arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2106.02550cs.LO
收藏

通过定义提取的认证DQBF求解

Certified DQBF Solving by Definition Extraction

Franz-Xaver Reichl, Friedrich Slivovsky, Stefan Szeider

摘要:

我们提出了一个新的决策程序依赖量化布尔公式(DQBF),使用基于插值的定义提取计算Skolem函数在反例引导归纳合成(CEGIS)循环。在每次迭代中,使用SAT求解器测试候选Skolem函数族的正确性,SAT求解器确定已经找到模型,或者返回通用变量的赋值作为反例。修复一个反例通常涉及改变具有不可比依赖集的多个存在变量的候选者。我们的过程引入了辅助变量--我们称之为仲裁变量--每个变量代表存在变量的值,用于其依赖集的特定分配。可能的修复被表示为这些变量上的子句,并且调用SAT求解器来找到处理所有先前看到的反例的赋值。添加仲裁器变量定义了Skolem函数的值,用于之前未定义的赋值,并且可能导致在后续迭代中通过定义提取来检测Skolem函数。所提出的过程的一个关键特征是,它是通过设计来验证的:对于真正的DQBF,模型可以以最小的开销返回。对于伪公式的证明,我们证明了子句可以在DQBF的基于扩展的证明系统中导出。在对标准基准集的实验评估中,实现能够匹配(在某些情况下,超越)最先进的DQBF求解器的性能。此外,可以为所有已解决的真实实例生成和验证模型。

英文摘要:

We propose a new decision procedure for dependency quantified Boolean formulas (DQBF) that uses interpolation-based definition extraction to compute Skolem functions in a counter-example guided inductive synthesis (CEGIS) loop. In each iteration, a family of candidate Skolem functions is tested for correctness using a SAT solver, which either determines that a model has been found, or returns an assignment of the universal variables as a counterexample. Fixing a counterexample generally involves changing candidates of multiple existential variables with incomparable dependency sets. Our procedure introduces auxiliary variables -- which we call arbiter variables -- that each represent the value of an existential variable for a particular assignment of its dependency set. Possible repairs are expressed as clauses on these variables, and a SAT solver is invoked to find an assignment that deals with all previously seen counterexamples. Adding arbiter variables defines the values of Skolem functions for assignments where they were previously undefined, and may lead to the detection of Skolem functions by definition extraction in subsequent iterations. A key feature of the proposed procedure is that it is certifying by design: for true DQBF, models can be returned at minimal overhead. Towards certification of false formulas, we prove that clauses can be derived in an expansion-based proof system for DQBF. In an experimental evaluation on standard benchmark sets, an implementation was able to match (and in some cases, surpass) the performance of state-of-the-art DQBF solvers. Moreover, models could be generated and validated for all true instances that were solved.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑