Scribd
Upload
0 views2 pages

Assign Examples

The document discusses the use of partial assignments in CNF (Conjunctive Normal Form) and their implications for enumeration strategies in AllSAT and AllSMT tasks. It emphasizes the benefits of using NNF (Negation Normal Form) combined with CNFPG for generating shorter assignments, which can lead to improved performance compared to traditional methods. Additionally, it outlines the experimental evaluation of various CNF encodings and the selection of solvers for effective comparison.

Uploaded by

seyl
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0 views2 pages

Assign Examples

The document discusses the use of partial assignments in CNF (Conjunctive Normal Form) and their implications for enumeration strategies in AllSAT and AllSMT tasks. It emphasizes the benefits of using NNF (Negation Normal Form) combined with CNFPG for generating shorter assignments, which can lead to improved performance compared to traditional methods. Additionally, it outlines the experimental evaluation of various CNF encodings and the selection of solvers for effective comparison.

Uploaded by

seyl
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

CNF assignment examples

As in Examples 1 and 2, consider the partial assignment 𝜇A def = {￢𝐴3, ￢𝐴4, ￢𝐴7} (4) which satisfies 𝜑.
First, it is
easy to see that 𝜇A satisfies also NNF(𝜑), in compliance with Lemma 2. Also, 𝜇A satisfies ∃
B.CNFPG(NNF(𝜑)),
because 𝜇A ∪ 𝜂B |=𝑝 CNFPG(NNF(𝜑)) where 𝜂B def = {￢𝐵+1, ￢𝐵+2, 𝐵−2, ￢𝐵+3, ￢𝐵−3, ￢𝐵+4 , 𝐵−4, 𝐵+5, 𝐵+6 ,
𝐵+7}.
The above assignment can be produced by adopting the strategy described above. First, we assign the
literals in 𝜇A,
which force to add also {￢𝐵+2, ￢𝐵+4 } due to (11f) and (11h) respectively. Then we add {𝐵+6 } by step
(a) on (11i), {𝐵−2 } by
step (b) on (11b), and {￢𝐵+1 , ￢𝐵−3, ￢𝐵+3 } by step (c) on (11a), (11c), (11g) respectively. These force to
add {𝐵+7 } and {𝐵+5 }
due to (11k) and (11j) respectively. Finally, we add {𝐵−4} by step (b) on (11d). Overall, this corresponds
to enumerate
only the partial assignment 𝜇A satisfying ∃B.CNFPG(NNF(𝜑)), with both disjoint and non-disjoint
enumeration:
{￢𝐴3, ￢𝐴4, ￢𝐴5} //{￢𝐵+1, ￢𝐵+2, 𝐵−2, ￢𝐵+3, ￢𝐵−3 , ￢𝐵+4, 𝐵−4 , 𝐵+5, 𝐵+6, 𝐵+7 }. ⋄
Notice that, the shorter is 𝜇A w.r.t. a total assignment, the higher is the chance that CNFTs and CNFPG force
the
production of a high number of extra assignments, the more beneficial is the usage of NNF+CNFPG which
avoids
it. This said, if the enumerator is able to produce short partial assignments 𝜇A satisfying the formula, we
expect a
high benefit from using NNF+CNFPG instead of CNFTs and CNFPG; vice versa, if the enumerator produces only
total or nearly-total assignments, we expect no or very-limited benefit respectively.
Implementation-wise, the strategy to assign the values of 𝜂B described above is difficult to implement
inside the
current enumerators. Therefore, there is no formal guarantee that a generic enumeration procedure
always finds
exactly the 𝜂B which prevents the generation of longer assignments. For example, the enumeration
procedure of
§2.2 finds a total truth assignment 𝜂A ∪ 𝜂B that satisfies the formula, and then finds 𝜇A ⊆ 𝜂A that is
minimal
w.r.t. that specific 𝜂B such that 𝜇A ∪ 𝜂B |=𝑝 CNFPG(NNF(𝜑)), so that the 𝜂B found is not guaranteed to be
the
one that allows for the most effective minimization of 𝜇A. Ad-hoc enumeration heuristics could be adopted.
Nevertheless, in §5 we show that a very simple heuristic —i.e., force the assignments of false values first
to decision
atoms— guarantees dramatic improvements w.r.t. previous approaches using two state-of-the-art
AllSAT/AllSMT
enumerators.
5 Experimental Evaluation
In this section, we experimentally evaluate the impact of different CNF-izations on the disjoint and non-
disjoint
AllSAT and AllSMT tasks. In order to compare them on fair ground, we have implemented a base version
of each
from scratch in PySMT [22], avoiding specific optimizations done by the solvers.
We have conducted a very extensive collection of experiments on AllSAT and AllSMT.We tested the
encodings
on MathSAT, which supports disjoint and non-disjoint AllSAT and AllSMT, and on TabularAllSAT and
TabularAllSMT, which support disjoint AllSAT and AllSMT, respectively. The choice of the solvers is
motivated
in §5.1; the benchmarks are described in §5.2, the information for the reproducibility of the experiments is
given in
§5.3, and the results are presented in §5.4; finally, in §5.5 we analyze and discuss the application of the
encodings
to related fields.
5.1 An Analysis of Available Solvers
In order to evaluate the different CNF encodings, we need an AllSAT/AllSMT solver that (i) is publicly
available,
(ii) takes as input a CNF formula, (iii) allows performing projected enumeration, (iv) allows enumerating
(disjoint
or non-disjoint) partial truth assignments; (v) also, the ability of producing minimal assignments is valuable
although not necessary.
In the literature, we found the following candidate solvers: for AllSAT, we found RELSAT [4], Grumberg
[26],
SOLALL [42], Jin [33, 34], clasp [24], PicoSAT [7], Yu [79], BC, NBC, and BDD [76], depbdd [75], Dualiza
[56],
BASolver [80], AllSATCC [43], HALL [21, 20], TabularAllSAT [72, 73], and model-graph [37]. For AllSMT,
the only candidates are MathSAT [15], aZ3 [63], and TabularAllSMT [73].

You might also like