table 3 were conducted on a cluster of Dual Intel Xeon Gold 6226 CPUs @ 2.7 GHz (24 cores/node).
Software Dependencies Yes We used the g++ compiler version 9.3.0 with option -O3 to compile the SAT solvers used.
Experiment Setup Yes In MATHCHECK, we implement a slight modification of traditional Cn C practices to resolve the above challenges. In addition, we use a new Cn C solver ALPHAMAPLESAT [Jha et al., 2024] that provides significant speedup for the cubing process. In our proposed method, the cubing solver of ALPHAMAPLESAT operates on a much smaller CNF instance obtained by omitting all non-colorability constraints (Section 4.4). This approach is substantiated by empirical evidence suggesting that the same set of cubes can be obtained even without the non-colorability constraints. ALPHAMAPLESAT iteratively generates cubes until the total number of cubes surpasses a predefined cutoff, which is based on available computational resources. Following cube generation, each subproblem is processed by the simplification solver (Ca Di Ca L with OG), and then it is passed to the conquering solver (Maple SAT with OG) and solved in parallel. To efficiently manage the termination of each subproblem, a termination strategy is implemented: if the proof size for a subproblem exceeds 7 Gi B, the problem is further divided into more cubes and solved accordingly.