Let $A$ be the subset of $\mathbb{R}^n$ defined by $A=\{x\in\mathbb{R}^{n}:|x_{1}-x_{n}|+\sum_{i=1}^{n-1}|x_{i+1}-x_{i}|\leq d\}$ for a given $d$. Next, sample a point $p$ uniformly in the unit cube, and let $B$ be the $\ell_1$ ball of fixed radius $r$ about $p$. Is there a good upper bound for the probability that $A\cap B$ is nonempty, in terms of $n$, $d$, and $r$? I am most interested in limiting behavior as $n\to\infty$ (in which case, obviously, $d$ and $r$ would have to depend on $n$).
$\begingroup$
$\endgroup$
4
-
$\begingroup$ You seem to use $\ell$ in two different senses... $\endgroup$– Joseph O'RourkeCommented Sep 1, 2017 at 19:04
-
$\begingroup$ Why isn't this the same question as taking $r=0$ and $d=d+r$? $\endgroup$– Anthony QuasCommented Sep 2, 2017 at 9:51
-
$\begingroup$ @AnthonyQuas, if that were the case, then it would also be the same as taking $r=d+r$ and $d=0$. If we take $d+r=1$ and have $p=(0,1)$, then your set is empty, whereas the new set I just described is not. $\endgroup$– Tom SolbergCommented Sep 2, 2017 at 18:08
-
$\begingroup$ Hmmm... Maybe I meant $d=d+2r$? If I'm right that there's a reformulation like this, it will clearly make your problem much easier. $\endgroup$– Anthony QuasCommented Sep 2, 2017 at 20:28
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
For $n=3$ this is easy. A triple is in $A$ iff its maximum and minimum are within $d/2$. So $A \cap B_p$ is non-empty iff the maximum and minimum coordinates of $p$ are within $c=\min(1,d/2+2r)$. This has probability $$6\left(\int_{x=0}^{1-c} \int_{y=x}^{x+c} \int_{z=y}^{x+c} dz\, dy\, dx + \int_{x=1-c}^1 \int_{y=x}^1 \int_{z=y}^1 dz\, dy\, dx \right) $$ $$=6\left(\frac{c^2-c^3}{2}+\frac{c^3}{6}\right)=3c^2-2c^3.$$
We'd need other descriptions of $A$ in higher dimensions to make this work for larger $n$.