## Sat-based compositional verification using lazy learning (2007)

Venue: | In CAV |

Citations: | 10 - 1 self |

### BibTeX

@INPROCEEDINGS{Sinha07sat-basedcompositional,

author = {Nishant Sinha and Edmund Clarke},

title = {Sat-based compositional verification using lazy learning},

booktitle = {In CAV},

year = {2007},

pages = {3--5}

}

### OpenURL

### Abstract

be interpreted as representing the official policies, either expressed or implied, of any sponsoring institution, the U.S. government or any other entity. Keywords: Assume-Guarantee Reasoning, SAT, SMT, BMC, Learning A recent approach to automated assume-guarantee reasoning (AGR) for concurrent systems relies on computing environment assumptions for components using the L ∗ algorithm for learning regular languages. While this approach has been investigated extensively for message passing systems, it still remains a challenge to scale the technique to large shared memory systems, mainly because the assumptions have an exponential communication alphabet size. In this paper, we propose a SAT-based methodology that employs both induction and interpolation to implement automated AGR for shared memory systems. The method is based on a new lazy approach to assumption learning, which avoids an explicit enumeration of the exponential alphabet set during learning by using symbolic alphabet clustering and iterative counterexample-driven localized partitioning. Preliminary experimental results on benchmarks in Verilog and SMV are encouraging and show that the approach scales well in practice. 1

### Citations

530 |
Learning regular sets from queries and counterexamples
- Angluin
- 1987
(Show Context)
Citation Context ...(their languages are regular), then a finite state assumption A exists. Therefore, the task of computing A is cast as a machine learning problem, where an algorithm for learning regular languages L ∗ =-=[5, 33]-=- is used to automatically compute A. The L ∗ learner computes a deterministic finite automaton (DFA) corresponding to an unknown regular language by asking queries to a teacher entity, which is capabl... |

212 | A fast linear-arithmetic solver for DPLL(T
- Dutertre, Moura
- 2006
(Show Context)
Citation Context ... Bit-vector variables are ”bit-blasted” currently. We check the correctness of the translation by monolithic SAT-based model checking on the translated models. We use the incremental SMT solver YICES =-=[3, 17]-=- as the main decision procedure. Interpolants are obtained using the library interface to the FOCI tool [1]. We represent states of a CFA explicitly while BDDs are used to represent transitions compac... |

113 | Learning assumptions for compositional verification
- Cobleigh, Giannakopoulou, et al.
(Show Context)
Citation Context ...scharge them on its environment (i.e., the other components). The primary bottleneck is that these approaches require us to manually provide appropriate environment assumptions. Recently, an approach =-=[15]-=- has been proposed to automatically generate these assumptions using learning algorithms for regular languages assisted by a model checker. Figure 1 shows a simplified view of this approach for an AGR... |

54 |
Orna Grumberg, Somesh Jha, Yuan Lu, and Helmut Veith. Counterexample-guided abstraction refinement
- Clarke
- 2000
(Show Context)
Citation Context ...rform more efficient counterexample analysis by differentiating positive and negative counterexamples (cf. Section 4). In contrast to the counterexample-guided abstraction refinement (CEGAR) approach =-=[23, 14, 7]-=-, the assumption languages may change non-monotonically across iterations of the learning algorithm. The CEGAR approach removes spurious behaviors from an abstraction by adding new predicates. In cont... |

42 |
Generating Abstract Explanations of Spurious Counterexamples in C Programs
- Ball, Rajamani
- 2002
(Show Context)
Citation Context ...rform more efficient counterexample analysis by differentiating positive and negative counterexamples (cf. Section 4). In contrast to the counterexample-guided abstraction refinement (CEGAR) approach =-=[23, 14, 7]-=-, the assumption languages may change non-monotonically across iterations of the learning algorithm. The CEGAR approach removes spurious behaviors from an abstraction by adding new predicates. In cont... |

27 | An Analysis of SATbased Model Checking Techniques in an Industrial Environment
- Amla, Du, et al.
- 2005
(Show Context)
Citation Context ...e found in [19, 27, 13]. SAT-based bounded model checking for LTL properties was proposed by Biere et al. [10] and several improvements, including techniques for making it complete have been proposed =-=[30, 4]-=-. All the previous approaches are non-compositional, i.e., they build a monolithic transition relation for the whole system. To the best of our knowledge, our work in the first to address automated co... |

15 | Breaking up is hard to do: an investigation of decomposition for assume-guarantee reasoning - Cobleigh, Avrunin, et al. - 2006 |

14 | SAT-based Induction for Temporal Safety Properties
- Armoni, Fix, et al.
(Show Context)
Citation Context ... involves checking for an accepting trace of any length. Several SAT-based approaches have been proposed to solve this problem [30]. In this paper, we consider two approaches, one based on kinduction =-=[34, 18, 6]-=- and the other based on interpolation [25]. The k-induction technique [34] tries to show that there are no accepting traces of any length with the help of two SAT checks corresponding to the base and ... |

14 | Regular inference for state machines with parameters
- Berg, Jonsson, et al.
(Show Context)
Citation Context ...them back iteratively. A learning algorithm for parameterized systems (alphabet consists of a small set of basis symbols, each of which is parameterized by a set of boolean variables) was proposed in =-=[9]-=-. Our lazy learning algorithm is different: we reason about a set of traces directly using a SAT-based model checker and perform more efficient counterexample analysis by differentiating positive and ... |

5 |
Proof rules for automated compositional verification
- Barringer, Giannakopoulou, et al.
- 2003
(Show Context)
Citation Context ...n STSs M1, M2 and CFA P , show that M1 � M2 � P , by picking an assumption CFA A, such that both (n1) M1 � A � P and (n2) M2 � A hold. The following circular rule has also been proposed in literature =-=[8, 27]-=-. Definition 3 Circular AGR (C) Show that M1 � M2 � P holds by picking an CFA assumption tuple, 〈A1,A2〉, such that each of the following hold: (c1) M1 � A1 � P (c2) M2 � A1 � P and (c3) A1 � A2 � P . ... |

4 |
O.: Optimized L* for assume-guarantee reasoning
- Chaki, Strichman
(Show Context)
Citation Context ...artitioning, the BDD-based approach may introduce unnecessary states in the assumptions. Recently, two approaches for improved learning based on alphabet under-approximation and iterative enlargement =-=[12, 19]-=- have been proposed. Our lazy approach is complementary: while the above techniques try to reduce the overall alphabet by under-approximation, our technique tries to compactly represent a large alphab... |

4 |
Prasanna Thati. Automated assume-guarantee reasoning for simulation conformance
- Chaki, Clarke, et al.
- 2005
(Show Context)
Citation Context ...ic algorithms in [31, 27]. The problem of whether it is possible to obtain good decompositions of systems for this approach has been studied in [16]. An overview of other related work can be found in =-=[19, 27, 13]-=-. SAT-based bounded model checking for LTL properties was proposed by Biere et al. [10] and several improvements, including techniques for making it complete have been proposed [30, 4]. All the previo... |

3 |
Zurab Khasidashvili, Amit Palti, and Roberto Sebastiani. Encoding rtl constructs for mathsat: a preliminary report
- Bozzano, Bruttomesso, et al.
(Show Context)
Citation Context ... be carried out using a SMT solver, provided it supports the theories over which the above formulas are defined. A particular mixed boolean/integer encoding of hardware RTL constructs can be found in =-=[11]-=-. Similarly, interpolation-based UMC may be carried out using an interpolating prover provided it can generate interpolants in the required theories. 3 Assume-Guarantee Reasoning using Learning Assume... |