Throughout, μ∗ is an outer measure on X, M is the family of Carathéodory measurable sets, and we use the splitting condition in the form: E∈M if and only if μ∗(A)≥μ∗(A∩E)+μ∗(A∖E) for every A⊆X, the reverse inequality being automatic from subadditivity (apply countable subadditivity to the sequence A∩E,A∖E,∅,∅,…, using μ∗(∅)=0).
Step 1 (M contains X and is closed under complements). X∈M since A∩X=A and A∖X=∅. The defining condition is symmetric in E and X∖E, because A∩(X∖E)=A∖E and A∖(X∖E)=A∩E; hence E∈M implies X∖E∈M (complements).
Step 2 (closure under finite unions and finite additivity). Let E,F∈M and A⊆X. Splitting A by E, then splitting both pieces by F:
μ∗(A)=μ∗(A∩E∩F)+μ∗((A∩E)∖F)+μ∗((A∖E)∩F)+μ∗((A∖E)∖F).
The first three sets cover A∩(E∪F) (indeed they partition it), so by subadditivity their outer measures sum to at least μ∗(A∩(E∪F)), while the fourth set is A∖(E∪F). Hence μ∗(A)≥μ∗(A∩(E∪F))+μ∗(A∖(E∪F)), so E∪F∈M. With Step 1, M is closed under finite unions, finite intersections, and differences. Moreover, if E,F∈M are disjoint, splitting A∩(E∪F) by E gives
μ∗(A∩(E∪F))=μ∗(A∩E)+μ∗(A∩F)for every A⊆X,
and by induction the analogous identity holds for finitely many pairwise disjoint members of M.
Step 3 (countable unions and countable additivity). Let (Em)m∈N be a sequence in M and E=⋃mEm. Replacing Em by Em∖⋃l<mEl — members of M by Step 2 with the same union — we may assume the Em are pairwise disjoint. Fix A⊆X and k∈N. Splitting A by Fk=⋃m≤kEm∈M and using the finite additivity identity of Step 2 and monotonicity (A∖Fk⊇A∖E):
μ∗(A)=μ∗(A∩Fk)+μ∗(A∖Fk) ≥ m≤k∑μ∗(A∩Em)+μ∗(A∖E).
Letting k→∞ (the partial sums are nondecreasing; sums in [0,∞] as in Measure, Measure Space, and Probability Measure) and then applying countable subadditivity to A∩E=⋃m(A∩Em):
μ∗(A) ≥ m∑μ∗(A∩Em)+μ∗(A∖E) ≥ μ∗(A∩E)+μ∗(A∖E) ≥ μ∗(A),
the last inequality again by subadditivity. Hence all inequalities are equalities: E∈M, and M satisfies the three properties of a σ-algebra, proving claim 1.
Taking A=E in the displayed chain of equalities gives
μ∗(E)=m∑μ∗(Em)
for every sequence of pairwise disjoint members of M with union E; together with μ∗(∅)=0, the restriction of μ∗ to M is a measure, proving claim 2. ■