the Caratheodory extension theorem

Here is the bootstrapping problem at the heart of building any measure: you know what size you want to give to simple sets — intervals get their length, rectangles get their area — but you need a measure defined on the whole vast sigma-algebra of Borel sets, including wildly complicated ones you cannot picture. How do you extend a sensible rule on simple pieces to a genuine measure on everything? The Caratheodory extension theorem is the machine that does this, and it is how Lebesgue measure and most other measures are actually constructed.

You start with a pre-measure: a set function defined only on a small, convenient family (an algebra closed under finite operations, like finite unions of intervals) that already obeys countable additivity wherever it makes sense there. The theorem then guarantees this pre-measure extends to a full measure on the entire sigma-algebra generated by that family. The mechanism is the notion of outer measure: to size an arbitrary set A, cover it by countably many simple sets whose total size you know, and take the smallest total over all such coverings. A set is declared measurable (in Caratheodory's sense) if it splits every other set cleanly with respect to this outer measure, and these well-behaved sets form a sigma-algebra on which the outer measure is genuinely countably additive. Paired with Dynkin's theorem, which makes the extension unique, you get existence and uniqueness together.

Concretely: define length on intervals as length([a,b]) = b - a, verify this little rule is a countably additive pre-measure on finite unions of intervals, and Caratheodory hands you Lebesgue measure on all the Borel sets — the rigorous foundation for areas, volumes, and the Lebesgue integral. The same theorem builds product measures and, ultimately, the probability spaces underlying infinite sequences. It is the existence engine: it is why you may assume the measures you want actually exist.