Question about the properties of Shnirelman density

104 Views Asked by At

In elementary methods in analytic number theory by Gelfond and Linnik, the claim is made that if $d(A) + d(B) > 1$, then we can find $A',B'$ where $A' \subseteq A$ and $B' \subseteq B$ such that $d(A') + d(B')$ is as close to $1$ as we want.

I am not clear why this would be true.

Here are the assumptions:

$A,B$ are infinite sequences of integers starting with $0$ with in sequential order such as $0, a_1, a_2, \cdots$ where $0 < a_1 < a_2 < \cdots$

Shnirelman density is defined as: $$d(A) = \inf\limits_{n}\frac{A(n)}{n}$$

where: $$A(n) = \sum\limits_{0<a_i\le{n}}{1}$$

So, it is clear that: $$0 \le \frac{A(n)}{n} \le 1$$

I would appreciate it if someone could explain why we can make the assumption that $d(A') + d(B')$ can be as close to $1$ as we wish.

1

There are 1 best solutions below

0
On BEST ANSWER

I think that I've figured out the reasoning here.

Here's my thinking:

(1) We can assume that $d(B) \le d(A) < 1$.

Note: If $d(A)=1$, then we set $B' = B - \{1\}$ to get $d(A)+d(B')=1$

(2) Let $\epsilon$ be any number such that $1 > \epsilon > 0$

(3) We can assume that $1 - d(A) > \epsilon$

Note: If not, then we set $B' = B - \{1\}$ to get $\epsilon > 1 - d(A)-d(B')$

(4) There exists integers $x,y$ such that $1 - d(A) - \epsilon \le \frac{x}{y} < 1 - d(A)$

(5) From the definition of density, we know that $d(B) \le \frac{B(y)}{y}$

(6) Using $y$, we can build $B'$ by removing $B(y)-x$ elements from $B$.

Note: $\frac{x}{y} < d(B) \le \frac{B(y)}{y}$ since $d(A) + d(B) > 1$ but $d(A) + \frac{x}{y} < 1$.

Please let me know if you can state the argument more concisely or if you see a problem in my reasoning.