Improved bounds on the probability of the union of events some of whose intersections are empty
Introduction
Computing the probability of the union of events is important in reliability theory, stochastic programming, and other sciences concerned with stochastic systems. In network reliability, consider a communication network with nodes and links, each with a probability of failure. The two-terminal reliability of a pair of nodes is the probability of the union of events, each of which occurs when a path between the two nodes consists of links without failure. The all-terminal reliability is the probability of the union of events, each of which occurs when a spanning tree of the network consists of links without failure. In probabilistic constrained stochastic programming, a joint probabilistic constraint for random variables specifies a lower bound on , which involves the probability of the union of events. Although it is very hard to compute the exact probability of the union of a large number of events, we can compute its approximation by using the probabilities of individual events and intersections of a small number of events.
Let be any set of events in an arbitrary probability space and be the set of all positive integers at most . For any finite set , we designate its cardinality (i.e., the number of elements in ) by . For any subset , we introduce a notation for the probability of the intersection of events for all as follows:
We introduce binomial moments for all as follows: Let denote the random number of events among that occur. Then we have the following relation for all : where we employ the extended definition of binomial coefficients, in which for all such that . The value is called the th binomial moment of .
The classical inclusion–exclusion principle [10] (see also Prékopa [21]) gives the probability of the union of events by using binomial moments for all as follows: However, this formula is impractical if the number of events is large, in which case the calculation of is intractable unless is small enough (close to 1) or large enough (close to ) since we need sums in (1). We can still calculate for a few small and compute lower and upper bounds on the probability. Let be the largest number of events the probability of whose intersection is used in approximating the probability of the union and be the set of all positive integers at most ; we use only for all such that , from which can be calculated by (1). In practice, we usually consider a small , mostly such as . The well-known Bonferroni inequalities (or bounds) [2] state that for any , These bounds are usually very weak, often out of [0,1]. Then the best possible (also called sharp) bounds using for a small have been found. In the case , the sharp lower and upper bounds expressed as closed forms in terms of are obtained by Dawson and Sankoff [8] and Kwerel [15], respectively (see also Prékopa [18] and Boros and Prékopa [3]). In the case , the sharp bounds expressed as closed forms in terms of are obtained by Boros and Prékopa [3]. In the case , the sharp upper bound expressed as a closed form in terms of is obtained by Boros and Prékopa [3]. For a general , Prékopa [18] observed that all these sharp bounds are optimal objective function values of certain linear programs, known as binomial moment problems. Furthermore, sharp bounds on the probabilities that exactly/at least events occur are given in Prékopa [19], and the bounds on the probabilities and expectations of convex functions of discrete random variables are given in Prékopa [20]. A binomial moment carries an aggregated information over event probabilities , using every one of which without aggregation we can obtain better bounds. Hailperin [14] formulated linear programs with an exponential number of variables, known as Boolean probability bounding problems, which give sharp bounds using disaggregated event probabilities.
A pair of binomial moment problems (also called the aggregated LP problems, for probability bounds of the union of events) is formulated as follows [18]: The optimal objective function values of these minimization and maximization problems are the sharp lower and upper bounds, respectively, on the probability of the union of events that can be computed using only the aggregated information for all .
A pair of Boolean probability bounding problems (also called the disaggregated LP problems, for probability bounds of the union of events) is formulated as follows [14]: where we define The optimal objective function values of these minimization and maximization problems are the sharp lower and upper bounds, respectively, on the probability of the union of events that can be computed using only the disaggregated information for all such that . Since such a set of event probabilities is almost always the most detailed information we can use, the bounds are the best possible we can expect in general. These problems are, however, impractical owing to an exponential number of the decision variables .
Although we cannot solve the disaggregated LP problems for a large number of events in practice, we may extract useful information from disaggregated event probabilities without dealing with an exponential size and use it with their aggregations to obtain improved bounds.
Section snippets
Improved bounds
We assume that all probabilities of intersections of small numbers of events (formally, for all such that ) are known and some of them are 0 (i.e., such intersections are empty). The smallest number of events that have an empty intersection is 2 since we may use only nonempty individual events when computing their union; hence, we assume that . For any nonempty subsets such that and any , we have the implication . In particular when , we have the
Numerical experiments
We carried out numerical experiments to compare probability bounds by our formulations to those by binomial moment problems with data sets that are expected to contain empty intersections of events. We used the Gurobi™ Optimizer (version 6.0) [12] for solving linear programs; when we solve binomial moment problems, for which numerical inaccuracy of the feasibility of solutions is sometimes encountered, we set the feasibility tolerance parameter (‘FeasibilityTol’) to its minimum value .
Applications
Approximating the probability of the union of events has applications in network reliability (Boros et al. [5], Prékopa and Boros [22], and Gao and Prékopa [11]; see also Ball et al. [1]) and system reliability (Habib and Szántai [13] and Unuvar et al. [24]; see also a survey by Chao et al. [7]).
In most real-world examples, not a small number of intersections of events are often empty. Many intersections of half-spaces used in expressing the complement of a Euclidean polyhedron are empty
Acknowledgments
We are grateful to Prof. Endre Boros for his valuable comments. This research was supported by NSF Grant CMMI-0856663.
References (24)
- et al.
Network reliability
- et al.
New bounds on the reliability of the consecutive -out-of--from-: F system
Reliab. Eng. Syst. Saf.
(2000) The discrete moment problem and linear programming
Discrete Appl. Math.
(1990)- et al.
Closed form two-sided bounds for probabilities that at least and exactly out of events occur
Math. Oper. Res.
(1989) - et al.
Probabilistic bounds and algorithms for the maximum satisfiability problem
Ann. Oper. Res.
(1989) - et al.
The use of binomial moments for bounding network reliability
- J. Bukszár, R. Henrion, M. Hujter, T. Szántai, Polyhedral inclusion-exclusion, Stochastic Programming E-Print Series...
- et al.
Survey of reliability studies of consecutive--out-of-:F and related systems
IEEE Trans. Reliab.
(1995) - et al.
An inequality for probabilities
Proc. Amer. Math. Soc.
(1967)
The Doctrine of Chances: Or, A method of Calculating the Probability of Events in Play
Cited by (8)
Bounds for probabilistic programming with application to a blend planning problem
2022, European Journal of Operational ResearchCitation Excerpt :Further Bonferroni-type inequalities and a summary of them can be found in the book (Bukszár, Mádi-Nagy, & Szántai, 2012; Galambos & Simonelli, 1996). Improved bounds on the probability of the union of events some of whose intersections are emptly are discussed in Yoda & Prékopa (2016). Subasi, Subasi, Binmahfoudh, & Prékopa (2017) improve the previous bounds using the shape information of the distribution of the random variable based on the knowledge of some binomial moments.
Sharp probability bounds for the binomial moment problem with symmetry
2018, Operations Research LettersNew bounds for the probability that at least k-out-of-n events occur with unimodal distributions
2017, Discrete Applied MathematicsCitation Excerpt :Probability bounds based on the probabilities of the individual events and their intersections, and graph structures also exist in literature [28,10,8,55,9]. The reader is referred to papers by Veneziani (2009) [56], Boros, Scozzari, Tardella, and Veneziani (2014) [7], Prékopa, Ninh, and Alexe (2016) [45], and Prékopa and Yoda (2016) [47] for recent linear programming based probability bounds. Other studies on probability bounding can be found in [13,25,34,48].
TIGHT PROBABILITY BOUNDS WITH PAIRWISE INDEPENDENCE
2023, SIAM Journal on Discrete MathematicsThe value of shape constraints in discrete moment problems: a review and extension
2022, Annals of Operations Research