4

Se cube

Given an $n$-sphere of radius $1$, we can construct the $n$-cube that exscribes it as having half-side-length $1$, and the $n$-cube that inscribes it as having a half-side-length of $1/\sqrt{n}$. Between these two $n$-cubes is one that splits the difference between them and best approximates the sphere (specifically, such that the average distance from the center to the cube over an directions given an even spherical weighting is as close as possible to $1$).

Some low-dimensional examples:

  • In one dimension: The $1$-cube with half-length $\ell$ has a mean radius of $$\bar{r} = \ell,$$ so the $1$-cube that best approximates the $1$-sphere has length $\ell=1$.

  • In two dimensions: The unit $2$-cube has an angular mean radius of $$ \bar{r} = \frac{1}{\pi/4}\int_{0}^{\pi/4} \frac{1}{\cos \theta}\, d\theta \approx 1.12,$$ so the $2$-cube that best approximates the unit two-sphere has $$\ell = \frac{1}{\bar{r}} \approx 0.89,$$ as illustrated in the diagram below.

Unit circle with inscribed, exscribed, and mean-scribed squares.

Is there a standard formula for the size of this cube? It seems like it should be possible to generalize the two-dimensional solution (at least to the point of providing an integral that can be evaluated numerically), but it’s a side problem to the work I’m actually doing, and is likely to have some traps in properly weighting the spherical integrals. I’d rather have something I can cite and avoid reinventing the wheel than to dive down this rabbit hole.

Searches haven’t come up with anyone even asking the question I’m looking at, so I’m not sure if it normally gets asked under different terminology.

RLH
  • 184
  • 1
    Interesting. Can you solve the problem for the plane? Three space? [edit] the question to show us some work. – Ethan Bolker Apr 09 '25 at 02:19
  • 1
    @EthanBolker: Added the n=1 and n=2 solutions. n=3 seems feasible, but more trig than I want to verify and debug if there's already a citable treatment of this somewhere. – RLH Apr 09 '25 at 04:19
  • It’s not hard to generalize the integral, but I’m pretty skeptical that it has a closed form solution. – Eric Apr 09 '25 at 05:10
  • If you use the $L^2$ norm/moment of inertia instead, I get that the average squared distance is $(n-1)\times 1/3+1=(n+2)/3$, so the radius of a n-cube is $\sqrt{(n+2)/3}$, so the half-side length of the closest to a unit sphere is $\sqrt{3/(n+2)}$ which isn’t that far off from your calculations - $1$ matches and 2 is $.87$. – Eric Apr 09 '25 at 05:18
  • @Eric I’d be happy with $L^2$-norming the circle/square discrepancy, especially with the concise expression you’ve provided. I think the simplicity comes from averaging over the area of the faces of the $n$-cube instead of using a spherical norm, but that’s also a trade off that I’m ok with in exchange for simplicity. Thanks! – RLH Apr 09 '25 at 05:37

0 Answers0