Entity and referential integrity are the most fundamental constraints that any relational database should satisfy. We re-examine these fundamental constraints in the context of incomplete relations, which may have null values of the types "value exists but is unknown" and "value does not exist" . We argue that in practice the restrictions that these constraints impose on the occurrences of null values in relations are too strict. We justify a generalisation of the said constraints wherein we use key families, which are collections of attribute sets of a relation schema, rather than keys, and foreign key families which are collections of pairs of attribute sets of two relation schemas, rather than foreign keys. Intuitively, a key family is satisfied in an incomplete relation if each one of its tuples is uniquely identifiable on the union of the attribute sets of the key family, in all possible worlds of the incomplete relation, and, in addition, is distinguishable from all the other tuples in the incomplete relation by its nonnull values on some element in the key family. Our proposal can be viewed as an extension of Thalheim's key set, which only deals with null values of type "value exists but is unknown" . The intuition behind the satisfaction of a foreign key family in an incomplete database is that each pair $(F_i, K_i)$ of attribute sets in the foreign key family takes the foreign key attribute values over Fi of a tuple in one incomplete relation referencing the key attribute values over Ki of a tuple in another incomplete relation. Such referencing is defined only in the case when the foreign key attribute values do not have any null values of the type "value does not exist" ; we insist that the referencing be defined for at least one such pair. We also investigate some combinatorial properties of key families, and show that they are comparable to the standard combinatorial properties of keys.

A PAC teaching model -under helpful distributions -is proposed which introduces the classical ideas of teaching models within the PAC setting: a polynomial-sized teaching set is associated with each target concept; the criterion of success is PAC identification; an additional parameter, namely the inverse of the minimum probability assigned to any example in the teaching set, is associated with each distribution; the learning algorithm running time takes this new parameter into account. An Occam razor theorem and its converse are proved. Some classical classes of boolean functions, such as Decision Lists, DNF and CNF formulas are proved learnable in this model. Comparisons with other teaching models are made: learnability in the Goldman and Mathias model implies PAC learnability under helpful distributions. Note that Decision lists and DNF are not known to be learnable in the Goldman and Mathias model. A new simple PAC model, where "simple" refers to Kolmogorov complexity, is introduced. We show that most learnability results obtained within previously defined simple PAC models can be simply derived from more general results in our model.

Branching programs are a well established computation model for Boolean functions, especially read-once branching programs have been studied intensively. In this paper the expressive power of nondeterministic read-once branching programs, more precisely the class of functions representable in polynomial size, is investigated. For that reason two restricted models of nondeterministic read-once branching programs are defined and a lower bound method is presented. Furthermore, the first exponential lower bound for integer multiplication on the size of a nondeterministic nonoblivious read-once branching program model is proven.

Digital trees or tries are a general purpose flexible data structure that implements dictionaries built on words. The present paper is focussed on the average-case analysis of an important parameter of this tree-structure, i.e., the stack-size. The stack-size of a tree is the memory needed by a storage-optimal preorder traversal. The analysis is carried out under a general model in which words are produced by a source (in the information-theoretic sense) that emits symbols. Under some natural assumptions that encompass all commonly used data models (and more), we obtain a precise average-case and probabilistic analysis of stack-size. Furthermore, we study the dependency between the stack-size and the ordering on symbols in the alphabet: we establish that, when the source emits independent symbols, the optimal ordering arises when the most probable symbol is the last one in this order.

We investigate the number of iterations needed by an addition algorithm due to Burks et al. if the input is random. Several authors have obtained results on the average case behaviour, mainly using analytic techniques based on generating functions. Here we take a more probabilistic view which leads to a limit theorem for the distribution of the random number of steps required by the algorithm and also helps to explain the limiting logarithmic periodicity as a simple discretization phenomenon.