Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I have the following problem:

Let $K$ and $K'$ be two convex compact sets in $\mathbb{R}^n$, with $n \geq 2$. Assume that $K$ and $K'$ both have a smooth ($\mathcal{C}^2$) boundary. Assume moreover that $K \cup K'$ is convex. Then is the boundary of $K \cup K'$ also $\mathcal{C}^2$?


Edit : I'll add some context.

The motivation of the question is that I would like to prove that curvature integrals are additive over the space of $\mathcal{C}^2$ convex compact sets, which basically boils down to this problem.

My guess is that the answer is positive. Here are some elements in two dimensions. Let $K$ and $K'$ be two convex compact sets in $\mathbb{R}^2$ with smooth boundary, and assume that $K \cup K'$ is also convex. Let $x \in \partial K \cap \partial K'$. Then $\partial K$ and $\partial K'$ have the same tangent space in $x$; otherwise, $K \cup K'$ would not be convex. Now, let us look at the curvature at $x$. If the curvature of $\partial K$ is strictly smaller than the curvature of $\partial K'$, then $K'$ is locally included into $K$, so $\partial K \cup K'$ is locally $\partial K$, which is $\mathcal{C}^2$. The same holds if we exchange $K$ and $K'$. The last possibility is that both curvatures are equal, which "trivially"* imply that the second derivative of any reasonable parametrization of $\partial K \cup K'$ in a neighborhood of $x$ is $\mathcal{C}^2$. In all cases, $\partial K \cup K'$ is a $\mathcal{C}^2$ curve.

This is sketchy, I am not really sure I can write it down neatly, let alone tackle higher dimensions...

$*$ this may be Jordan-theorem-like trivial: getting a rigorous proof does not look that easy...

share|cite|improve this question
One special case of the question seems to be: let $f$ and $g$ be $C^2$ convex functions from $\mathbb R$ to $\mathbb R$, and let $h(x) = \min\{ f(x),g(x) \}$. If $h$ is convex, then is $h$ a $C^2$ function? - Can this question be resolved? It feels equivalent to the general two-dimensional case. – Greg Martin Mar 29 '12 at 17:25
@GregMartin: yes, this is equivalent and generalizes to the n-dimensional case. The boundary of a set is a manifold of class $C^2$ iff it is locally representable as the graph of a $C^2$-function. – user20266 Mar 29 '12 at 17:39
up vote 4 down vote accepted

The problem is the following: if two $\mathcal{C}^2$ functions $f$ and $g$ from $]-\epsilon, \epsilon[$ to $\mathbb{R}$ are convex with $f(0)=g(0)=0$ and if $h=\min(f,g)$ is convex, is $h$ a $\mathcal{C}^2$ function too?

If $f'(0)\neq g'(0)$, $h$ is not convex. So $f'(0)=g'(0)$.

If $f''(0)> g''(0)$, $f(x)\geq g(x)$ in a neighborhood of $0$. So $h=g$ in this neighborhood and $h$ is $\mathcal{C}^2$ in this neighborhood.

So we may suppose $f''(0)=g''(0)$.

Let $I= \{x \in ]-\epsilon, \epsilon[ | f(x)=g(x)\}$,

$I^+= \{x \in ]-\epsilon, \epsilon[| f(x)>g(x)\}$,

$I^-= \{x \in ]-\epsilon, \epsilon[| f(x)<g(x)\}$

(1) If $x \in I^-$: $I^-$ is open so $h=f$ in a neighborhood of $x$, so $h''(x)=f''(x)$ and $h'(x)=f'(x)$.

(2) If $x \in I^+$: $I^+$ is open so $h=g$ in a neighborhood of $x$, so $h''(x)=g''(x)$ and $h'(x)=g'(x)$.

(3) If $x \in I$: $f(x)=g(x):=a$. $h$ is convex, so $f'(x)=g'(x):=b$.

(3.i) if $f''(x)> g''(x)$, as $f'(x)=g'(x)$ ,$f\geq g$ in a neighborhood of $x$, so $h''(x)=g''(x)$ and $h'(x)=g'(x)$.

(3.ii) if $f''(x)<g''(x)$, as $f'(x)=g'(x)$, $f \leq g$ in a neighborhood of $x$, so $h''(x)=f''(x)$and $h'(x)=f'(x)$.

(3.iii) if $f''(x)=g''(x):=c$, for all $u$ in a neighborhood of $0$, $h(x+u)=\min(f(x+u),g(x+u))$ $=\min(f(x)+f'(x)u+(c/2)u^2+ o_1(u^2), g(x)+g'(x)u+ (c/2)u^2+o_2(u^2))=a+bu+(c/2)u^2+o_3(u^2)$. where $\lim_{u \mapsto 0} o_i(u^2)/u^2=0.$ So $h'(x)=b=f'(x)=g'(x)$.

So $\forall x \in ]-\epsilon, \epsilon[, h'(x)=f'(x)$ or $h'(x)=g'(x)$. And $h'(0)=f'(0)=g'(0)$.

So $\forall x \in ]-\epsilon,\epsilon[, \frac{h'(x)-h'(0)}{x} = \frac{f'(x)-f'(0)}{x}$ or $\frac{h'(x)-h'(0)}{x} =\frac{g'(x)-g'(0)}{x}$.

So the limit of $\frac{h'(x)-h'(0)}{x}$ when $x \mapsto 0$ is $f''(0)=g''(0)$.

So $h''(0)=f''(0)=g''(0)$.

By the same way, we show that if $f(z)= g(z)$, then $h''(z)=f''(z)=g''(z)$.

If $f(z) \neq g(z)$, then $h''(z)=f''(z)$ or $h''(z)=g''(z)$.

(1) if $f(z) < g(z)$, it's easy to see that $h(z)=f(z)$ so $h$is $\mathcal{C}^2$ in the neighborhood of $z$.

(2) idem if $f(z)>g(z)$.

(3) if $f(z)=g(z)$, then $f'(z)=g'(z)$.

(3.i) if $f''(z) < g''(z)$, $f \leq g$ in a neighborhood of $z$, so $h=f$ in this neighborhood an is $\mathcal{C}^2$ in this neighborhood.

(3.ii) idem if $f''(z)> g''(z)$.

(3.iii) if $f''(z)=g''(z)$, $h''(z)=f''(z)=g''(z)$.And $\forall y$, $h''(y)=f''(y)$ or $h''(y)=g''(y)$. So $\lim_{y \mapsto z} h''(y)=h''(z)$. So $h''$ is continuous at $z$.

So $h$ is a $\mathcal{C}^2$ function.

[Edit: I believed the problem was in $\mathbb{R}^2$] ]

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.