The "classical Beurling density" of a subset of the natural numbers is $d(A)=lim_{n\rightarrow\infty}\frac{|A\cap[1,n]|}{n}$, when it exists. It defines a finitely additive probability measure on the natural numbers which is invariant with respect to the sum. Here is my question: does there exist a "nice formula" to describe a finitely additive probability measure on $\mathbb N$ which is invariant with respect to the multiplication?

A couple of remarks: I don't know if it is trivial that such a measure exists, but anyway it follows from the application of a general result of Vern Paulsen (on arxiv "Syndetic sets and amenability"). Another problem would be that of finding the measure of particular sets. What about the measure of {$1!,2!,3!,4!...$}? Sets with measure differente from 0 and 1? For example the set of numbers whose first digit through 4 to 9 seems to have measure $=\log_{10}4$... any other?

Thanks in advance, Valerio

  • $\begingroup$ when you say invariant with respect to the sum do you mean that $\lbrace k+2 \mid k \in A \rbrace$ has the same measure as $A$? $\endgroup$ – Aaron Meyerowitz Mar 21 '11 at 21:52
  • $\begingroup$ Any finitely additive probability measure (also called a content) is defined on a field. But the collection of subsets $A$ of the natural numbers having a density $d(A)$ is not a field. $\endgroup$ – Did Mar 21 '11 at 22:37
  • $\begingroup$ That is, one can write down two sets that have density, but whose intersection does not have density. $\endgroup$ – Gerald Edgar Mar 21 '11 at 23:48
  • $\begingroup$ One example would be a measure assigning $\mu(A)=1$ if $A$ contains $0$, and $\mu(A)=0$ otherwise. $\endgroup$ – Daniel Litt Mar 22 '11 at 0:59
  • $\begingroup$ @Daniel, for many of us, $0$ is not a natural number. $\endgroup$ – Gerry Myerson Mar 22 '11 at 4:31

The natural thing to do here is to replace the intervals $[1,n]$ (and $n = |[1,n]|$) in the definition of $d(A)$ with a sequence $F_n$ of subsets of $\mathbb{N}$ which is multiplicatively asymptotically invariant (or, in other words, a Folner sequence for the semigroup $(\mathbb{N},\cdot)$). For an exploration of this idea, as well as applications, see for instance this article by Vitaly Bergelson:

Multiplicatively large sets and ergodic Ramsey theory, Israel Journal of Mathematics 148 (2005), 23-40.

EDIT: One particular example (mentioned in the article) is to take $F_n$ to be the set of all positive integers which can be written as a product of powers of the first $n$ primes, where the powers are allowed to be any non-negative integer which is less than or equal to $n$. The motivation for choosing $F_n$ this way is that just as $1$ generates the additive semigroup $(\mathbb{N},+)$, the primes generate the multiplicative semigroup. Think of balls in the corresponding Cayley graphs with radius getting larger and larger. The article contains many other examples of such $F_n$, and each of them gives a notion of "multiplicative density" by setting $d(A) = \lim_{n \to \infty} \frac{|A \cap F_n|}{|F_n|}$ (if the limit exists. Otherwise one usually considers the limsup and liminf).

  • $\begingroup$ Perhaps I am misunderstanding you, but I don't think this works; this is what Gerald Edgar attempts in his answer. It seems to me that $\mathbb{N}$ still has density $1$, whereas $2\mathbb{N}$ has density $1/2$. $\endgroup$ – Daniel Litt Mar 22 '11 at 19:48
  • $\begingroup$ No, $A_k = [1,k!]$ is not a Folner sequence for $(\mathbb{N},\cdot)$, since $2[1,k!]=[2,2 \cdot k!]$ and $[1,k!] \triangle [2,2 \cdot k!]$ has cardinality $1+ k!$, which is not $o(|A_k| = k!)$. $\endgroup$ – Mark Mar 22 '11 at 21:01
  • $\begingroup$ Ah, I see. (Btw, $2[1, k!]$ is only the evens in $[2, 2k!]$, but your computations still works). What is an example of a Folner sequence for $\mathbb{N}$? $\endgroup$ – Daniel Litt Mar 22 '11 at 22:09
  • $\begingroup$ Right. The notation confused me with intervals in the real numbers. As for your question, I've edited my answer to include such an example. $\endgroup$ – Mark Mar 22 '11 at 23:09
  • $\begingroup$ Thanks a lot, Mark. Even if I can't open the article, from Bergelson's home page directly either.. don't understand why! $\endgroup$ – Valerio Capraro Mar 25 '11 at 8:10

Here is an example. Let $d(A)$ be as you define it in the question, namely $$d(A)=\lim_{n\to \infty} \frac{|A\cap [1,n]|}{n}.$$ Let $U_n=\{k\in\mathbb{N} \text{ such that } k \text{ is a mutiple of } n!\}$. Then define $$\mu(A)=\lim_{n\to \infty} n!\cdot d(A\cap U_n).$$ Then I claim that $\mu$ is finitely additive and multiplicatively invariant. Finite additivity is obvious. For multiplicative invariance, note that for $s>k$, we have $(s+k)!\cdot d(kA\cap U_{s+k})=s!\cdot d(A\cap U_{s})$, unless I've screwed something up.

EDIT: Note by the way that one can replace the limit in the definition with $d$ with the Cesaro mean, for example, giving a much broader class of sets with defined measure. For example, with this addition, the set of natural numbers with a fixed leading digit in a fixed prime base $p$ has density $1/(p-1)$.

  • $\begingroup$ What would be an example of a set with density strictly between zero and one? $\endgroup$ – Gerry Myerson Mar 22 '11 at 6:08
  • $\begingroup$ It seems to me that numbers whose first digit is $1$ in base $3$ works, for example. $\endgroup$ – Daniel Litt Mar 22 '11 at 6:14
  • $\begingroup$ (That is, leading, not trailing.) $\endgroup$ – Daniel Litt Mar 22 '11 at 6:15
  • 1
    $\begingroup$ Another, more interesting example, seems to be--natural numbers $n$ such that the exponent of the greatest power of $2$ dividing $n$ equals twice that of the greatest power of $3$ dividing $n$, say $\pm 2$. $\endgroup$ – Daniel Litt Mar 22 '11 at 8:03

How about: $$ \lim_{k\to\infty} \frac{|A \cap [1,k!]|}{k!} $$ when it exists?

  • $\begingroup$ Then all the positive integers have density 1, but the even integers still have density 1/2. Right? $\endgroup$ – JBL Mar 22 '11 at 0:13
  • $\begingroup$ (In fact, your proposed definition must agree with the usual one whenever the usual one exists.) $\endgroup$ – JBL Mar 22 '11 at 0:14
  • $\begingroup$ I would argue that the two measures disagree. Take the set of integers $A:=\cup_{n\in\mathbb{N}}[n!,\mathrm{E}(\sqrt{n}n!)]$. If I'm not mistaken, you should have that the sequence $\frac{A\cap[1,k]}{k}$ has both 0 and 1 as adherent values. $\endgroup$ – Olivier Bégassat Mar 22 '11 at 0:36
  • $\begingroup$ while $\frac{|A\cap [1,k!]|}{k!}$ tends to $0$. $\endgroup$ – Olivier Bégassat Mar 22 '11 at 0:38
  • $\begingroup$ Which is consistent with what I wrote: Gerald Edgar's density is the limit of a subsequence of the sequence whose limit is the usual density. $\endgroup$ – JBL Mar 22 '11 at 0:48

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.