The Qualitative Process Engine

Efficient qualitative simulators are crucial to continued progress in qualitative physics. Assumption-based truth maintenance systems (ATMS) were developed in part to simplify writing such programs. This paper identifies several general abstractions for organizing ATMS-based problem-solvers which are especially useful for envisioning. In particular, we describe the many-worlds database, which avoids complex temporal reference schemes; how to organize problem-solving into justify/assume/interpret cycles which successively construct and extend partial solutions; and closed-world tables, which provide a mechanism for making closed-world assumptions. We sketch the design of the Qualitative Process Engine, QPE, an implementation of Qualitative Process theory, to illustrate the utility of these abstractions. On the basis of our experience in developing QPE and analyzing its performance, we draw some general conclusions about the advantages and disadvantages of assumption-based truth maintenance systems.