1
$\begingroup$

Let $S$ be a set of $n$ points in the plane in general position (no 3 on a line), $n$ even.

A halving line is a line through $2$ points of $S$ that partitions $S$ into 2 equal parts ($(n-2)/2$ points on each side).

A halving edge is the restriction of a halving line to the segment between the two points it connects.

A half-set of $S$ is a subset of the form $T=S∩H$ for some halfspace $H,$ such that $|T|=n/2.$ See this image.

Halving edges are more convenient to work with than halving lines, as we can see in figure.

Let $e=pq$ be a halving edge, with $p$ left of $q.$

Define next⁡(e) as follows: Let $ℓ$ be a line through $e.$ Rotate $ℓ$ CW around $q$ until we hit the next halving edge of $q$. If $ℓ$ becomes vertical before we hit any halving edge, next⁡(e) is undefined. We see the interpretation where next⁡(e'′) is undefined.

Defining prev⁡(e): Rotate $ℓ$ CCW around $p$. See also this.

Given a halving edge $e$, the convex chain of $e$ is obtained by repeatedly applying next⁡() and prev⁡() as much as possible. For visualization see here.

My question is how to Prove that every point in the left half starts exactly one convex chain, and every point in the right half ends exactly one convex chain? And is it possible to conclude that the halving edges are partitioned into exactly $n/2$ convex chains?

$\endgroup$
14
  • 2
    $\begingroup$ Are vertical edges allowed in convex chains? Or does "general position" also include that there is no vertical line through points in $S$? $\endgroup$ Commented Jul 26, 2024 at 16:55
  • 3
    $\begingroup$ @PietroMajer vertical lines maybe not allowed in convex chains, I never seen about vertical lines and convex chains in any research paper. Also S doesn't have no vertical lines, and no two points of S have same x-coordinate. $\endgroup$ Commented Jul 26, 2024 at 17:16
  • $\begingroup$ I'll post an answer as soon as I'm back from wilderness :) I also wonder what happens if one looks for closed convex chains, that is, allowing next(e) to cross the vertical and turn back. $\endgroup$ Commented Jul 30, 2024 at 12:44
  • 3
    $\begingroup$ @domotorp please tell me the any resource of discrete geometry, and how can you say it's homework, after long researching in internet, I found these information mentioned in the question. Please tell good resource of convex chain? $\endgroup$ Commented Jul 31, 2024 at 6:46
  • 2
    $\begingroup$ @domotorp I have collected these information only, I don't have much sufficient background information. If you have, please provide.i have collected from this article link.springer.com/article/10.1007/PL00009490 $\endgroup$ Commented Jul 31, 2024 at 18:57

1 Answer 1

3
+100
$\begingroup$

Adopting complex notations, let's consider the open half-planes $H_0:=\{z\in \mathbb C: \Re z>0 \}$ and $H_\theta:=e^{i\theta}H_0$ for $\theta\in[0, 2\pi]$. For $p\in S$, we are interested in the cardinality $\nu(p,\theta):=\big| (p+H_\theta) \cap S\big|.$ When $\theta$ varies from $0$ to $2\pi$, this cardinality jumps by $+1$ or $-1$ each time the boundary line $\partial(p+H_\theta)=p+ ie^{i\theta}\mathbb R$ hits a point of $S$ lying on the left, resp. on the right to $p$. Let $\{q_1,\dots,q_m\}$ be the points of $S$ on the left to $p$, listed in counterclockwise cyclic order, that is corresponding to angles $0<\theta_1<\dots<\theta_m<\pi$ (meaning $q_k\in \partial(p+H_{\theta_k})$. Thus for $\theta\in[\theta_k,\theta_{k+1}]$ we have $\nu(p,\theta) \le\nu(p,\theta_k)+1$. To exploit the above property, it is convenient to state the following Lemma (a discrete version of the Intermediate Value Theorem; see below for the simple proof).

Lemma. Let $f$ be an integer valued function defined on an interval of integers $[a,b]\subset\mathbb Z$ such that $f(a)\le f(b)$ and $f(k+1)\le f(k)+1$ for every $a\le k<b$. Then $f([a,b])$ is an interval of integers.

As a consequence one has the following facts:

1. Every point $p$ on the right half of $S$ ends at least one convex chain;

2. Two convex chains share no edge;

3. Every point $p$ on the left half of $S$ ends no convex chain;

that is, the convex chains are a partition of the halving edges, and are exactly indexed by their ending points, which are exactly the $n/2$ points of the right half of $S$.

The arguments to prove these facts are similar. Here is the proof of the first one (I'll write down the others too, if needed).

Proof of Claim 1. Let $p$ be on the right half of $S$, so $\nu(p,\theta_1)\le \nu(p,0)\le\frac {n-2}2 $ and $\nu(p,\theta_m)\ge \nu(p,\pi)-1\ge\frac{n-2}2 $. By the Lemma this implies that for at least one index $1\le k\le m$ one has $\nu(p,\theta_k)=\frac {n-2}2 $, that is, $q_kp$ is halving. Let $k_*$ be the minimum of these indices, so $\nu(p,\theta_{k_*})=\frac {n-2}2 $ and again by the Lemma, $\nu(p,\theta_k)<\nu(p,\theta_{k_*})$ for all $1\le k<k_*$. This implies that in fact no line $\partial(p+H_\theta)$ can be halving for $0\le \theta\le \theta_{k_*}$, that is $q_{k_*}p$ has no next.

Proof of the Lemma: Given an integer $ f(a)\le v\le f(b)$, let $u$ be the maximum of the non-empty set $\{k\in[a,b]: f(k)\le v\}$. Then $u<b$ and $v<f(u+1)\le f(u)+1\le v+1$ whence $f(u)=v$.

$\endgroup$
6
  • $\begingroup$ I really don't understand why you answered after the discussion above. This enables people to misuse the site, in particular this user D. S. is particularly suspicious. $\endgroup$ Commented Aug 4, 2024 at 4:29
  • $\begingroup$ I do believe there should be online references in the Internet with a complete answer as you claimed, and I'd liked better to see one than writing it. However, nobody furnished a reference, so I wrote one, as kindly required by the OP. As to suspects and unproven claims, I think we should better avoid them :) $\endgroup$ Commented Aug 4, 2024 at 7:36
  • 1
    $\begingroup$ Nobody wrote a reference because the OP is hiding information, as I pointed out in my above comments. For your sake, let me provide one so that you can see where this all comes from: jeffe.cs.illinois.edu/open/half.html This is standard textbook material that we teach and not about research level math, so it is not on scope here and should not be answered. $\endgroup$ Commented Aug 4, 2024 at 8:30
  • 2
    $\begingroup$ Thank you! Yet I think it is debeatable if the OP was really hiding information. Also note that (even if this is possibly not the case) it may happen that, doing research in a field X one needs or becomes interested in basic material, yet not easily accessible, from a far field Y. The opportunity to ask experts from different areas is a great feature of MO. $\endgroup$ Commented Aug 4, 2024 at 9:37
  • 3
    $\begingroup$ If you think there is a general problem, you should really better post a meta. A generic complaint in comments may even result in unfair harms to questioners and answerers $\endgroup$ Commented Aug 5, 2024 at 12:17

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.