10
$\begingroup$

Let us consider a bounded, Borel function $F\colon \mathbb R^d \to \mathbb R^d$. Assume it satisfies the following Osgood-like condition: $$\tag{O} \boxed{\vert \langle F(x) - F(y), x-y \rangle\vert \le \Vert x-y \Vert \rho( \Vert x-y \Vert)} \qquad \forall x,y \in \mathbb R^d, $$ where $\rho \colon [0,1) \to \mathbb [0,+\infty) $ is an Osgood modulus of continuity, i.e. a continuous, non-decreasing function with $\rho(0)=0$ and $$ \int_0^1 \frac{1}{\rho(s)} \, ds = +\infty. $$

Q. Is it true that $F$ is continuous? Or, more precisely, is it true that $F$ is equivalent, up to null sets, to a continuous function?

Beside the link above, something (mildly) related can also be found here.

Of course the case $d=1$ is trivial and the question makes sense essentially for $d \ge 2$. Thanks.

Update (the "linear" case). Quoting from here (pag. 2)

[...] in the case when the modulus $\rho$ is linear, (O) implies that the symmetric part of the distributional derivative is bounded, hence Korn’s inequality gives that $F$ is equivalent, up to Lebesgue negligible sets, to a continuous function.

I honestly do not see how to show rigorously this claim in the case when $\rho$ is linear. How to use Korn's inequality (I suspect a variant of this $L^2$ inequality is needed, with estimates in $L^\infty$ but do not know)?

$\endgroup$
  • $\begingroup$ Can you prove this in the special case when $C$ is a constant function? Or perhaps separate continuity in one the variables? $\endgroup$ – erz Jan 20 '18 at 1:31
  • 1
    $\begingroup$ @erz Thanks for the interest. Actually, you are right, time-dependence is not so important, I have edited the question accordingly. Do you have any hints on how to approach it? $\endgroup$ – Y.B. Jan 20 '18 at 10:06
9
+50
$\begingroup$

You do not need such heavy high-tech as Korn's inequality or even Lebesgue measure theory for an elementary geometry homework. Let's say $F(0)=0$.

The first claim is that $F$ is bounded in some neighborhood of the origin. Indeed, just choose finitely many points $x_j$ such that $x_i-x_j$ span the space. Then for any $x$ in some fixed ball we have $\langle F(x)-F(x_j), x-x_j\rangle$ bounded by some constant whence $\langle F(x), x-x_j\rangle$ are bounded by a constant, whence $\langle F(x), x_i-x_j\rangle$ are bounded too and that is the end of this story.

Now suppose $x_j\to 0$ with $|F(x_j)|\ge\delta>0$. Then WLOG $F(x_j)\to u\ne 0$. Take any $x\ne 0$ close to $0$. We have $$ |\langle F(x)-0,x-0\rangle|\le |x|\rho(|x|) $$ and $$ |\langle F(x)-F(x_j),x-x_j\rangle|\le |x-x_j|\rho(|x-x_j|) $$ which, after passing to the limit, becomes $$ |\langle F(x)-u,x\rangle|\le |x|\rho(|x|) $$ Subtracting, we get $$ |\langle u,x\rangle|\le 2|x|\rho(|x|) $$ but this is absurd if $x$ is collinear with $u$ and $2\rho(|x|)<|u|$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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