5
$\begingroup$

Let $H$ be a $r \times r$ grid. Suppose that at most $r/10^5$ vertices of this grid are colored red. For every vertex $v \in V(H)$, let $B_i(v)$ be the ball of radius $i$ centered at $v$. (Or for simplicity you can consider the $2i+1 \times 2i+1$ sub-grid centered at $v$.). Let $C_i(v) = B_i(v) \setminus B_{i/2}(v)$, for $i \in I:=\{2^k\colon k\in\mathbb{Z}, 1\leq k\leq\frac{\log r}{\log 2} \}$; that is the annulus of radius $i$ around $v$. Is the following claim true: There exists a vertex $v \in V(H)$ such that for every $i \in I$, we have that the number of red vertices in $C_i(v)$ is at most $i$?

$\endgroup$
5
  • $\begingroup$ Dear @user113298: If one interprets your question to mean that towards the boundaries you prefer to just truncate the annuli, then writing down the details becomes significantly more complicated. I, and, I think, many others, would prefer if you stipulated a toroidal topology for your problem, in the sense which makes every annulus exactly look like an annulus. I think if you declared that treating your grid as a grid on a torus, more people would be willing to do the analysis. Would you please at least state if you insist on a grid in a plane, preferably with a reason? $\endgroup$ Commented Aug 12, 2017 at 17:19
  • $\begingroup$ This is a very nice question, looks like some discrete version of well studied fractal dimensions. Do you have any related results? What happens in one dimension? Why do you need to consider the annulus instead of the disk? $\endgroup$
    – domotorp
    Commented Aug 12, 2017 at 19:16
  • $\begingroup$ Another trivial observation, hopefully useful to anyone thinking about answering the OP's question negatively via the obvious 'equispaced red dots on a large toroidal grid' example: the OP has made sure that the claim survives that example, with room to spare: assuming the 'usual circular annuli'-interpretation of the question, it follows that it is necessary for the claim to be true that $\delta\leq\frac{4\log 2}{3\pi}\frac{1}{\log r}\qquad$ (N), where $\delta$ denotes red-dot-density. Since the OP demands $\delta\leq\frac{1}{10^5}\frac{1}{r}$, (N) is satisfied with room to spare. $\endgroup$ Commented Aug 13, 2017 at 5:16
  • 1
    $\begingroup$ Recently noticed that this question was also posted here. $\endgroup$ Commented Aug 14, 2017 at 5:42
  • $\begingroup$ It was pointed out by domotorp that there may be natural connections between the OP and Minkowski–Bouligand dimension. $\endgroup$ Commented Aug 14, 2017 at 5:55

1 Answer 1

1
$\begingroup$

(This is not an answer to the OP's question1, under any of its interpretations, only an extended comment which the comment box is too small to contain, a comment thought to be helpful to the OP. It is an illustrated summary of what the OP---who so far has not responded to even the more important requests for clarification in the comments--- seems to be asking for.)

Beginning of extended comment on OP.

One-sentence summary of what the OP seems to be asking for in essence. What is the lowest density such that one can still find a 'mass distribution' in a 2-dimensional grid such that around each grid-point there exists at least one 'balancedly-hollow shell'2 of density $\Omega(\frac{1}{\text{radiusoftheshell}})$?

Detailed summary of what the OP is asking(with a little interpretation on my part, in particular assuming toroidal topology on the grid). The OP in essence seems to be asking for what is the smallest $\delta\in[0,1]$ such that it is still possible to choose the following data:

  • a number $r\in\omega$,
  • a function $\mathrm{c}\colon (\mathbb{Z}/r\mathbb{Z})^2\rightarrow\{0,1\}$,
  • a function $\mathrm{k}\colon (\mathbb{Z}/r\mathbb{Z})^2\rightarrow\mathbb{Z}$,

such that the following axioms are satisfied:

  1. $\frac{\lvert\mathrm{c}^{-1}(1)\rvert}{r^2}\leq \delta$,
  2. for each $v\in (\mathbb{Z}/r\mathbb{Z})^2,\qquad$ $1\leq\mathrm{k}(v)\leq\frac{\log r}{\log 2}$,
  3. for each $v=(x_0,y_0)\in (\mathbb{Z}/r\mathbb{Z})^2,\qquad$ ${\small\lvert\{ (x_1,y_1)\in (\mathbb{Z}/r\mathbb{Z})^2\colon\qquad { \small 2^{\mathrm{k}(v)-1}<\lvert x_1-x_0\rvert_{\mathrm{mod}\ r}\leq 2^{\mathrm{k}(v)},\ 2^{\mathrm{k}(v)-1}<\lvert y_1-y_0\rvert_{\mathrm{mod}\ r}\leq 2^{\mathrm{k}(v)},\ \mathrm{c}(x_1,y_1)=1\}\rvert>2^{\mathrm{k}(v)}}}$

Remarks.

  • The $\mathrm{c}$-value $1$ is interpreted as the color 'red'.

  • For each $r\in\omega(r)$ there is a smallest $\delta_{\text{user113298}}(r)\in[0,1]$ such that a 'mass-distribution'(i.e. function $\mathrm{c}$) like the above exists, since in the trivial case $\delta:=1$, upon choosing $\mathrm{c}(v)=1$ for all $v$ and $\mathrm{k}(v):=1$, one finds that 1. and 2. hold by definition, and that 3. is true in the form of $16>2$.

  • The OP in particular seems to be asking for something similar to whether $\delta_{\text{user113298}}(r) > \frac{1}{10^5}\frac{1}{r}$ ($= \frac{r/10^5}{\lvert r\times r \rvert}$). (This seems the greatest distortion the present interpretation causes compared with the OP's original formulation; I have strengthened a formulation one obtains when applying the usual negation-rules to the OP's original question which has quantifier structure

$\exists v \forall i\quad \text{$C_i(v)$ sparse}$

whose negation of course is

$\forall v \exists i\quad \text{$C_i(v)$ not sparse}$. )

  • Like 'domotorp' suggested, there are generalizations of the above to an arbitrary dimension $d\geq 1$, suggesting an analogously defined 'physical constant' $\delta_{\text{user113298}}(d,r)\in [0,1]$.

  • There is the followig reason for the $\Omega(\frac{1}{\text{outerradius}})$ in the above one-sentence-rendition. Evidently, abbreviating $k:=\mathrm{k}(v)$, there are exactly $(2^{k+1}+1)(2^{k+1}+1)-(2^k+1)(2^k+1) = 2^{k+1}\cdot(3\cdot 2^{k-1}+1)$ grid-points in the set ${\{ (x_1,y_1)\in (\mathbb{Z}/r\mathbb{Z})^2\colon\qquad { \small 2^{\mathrm{k}(v)-1}<\lvert x_1-x_0\rvert_{\mathrm{mod}\ r}\leq 2^{\mathrm{k}(v)},\ 2^{\mathrm{k}(v)-1}<\lvert y_1-y_0\rvert_{\mathrm{mod}\ r}\leq 2^{\mathrm{k}(v)}\}}}$

and hence the red-dot-density ($=$'mass density') of the 'shell' picked by the 'witness function' $\mathrm{k}(v)$ is required to be $>\frac{1}{3\cdot 2^k + 2} \in \Theta_{i=2^k\rightarrow\infty}(\frac{1}{i})$.

An illustration relevant to the OP. In the following picture one sees a representation of an 100x100 grid, which I would suggest to be equipped not with the topology of the plane, but rather with the quotient topology w.r.t. identifying opposite points of its boundary (aka 'toroidal topology'). In particular, the 'overlapping' annulus is to be thought of as 'wrapping around the torus'.

Moreover, one sees a representation of one "annulus of radius $i$" in the OP, using the "$2i+1\times 2i+1$ subgrid"-interpretation permitted by the OP, for $i:=\mathrm{k}(v):=2^5$. (Note that the annulus has exactly $(2^6+1)(2^6+1)-(2^5+1)(2^5+1)=3136$ black grid-points.) Instead of the center-vertex $v$ I have displayed the 'outer radius' $2^5$ of the 'grid annulus' in the center of the annulus.

Moreover, I have displayed exactly $33>2^5$ randomly selected red vertices in said annulus, which, by itself make this annulus violate the property that the OP seems to be hoping for.

Here, axiom 3. holds in the instance of this particular annulus-center in the form of $33>2^5$.

Needless to say, this is not a counterexample to the OP's question, already because this is but one of the $100\times 100 = 10^4$ such annuli one would have to display if one were to give an explicit counterexample in such a way.

enter image description here

1 Which appears (to me) to have some non-trivial substance and not to be answerable straight out of the literature.

2 Where 'balancedly-hollow shell' is the shortest sort-of-self-explanatory term I can think of for 'hollow shell of outer radius a power of two and inner radius equal to half of the outer radius. Moreover, arguably the essence of the OP's formulation is not that the inner radius be half the outer radius, rather the asymptotic statement that $\mathrm{innerradius}\sim_{r\to\infty}\mathrm{outerradius}$. Because of this, and for brevity, I replaced '$\Omega(\frac{1}{\mathrm{outerradiusofshell}})$ with $\Omega(\frac{1}{\mathrm{radiusofshell}})$ in the 'one-sentence summary above.

End of extended comment on OP.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .