 |
|
|
|
|
 |
 |
|
Testing the "No Free Lunch Theorem" for Circle and Sphere Packings with MathOptimizer Professional
|
|
|
|
|
|
| Organization: | Physicist at Large Consulting |
|
|
|
|
|
|
2007 Wolfram Technology Conference
|
|
|
|
|
|
Champaign, IL
|
|
|
|
|
|
Abstract
The "No Free Lunch Theorem" in global optimization (see wikipedia.org/wiki/No_free_lunch_in_search_and_optimization) states that roughly the same number of function evaluations must be performed to find a solution of a given (guaranteed) quality, no matter what approach is taken. We test this idea for circle and sphere packing problems by performing an initial optimization and then using local search techniques to improve the initial solution. These local search techniques involve swapping the position of the circles or spheres in order to improve the objective function, the radius of an enclosing circle or sphere. The different variations studied include swapping all pairs of adjacent sized circles or spheres, swapping all pairs, swapping adjacent sized pairs starting from the smallest pair, and performing the previously described swapping techniques either once or repeatedly until the objective function stops improving. The various techniques are themselves performed multiple times and the best objective function result is plotted versus time. The initial optimization and local searches are performed using MathOptimizer Professional, taking advantage of its ability to reuse machine-compiled code for function evaluation. The new graphics features in Mathematica 6, especially for three-dimensional plots, are employed to display our results.
|
|
|
|
|
|
|
|
|
|
|
|
http://www.wolfram.com/news/events/techconf2007/
|
|
|
 |
|
|
| TestingTheNoFreeLunchTheorem.nb (992.2 KB) - Mathematica Notebook [for Mathematica 6.0] |
|
|
For the newest resources, visit Wolfram Repositories and Archives »
