It is known from Blaschke and later Sas that any convex region $P$ with area $1$ admits an inscribed triangle with area at least $\frac{3\sqrt{3}}{4\pi}$. What if we require the triangle to have a given boundary point of P as vertex? The optimal constant should be between $\frac{1}{4}$ and $\frac{1}{3}$, but I can't tighten these bounds by too much.
-
$\begingroup$ "a given point": A given point anywhere in $P$, or a given point on the boundary of $P$? $\endgroup$– Joseph O'RourkeCommented Apr 28, 2017 at 11:33
-
$\begingroup$ Thanks, I mean a given point on the boundary. The question has been edited. $\endgroup$– Xiaosheng MuCommented Apr 28, 2017 at 18:45
-
$\begingroup$ First, your restricting to "extreme point" (in the question) instead of "boundary point" (in your comment and Joseph's) makes no difference since a small perturbation can make any boundary point an extreme point and not change the answer by more than epsilon. Second, if I calculated correctly, for the semicircle and the center of its straight side, the answer is 1/π, which improves your upper bound. $\endgroup$– Yoav KallusCommented Apr 28, 2017 at 19:55
-
1$\begingroup$ In fact, I would not be surprised if that is the optimal constant. For the region P and point x that achieve the optimal constant, it should be the case that every boundary point is in a triangle xyz that achieves the largest area. Otherwise, we can enlarge P around that point without creating a larger triangle and thus contradict the constant being optimal. $\endgroup$– Yoav KallusCommented Apr 28, 2017 at 20:13
1 Answer
The answer is indeed $1/\pi$, equality achieved in the case of $P$ a semicircle and the point its center.
For an arbitrary convex region $P$ and a point $X$ on its boundary, consider the convex hull of $P$ and $2X - P$, and call it $S$. Then any $S$ is clearly symmetric with respect to $X$. Any triangle in $P$ having $X$ as a vertex with area $a$, induces a parallelogram in $S$ with center $X$ and area $4a$. Also we have the trivial inequality $\operatorname{area}(S) \ge 2 \operatorname{area}(P)$. So it suffices to show that, given a centrally symmetric convex region $S$ with respect to $O$, there exists a parallelogram in $S$ with center $O$ and area at least $(2/\pi) \operatorname{area}(S)$.
This can be proved by essentially the same technique used by Sas. After scaling and rotation, let $S$ be contained in the unit disc, with $(1, 0), (-1, 0) \in S$. Parametrize the boundary of $S$ by $\pm (\cos \theta, f(\theta) \sin \theta)$, where $f : (0, \pi) \to [0, 1]$. Then the area of $S$ is $$ \operatorname{area}(S) = 2 \int_{0}^{\pi} f(\theta) \sin^2 \theta.$$ The area of the parallelogram with vertices $\pm (\cos \theta, f(\theta) \sin \theta)$ and $\pm (-\sin \theta, f(\theta + \pi/2) \cos \theta)$ is $$\operatorname{area}(\mathrm{Par}_\theta) = 2 \Bigl( f(\theta) \sin^2 \theta + f\Bigl( \theta + \frac{\pi}{2} \Bigr) \sin^2 \Bigl( \theta + \frac{\pi}{2} \Bigr) \Bigr).$$ Averaging this area over $0 \le \theta \le \pi / 2$ shows that there exists some $\theta$ such that $$\operatorname{area}(\mathrm{Par}_\theta) \ge \frac{2}{\pi} \operatorname{area}(S). $$
-
$\begingroup$ "any triangle in $P$ having $X$ as a vertex with area $a$, induces a parallelogram in $S$ with center $X$ and area $4a$": but the converse is not true I think. That is, a parallelogram in $S$ does not necessarily correspond to a triangle IN $P$ $\endgroup$ Commented Apr 29, 2017 at 4:01
-
1$\begingroup$ But the one with maximal area is indeed supported on $P$ and $2X-P$, so your argument does work :) $\endgroup$ Commented Apr 29, 2017 at 4:06