Skip to main content

The Granularity Concept in Mixed-Integer Optimization

Abstract

Granularity of optimization problems relaxes the difficulties imposed by integrality conditions and often
provides ways for determining good feasible points of mixed-integer optimization problems at low computational cost.
It thus provides so-called primal heuristics while, in fact, it is not based on heuristic ideas, but on transparent
geometric considerations.

Starting from error bound results for roundings in mixed-integer linear optimization, we illustrate
how the granularity concept unfolds to provide algorithms for the computation of feasible points in mixed-integer
linear, convex and nonconvex optimization. We also comment on the treatment of equality constraints and
explain the integration of the granularity idea into branch-and-bound frameworks.

Time: Thu 2022-06-09 11.00 - 12.00

Location: Seminar room 3721

Video link: Zoom ID 63658381373

Language: English

Lecturer: Oliver Stein, Karlsruhe Institute of Technology (KIT), stein@kit.edu