Next: octave_sqp, Previous: nonlin_min, Up: Scalar optimization [Index]
A Levenberg/Marquardt-like algorithm, attempting to honour constraints
throughout the course of optimization. This means that the initial
parameters must not violate constraints (to find an initial feasible set
of parameters, e.g. core Octaves sqp can be used
(
see
octave_sqp), by specifying an objective function which is
constant or which returns a norm of the distances to the initial
values). The Hessian is either supplied by the user or is approximated
by the BFGS algorithm. Core Octaves sqp performed better in some
tests with unconstrained problems.
Returned value cvg will be 2 or 3 for success and
0 or -4 for failure (
see
nonlin_min for
meaning). Returned structure outp will have the fields
niter, nobjf, and user_interaction.
Backend-specific defaults are: MaxIter: 20, fract_prec:
zeros (size (parameters)), max_fract_change: Inf
for all parameters. The setting TolX is not honoured.
Interpretation of Display: if set to "iter", currently
only information on applying max_fract_change is printed.