Experimental errors (and hence weighting  factors for least squares fit)
may be specified by command FERR to be
undefined (corresponding to constant weighting factors),
statistical (square root of data value),
proportional (given by a percentual amount of the data value)
or by the data themselve.
Initially errors are undefined.
Symmetric errors and correlations of fit parameters are taken from the
matrix of second derivatives; in addition an extended error analysis
generally yielding asymmetric errors may be demanded for particular
parameters (commands FPAR, FAREA, FPOS, FWIDTH).
Errors for fit data (commands FLIST, FSTORE) or net integrals of raw data
(command FSUM) are derived from symmetric errors and correlation coefficients
(which are essential for considering error propagation) and listed.