6

Let $0 < c < 2/(n-1)$. For $p_1,...,p_n \in \mathbb{R}^2$, consider the property $$ |p_{k+1}-p_k| = c, \ k=1,...,n-1 \tag 1 $$ where $| \cdot |$ is the Euclidean distance. Denote $C_k$ for the unit circle centered at $p_k, \ k=1,...,n$.

$\textbf{Question}$: For what geometric configuration of $p_1,...,p_n \in \mathbb{R}^2$ satisfying $(1)$ is the area of $C_1 \cap \cdots \cap C_n$ minimized? (note $c$ is fixed)

Comment: I've been trying to prove the solution is a straight line, as it appears to lead to a solution for the following: if you are placed at a random position in a circle, which path minimizes the average time it takes for you to escape the circle? (something I really want to solve)

$\textbf{Update} \ (2/19)$: I solved the alternate question, that the area of $C_1 \cup \cdots \cup C_n$ is maximized when $p_k$ are in line. The proof is below.

Write $\mu(A)$ for the area of a set $A$. Write $B_1 = C_1 \cup \cdots \cup C_{n-1}$ and $B_2 = C_2 \cup \cdots \cup C_n$. Then

\begin{align} \mu(C_1 \cup \cdots \cup C_n) &= \mu(B_1 \cup B_2) \\ &= \mu(B_1) + \mu(B_2) - \mu(B_1 \cap B_2) \\ &= \mu(B_1)+\mu(B_2) - \mu\big((C_1 \cap C_n) \cup (C_2 \cup \cdots \cup C_{n-1})\big) \\ &\leq \mu(B_1) + \mu(B_2) - \mu(C_2 \cup \cdots \cup C_{n-1}) \tag 2 \\ &= \mu(B_1 \backslash (C_2 \cup \cdots \cup C_{n-1})) + \mu (B_2) \\ &= \mu(C_1 \backslash (C_2 \cup \cdots \cup C_{n-1})) + \mu(B_2) \\ &\leq \mu(C_1 \backslash C_2) + \mu(B_2) \tag 3 \end{align}
We will now prove by induction $\mu(C_1 \cup \cdots \cup C_n)$ is maximized when $p_k$ are in a sequential line. For case $n=3$, notice $\mu(B_1)$ and $\mu(B_2)$ only depend on $c$, therefore are constant. Let $p_1, p_2, p_3$ be in a line. One can prove with law of consines that $C_1 \cap C_3 \subset C_2$. So equality holds at $(2)$, which shows the result.

For case $n > 3$, assume case $n-1$ holds. Let $p_1,...,p_n$ be in a line. By the induction hypothesis, $\mu(B_2)$ is maximized. $\mu(C_1 \backslash C_2)$ only depends on $c$, so is constant. So $(3)$ is maximized. Also, by law of cosines one can show $C_1 \cap C_n \subset C_2 \cup \cdots \cup C_{n-1}$ and $C_1 \backslash C_2 = C_1 \backslash (C_2 \cup \cdots \cup C_n)$. So equality holds at $(2)$ and $(3)$. This proves case $n$.

  • 2
    What is the context? What have you tried? Why not just put them in a straight line? – Ross Millikan Feb 18 '24 at 05:05
  • The context is Bellman's lost in a forest problem, but a version which hasn't been worked on. I believe its a straight line. – Stephen Harrison Feb 18 '24 at 05:07
  • 1
    I suspect you can show that $ C_1 \cap C_n \subset C_i$ hence $C_1 \cap C_n = \cap C_i$, and this area is minimized when their centers are as far apart as possible, so then $p_i$ are points on a straight line. – Calvin Lin Feb 18 '24 at 06:18
  • I don't see a way to show that, but instead considered the two circles which are maximum distance apart and it appears promising, albeit crude. My other idea is the inclusion exclusion formula. – Stephen Harrison Feb 18 '24 at 22:54
  • You might try strong induction: If the cases for all $k < n$ are true, prove the result $n$ is true as well. – Paul Sinclair Feb 19 '24 at 20:45
  • Please [edit] the question to clarify what you are asking. What does $p_i$ represent? A point on the 2-dimensional plane? A point in 3 dimensions? A point in 1 dimension? What does $|\cdots|$ refer to? The absolute value? A norm? Which norm? What does "the alternate question" refer to? Some context seems missing. Please don't use "Update". Instead, revise the question so it reads well for someone who encounters it for the first time. No need to mark what has changed. Please don't put clarifications in the comments. Instead, revise the question. – D.W. Feb 20 '24 at 06:56
  • Is it okay now? – Stephen Harrison Feb 20 '24 at 07:40

1 Answers1

0

$\forall x \in \mathbb{R}^2$, let $B_1(x)$ be the closed unit ball in $\mathbb{R}^2$ with center $x$. Let $\mu$ be the Lebesgue measure.

We will outline a proof of the stronger statement (which hopefully can make you a believer, because many details are needed to make the proof rigorous):

For all paths $\gamma: [0,L] \rightarrow \mathbb{R}^2$ parameterized by arclength with $\gamma(0) = 0$ and length $L < 2$, $$ \mu\Big(\bigcap_{0 \leq s \leq L} B_1(\gamma(s))\Big) \geq \mu\Big(B_1(0) \cap B_1((L,0)) \Big). \tag 4 $$

It is easy to show with law of cosines that equality holds when $\gamma$ is a straight line.

$(4)$ solves the question asked, since one can take $L = c(n-1)$ and $\gamma$ to be a polygonal path with vertices $p_1,...,p_n$, making the LHS of $(4) \leq \mu(C_1 \cap \cdots \cap C_n)$, whereas if $p_1,...,p_n$ are put in a sequential line one has $\mu(C_1 \cap \cdots \cap C_n) = $ RHS of $(4)$.

Fix $L$ and a $\gamma$ satisfying the conditions of $(4)$. Denote $$ I(s) = \bigcap_{0 \leq s' \leq s} B_1(\gamma(s')), \ \forall s \in [0,L]. \\ $$ Note $\forall s' \in [0,s], \ B_1(\gamma(s')) \cap \partial B_1(\gamma(s))$ is an arc of $\partial B_1(\gamma(s))$. So $I(s) \cap \partial B_1(\gamma(s))$ is an intersection of arcs, hence also an arc of $\partial B_1(\gamma(s))$.

There are two main ideas:

(i): $\frac{d}{ds} \mu(I(s))$ is directly limited by the length of the arc $I(s) \cap \partial B_1(\gamma(s))$. This is explained as follows: $x \in I(s)$ gets removed from $I(s)$ as $s$ increases iff the arc $I(s) \cap \partial B_1(\gamma(s))$ is "dragged over" $x$ as $s$ increases. So a shorter arc leads to a slower decrease of area.

Similarly $\frac{d}{ds} \mu\big(B_1(0) \cap B_1((s,0))\big)$ is directly limited by the length of the arc $B_1(0) \cap \partial B_1((s,0))$.

(ii): Let $A$ be any arc of any unit circle. If $I(s) \supset A$ then (due to $I(s)$ being an intersection of closed unit balls) the smallest $I(s)$ can be is the intersection of all closed unit balls which contain $A$. i.e. $$ I(s) \supset A \implies I(s) \supset \bigcap_{\substack{x \in \mathbb{R}^2 \\ B_1(x) \supset A}} B_1(x). \tag 5 $$ It can be shown $$ \bigcap_{\substack{x \in \mathbb{R}^2 \\ B_1(x) \supset A}} B_1(x) = B_1(x_1) \cap B_1(x_2) \tag 6 $$
where $x_1$ is such that $A \subset \partial B_1(x_1)$ (recall $A$ is an arc of a unit circle) and $x_2$ is such that $\partial B_1(x_2)$ intersects $A$ at precisely the two endpoints of $A$.

Now, if for some $s \in [0,L]$, $$ \frac{d}{ds} \mu(I(s)) < \frac{d}{ds} \mu(B_1(0) \cap B_1((s,0)), \tag 7 $$ then by (i) the arc $I(s) \cap \partial B_1(\gamma(s))$ must be longer than the arc $B_1(0) \cap \partial B_1((s,0))$.

The intersection of all closed unit balls containing $B_1(0) \cap \partial B_1((s,0))$ is precisely $B_1(0) \cap B_1((s,0))$. Since $I(s) \cap \partial B_1(\gamma(s))$ is longer than $B_1(0) \cap \partial B_1((s,0))$, by $(5)$ and $(6)$ this implies $I(s)$ is a larger area than $B_1 \cap B_1((s,0))$.

In summary, $(7) \implies \mu(I(s)) > \mu\big(B_1(0) \cap B_1((s,0))\big)$. Denote $$ s_0 = \sup \big\{s \in [0,L] : \mu(I(s)) > \mu\big(B_1(0) \cap B_1((s,0))\big) \big\}. $$ So $s > s_0$ implies $(7)$ does not hold. It follows \begin{align} \mu(I(L)) & = \mu(I(s_0)) + \int_{s_0}^L \frac{d}{ds} \mu(I(s)) \ ds \\ & \geq \mu\big(B_1(0) \cap B_1((s_0,0))\big) + \int_{s_0}^L \frac{d}{ds} \mu\big(B_1(0) \cap B_1((s,0))\big) \ ds \\ &= \mu\big(B_1(0) \cap B_1((L,0)) \big) \end{align} which is $(4)$.