Multivalued Dependencies With Null Values In Relational Data Bases

The role of null values in a relational data base is considered within the framework of multivalued dependencies. When a relation contains null values, a different treatment of data dependencies and different relational operations such as projection and join, are necessary. This paper develops a complete axiomatization for the revised multivalued dependencies. In particular, complementation, reflexivity, augmentation, and union are shown to be a complete set of inference rules. In contrast with the conventional multivalued dependencies, the transitivity rule cannot be used. These results provide a framework for the use of the revised multivalued dependencies in choosing relations and their attributes for a feasible data-base design.