Subject: ANNOUNCEMENT: The TOMLAB 1.0 Optimization Environment in Matlab From: Kenneth =?iso-8859-1?Q?Holmstr=F6m?= Date: Sat, 12 Dec 1998 01:28:24 -0800 Newsgroups: sci.op-research,sci.math.num-analysis,comp.soft-sys.matlab,sci.stat.math ANNOUNCEMENT: The TOMLAB 1.0 Optimization Environment in Matlab The first official release of the Matlab 5 based optimization environment TOMLAB is available at http://www.ima.mdh.se/tom, the home page of the Applied Optimization and Modeling group (TOM) at Malardalen University, Vasteras, Sweden. A 160 page User's Guide in postscript format is available. TOMLAB 1.0 is free for academic use. Much effort has been put on advanced design utilizing the power and features of Matlab. The design principle is: *** Define your problem once, optimize using any suitable solver. *** This is possible using general gateway and interface routines, global variables, string evaluations and structure arrays. A stack strategy for the global variables makes recursive calls possible, in e.g. separable nonlinear least squares algorithms. One structure holds all information about the problem and one holds the results. TOMLAB should be seen as a proposal for a standard for optimization in Matlab. TOMLAB offers about 65 numerically robust algorithms for linear, discrete, nonlinear, global optimization and constrained nonlinear parameter estimation. It includes an advanced graphical user interface (GUI), menus, graphics and a lot of predefined test problems. New user-defined problems are easily added. TOMLAB has interfaces to C, Fortran, MathWorks Optimization TB, CUTE and AMPL. Automatic differentiation is easy using an interface to ADMAT/ADMIT TB and four types of numerical differentiation are included. Currently, MEX-file interfaces have been developed for the commercial solvers MINOS, NPSOL, NPOPT, NLSSOL and QPOPT. A problem is solved either by direct call to a solver or a general multi-solver driver routine, or interactively, using the graphical user interface (GUI) or a menu system. TOMLAB is very easy to use for the occasional user and student, directly giving access to large set of solvers and algorithms. For the optimization algorithm developer and the applied researcher in need of optimization tools it is very easy to compare different solvers or do test runs on thousands of problems. A tool is provided to automate the tedious work of making computational result tables from test runs, directly making LaTeX tables for inclusion in papers. The TOMLAB solvers all explicitly handle bounds and linear constraints, with an input model upper/lower bound format like NPSOL. Implemented solver algorithms for general NLP problem are SQP type algorithms like Schittkowski SQP, Fletcher-Leyffer Filter SQP and Han-Powell SQP. A structural trust region algorithm for partially separable functions on convex sets (Conn et.al) is also implemented. For nonlinear least squares, Gauss-Newton with subspace minimization, Fletcher-Xu, Al-Baali-Fletcher and Huschens TSSM method are implemented, together with an active set strategy to handle bounds and linear constraints. The most common unconstrained algorithms are implemented: Newton algorithms, and several Quasi-Newton and conjugate gradient methods. Global optimization problems without derivatives are solved using the DIRECT and EGO algorithms (Jones et.al). Quadratic programming problems are solved with a standard active set method, using eigenvalue information for indefinite problems. For linear programming, different types of simplex algorithms are implemented as well as the dual simplex algorithm. A branch and bound and a cutting plane algorithm are solving MIP problems. Solvers for different types of network programs and dynamic programs are available. Happy computing with TOMLAB! (Feedback is welcome) Kenneth Holmstrom