3 ms·
This is not true! Nearly all of the hypervolume of a high-dimensional n-ball is located near its surface. A popular way of stating this is "If you peel a high-d
by goodside 8y ago
This is not true! Nearly all of the hypervolume of a high-dimensional n-ball is located near its surface. A popular way of stating this is "If you peel a high-dimensional orange, there will be almost nothing left." The ratio of "rind" to "pulp" increases very, very quickly.
A good intuition for why this happens is that the distance from the center of an n-dimensional hypercube to any of its corners is `r * sqrt(n)`, but the distance from the center of a ball to its surface is always just `r`. So if `r` is fixed, and `n` keeps increasing, the corners get "spikier" and farther from the center. Very quickly, almost all of the hypervolume of a hypercube is located in these remote corners, and almost none of it is in the ball at the center.
See https://en.wikipedia.org/wiki/Curse_of_dimensionality https://en.wikipedia.org/wiki/Curse_of_dimensionality
- drb91 8y agoSeems to me it’s still premature optimization if you expect to use this primarily for low values of n (eg three).
- goodside 8y agoThe comment I was replying to stated that n was a "huge number", presumably bigger than 3.
- Dylan16807 8y agoI'm pretty sure you're reading it wrong. I interpret the comment as saying "it's premature to worry about (finding points in n dimensions where n is large)". This interpretation implies n is not large. Not "it's (premature to worry about finding points in n dimensions) where n is large". This interpretation implies n is large, but that you don't care.