JAKEF, Gradient or Jacobian Function from Objective Function or Vector Function

1 - Description of program or function: JAKEF is a language processor that accepts as input a single- or double-precision ANSI standard 1977 FORTRAN subroutine defining an objective function f(x), or a vector function F(x), and produces as output a single- or double- precision ANSI standard 1977 FORTRAN subroutine defining the gradient of f(x), or the Jacobian of F(x). 2 - Method of solution: JAKEF is a four-pass compiler consisting of a lexical preprocessor, a parser, a tree-building and flow analysis pass, and a differentiator and output construction pass. The lexical preprocessor reworks the input FORTRAN program to give it a recognizable lexical structure. The parser transforms the pre-processed input into a string of tokens in a post-fix representation of the program tree. The tree-building and flow analysis pass constructs a tree out of the post-fix token string. The differentiator identifies relevant assignment statements; then, if necessary, it analyzes them into component statements governed by a single differentiation rule and augments each of these statements with a call to a member of the run-time support package which implements the differentiation rule. After completing the construction of the main body of the routine, JAKEF inserts calls to support package routines that complete the differentiation. This results in a modified program tree in a form compatible with FORTRAN rules. 3 - Restrictions on the complexity of the problem: Statement functions and Equivalence's that involve the independent variables are not handled correctly. Variables, constants, or functions of type COMPLEX are not recognized. Character sub-string expressions and alternate returns are not permitted