I want to know the smoothness of convex set in ${\bf R}^n$. Recall the following definition. Definition : $X$ is a bounded closed convex set in ${\bf R}^n$ if for $x$, $y\in X$, the any $d$-minimizing geodesic from $x$ to $y$ lies in $X$ where $d$ is a distance function of $X$.

That is, if $Y= S^{n-1}(1)$ and $X$ is convex then for $a$, $b\in X$, then $\frac{sa + (1-s)b}{|sa + (1-s)b |}$ is in $X$ for $0< s<1$

Question 1) Does the boundary of $m$-dimensional bounded convex set has dimension $m-1$ ?

Question 2) Is the following opinion is right ?

$(\ast)$ My thought : Let $m\geq 2$. A $(m-1)$-dimensional boundary of a $m$-dimensional bounded closed convex set $X$ is smooth except some $(m-2)$-dimensional set.

The motivation of this is as follows: In some paper, the Hausdorff measure of convex set in $S^{n-1}(1)$ is considered.

That is, in my thought convex set may be a set of noninteger Hausdorff dimension. Am I right?

If $\ast$ is right, then why does one consider the Hausdorff measure of convex set?

Thank you in advance.

[paper's content]-----------------------------------------------------

3.1 Proposition : $X$ is a closed convex set in $S^{n-1}(1)$ and $u$ is a point in $X$ Then area $ (X\cap {\bf H}_u) \geq \frac{1}{2} $ area $ (X)$ where ${\bf H}_u = \{ p\in S^{n-1}(1) | p\cdot u \geq 0\}$

3.2 Note : If $X \subset S^{n-1}(1)$ is a convex spherical set of Hausdorff dimension $d$, then $H^d(X\cap {\bf H}_u) \geq \frac{1}{2} H^d(X)$ where $H^d$ is the $d$-dimensional Hausdorff measure.

Here there is the word "spherical". I think that if we omit the word, then it is also fine.

  • 1
    $\begingroup$ If $m< n$, the boundary of an $m$-dimensional closed set (convex or otherwise) is the set itself, as it has empty interior. And what is $S^{n-1}(1)$? A sphere has no convex subset with more than one point. In any case, a convex set has the same Hausdorff dimension as its affine hull, which is an integer. $\endgroup$ Nov 23, 2012 at 16:34
  • 1
    $\begingroup$ Perhaps this discussion would have more point if the OP told us which paper he was looking at ?! $\endgroup$
    – Igor Rivin
    Nov 23, 2012 at 20:14
  • 1
    $\begingroup$ A geodesic ball in $S^{n-1}(1)$ is not convex according to the definition you’ve given. If $x,y\in X\subseteq S^{n-1}(1)$, $x\ne y$, then every point $tx+(t-1)y$ with $0< t< 1$ has norm strictly less than $1$, and thus is not an element of $X$. $\endgroup$ Nov 26, 2012 at 12:39
  • 1
    $\begingroup$ I see that the paper is using a different definition: they say that an $X\subseteq S^{n-1}$ is convex if any two points of $X$ can be joined by a distance-minimizing geodesic which lies in $X$. $\endgroup$ Nov 26, 2012 at 12:44
  • 1
    $\begingroup$ This question formulation was highly unclear to me. $\endgroup$ Apr 15, 2014 at 21:56

2 Answers 2

  1. Yes. The boundary even has a locally finite Haudsorff $(m-1)$-measure.

  2. No. A convex function of 1 variable has increasing derivative, but this derivative can have a dense set of jumps.

In general, the function describing the boundary is only Lipschitz (and differentiable almost everywhere).

For all these facts, you may consult a nice book Hormander, Notions of convexity, Chap II.

On your other questions. Of course, there is no reason to consider Hausdorff measure of a convex set: it is ordinary Lebesgue measure in the linear span of this set. I guess the paper you mention considers Hausdorff measure on the BOUNDARY of a convex set. As I said in 1, it has integer dimension. But so what? It is not a smooth surface. What other measure you propose to consider on it?

  • 3
    $\begingroup$ In fact one can say much more about regularity of the Lipschitz functions that define boundary of a convex set. $\endgroup$ Mar 15, 2020 at 20:30

In fact one can say quite a lot of regularity of the boundary of a convex set. Assume that $X\subset\mathbb{R}^n$ is a bounded convex set with non-empty interior.

  1. Convex functions are locally Lipschitz and therefore the boundary of a convex set is locally a graph of a Lipschitz functions defined on a ball in $\mathbb{R}^{n-1}$. It follows that $\mathcal{H}^{n-1}(\partial X)<\infty$.

  2. Locally Lipschitz functions are differentiable almost everywhere so convex functions are differnetiable almost everywhere. However, one can say much more.

The following beautiful result is due to Anderson and Klee [AK] and Zajíček [Z]:

Theorem. If $f:\mathbb{R}^{n-1}\to\mathbb{R}$ is convex, then there are counrably manly Lipschitz functions $ g_i:\mathbb{R}^{n-2}\to\mathbb{R}^{n-1}, \quad i=1,2,\ldots $ such that $f$ is differentiable in $$ D=\mathbb{R}^{n-1}\setminus \left(\bigcup_{i=1}^\infty g_i(\mathbb{R}^{n-2})\right). $$ Moreover, $\nabla f:D\to\mathbb{R}^{n-1}$ is continuous.

In fact Anderson and Klee did not discuss points of non-differentiability of a function, but the above result follows from what they proved. On the other hand Zajíček proved a stronger result than the one above since he showed that the functions $g_i$ can be represented as a differennces of two convex fucntions, see also [H] for an easy to follow adaptation of the original proof due to Zajíček.

This theorem generalizes to convex functions defined on convex domains. As a corollary we obtain:

Corollary. Let $X\subset\mathbb{R}^n$, be a bounded convex domain with nonempty interior and let $B^{n-2}$ be the unit ball in $\mathbb{R}^{n-2}$. Then $\mathcal{H}^{n-1}(\partial X)<\infty$, and there are countably many Lipschitz functions $g_i:B^{n-2}\to\partial X$ such that $\partial X$ has a tangent space at every point of $$ D=\partial X\setminus \left(\bigcup_{i=1}^\infty g_i(B^{n-2})\right) $$ and the tangent space changes continuously along $D$.

Therefore, in a sense, the boundary of $X$ is smooth away from a set of $\sigma$-finite Hausdorff measure $\mathcal{H}^{n-2}$.

Yet another result was proved in [AH].

Theorem. If $X\subset\mathbb{R}^n$ is a convex set with nonempty interior, then for any $\epsilon>0$ there is a convex set $X_\epsilon\subset X$ with the boundary of class $C^{1,1}$ such that $\mathcal{H}^{n-1}(\partial X\setminus\partial X_\epsilon)<\epsilon$.

Here $C^{1,1}$ means the class of functions with Lipschitz derivative so the boundary of class $C^{1,1}$ means that the boundary is locally a graph of a $C^{1,1}$ function.

[AH] D. Azagra, P. Hajłasz, Lusin-type properties of convex functions and convex bodies.
J. Geom. Anal. 31 (2021), 11685–11701.

[AK] R. D. Anderson, V. L. Klee, Jr., Convex functions and upper semi-continuous collections. Duke Math. J. 19 (1952), 349–357.

[H] P. Hajłasz, On an old theorem of Erdős about ambiguous locus. Colloq. Math. 168 (2022), 249–256.

[Z] L. Zajíček, On the differentiation of convex functions in finite and infinite dimensional spaces. Czechoslovak Math. J. 29 (1979), 340–348.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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