Performance of DLP on Random Modal Formulae

A system that correctly reasons with information in an expressive description logic includes a decision procedure for an expressive propositional modal logic. This decision procedure must be heavily optimised if it is to be able to usefully reason with knowledge bases of any complexity. Description logic systems that incorporate a heavily-optimised propositional modal logic decision procedure include FACT [5], DLP [7], and HAM-ALC [4]. Therefore, a system that efficiently reasons with information in an expressive description logic can be used as a fast decision procedure for an expressive propositional modal logic. The current status is that not only are the fastest reasoners for expressive description logics fast reasoners for propositional modal logics, but for many classes of formulae they are the fastest such reasoners. We have performed numerous experiments with DLP, showing that it is competitive with other reasoners for propositional modal logics, including comparisons presented at recent Tableaux conferences where DLP was the fastest system [1; 8]. In more recent tests we have compared DLP with KSAT [3], TA [6], and KSATC [2] on various collections of random formulae. The two fastest of these systems by a considerable amount are DLP and KSATC. DLP is an experimental description logic system available from Bell Labs at http://www.bell-labs.com/user/pfps/dlp. It implements a very expressive description logic, including full regular expressions on roles. KSATC is a reasoner for K(m), built on a fast Davis-Putnam-Logemann-Loveland decision procedure for propositional logic. KSATC is available at ftp://ftp.mrg.dist.unige.it/pub/mrgsystems/KR98-sources/KSat-source /KSatC.