6
$\begingroup$

A polytope in $\mathbb R^n$ is the convex hull of a nonempty finite set in $\mathbb R^n$.

Let $C$ be a polytope in $\mathbb R^n$. We shall say that a hyperplane $H\subseteq \mathbb R^n$

$\bullet$ weakly supports $C$ if the set $C\setminus H$ is connected;

$\bullet$ strongly supports $C$ if $C\setminus H$ is connected and $\dim(C\cap H)=\dim(C)-1$.

Problem. Let $A,B$ be two disjoint polytopes in $\mathbb R^n$. Is there a hyperplane $H$ in $\mathsf R^n$ such that

  1. $H$ weakly supports both $A$ and $B$;
  2. $H$ strongly supports $A$ or $B$;
  3. $H$ separates $A$ and $B$ in the sense that for any points $a\in A$ and $b\in B$ the segment $[a,b]$ intersects the hyperplane $H$?
$\endgroup$
11
  • $\begingroup$ The convex hull of a finite set of points is usually called a "Polytope". A "Polyhedron" is usually defined as the intersection of finitely many half-spaces. $\endgroup$
    – pinaki
    Commented Jul 2, 2023 at 12:50
  • $\begingroup$ Also do you consider empty intersection as weakly supporting? $\endgroup$
    – pinaki
    Commented Jul 2, 2023 at 12:52
  • 1
    $\begingroup$ @pinaki you mean these two notions do not coincide? $\endgroup$ Commented Jul 2, 2023 at 13:35
  • 2
    $\begingroup$ @PietroMajer a single half-space would count as a "polyhedron" but not a "polytope" according to pinaki's definitions. I don't know whether those are the standard definitions, though. $\endgroup$
    – Nik Weaver
    Commented Jul 2, 2023 at 18:45
  • 2
    $\begingroup$ @PietroMajer Yes, this is true even in the betweeness geometry: a set is a polytope if and only if it is a bounded polyhedron. But the proof is not easy. $\endgroup$ Commented Jul 2, 2023 at 19:46

3 Answers 3

5
$\begingroup$

Edited to give a more symmetric and easily visualized solution.

Counterexample in $\mathbb{R}^3$. $A$ and $B$ will be two tetrahedrons. $A$ has vertices at $(1,-1,0)$, $(-1,-1,0)$, $(0, -1, .1)$ and $(0,-.1,0)$. $B$ has vertices at $(0, 1, 1)$, $(0,1, -1)$, $(.1, 1, 0)$ and $(0, .1, 0)$. $A$ is wide and flat, $B$ is tall and thin.

$A$ is nearly contained in the $xy$ plane and $B$ is nearly contained in the $yz$ plane. Both have an outer side which is parallel to the $xz$ plane; their hyperplanes lie to the same side of $A$ and $B$. The hyperplanes of the other three sides of $A$ are nearly parallel to the $xy$ plane and slice through the middle of $B$, while the hyperplanes of the other three sides of $B$ are nearly parallel to the $yz$ plane and slice through the middle of $A$.

$\endgroup$
3
  • $\begingroup$ Because of my weak 3D imaging skills, it is hard for me to follow your proof. If it is not too much to ask, can you present it in an algebraic/analytic form? $\endgroup$ Commented Jul 2, 2023 at 17:02
  • $\begingroup$ @IosifPinelis yeah, I had trouble visualizing it too. Maybe the simplest thing would be to use Mathematica to draw some pictures, but unfortunately I don't know how to use Mathematica. $\endgroup$
    – Nik Weaver
    Commented Jul 2, 2023 at 17:03
  • $\begingroup$ I am still unable to visualize this example, but calculations do show that your example is correct. +1 $\endgroup$ Commented Jul 2, 2023 at 18:39
3
$\begingroup$

This is very similar to Nik Weaver's answer but may be easier to visualise. Fix a small $\epsilon>0$. Let $A$ have vertices $(\pm 1,0,-1)$ and $(0,\epsilon^2,-1)$ and $(0,0,\epsilon-1)$. This means that $A$ is close to a line segment $A'$ parallel to the $x$-axis, the bottom face is horizontal, and the other three faces are close to the plane $y=0$. Let $B$ be the image of $A$ under $(x,y,z)\mapsto (y,x,-z)$, so $B$ is close to a line segment $B'$ parallel to the $y$-axis, the top face is horizontal, and the other three faces are close to the plane $x=0$. Both $A$ and $B$ lie above the horizontal face of $A$ and below the horizontal face of $B$. The plane $y=0$ cuts the line segment $B'$ at $(0,0,1)$, and similarly, any of the three nearly-vertical supporting planes of $A$ cuts $B$. By symmetry, any of the three nearly-vertical supporting planes of $B$ cuts $A$. By plotting you can check that $\epsilon=1/2$ is (just) small enough, but smaller $\epsilon$ makes the picture clearer.

Maple code is as follows:

with(plots): with(plottools):
e := 1/3;
A := [[-1,0,-1],[1,0,-1],[0,e^2,-1],[0,0,e-1]];
B := map(u -> [u[2],u[1],-u[3]],A);
xp := (u,v) -> [u[2]*v[3]-u[3]*v[2],u[3]*v[1]-u[1]*v[3],u[1]*v[2]-u[2]*v[1]]:
dp := (u,v) -> u[1]*v[1]+u[2]*v[2]+u[3]*v[3]:
dsc := proc(u)
 local u0,v0,w0;
 u0 := (u[1] +~ u[2] +~ u[3])/~3.;
 v0 := u[1] -~ u0;
 v0 := v0 /~ sqrt(dp(v0,v0));
 w0 := u[2] -~ u0;
 w0 := w0 -~ dp(w0,v0) *~ v0;
 w0 := w0 /~ sqrt(dp(w0,w0));
 plot3d(u0 +~ (r * cos(t)) *~ v0 +~ (r * sin(t)) *~ w0,r=0..2.5,t=0..2*Pi,style=wireframe,colour=gray,scaling=constrained,axes=none);
end:
display(
 polygon([A[1],A[2],A[3]]),
 polygon([A[1],A[2],A[4]]),
 polygon([A[1],A[3],A[4]]),
 polygon([A[2],A[3],A[4]]),
 polygon([B[1],B[2],B[3]]),
 polygon([B[1],B[2],B[4]]),
 polygon([B[1],B[3],B[4]]),
 polygon([B[2],B[3],B[4]]),
 dsc([A[2],A[3],A[4]]),
 scaling=constrained,axes=none
);

Here is a picture with $\epsilon=1/2$ (but it is much easier to see if you use Maple or similar to make a plot that you can rotate with your mouse). enter image description here

$\endgroup$
0
$\begingroup$

Another visualization with Maple $\ge$ 2015 is as follows.

with(PolyhedralSets):
ps1 := PolyhedralSet([[1, -1, 0], [-1, -1, 0], [0, -1, 1/10], [0, -1/10, 0]]):
ps2 := PolyhedralSet([[0, 1, 1], [0, 1, -1], [1/10, 1, 0], [0, 1/10, 0]]):
Plot([ps1, ps2], scaling = constrained);

enter image description here

$\endgroup$
1

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .