# Why define measures on $\sigma$-rings?

I have the impression that modern texts deal almost excusively with measures on $\sigma$-algebras, while older texts, such as the one of Halmos, deal mainly with measures defined on $\sigma$-rings. I'm curious what motivated this change and in what context are $\sigma$-rings more natural domains for measures?

-
This proposed innovation by Halmos never caught on. – GEdgar Jun 25 '12 at 14:10
According to J.-P. Pier, Histoire de l'intégration, $\sigma$-rings were introduced by Fréchet. They appear on p. 250 of Sur l'intégrale d'une fonctionnelle étendue à un ensemble abstrait (1915) under the name familles additives d'ensembles. On the other hand, the book by Saks (1937) works with $\sigma$-algebras "additive classes of sets". I can't check in Carathéodory's work at the moment. – t.b. Jun 25 '12 at 16:04
I'm pretty sure, you now know something to answer this question yourself. Also, on page 10 of Dubins and Savage gambling book, they mention several paper by De Finetti that compare $\sigma$-additive and finitely-additive approach. – Ilya Jul 16 '13 at 13:11

Being unfamiliar with the older text, I can only speculate. One explanation is that one prefers to work with sets of $\sigma$-finite measure: those that can be written as a countable union of sets of finite measure. For example, sets of $\sigma$-finite length ($1$-dimensional Hausdorff measure) in the plane form a $\sigma$-ring, not a $\sigma$-algebra. It is rather fruitless to think about 1-dimensional measure of the complement of a line, so removing such sets from consideration seems reasonable.
I guess that interesting measure spaces are always measurable, and that's why modern books tend to use $\sigma$-algebras rather than $\sigma$-rings. For example, on $\sigma$-rings it may be impossible to integrate over the whole space, and this is usually a useless restriction.