CONTENTS:
A primal-dual interior method for solving
linearly constrained optimization problems
with a convex objective function (preferably separable):
minimize phi(x) + 1/2 ||D1*x||^2 + 1/2 ||r||^2
subject to A*x + D2*r = b and bl <= x <= bu,
where the m x n matrix A may be sparse
or represented by an M-file for computing Ax and A'x,
and D1, D2 are diagonal matrices that provide
primal and dual regularization. D2 determines whether
each row of Ax = b should be satisfied accurately
or in the least-squares sense. (Each element of D2
is typically in the range [1e-4,1].)
REFERENCES:
Documented in the MATLAB and LaTeX files below.
Uses established primal-dual technology, with choice
of direct or iterative method for computing search directions.
Special feature: Iterative (and inexact) computation
of search directions using LSQR.
RELEASE:
16 Oct 2002: First version, derived from PDSCO.
D1, D2 and general bounds implemented.
23 Sep 2003: Revised.
Maximum entropy test problems included
(joint work with John Tomlin, IBM Almaden).
11 Feb 2005: Slight changes for MATLAB 7.0.
11 Feb 2005: Removed ENTROPY.big from zip file
(too big for those who don't want it). Link is below.