The S version of the simplex minimization works identically to the C
version (as it should), off by a digit or so in the parameter values at
the function minimum for a 6 free parameter nonlinear function (for
bi-bi ordered enzyme mechanism , with product inhibition and EAP
abortive complex).

But.., 

The C --> S calls and calculations within S to return the function
value at the current parameter estimates are slow.  Fast enough,
however, to make the mixed C/S version convenient to use, considering
ease of writing the function in S and the ease of manipulating 
input and output, including plots, in S.

Importantly, there appears to be a memory leak.  Each iteration, the
program size increases by at least 4 kbytes (generally more).
Apparently associated with the repeated C --> S calls for function
evaluation.  Once swapping starts, the program for all practical
purposes locks up.  However, for even a small machine (16 meg), one has
available about 1000 iterations, including search failures, attempted
quadratic fits, etc.  Fortunately, in most cases, 1000 iterations is
enough.


to install in S, read README