## Asymptotic Conditional Probabilities for First-Order Logic (1992)

Venue: | In Proc. 24th ACM Symp. on Theory of Computing |

Citations: | 13 - 7 self |

### BibTeX

@INPROCEEDINGS{Grove92asymptoticconditional,

author = {Adam J. Grove and Joseph Y. Halpern and Daphne Koller},

title = {Asymptotic Conditional Probabilities for First-Order Logic},

booktitle = {In Proc. 24th ACM Symp. on Theory of Computing},

year = {1992},

pages = {294--305}

}

### OpenURL

### Abstract

Motivated by problems that arise in computing degrees of belief, we consider the problem of computing asymptotic conditional probabilities for first-order formulas. That is, given first-order formulas ' and `, we consider the number of structures with domain f1; : : : ; Ng that satisfy `, and compute the fraction of them in which ' is true. We then consider what happens to this probability as N gets large. This is closely connected to the work on 0-1 laws that considers the limiting probability of first-order formulas, except that now we are considering asymptotic conditional probabilities. Although work has been done on special cases of asymptotic conditional probabilities, no general theory has been developed. This is probably due in part to the fact that it has been known that, if there is a binary predicate symbol in the vocabulary, asymptotic conditional probabilities do not always exist. We show that in this general case, almost all the questions one might want to ask (such as d...

### Citations

10886 |
Computers and Intractability: A Guide to the Theory of NP-Completeness
- Garey, Johnson
- 1979
(Show Context)
Citation Context ...erested in is Pr w 1 ('j`), which is not integer valued. Nevertheless, we want to show that it is in #EXP. In the spirit of similar definitions for #P (see, for example, [Val79b, PB83]) and NP (e.g., =-=[GJ79]-=-) we extend the definition of #EXP to apply also to noninteger valued functions. Definition 5.5 : An arbitrary function f is said to be #EXP-easy if there exists an integer function f 0 in #EXP and a ... |

1498 |
The temporal logic of reactive and concurrent systems
- Manna, Pnueli
- 1992
(Show Context)
Citation Context ... persistent limit, denoted 32Pr w 1 ('j`), and the intermittent limit, denoteds23Pr w 1 ('j`) (the prefixes stand for the temporal logic representation of the persistence and intermittence properties =-=[MP92]-=-). In either case, we say that the limiting probability is either not well-defined, does not exist, or is some fraction between 0 or 1. The only difference between the two notions lies in when the lim... |

837 |
Theory of recursive functions and effective computability
- Rogers
- 1967
(Show Context)
Citation Context ...s of the arithmetic hierarchy. This is a hierarchy that extends the notions of r.e. (recursively enumerable) and co-r.e. sets. We briefly review the relevant definitions here, referring the reader to =-=[Rog67] for fur-=-ther details. Consider a formula �� in the language of arithmetic (i.e., using 0; 1; +; \Theta) having k free variables. The formulas��, interpreted over the natural numbers, is said to define... |

502 |
The complexity of computing the permanent
- Valiant
- 1979
(Show Context)
Citation Context ... checking whether the limit exists is NEXPTIME-complete, while the problem of computing the probability if it is exists is essentially #EXP-complete (where #EXP is the exponential time analogue of #P =-=[Val79a]-=-). Somewhat surprisingly in light of our other results, these results hold even if we restrict to formulas of quantification depth two. For unquantified formulas or depth one quantification, things be... |

418 | The complexity of enumeration and reliability problems - Valiant - 1979 |

358 | 1921�, “A Treatise on Probability
- Keynes
- 1948
(Show Context)
Citation Context ... same general approach. The approach is an old one, going back to Laplace [Lap20]. It is essentially what has been called the principle of insufficient reason [Kri86] or the principle of indifference =-=[Key21]. The idea-=- is to assign equal degree of belief to all basic "situations" consistent with the known facts. We consider two embodiments of this principle, which we call the random-worlds method and the ... |

142 |
Where do we stand on maximum entropy
- Jaynes
- 1978
(Show Context)
Citation Context .... We have already mentioned its relevance to computing degrees of belief. There are well-known results from physics that show the close connection between the random-worlds method and maximum entropy =-=[Jay78]-=-; some formalization of similar results, but in a framework that is close to that of the current paper, can be found in [PV89, GHK92]. Essentially, the results say that in certain cases the asymptotic... |

126 | The complexity of counting cuts and of computing the probability that a graph is connected - Provan, Ball - 1982 |

122 |
Probabilities on finite models
- Fagin
(Show Context)
Citation Context ...is work on special cases, to the best of our knowledge there has been no work on the general problem of asymptotic conditional probabilities. Perhaps the main reason for this is that it is well known =-=[Fag76]-=- that we do not always get convergence of conditional probabilities. In this paper, motivated by our use our own terminology for these methods, partly because we feel it is more descriptive, and partl... |

76 |
The Decision Problem. Solvable Classes of Quantificational Formulas
- Dreben, Goldfarb
- 1979
(Show Context)
Citation Context ... the fact that first-order logic over a vocabulary with only unary predicates (and constant symbols) is decidable, while if we allow even a single binary predicate symbol, then it becomes undecidable =-=[DG79, Lew79]-=-. This similarity is not coincidental; some of the techniques used to show that first-order logic over a vocabulary with unary predicate symbols is decidable are used here to show that asymptotic limi... |

55 | Range and degree of realizability of formulas in the restricted predicate calculus - Glebskii, Kogan, et al. - 1969 |

49 | Random worlds and maximum entropy
- Grove, Halpern, et al.
- 1994
(Show Context)
Citation Context ...arise with statistical assertions are significantly different from those that arise with first-order assertions, in this paper we restrict attention to the pure first-order case; in a companion paper =-=[GHK92]-=- we consider the case of statistical assertions. To readers familiar with the work on asymptotic probabilities for various logics, our questions will seem quite familiar. For example, in the context o... |

46 |
0-1 laws in logic and combinatorics
- Compton
- 1988
(Show Context)
Citation Context ...nguage should only be counted once. 1 1 The random-worlds method considers what has been called in the literature labeled structures, while the random-structures method considers unlabeled structures =-=[Com88]-=-. We choose to We often do not know the precise domain size N , but do know that it is large. Thus, we are particularly interested in knowing the asymptotic behavior of the degree of belief. Thus, we ... |

46 |
Complexity results for classes of quantificational formulas
- Lewis
- 1980
(Show Context)
Citation Context ...rom this number we can easily calculate the asymptotic conditional probability, using the formula in Theorem 4.9. We leave details to the full paper. For the lower bound, we use a result due to Lewis =-=[Lew80]-=-; he shows that the satisfiability problem for L \Gamma (Q) is complete for NEXPTIME. Lewis' proof involves a straightforward reduction of NEXPTIME to L \Gamma (Q): given a nondeterministic Turing mac... |

43 | From statistics to beliefs
- Bacchus, Grove, et al.
- 1992
(Show Context)
Citation Context ...hese methods, partly because we feel it is more descriptive, and partly because there are other variants of the approach that do not fit into the standard labeled /unlabeled structures dichotomy (see =-=[BGHK92]-=-) . desire to apply these methods to computing degrees of belief, we consider the general question of when asymptotic probabilities exist for first-order logic, and how to compute them if they do. We ... |

38 |
A method of computing generalized Bayesian probability values for expert systems
- Cheeseman
- 1983
(Show Context)
Citation Context ...redicates). For example, in physics applications we are interested in such predicates as quantum state (see [DD85]). Similarly, AI applications and expert systems typically use only unary predicates (=-=[Che83]-=-) such as symptoms and diseases. In these examples, the vocabulary typically used is relatively small, and quantifiers are rarely deeply nested. This, of course, is precisely the case where our linear... |

25 |
Entropy in Relation to Incomplete Knowledge
- Denbigh, Denbigh
- 1985
(Show Context)
Citation Context ...ivalently, unary functions with a finite range, which of course can be encoded using unary predicates). For example, in physics applications we are interested in such predicates as quantum state (see =-=[DD85]-=-). Similarly, AI applications and expert systems typically use only unary predicates ([Che83]) such as symptoms and diseases. In these examples, the vocabulary typically used is relatively small, and ... |

22 | Unsolvable classes of quantificational formulas - LEWIS - 1979 |

18 | On the applicability of maximum entropy to inexact reasoning - Paris, Vencovská - 1989 |

17 |
Almost sure theories
- Lynch
(Show Context)
Citation Context ... results can be viewed as special cases of the problem that we are interested in: computing asymptotic conditionalsprobabilities relative to structures satisfying the constraints of a database. Lynch =-=[Lyn80]-=- showed that, for the random-worlds method, asymptotic limits exist for first-order formulas involving unary functions, although there is no 0-1 law. (Recall that the original 0-1 result is specifical... |

14 |
Essai Philosophique sur les Probabilités. English trans ˙ by
- Laplace
- 1951
(Show Context)
Citation Context ...principled manner. In this paper we investigate the properties of two formalisms for computing degrees of belief, based on the same general approach. The approach is an old one, going back to Laplace =-=[Lap20]. It is es-=-sentially what has been called the principle of insufficient reason [Kri86] or the principle of indifference [Key21]. The idea is to assign equal degree of belief to all basic "situations" c... |

8 |
Summability Theory and Applications
- Powell, Shah
- 1972
(Show Context)
Citation Context ...oof of Theorem 3.1), the regular limit does not exist, but the Ces`aro limit does, and is 1=2. Unfortunately, we can show that for any definition of limit satisfying some very basic restrictions (see =-=[PS72]-=- for some definitions), the limit of the conditional probabilities may not exist. That is, for any notion of limit in a large class, there exist sentences '; ` 2 L(fSg) (for a binary predicate S), suc... |

5 |
Kries. Die Principien der Wahrscheinlichkeitsrechnung und Rational Expectation
- von
- 1886
(Show Context)
Citation Context ...computing degrees of belief, based on the same general approach. The approach is an old one, going back to Laplace [Lap20]. It is essentially what has been called the principle of insufficient reason =-=[Kri86] or the pr-=-inciple of indifference [Key21]. The idea is to assign equal degree of belief to all basic "situations" consistent with the known facts. We consider two embodiments of this principle, which ... |

4 |
The number of finite relational structures. Discrete Mathematics
- Fagin
- 1977
(Show Context)
Citation Context ...with no constant or function symbols, then (for both methods) its asymptotic probability is either 0 or 1 [GKLT69, Fag76]. Moreover, it is known that both methods give the same asymptotic probability =-=[Fag77]-=-; as we shall see, this is not the case once we move to conditional probabilities. After the initial results, 0-1 laws were proved for special classes of first-order structures (such as graphs, tourna... |