5
$\begingroup$

Let $P$ be an opaque convex polyhedron containing the origin in $\mathbb{R}^3$, and let $S$ be an origin-centered sphere strictly containing $P$: $S \supset P$. For a point $x$ on $S$, let $\sigma(x)$ be the area of the shadow of $P$ cast from a light at $x$ onto the plane tangent to $S$ at $-x$:


          enter image description here
My question is:

Q. What is the differentiability class $C^k$ of $\sigma(x): S \to \mathbb{R}$?

I would be surprised if $\sigma$ is a smooth map, $C^\infty$, but it seems to be at least $C^1$...

The question makes sense for a convex polytope in $\mathbb{R}^d$ for $d \ge 2$, with $\sigma(x)$ the $(d{-}1)$-volume of the shadow cast on a $(d{-}1)$ hyperplane.

$\endgroup$
2
  • $\begingroup$ For $d = 2$ and $P$ a regular polygon, $\sigma$ can be computed explicitly in principle. Do you have that computation done? $\endgroup$ Jun 9, 2015 at 12:44
  • $\begingroup$ @WillieWong: Good point. No, I have not performed that computation. I can see it would be worthwhile. (Cannot attend now.) $\endgroup$ Jun 9, 2015 at 12:46

2 Answers 2

4
$\begingroup$

Let me try to give a computation free (sketch of) proof that the shadow area is not $C^1$. The basic idea is the same as in the answer of Willie Wong: a problem happen when a corner shows up.

Consider a position $x_0$ where some corner $c$ of the polytope is projected right on a facet of the shadow. Then move $x_0$ along a smooth path $x_t$ such that for small negative $t$, $c$ is projected in the interior of the shadow, and for small positive $t$ it is projected "outside the shadow", i.e. it is projected to a vertex of the shadow. Then the area $\sigma(x_t)$ is given by the sum of a smooth function (corresponding to the shadow of the facets already contributing to the shadow at $x_0$) and a function which is zero for $t<0$ (when the facets of $c$ do not contribute to the shadow) and is linear for $t>0$ (when the facets of $c$ start contributing to the shadow). Thus $\sigma$ is not $C^1$ (but it is certainly Lipschitz).

$\endgroup$
2
  • $\begingroup$ Credit to Willie for seeing it first, but this is an illuminating "computation-free" answer! $\endgroup$ Jun 10, 2015 at 1:36
  • $\begingroup$ In fact, the same idea appears when one tries to make a Riemannian metric out of a Finsler metric by using the John ellipsoid, which is something I came across a few years ago. $\endgroup$ Jun 10, 2015 at 8:47
3
$\begingroup$

Just doing this very quickly, so maybe computational errors:

In the case $d = 2$, take $P$ as an equilateral triangle centered at the origin inscribed in the unit circle, and $S$ the circle of radius 2, the problem is equivalent to computing the order of tangency of the two functions $$f_1(\theta) = \frac{\sin\theta}{2 - \cos \theta} \quad \text{ and } f_2(\theta) = \frac{\sin(\pi/3 - \theta)}{2 + \cos(\pi/3 -\theta)}$$ at their first positive intersection point.


  WilliePlot
  (Image added by J.O'Rourke. They cross at $\theta \approx 0.2709$.)


If I did the computations right, this is equivalent to asking for the order of vanishing of the function

$$ 6 \sin \theta + \sqrt{3} - 2 \sqrt{3} \cos\theta $$

at its first positive zero. But using the monotonicity of the $\sin$ and $\cos$ functions it appears that the derivative is strictly positive at the first zero, so that $\sigma$ is not even $C^1$. This suggests that in the $d = 2$ case the answer is Lipschitz + piecewise smooth.

$\endgroup$
0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.