ProofComplete

Signed Measures and Radon-Nikodym - Key Proof

ProofProof Sketch of the Hahn Decomposition Theorem

We outline the proof of the Hahn Decomposition Theorem for a signed measure ν\nu.

Goal: Find disjoint sets P,NP, N with PN=XP \cup N = X such that ν\nu is non-negative on subsets of PP and non-positive on subsets of NN.

Proof (assuming ν\nu takes no value -\infty):

Step 1: Define: α=sup{ν(A):AF}\alpha = \sup\{\nu(A) : A \in \mathcal{F}\}

Since ν\nu takes no value -\infty, we have α[0,]\alpha \in [0, \infty].

Step 2: Choose a sequence {An}\{A_n\} in F\mathcal{F} with ν(An)α\nu(A_n) \to \alpha. Define: P=n=1AnP = \bigcup_{n=1}^{\infty} A_n

Step 3: We claim PP is a positive set. Suppose not: there exists EPE \subseteq P with ν(E)<0\nu(E) < 0.

Then for large nn, considering AnEA_n \cup E: ν(AnE)=ν(An)+ν(EAn)ν(An)+ν(E)\nu(A_n \cup E) = \nu(A_n) + \nu(E \setminus A_n) \geq \nu(A_n) + \nu(E)

If ν(EAn)0\nu(E \setminus A_n) \geq 0, this gives ν(AnE)ν(An)+ν(E)>ν(An)\nu(A_n \cup E) \geq \nu(A_n) + \nu(E) > \nu(A_n), contradicting α=supν(A)\alpha = \sup \nu(A) as nn \to \infty.

Step 4: More carefully, one shows that PP can be chosen so that for any EPE \subseteq P, we have ν(E)0\nu(E) \geq 0. This requires an iterative construction that removes negative subsets.

Step 5: Set N=PcN = P^c. We must show ν(E)0\nu(E) \leq 0 for all ENE \subseteq N.

If ν(E)>0\nu(E) > 0 for some ENE \subseteq N, then PEP \cup E would give ν(PE)>ν(P)=α\nu(P \cup E) > \nu(P) = \alpha, contradicting the definition of α\alpha.

Step 6: Thus (P,N)(P, N) is a Hahn decomposition.

Remark

Key insights in the proof:

  1. Supremum argument: The proof uses the supremum of ν\nu over all measurable sets to construct PP. This is a variational approach.

  2. Countable unions: Taking P=AnP = \bigcup A_n ensures PP "captures" the supremum in the limit, using countable additivity of ν\nu.

  3. Iterative refinement: The actual proof requires iteratively removing bad subsets from candidates for PP until a true positive set is achieved.

  4. Non-uniqueness: The decomposition is not unique. If (P,N)(P, N) is a Hahn decomposition and EE is a null set (with ν+(E)=ν(E)=0\nu^+(E) = \nu^-(E) = 0), then (PE,NE)(P \cup E, N \setminus E) is also a Hahn decomposition.

  5. Jordan decomposition: Once the Hahn decomposition is established, the Jordan decomposition ν=ν+ν\nu = \nu^+ - \nu^- follows immediately by defining ν+(A)=ν(AP)\nu^+(A) = \nu(A \cap P) and ν(A)=ν(AN)\nu^-(A) = -\nu(A \cap N).