QL solves quadratic programming problems with a positive definite
objective function matrix and linear equality and inequality
constraints.
Numerical Method:
The algorithm is an implementation of the dual method of Goldfarb
and Idnani and a modification of the original implementation of
Powell. Initially, the algorithm computes a solution of the
unconstrained problem by performing a Cholesky decomposition and
by solving the triangular system. In an iterative way, violated constraints
are added to a working set and a minimum with respect to the
new subsystem with one additional constraint is calculated.
Whenever necessary, a constraint is dropped from the working set.
The internal matrix transformations are performed in numerically
stable way.
Program Organization:
QL is a FORTRAN subroutine where all data are
passed by subroutine arguments.
Special Features:
separate handling of upper and lower bounds
initially given Cholesky decomposition exploited
full documentation
FORTRAN source code (close to F77, conversion to C by f2c
possible)
Applications:
As an essential part of the nonlinear programming routine
NLPQLP,
QL solves the internal quadratic programming subproblem of
the SQP-method and has therefore the same domain of application
as NLPQLP.
Reference:
M.J.D. Powell, On the quadratic programming algorithm of Goldfarb and
Idnani,
Report DAMTP 1983/Na 19, University of Cambridge, Cambridge (1983)