[segner]

Another application area for switching data representations explicitly is Computational Geometry. The abstract reason for this is that the initial representations of geometric objects are highly implicit.

As an example, consider a set of points in 2D specified as a list of pairs of numbers. When you need to find the point in the set closest to a given location the easiest and fastest method is to run through the list and look for the minimal distance. However, if you repeat such a query for different locations while the set of points remains the same every time there are much faster (and complicated) ways to do it: geometric search trees and Delaunay triangulations. These make the implicit structure of the geometry of a set of points explicit. Whether this is actually worth it is a matter of understanding the application and its profiling data.

This can also be seen as 'acceleration data structures'.