Generating Quasi-Symbolic Representation of Three-Dimensional Flow

Abstract

nishidaCis.aist-nara.acjp Understanding flow in the three-dimensional phase space is challenging both to human experts and to current computer science technology. To break through the barrier, we are building a program called PSX3 that can autonomously explore the flow in a three-dimensional phase space, by integrating Al and numerical techniques. In this paper, I point out that quasi-symbolic representation called flow mappings is effective as a means of capturing qualitative aspects of threedimensional flow and present a method of generating flow mappings for a system of ordinary differential equations with three unknown functions. The method is based on a finding that geometric cues for generating a set of flow patterns can be classified into five categories. I demonstrate how knowledge about interaction of geometric cues is utilized for intelligently- controlling numerical computation.