It is well known that if $K_n$ are compact sets in $\mathbb{R}^n$ converging in Hausdorff distance to $K$ compact as well, then it does not follow that their Lebesgue measures converge (even if the Lebesgue measure of the $K_n$ is constant).

Given this, I would like to ask the following question: If we assume that $K_n$ does not only converge in HD to $K$ but also in measure $\lambda(K_n) \rightarrow \lambda(K)$ and let $F$ be a Lipschitz function, does it then follow that $\lambda(F(K_n)) \rightarrow \lambda(F(K))$?

This question is related to the previous since Lipschitz continuity still preserves metric convergence of $F(K_n)$ to $F(K)$, but now we have also the information that $\lambda(K_n) \rightarrow \lambda(K)$. Does this allow us to deduce the claim?

  • 1
    $\begingroup$ There are counter examples if $F$ is allowed to change dimension e.g. $F(x,y)=x$. $\endgroup$ – Anthony Quas Jan 8 '18 at 2:19

Note the following lemma: under the given assumptions, we have $\lambda(K_n \triangle K) \to 0$. Fix $\epsilon > 0$ and choose a $\delta$ so small that the $\delta$-thickening $K^\delta$ of $K$ has measure $\lambda(K^\delta) < \lambda(K) + \epsilon$. (This is possible since $K$ is compact and hence $\bigcap_i K^{1/i} = K$.) Now if $n$ is so large that $K_n \subset K^\delta$ (by Haudsdoff convergence) then $$\lambda(K_n \setminus K) \le \lambda(K^\delta) - \lambda(K) < \epsilon$$ and if we further take $n$ large enough that $\lambda(K_n) > \lambda(K) - \epsilon$, then we also get $$\lambda(K \setminus K_n) \le \lambda(K^\delta) - \lambda(K_n) < 2 \epsilon$$ proving the lemma.

Now note that $F(K_n) \triangle F(K) \subseteq F(K_n \triangle K)$. As such, we have $$\lambda(F(K_n) \triangle F(K)) \le \lambda(F(K_n \triangle K)) \le C_F \lambda(K_n \triangle K) \to 0$$ where $C_F$ is the Lipschitz constant of $F$. In particular $\lambda(F(K_n)) \to \lambda(F(K))$.

(Thanks to Iosif Pinelis for suggesting this argument which is much simpler than what was here before.)

| cite | improve this answer | |
  • $\begingroup$ I wanted to give a similar answer, but you did it first! Nice answer. However, it looks like the change of variables formula you referred to works only for bi-Lipschitz maps. However, it is easy to see that for any Lipschitz map $F$ there is a real number $C_F$ such that $\lambda(F(A))\le C_F\lambda(A)$ for all Lebesgue-measurable sets $A$ (cover $A$ tightly enough by countably many boxes, say). $\endgroup$ – Iosif Pinelis Jan 8 '18 at 5:37

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.