.-
help for ^simplex^                                             (STB-32: sg56)   
.-

Simplex algorithm for function minimization
-------------------------------------------

    
    ^simplex^ varlist [^if^ exp] [^in^ range] ^,^ ^from(^matrix^)^ 
        ^lf(^function^)^ [^other-ml-options^]


^simplex^ uses Nelder and Meads amoeba algorithm for function minimization.
This is a simplex method that does not require derivative calculations.

Options
-------

^from^ specifies a matrix of initial values.  Unlike the usual Stata default,
    the column names for this matrix are assumed to correspond to those given
    in the variable list and need not be specified before the matrix is
    supplied.  Any column names supplied will be ignored.

^iterate^ specifies a maximum number of iterations to compute before the program
    gives up.  The default is 16,000.

^lf^ specifies the likelihood function to be used.  It is assumed to be of the 
     ^deriv0^ type as specified in the Stata manual.

^noconstant^ specifies that there will not be a constant term in the estimation
    (the default is to include one).  If a constant term is specified, an
    initial value for it must be specified as the last column of the matrix of
    initial values.

^step^ specifies how finely the program looks along the simplex path for
    higher likelihood values.  A step value close to zero causes the program
    to look more frequently, while a value close to one causes it to only look
    once.  The default is .1.  Choosing finer values will probably only
    increase computational time per iteration, however if the function is
    non-concave in small intervals, a smaller value for step may pick up peaks
    that would otherwise be missed.

^span^ The default initial simplex consists of the vector specified in the
    ^from^ option, plus a vector corresponding to each parameter which is the
    same as the ^from^ vector, except that the initial value for that
    parameter in these vectors is the value specified in ^from^ plus that
    specified in ^span^.  For example, if there are two parameters, and the
    user specifies from as (1,1), the program will use the three points (1,1),
    (1+span,1), and (1,1+span) as the initial simplex.  The default is .1.

^tolerance^ specifies how close the different simplex points must be in terms
    of the largest norm difference before the program considers them
    essentially the same and stops working and reports its estimates.  The
    default is .00001.

^trace^ outputs more information about the path of the simplex than anyone
    probably wanted to know.


Author
------

        Tavis Barr                               
        Department of Economics, Columbia University