Server: Netscape-Commerce/1.12 Date: Tuesday, 26-Nov-96 00:07:00 GMT Last-modified: Thursday, 15-Jun-95 00:37:21 GMT Content-length: 3111 Content-type: text/html Project on Mathematics and Computation

Harold Abelson,
Class of 1922 Professor of Computer Science and Engineering
Gerald Sussman,
Matsushita Professor of Electrical Engineering

Solving scientific problems increasingly depends on high-speed computation, careful planning of numerically based experiments, and high-level, qualitative interpretation of large volumes of quantitative data. Working jointly with the Artificial Intelligence (AI) laboratory, researchers within MIT's Project on Mathematics and Computation are developing a wide range of intelligent computation tools to help scientists and engineers understand mathematical models of physical processes. Some of the Project's goals include efficient new algorithms, specialized hardware, and "smart" programs that comprehend measured or numerical data. Much of our work in scientific computation is based on AI methods. We have used computer-vision techniques, for example, to create programs that "look at" and qualitatively interpret graphical results of numerical experiments. Other programs can construct numerical simulation systems of dynamically complex processes (such as oscillating chemical reactions and nonlinear vibrating beams), then automatically incorporate them into qualitative-analysis programs. Such programs can also generate high-level summary descriptions -- graphically and in English -- of numerical experiments, similar to the descriptions that appear in published scientific and engineering papers.

To support the automatic construction of numerical procedures, we are seeking ways to express numerical algorithms in terms of high-order procedural abstractions. Sophisticated numerical routines thus can be assembled by mixing and matching components from a numerical library. The large library of routines we are now assembling will contain symbolic methods as well as numeric ones.

Our work rests on the Scheme dialect of Lisp and the Scheme programming environment, which we invented and continue to develop. We have joined with other Scheme users and developers to formally standardize this dialect through the IEEE. We expect that our freely available native-code Scheme compiler will offer performance comparable to that of the best commercial Lisp compilers.