4
$\begingroup$

I am looking for a reference (name, source) of the following elementary geometric combinatorial problem:

Inside a unit square, given a smooth curve of length $L$. Then there exists a horizontal or vertical line that intersects the given curve at at least $L/2$ points.

Also, I am looking for a higher dimensional version of this result. Thanks a lot!

$\endgroup$
2
  • $\begingroup$ The energy of the path decomposes by the Pythagorean theorem as the sum of the energies of the projections to the two axes. The projection with the larger energy (at least half the energy of the original path) will provide the desired line by the pigeonhole principle. Similarly for paths in 3-space where the constant will be 3 instead of 2. $\endgroup$ Commented Sep 6, 2023 at 13:13
  • $\begingroup$ Consider the smooth curve of length $0$ defined by $\gamma(t) = (1/2, 1/2)$ for $t \in [0, 1]$. How many intersection points with $\{1/2\} \times [0, 1]$ should we count? Infinitely many or just one? Related to this: is your curve assumed to be regular? Is the Lebesgue measure of the set of critical points of the curve assumed to be zero? $\endgroup$
    – Luc Guyot
    Commented Sep 11, 2023 at 11:19

2 Answers 2

4
$\begingroup$

Let $\varphi: [0,L] \to [0,2\pi)$ give the direction of the curve.
Then the number of intersected horizontal lines (with multiplicities) is given by $\int_0^L |\sin\varphi(t)| dt$.
Similarly, the number of intersected vertical lines (with multiplicities) is $\int_0^L |\cos\varphi(t)| dt$.
Since $|\sin x| + |\cos x|\ge 1$ for every $x$, the result follows.

$\endgroup$
2
  • 1
    $\begingroup$ Did you assume that the curve is regular, or at least, that the Lebesgue measure of its critical points is zero? If not, how would you re-parametrise the curve according to arc length? $\endgroup$
    – Luc Guyot
    Commented Sep 11, 2023 at 11:27
  • $\begingroup$ @Luc: Good point. $\endgroup$
    – domotorp
    Commented Sep 11, 2023 at 19:44
1
$\begingroup$

The following claim and its proof answer OP's question, but only under an additional hypothesis on the critical points of the coordinate functions $\gamma_i$ of the curve $\gamma = (\gamma_1, \dots, \gamma_n)$.

Claim. Let $n \ge 1$ and let $\gamma = (\gamma_1, \dots, \gamma_n): [0, 1] \rightarrow [0, 1]^n$ be a curve of class $C^1$. Assume moreover that $\gamma_i$ has at most countably many critical points for every $i \in \{1, \dots, n\}$. Then there is $j \in \{1, \dots, n\}$ and $c \in [0, 1]$ such that $$\vert \{ t \in [0, 1] \, \vert \, \gamma_j(t) = c \} \vert \ge L(\gamma) /n$$ where $L(\gamma)$ is the arc length of $\gamma$.

I'll be grateful to contributors willing to answer the following question:

Question. Is it possible to remove, or at least weaken, the extra hypothesis on critical points?

The proof of the above claim relies on the next two lemmas.

Lemma 1. Let $\gamma = (\gamma_1, \dots, \gamma_n): [0, 1] \rightarrow [0, 1]^n$ be a rectifiable curve. Then $\gamma_i$ is rectifiable for every $i$ and there is $j \in \{1, \dots, n\}$ such that $L(\gamma_j) \ge L(\gamma) / n$.

Proof. It follows easily from the definition of the arc length of a curve, and from the inequalities $\vert x_i \vert \le \sqrt{x_1^2 + \cdots + x_n^2} \le \sum_{j = 1}^n \vert x_j \vert$ that $$L(\gamma_i) \le L(\gamma) \le \sum_{j = 1}^n L(\gamma_j)$$ for every $i \in \{1, \dots, n\}$. Thus there is $j \in \{1, \dots, n\}$ such that $L(\gamma_j) \ge L(\gamma)/n$.

Lemma 2. Let $\gamma: [0, 1] \rightarrow [0, 1]$ be a curve of class $C^1$. Assume moreover that the set of critical points of $\gamma$ is at most countable. Then we have $$\deg(\gamma) \ge L(\gamma)$$ where $$\deg(\gamma) := \sup_{y \in [0, 1]} \vert \{ \gamma^{-1}(\{y\})\} \vert.$$

Proof. Since $\gamma$ is of class $C^1$, it is rectifiable, i.e., its arc length $L = L(\gamma)$ is finite. Thus we can assume, without loss of generality, that $\deg(\gamma) = d < \infty$. Define $s: [0, 1] \rightarrow [0, L]$ by $s(t) = \int_0^t \vert \gamma'(t) \vert dt$. Clearly, the function $s$ is of class $C^1$. Since the set of critical points of $\gamma$ has empty interior, the function $s$ is strictly increasing and is therefore an homeomorphism. Besides, the function $s$ is differentiable at $s(t)$ if and only if $t$ is not a critical point of $\gamma$, i.e., if and only if $\gamma'(t) \neq 0$. Let $\gamma_1 = \gamma \circ s^{-1}$. Then $\gamma_1$ is a continuous curve which is locally an isometry, i.e., for every $t \in [0, 1]$ such that $\gamma'(t) \neq 0$, there is an open interval $I = I_t$ containing $s(t)$ such that the restriction of $\gamma_1$ to $I$ is an isometry onto its image. Let $O \subseteq [0, 1]$ be the complement of the critical values of $\gamma$. As $O$ is open, it is the union of at most countably many disjoint intervals with non-empty interior. Let $J$ be one of these intervals and denote by $\mathring{J}$ its interior. We claim that $\gamma_1^{-1}(\mathring{J})$ is the union of at most $d$ disjoint intervals which are all isometric to $\mathring{J}$. It follows from our claim that $[0, L] = \gamma_1^{-1}([0, 1])$ is the disjoint union of an open set of Lebesgue measure at most $d$, that is $\gamma_1^{-1}(O)$, and of a set which is at most countable. Therefore $L \le d$.

Proof of the claim. Combine Lemmas 1 and 2.

$\endgroup$

You must log in to answer this question.

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