Let $\mu$ be a finite Borel measure on $\mathbb{R}^n$. I am interested in how does $\mu(B(x,r))$ behave, where $B(x,r)$ is the open ball of radius $r$ centered at $x$. For instance, as far as I recall, for each $\alpha \in [0,n]$, there exists finite $\mu$ so that $\mu(B(x,r)) \sim r^{\alpha}$ for $\mu$-a.e. $x$, which are called dimensionally-regular measures (with constant local dimension $\alpha$).

Here is my question: does there exist finite Borel measure$\mu$ such that that $\mu(B(x,r))$ vanishes superpolynomially fast, say, $\sim e^{-\frac{1}{r}}$, i.e.,

$\mu\left(\left\{x: \liminf_{r \to 0} r |\log \mu (B(x,r)) | > 0 \right\}\right) > 0$?


If I am not wrong, the answer is no (even for $\sigma$-finite measures). We can assume we are working with a finite measure $\mu$ in $[0,1]^d$.

Consider the sets $A_k =${$ x \ : \ \liminf_{r\to 0} |\log \mu(B(x,r))| r > 1/k$}. It is enough to show that the sets $A_k$ have zero measure.

Now, $A_k = \bigcap_N \bigcup_{n>N} B_n$ where $B_n =$ {$x \ : \ \mu(B(x,1/n))< e^{-n/k} $}.

To see this, notice that for every $x\in A_k$, there exists $n_0$ such that for every $n>n_0$ we have that $1/n |\log \mu(B(x,1/n))| > 1/k$.

So, it is enough to show that $\sum_{n>N} \mu(B_n)\to 0$ as $N\to \infty$.

For every $n>0$ we can divide $[0,1]^d$ in $C n^d$ cubes whose interiors cover $[0,1]^d$ and such that any of them is has diameter smaller than, say, $1/(10n)$.

Now, $\mu(B_n) \leq \sum_{i\in I} \mu(Q_i)$ where the $Q_i$ are the cubes which intersect $B_n$. Now, for each $Q_i$ with $i\in I$ we have that $$\mu(Q_i) \leq e^{-n/k}$$

since it is contained in $B(x,1/n)$ of a point $x\in B_n$.

So, $\mu(B_n) \leq C n^d e^{-n/k} \to 0$ and this implies that $\sum_{n>N} \mu(B_n) \to 0$ as $N\to \infty$.

| cite | improve this answer | |
  • 1
    $\begingroup$ Why is it that $A_k=\bigcup B_n$, and even it were so, why is it enough to show that $\mu(B_n)\to 0$? $\endgroup$ – R W Nov 11 '10 at 16:32
  • $\begingroup$ Sorry, there were some holes in the proof. I hope now it is correct. $\endgroup$ – rpotrie Nov 11 '10 at 16:58
  • $\begingroup$ Yes - it's OK now - except for one detail. In the last line you, of course, wanted to refer to summability of the series $\sum n^d e^{-n/k}$ (which is the easy direction of the Borel-Cantelli Lemma). $\endgroup$ – R W Nov 11 '10 at 17:51
  • $\begingroup$ Great. The same proof shows that it is impossible to have $\mu(B(x,r)) \sim r^{\alpha}$ for $\alpha > d+1$. I think the tight result should be $\alpha > d$. $\endgroup$ – gondolier Nov 12 '10 at 0:19

I think in order to prove the sharp result you need to use a covering theorem. The best option seems to be Vitali's covering theorem for Radon measures, which says the following: let $\mu$ be a Radon measure on $\mathbb{R}^d$, let $A\subset\mathbb{R}^d$ be a set, and let $\mathcal{B}$ be a family of balls such that each point of $A$ is the center of balls in $\mathcal{B}$ of arbitrarily small radius. Then there is pairwise disjoint collection $\{ B_i\}\subset\mathcal{B}$ such that $\mu(A\backslash \bigcup_i B_i)=0$. For a proof, see e.g. Mattila's book, Theorem 2.8.

With this theorem it is easy to prove the following: if $\mu$ is a Radon measure on $\mathbb{R}^d$, then $$ \liminf_{r\to 0}\frac{\mu(B(x,r))}{r^d} > 0 \quad \text{for } \mu\text{-a.e. } x. $$

To prove this, suppose $\mu$ is a measure such that the claim fails. We may assume without loss of generality that $\mu$ has bounded support (otherwise, let $\mu_N$ be the restriction of $\mu$ to $B(0,N)$; the theorem holds for all $\mu_N$ so it also holds for $\mu$).

Fix a small $\varepsilon>0$. Let $\mathcal{B}$ be the family of all balls $B$ of radius $r\le 1$ such that $\mu(B) < \varepsilon r^d$. Moreover, let $A$ be the set of all $x$ such that $\mu(B(x,r_i)) < \varepsilon r_i^d$ for a sequence $r_i\to 0$. Then $A$ satisfies the hypotheses of Vitali's covering lemma for the family $\mathcal{B}$. Also note that $A\supset A'$, where $$ A' = \{x\in\mathbb{R}^d: \liminf_{r\to 0}\frac{\mu(B(x,r))}{r^d} = 0\}. $$ By assumption, $\mu(A')>0$.

Now let $\{ B_i\}$ be the disjoint collection of balls provided by Vitali's covering lemma. On one hand, we have $$ \mu(\bigcup_i B_i) \ge \mu(A) \ge \mu(A') >0. $$ On the other hand, by definition of the family $\mathcal{B}$ and the fact that $\{B_i\}$ is disjoint, $$ \mu(\bigcup_i B_i) = \sum_i \mu(B_i) \le \varepsilon \sum_i r_i^d < C \varepsilon, $$ where $r_i$ is the radius of $B_i$ and $C$ is a constant that depends only on the size of the support of $\mu$. By taking $\varepsilon$ small enough we obtain a contradiction.

As a corollary, we find that, indeed, it is impossible to have $\mu(B(x,r))\sim r^\alpha$ for any $\alpha>d$.

| cite | improve this answer | |
  • $\begingroup$ thanks Pablo! Now I remembered I've seen this theorem from Federer's book before. Somehow I escaped my mind... $\endgroup$ – gondolier Nov 23 '10 at 6:33

If you are willing to replace $\mathbb{R}^n$ by a Banach space $B$, then it is perfectly possible. For example, the Wiener measure $\mu$ on the space of continuous function from $[0,1]$ to $\mathbb{R}$ satisfies that for $\mu$-almost every $x$, $\mu(B(x,r)) \sim \exp(-1/r^2)$ for small $r$. (The keyword for this is "small ball estimates".) If you are willing to forgo separability of $B$, then you can even construct probability measures with the property that, for every $x \in B$, $\mu(B(x,1)) = 0$!

The canonical example for this is to take for $\mu$ the law of a sequence of i.i.d. normal random variables and for $B$ the space of sequences $x = (x_n)$ such that $$ \|x\| := \sup_{n\ge 1} \Big(2^{-n}\sum_{k=2^{n-1}}^{2^n-1} x_k^2\Big)^{1/2} < \infty\;. $$ The claim then follows from the strong law of large numbers.

| cite | improve this answer | |

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.