Mixed-Integer Nonlinear Programming: Optimization

You might also like

Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 3

Mixed-integer nonlinear programming

(Redirected from MINLP)

Mixed-integer nonlinear programming refers to optimization problems of the form: minimize with respect to subject to

where n and m are positive integers, and are convex functions of the vectors and and

is the set of integers. If , and if

and

is an affine function of the vectors

, then the problem is a convex MINLP.


Contents
[hide]

1 Theory 2 Methods 3 Software

3.1 Black box solvers

4 References

Theory
This article has been marked as incomplete and requiring attention. You can help DDWiki by expanding it.

Methods
Because nonconvex MINLPs may in general have multiple local minima, methods that ensure global solutions employ branch and bound methods that rely on convex underestimation of the objective and constraint functions. Global optima can generally only be found when valid convex underestimators can be generated automatically, such as for factorable functions. Branch and bound

Generally, local minima to MINLPs can be found using modified versions of methods for solving convex MINLPs. See convex MINLP

Software
BARON Author: Nick Sahinidis and Mohit Tawarmalani Method: Branch and reduce optimization navigator is the integration of range reduction techniques using constraint programming and duality within the branch and bound framework for global optimization of general nonconvex MINLPs. Lower bounds are provided using factorable programming for generating convex underestimators followed by polyhedral outer approximation to linearize convex lower bounds enabling use of reliable LP solvers. All functions should be differentiable; supported functions include polynomials, power function, exponential and logarithm, other forms such as trigonometric terms are not allowed; moreover, all nonlinear terms should be bounded appropriately to avoid numerical errors. Links: BARON home GAMS/BARON manual

LINDOGlobal Author: Lindo Systems, Inc. Method: Finds guaranteed globally optimal solutions to general nonconvex MINLPs using branch and cut. It supports most of the nonlinear functions plus nonsmooth (e.g. abs(.)), nondifferentabile (e.g. floor(.)) and trigonometric terms. Links: Lindo Systems, Inc. GAMS/LINDOGlobal manual

Lago Author: Ivo Nowak and Stefan Vigerske ,Humboldt-University Berlin Method: A software-package which finds guaranteed globally optimal solutions of mixed-integer all quadratic programs (MIQQP). The source could be obtained via coin-OR Links:

LaGo in coin-OR presentation slides about LaGo

AlphaECP Author: Tapio Westerlund and Toni Lastusilta, Abo Akademi University, Finland Method: A MINLP solver based on the extended cutting plane (ECP) method. global optimality is guraunteed for pseudo-convex MINLP problems. Links: AlphaECP at Prof. Westerlund homepage GAMS/AlphaECP manual

Black box solvers


LGO Author: Pintr Consulting Services, Inc. Method: The LGO solver integrates algorithms of global and local scope (branch and bound based global search, global adaptive random search, multistart global search, exactpenalty function based local search, constrained local optimization). It serves to analyze and solve complex nonlinear models, using only function values. Links: Lago LGO home GAMS/LGO manual

You might also like