Is the osculating circle at a point of maximum curvature enclosed by a convex closed curve?

199 Views Asked by At

I'm trying to solve the following problem:

(1) Given a convex, closed plane curve $\alpha$ without cusps that encloses an area $A$, prove that there exists a point $t$ such that the (unsigned) curvature $k(t)$ satisfies $k(t) \geq \sqrt{\dfrac{\pi}{A}}$.

My approach thus far: assuming that $\alpha$ is differentiable enough and parametrized by arclength, we have that $k(s)=|\alpha''(s)|$ is continuous in a compact set (since the curve is closed) and thus reaches a maximum $k(s_0)=K:=\displaystyle \max_s k(s)$. If we can show that the osculating circle at $\alpha(s_0)$ lies entirely inside the area $A$, we're done since this circle has a radius $R = 1/K$ and encloses an area $A_\circ=\pi R^2 \leq A$, and then we have: $$k(s_0)=K=\sqrt{\dfrac{\pi}{A_\circ}} \geq \sqrt{\dfrac{\pi}{A}}.$$

Therefore, problem (1) boils down to the following:

(2) Given a convex, closed plane curve $\alpha$ without cusps that encloses an area $A$ and reaches its maximum curvature at a point $P$, does the osculating circle at $P$ lie completely within $A$?

This intuitively feels true since $k$ determines the curve up to isometries, so in points/intervals where $k(s)=K$ the curve should be locally isometric to a section of the osculating circle, and if $k(s)<K$ the curve won't cut a tangent copy of the same circle. I also know that $k(t)=\theta'(t)$, but either way I can't seem to articulate any of this to produce a proof.

Two related questions I found on this site are:

which are sort of "dual" to what I want to prove, but a similar approach (e.g. involving the total curvature formula and the isoperimetric inequality) doesn't seem to work directly here.

How can I prove (2), if it is true? Is my strategy for (1) good or does it follow from a simpler statement?


Note that (1) is not necessarily true if we allow the plane curve $\alpha$ to have cusps: a convex polygon has constant curvature $0$ and encloses a non-trivial area, and a semicircle of radius $R$ has points of curvature $k(t)=0$ or $k(t)=1/R$ and encloses an area $A=\pi R^2/2$, and thus in this case we have $k(t)<\sqrt{\dfrac{\pi}{A}}$ for all $t$. It doesn't even suffice to assume non-zero curvature: imagine a convex polygon where the edges are replaced by circle sections of a large enough radius (i.e. small enough curvature). I imagine it's also easy to produce counterexamples if the curve isn't convex, which might be why the approach in the linked questions doesn't work here.