Bounded-variable least-squares: an algorithm and applications (Q1965978)

The bounded-variable least-squares algorithm (BVLS) which solves linear least-squares problems with upper and lower bounds on the variables is described. BVLS is used also to find bounds for linear functionals of a model constrained to satisfy, in approximate \(l_p\)-norm sense, a set of linear equality constraints in addition to upper and lower bounds. It is shown how to use BVLS to solve that problem when \(p=1,2\) or \(\infty\), and to solve minimum \(l_1\) and \(l_\infty\) fitting problems. A variety of applications of BVLS is described and such features of this algorithm as numerical stability and computational efficiency are emphasized. The BVLS algorithm is implemented as a Fortran subroutine and is available from the \textit{statlib gopher} at Carnegie Mellon University.