Structural and Computational Properties of Possibilistic Armstrong Databases

We investigate structural and computational properties of Armstrong databases for a new class of possibilistic functional dependencies. We establish sufficient and necessary conditions for a given possibilistic relation to be Armstrong for a given set of possibilistic functional dependencies. We then use the characterization to compute Armstrong databases for any given set of these dependencies. The problem of finding an Armstrong database is precisely exponential in the input, but our algorithm computes an output whose size is always guaranteed to be at most quadratic in a minimum-sized output. Extensive experiments indicate that our algorithm shows good computational behavior on average. As our possibilistic functional dependencies have important applications in database design, our results indicate that Armstrong databases can effectively support business analysts during the acquisition of functional dependencies that are meaningful in a given application domain.