SDLS: a Matlab package for solving conic least-squares problems

Didier Henrion, Jérôme Malick.

SDLS is a Matlab freeware allowing to solve approximately convex conic least-squares problems. Geometrically, these problems amount to finding the projection of a point onto the intersection of a symmetric convex cone with an affine subspace. SDLS solves the dual problem with a quasi-Newton minimization algorithm, using an implementation of the BFGS algorithm. The other key numerical component is eigenvalue decomposition for symmetric matrices, achieved by Matlab's built-in linear algebra functions. Note that SDLS may not be the most competitive implementation of this algorithm. Our first goal is to provide a simple, user-friendly software for solving and experimenting with general conic least-squares. Up to our knowledge, no such freeware existed when releasing the first version of SDLS.

Version history:

  • SDLS 1.0 - 27 June 2007;
  • SDLS 1.1 - 26 May 2008 - data scaling and improved stopping criterion;
  • SDLS 1.2 - 28 January 2009 - improved stopping criterion, detection of marginal feasibility and infeasibility with approximate Farkas certificate;

    SDLS is developed for Matlab 7 and above, and distributed freely as a tar.gz archive, for research and academic purposes. Please contact the authors. The software is described in this documentation.

    SDLS uses the BFGS algorithm of the HANSO package, which must be installed.

    Last modified on 13 February 2009.