Many measures begin with a small collection of sets whose sizes are easy to prescribe. Length is first specified on intervals; a product measure is first specified on rectangles. A useful construction must answer two separate questions: do these initial values extend to a countably additive measure, and if so, is that extension the only one? Countable coverings answer the first question. A finite-measure exhaustion supplies a standard answer to the second.
The data and the theorem
Definition 3.6.1
(Algebra of sets and premeasure)
An algebra of subsets of a set contains and is closed under complements and finite unions. A function is a premeasure if and
whenever the are pairwise disjoint and their union also belongs to .
An algebra need not contain every countable union; the condition applies precisely when the union is in the domain. Finite additivity follows by padding a finite family with empty sets, and then monotonicity follows by writing for in .
Common mistake
(Applying the premeasure rule to a union outside the algebra)
Countable additivity of says nothing about a countable union that is not in . In the algebra of finite and cofinite subsets of , the set of even numbers is the disjoint union of the singletons , all in , but it is neither finite nor cofinite, so has no value there. Assigning it is exactly what the extension theorem must justify, not something the hypothesis provides.
Theorem 3.6.2
(Extension and -finite uniqueness)
Every premeasure on extends to a measure on . If is a countable union of sets with , this extension is unique on . The construction also provides a complete measure on a possibly larger -algebra.
The complete measure is obtained by the outer-measure splitting test. We use the general outer-measure theorem: for any outer measure on any set, its Carathéodory-measurable sets form a -algebra, and restriction to them is a complete measure. That proof uses only monotonicity, countable subadditivity and the splitting test, so it applies on an arbitrary . We also use the finite-measure uniqueness theorem for a generating -system in the final step.
Build the outer measure
Lemma 3.6.3
(The covering outer measure)
For set
Then is an outer measure on .
Proof
(Pay for each set)
The cover always exists: use , followed by empty sets. Its cost can be infinite. The empty set has cost zero, and monotonicity follows because any cover of a larger set also covers a smaller one. For countable subadditivity, if , choose for each a cover with cost below . Enumerating all the covering sets gives
Let . If the sum is infinite, the inequality is immediate.
The original values are preserved
Lemma 3.6.4
(The outer measure agrees with the premeasure)
For every , .
Proof
(Disjointize an arbitrary cover inside the algebra)
For , the one-set cover gives . For the reverse inequality, take any cover by algebra sets and define
These lie in , are pairwise disjoint, and have union . The premeasure property and monotonicity give
This is valid even if a sum is infinite; no subtraction of infinite measures occurs. Taking the infimum over covers shows .
Proof X-ray
Why the premeasure hypothesis matters
The easy inequality uses nothing. The reverse inequality is the step where countable additivity, not merely finite additivity, is needed, and it is used exactly once. The key move is the disjointization: each removes only the finitely many earlier sets , so stays inside the algebra even though the cover is infinite. Because the have union exactly , which is in , the union clause in the definition of a premeasure is satisfied and follows. Monotonicity on the algebra then compares each , and the infimum over covers closes the argument. With finite additivity alone one only gets for each , which is the wrong direction; the worked example of a finitely additive below shows that the conclusion then really fails.
Algebra sets pass the splitting test
Lemma 3.6.5
(Algebra sets are Carathéodory-measurable)
Every satisfies for all . Consequently restricted to is a measure extending .
Proof
(Split each covering set by )
Fix and any . For a cover , both and are algebra sets and cover and respectively. Finite additivity gives
Infimizing yields . Outer subadditivity gives the reverse inequality. Therefore is Carathéodory-measurable. The complete outer-measure theorem now gives a measure on the measurable -algebra, which contains ; restriction to extends by the preceding lemma.
This proves existence without a -finiteness assumption. The restriction to need not itself be complete.
Proposition 3.6.6
(The Carathéodory -algebra is the completion)
Under the -finiteness hypothesis of the theorem, the Carathéodory -algebra of is the completion of for the extended measure.
Proof
(Approximate from outside by countable intersections of covers)
First suppose a Carathéodory-measurable has finite outer measure. For each , choose an algebra cover of of cost less than , and let be the union of that cover. The set belongs to , contains , and satisfies . Since is measurable, . For general , intersect it with each finite-premeasure covering set and apply the same argument; the countable union of the resulting supersets differs from by a null set. To place this difference in the completion, note that every outer-null set has a -measurable null superset: for each cover by algebra sets of total cost below and intersect the resulting countable unions. Conversely, the Carathéodory -algebra contains the completion because its measure is complete.
Without -finiteness, we make no general completion claim here.
Why finite blocks give uniqueness
Proof
(Uniqueness on finite blocks)
Assume with and . Replace by the increasing finite union ; finite subadditivity gives . Let and be two extensions on . For each , their restrictions to are finite measures. They agree on the -system
which contains and generates the trace -algebra . The finite-measure -system uniqueness theorem therefore gives for every . Since , continuity from below yields , including when the common value is infinite.
Proof X-ray
Where the finite blocks are used
The hypothesis enters exactly once: it makes and finite on each , which the -system uniqueness theorem needs, because its Dynkin-class argument subtracts measures of nested sets. Two properties of the algebra make the trace family usable: it is closed under intersection, so the family is a -system, and it contains , so the agreement is part of the data. Replacing the by the increasing unions is the gluing device: a single limit then closes the argument, even when that limit is .
Worked example 3.6.7
(A point mass on the dyadic algebra)
Let and let be the finite unions of dyadic half-open intervals inside . Define if and otherwise. If in , then lies in at most one , and it lies in exactly when it lies in some ; so both sides of the premeasure identity equal . Here , so the single set is a finite-premeasure cover and the theorem gives a unique extension to . The Dirac measure is one extension, hence the only one.
A counterexample when -finiteness is absent
The same dyadic algebra is an algebra because unions and complements can be computed after refining to a common dyadic level, and it generates the Borel -algebra on .
Counterexample mode
Two different extensions of an infinite premeasure
Claim: every premeasure has a unique extension to . On the dyadic algebra of define and for every nonempty . This is a premeasure: if a disjoint union in the algebra is nonempty, at least one term is nonempty, so both sides of the additivity equation are infinite. Counting measure and twice counting measure , restricted to the Borel sets, both extend , because every nonempty dyadic-algebra set contains infinitely many points. But and , so the extensions differ. No countable collection of finite- algebra sets covers , since the only such set is empty. The repair is the -finiteness hypothesis; the example does not say uniqueness always fails without it, only that the hypothesis cannot simply be dropped.
Practice and solutions
Worked example 3.6.8
(A finitely additive set function that is not a premeasure)
On , let be the finite and cofinite sets. Put for finite and for cofinite . Two disjoint cofinite sets cannot exist in the infinite set , so a disjoint union of two algebra sets is finite plus finite or finite plus cofinite, and is finitely additive. It is not a premeasure: is the disjoint union of its singleton sets, each with value zero, while . Consequently no countably additive measure can extend these values.
Worked example 3.6.9
(Counting measure from the finite–cofinite algebra)
For the same algebra on , let be the number of points of , allowing . This is a premeasure because counting disjoint unions is countably additive. The singleton sets cover and have finite value, so uniqueness applies. Since all singletons are in , every subset of is a countable union of them; hence . The unique extension is counting measure on all subsets.
Quick check
(For the counting premeasure on the finite–cofinite algebra of , is the set of even numbers in , and what is ?)
Solution · Answer
is neither finite nor cofinite, so . A cover of either uses a cofinite set, of cost , or consists of finite sets whose sizes add up to at least the number of points of , which is infinite. Hence , the counting measure of .
Quick check
(Why does the point-mass premeasure on the dyadic algebra have a unique extension, while the infinite premeasure on the same algebra does not?)
Solution · Answer
For the point mass, itself is an algebra set of finite value, so the -finiteness hypothesis holds. For the infinite premeasure, the only algebra set of finite value is , so no countable family of such sets covers .
Quick check
(Let be a subset of a -measurable set with . Must belong to ? Is it measurable for the complete outer-measure extension?)
Solution · Answer
The original -algebra need not contain ; completeness is a separate property. But , so the outer-measure theorem makes measurable and null in the larger complete -algebra. For example, the Borel -algebra of the real line omits some subsets of Borel null sets, while its Lebesgue completion contains them.