nLab density of a subset


The density of a subset SXS \subseteq X is a measure of how “large” the subset is, or what proportion of XX one can consider SS to take up.


Density is usually defined for subsets SS \subseteq \mathbb{N} of the natural numbers, or other countable sets, but the concept generalises to sets XX equipped with an exhaustive countable filtration F 1F 2X\emptyset \subseteq F_1 \subseteq F_2 \cdots \subseteq X (i.e., one such that nF n=X\bigcup_{n\in \mathbb{N}} F_n = X), where each F nF_n is equipped with a measure μ n\mu_n satisfying μ n(F n)<\mu_n(F_n) \lt \infty and where each inclusion F nF n+1F_n \to F_{n+1} is measure-preserving. We call such a filtration and system of measures ={F n,μ n}\mathcal{F} = \{F_n,\mu_n\} a measured filtration on XX.

The key example is of a totally ordered countable set A={a 1,a 2,}A = \{a_1,a_2,\ldots\} equipped with the filtration F n={a 1,a n}F_n = \{a_1,\ldots a_n\} and the counting measure. For instance, AA could be the natural numbers \mathbb{N}, or any countable subset with the inherited ordering, such as the primes.

Notice that any finite set can also be given the structure of a measured filtration, with a filtration that stabilises after finitely many steps, and with the counting measure.

In the other direction, a σ\sigma-finite measure space (M,μ)(M,\mu) can be equipped (by definition) with a measured filtration by measurable subsets, with the induced measures.


Given a measured filtration (X,)(X,\mathcal{F}), a subset SXS \subseteq X with the property that SF nS\cap F_n is measurable for all nn, the \mathcal{F}-density of SS is defined to be the number

d (S)=lim nμ n(SF n)μ n(F n)[0,1], d_\mathcal{F}(S) = \lim_{n\to \infty} \frac{\mu_n(S\cap F_n)}{\mu_n(F_n)} \in [0,1],

if this limit exists.

For a finite set B={b 1,,b n}B=\{b_1,\ldots,b_n\} equipped with any measured filtration \mathcal{F}, the \mathcal{F}-density of a subset SBS \subseteq B agrees with the simple ratio |S|/|B||S|/|B|.

Likewise, for any measure space (M,μ)(M,\mu) of finite total measure and SS a measurable subset, we can relate d (S)d_\mathcal{F}(S) to μ(S)/μ(M)\mu(S)/\mu(M) for various choices of measured filtration on MM (DR: does this depend on the choice of filtration?)


Density arguments are of importance in counting prime numbers with certain properties, and have shown great utility in proving partial results about the Birch and Swinnerton-Dyer conjecture on elliptic curves (namely that it is true for a set of elliptic curves over \mathbb{Q} with density roughly 67%). In this latter example, there are a number of filtrations on can place on the set of elliptic curves up to isomorphism, which only makes small changes to final calculated densities.

Any σ\sigma-finite measure μ\mu of a set MM is equivalent, as a measure, to a probability measure. This is constructed by considering a filtration arising from a countable disjoint partition into measurable sets of finite measure.

Upper and lower asymptotic density

One can also consider, instead of the limit in the definition of density, which may or may not exist, the lim sup or the lim inf. These give the upper and lower (asymptotic) densities, respectively.


  • Wikipedia, Natural density (which deals only with the case of sets of natural numbers)

See also

A great number of other definitions of densities of sets of natural numbers (clearly generalisable to other countable ordered sets) are given at

Created on May 2, 2016 at 00:45:22. See the history of this page for a list of all contributions to it.