In order to provide for full generality in the user-defined function
value, the user is allowed to define a normalization factor known
internally as UP
and defined by the Minuit user on an
[SET ERRordef]ERROR DEF command card.
The default value is one. The Minuit error on a
parameter is defined as the change of parameter which would
produce a change of the function value equal to UP
.
This is the most
general way to define the error, although in statistics it is more
usual to define it in terms of the second derivative of the
function -- with respect to the parameter in question. In the
simplest linear case (when the function is exactly parabolic at the
minimum), the value UP=1.0
corresponds to defining the error as the
inverse of the second derivative at the minimum. The fact that
Minuit defines the error in terms of a function change does not mean
that it always calculates such a function change. Indeed it
sometimes ([HESse]HESSE) calculates the second derivative matrix and
inverts it, assuming a parabolic behaviour. This distinction is
discussed in section .
The purpose of defining errors by function changes is threefold: