Kaj Madsen
Head of Department
DTU Informatics
Professor, dr. techn.
km@imm.dtu.dk
Tel.: +45 4525 3370
Fax.: +45 4593 0360

• Front page
• Curriculum Vitae
• Publications since 1990
• Publications before 1990
• Primary fields of research
  • Non-linear Optimization and Space Mapping
  • Package for Optimization
  • Interval Analysis

Last modified Monday, June 16, 2008

MATLAB toolbox for Space Mapping optimization

The space mapping principle is an exiting new way to solve problems with very costly function evaluations - trying to find the connection (a "space mapping") between two different models of the same problem: An easy model, and a more accurate, expensive model.

The problems to be solved by the optimization algorithms in this toolbox have two models available: One model denoted the fine model, being the model of primary interest, and the other denoted the coarse model. The fine model is often expensive to evaluate, though this is not always the case with the simple test problems in this toolbox. It is expected that the coarse model somehow resembles the behaviour of the fine model. Further, it is expected that the coarse model is cheaper to evaluate than the fine model, and therefore it is most likely less accurate than the fine model.

Availability

The package is available from here.

Robust subroutines for non-linear optimization

Package description

A software package for optimization is available, consisting of easy-to-use Fortran subroutines for solving unconstrained and constrained non-linear optimization problems. The intention is that the routines should use the currently best algorithms available. All routines have standardized calls, and the user does not have to worry about special parameters controlling the iteration. For convenience we include an option for numerical checking of the user's implementation of the gradient.

• Contact Hans Bruun Nielsen to get access to the package:
  
  
• The package is freely available for employees and students at DTU.
• For other universities and non-commercial research there is a fee of $500 to access the full package.
• For commecial use a special contract must be made.

Short descriptions of the individual subroutines

Unconstrained optimization

• MINF minimization of a scalar function
   
• MINL2 minimization of the L2-norm of a vector function (least squares)
   
• MINL1 minimization of the L1-norm of a vector function
   
• MININF minimization of the infinity-norm of a vector function
   

Constrained optimization

• MINCF generally constrained minimization of a scalar function
   
• MINCL1 linearly constrained minimization of the L1-norm of a vector function
   
• MINCIN linearly constrained minimax optimization of a vector function
   

Manual