Computational models of physics problem solving

Solving typical textbook physics problems, such as those found in books used in high school and first year college physics courses, involves several subtasks. These subtasks can be described in terms of the way the problem is represented. Computational approaches to physics problem solving can be distinguished by the subtasks that they address and the types of representations that they use. A general descriptive framework of physics problem solving in terms of five different types or levels of representation that can be used in understanding and solving a physics problem is presented. Six computer programs that investigate various aspects of physics problem solving are presented and compared against the general descriptive framework. This comparison against a common framework makes clear certain differences among the reviewed programs in terms of what subtasks each is addressing. Important issues not reflected in the framework and only briefly addressed in the paper include learning and organization of knowledge. The systems reviewed include Novak's ISAAC, Bundy, Byrd, Lueger, Mellish, and Palmer's MECHO, de Kleer's NEWTON, Larkin and Simon's ABLE, Shavlik and de Jong's PHYSICS101, and Larkin, Reif, Carbonell, and Cheng's FERMI. These systems were chosen because they explicitly deal with problems typical of beginning physics textbooks. Related work on naive physics and qualitative reasoning about physical mechanisms and processes is not addressed in this paper