The fit package is based on the algorithms described in Bevington and Brandt.
In particular Marquardt's algorithm is used for function optimization. 
The likelihood function is given by the weighted sum of squares ('chi-square').
The number of function parameters is limited only by the available computer memory.
Parameters may be fixed and limits for their values may be specified.
A special error analysis may be requested to derive asymmetric errors
from the true variation of the likelihood function near the optimum.
In addition, any fit parameter may be correlated to a reference parameter
by fixing the corresponding difference, sum, quotient or product.
 


 
The following fit functions are directly supported by commands:

polynomials, in combination with square root and exponentials,
Legendre series (for angular distributions),
linear exponentials (decay curves) and symmetric Gaussian and Lorentzian peaks.
 
In addition to  this set of standard  functions the user may define his
own fit function, either as a command line formula or as an external
C subroutine.
 
The total fit function may be specified as the sum of partial functions.
This function stacking provides means for background fitting
since any function may act as background for a previous one.
 
Fit utilities include the following:
 
- Definition of data windows to be exclusively considered by the fit
- Linear calibration of the x-axis
- Display of the total fit function, single components or single peaks
  with guessed or fitted parameter values
- Listing of raw and fit data including errors considering correlations between fit parameters
- Calculation of sum and first to third moments of raw data
  (or raw data minus fit data) in a window