An Algebraic Theory of Relational Databases

Abstract

In this paper we present a theory of relational database systems based on the partition lattice, which represents a new mathematical approach to the structure of relational database systems. A partition lattice can be defined for any given relation. This partition lattice is shown to be a meet-morphic image of the Boolean algebra of subsets of the attribute set. The partial ordering in the lattice is proved to be equivalent to the concept of functional dependency, and thus Armstrong's axioms for functional dependencies are proved. We solve the problem of finding the list of all keys by seeking the prime implicants of the Boolean function associated with the principal ideals generated by the attributes. We demonstrate the properties of the Boyce-Codd Normal Form (BCNF), and give a modified algorithm for synthesizing an information-lossless BCNF based on the principal filter. The necessary and sufficient conditions for multivalued dependency (MVD) are given in terms of a lattice equation, and the inference rules of MVD are proved. The necessary and sufficient conditions for join dependency (JD) are given; consequently, we can prove the known result that acyclic join dependency (AJD) is equivalent to a set of MVDs. The concept of data independence is introduced, and is extended to conditional independence and mutual independence. We established this algebraic theory of relational databases in the same spirit that the theory of probability was constructed. We present a comparison that demonstrates the similarities.