I'm interested in the question of finding the maximum area of $A\subset S^{d-1}$, such that, for all $x,y \in A, \left<x,y\right>\ge 0$. The portion of the sphere lying in the positive orthant seems like a reasonable guess. Can any subset of the sphere that satisfies the positive inner product property be rotated to fit in the positive orthant? I seem to have some evidence that the answer to this last question is "no" but I would like to know for certain.
-
3$\begingroup$ What's interesting is that asking for the biggest "opposite" kind of set, a set $A$ with $\langle x, y \rangle < 0$ for all $x\neq y \in A$, has a well-known and very clean answer: take the vertices of a regular simplex; see mathoverflow.net/questions/31436. $\endgroup$– Sam HopkinsCommented Jan 12 at 2:07
1 Answer
You can go slightly beyond the positive orthant. In the case of a 2-sphere in 3-dimensional space, you can render a cap with angular radius $45°$. This has a surface area of $2\pi(1-\sqrt{1/2})\approx(2\pi)\color{blue}{(0.293)}$ steradians, whereas the positive orthant has a surface area of $\pi/2=(2\pi)\color{blue}{(0.25)}$ steradians.
Although the cap is larger in area, the positive orthant cannot fit inside it. This is essentially a spherical analogue of the planar geometry result that a circle of unit diameter is larger than an equilateral triangle of unit side (or, for that matter, any regular odd-sided polygon whose longest chord measures one unit), but the triangle/odd-gon does not fit in the circle.
The counterpart to the cap also beats the positive orthant in spatial dimensions greater than 3; in fact the ratio between the two eventually becomes arbitrarily large with sufficiently many dimensions.
Let a $(d-1)$-sphere in $d$-dimensional space be defined by
$\sum\limits_{i=1}^d x_i^2=1$
The positive orthant is then given by $\min(x_i)\ge0$, while the cap may be defined by $x_1\ge\sqrt{1/2}$. In the latter case two antipodal points $(\sqrt{1/2},x_2,x_3,...x_d)$ and $(\sqrt{1/2},-x_2,-x_3,...-x_d)$ on the boubdary of the cap give the inner product
$1/2-\sum\limits_{i=2}^d x_i^2=0,$
where the equation for the sphere combined with defining $x_1=\sqrt{1/2}$ imply the second equality.
The fraction of the sphere covered by the positive orthant is then $2^{-d}$, while the fraction covered by the cap is given by a ratio of integrals:
$\dfrac{\int_{\sqrt{1/2}}^1(1-x^2)^{(3-d)/2}dx}{\int_{-1}^1(1-x^2)^{(3-d)/2}dx}.$
With $d=3$ this gives (to four decimal places) $0.1464$ for the cap versus $0.1250$ for the orthant. With four-dimensional space the corresponding fractions are $0.1051$ for the cap and $0.0625$ for the orthant. As $d\to\infty$, the integrals in the fraction for the cap become Laplace integrals and can be evaluated asymptotically by the appropriate method for such integrals; the result is that the fraction covered by the cap has the controlling factor $(\sqrt2)^{-d}$ versus $2^{-d}$ for the positive orthant. As $\sqrt2<2$, the cap ultimately becomes exponentially larger than the positive orthant.
-
1$\begingroup$ This is interesting. Do you have any idea about what happens in general dimension? $\endgroup$ Commented Jan 12 at 0:24
-
2$\begingroup$ @SamHopkins One can still define a cap with angular radius 45° in a sphere of any dimension. The proof that <x,y> >=0 should be the same as in 3-d. Computing the volume of this cap is a little more complicated. Intuition on whether it remains bigger than the orthogonal cap is tricky. Proving optimality seems hard and I have no idea whether it is even true. $\endgroup$– quaragueCommented Jan 12 at 7:28
-
3$\begingroup$ This is great, thanks! In case anyone is interested, I wrote a program that randomly sampled points from the from 2-sphere and added the points to a collection if its inner product with all points in the collection was positive. So it in some ways generated a random collection with this inner product condition. Its geomtery seemed to vary quite a bit. So I guess there are a lot of different maximal sets, as in you cannot add more volume without violating the inner product property, ignoring congruence via rotation. $\endgroup$ Commented Jan 12 at 11:46
-
2$\begingroup$ There are infitely many maximal sets. In particular, you can draw an equilateral triangle, regular pentagon, regular heptagon etc with longest chord = 90°; they are all maximal for their respective polygonal shapes and none fit inside each other. Nor do they fit inside the spherical cap described above, which beats all the polygons in area. $\endgroup$ Commented Jan 12 at 13:25
-
3$\begingroup$ @OscarLanzi: Do you claim that the cap is the set with the largest area? I guess that it is. Further, it would be interesting to look at all sets which are maximal under inclusion and then find the one with the minimal area. I guess that it is the "triangle" (positive orthant). $\endgroup$– gerwCommented Jan 12 at 14:04