CERN users' requirements for extensions of the NAG C family of minimizers
 - Z. Szkutnik, Oct. 97 -
 

The following set of requirements consists of those user requirements for the Minuit++ project, which relate to the minimization package and are not satisfied by version 4.0 of the NAG C library. It covers that part of Minuit functionality, which is not provided by the NAG C library and which is seen as indispensable by the physics community.

 

MUR 1. Error estimates for the general minimizers.

A routine similar to nag_opt_lsq_covariance, which can be called after nag_opt_nlp, nag_opt_bounds_no_deriv, nag_opt_bounds_deriv e.t.c. A user-defined scale coefficient will adjust the errors according to the objective function type (chisquare, log-likelihood, ...) and the confidence level requested.
  MUR 2. Bound constrained nonlinear least squares. Modified nag_opt_lsq_no_deriv and nag_opt_lsq_deriv which will allow for setting lower and upper bounds for selected parameters, including a possibility to fix a subset of parameters.
  MUR 3. MINOS-type errors for a selected parameter. See UR 4.5 and Section 2.1 in Minuit++ URD.
  MUR 4. MINOS-type confidence regions for a selected pair of parameters. For a given confidence level and objective function type (chisquare, log-likelihood, ...), the user shall define an UP parameter and this routine shall return a user defined number of consecutive points from the region's boundary (c.f. UR 4.4 and Section 2.1 in Minuit++ URD).
  MUR 5. Elliptical confidence regions for a selected pair of parameters. These are standard confidence regions based on inverted Hessian. For a given confidence level and objective function type (chisquare, log-likelihood, ...), the user shall define a scale coefficient for the inverted Hessian and this routine shall return a user defined number of consecutive points from the region's boundary.
  MUR 6. Function contours. For a selected pair of parameters and with the remaining parameters fixed, this routine shall return a user defined number of consecutive points from the contour corresponding to a given function value.