Essentials of Analysis and Probability, 1st Semester 2009-2010 Proof of Carath eodory extension theorem (Non-examinable)

We suppose we have an algebra A of subsets of and a countable additive set function on A. Associated with we have an outer measure which is a set function dened for all subsets of as follows:

(E ) = inf
k=1

(Ak ) : Ak A & E k=1 Ak

The following proposition summarises the properties of outer measure that we need: Proposition. (i) for any sequence E1 , E2 , of sets we have

( k=1 Ek )

k=1

(Ek )

(ii) if A A then (A) = (A) (iii) if A A and E is any set then (A E ) + (Ac E ) = (E ) Proof. (i) Let > 0. Then for each k we can nd a sequence Ak1 , Ak2 , of sets in A which k cover Ek and such that j =1 (Akj ) < (Ek ) + 2 . Then the countable family (Akj ) covers the union Ek and k,j (Akj ) k {(Ek ) + 2k } = + k (Ek ). As is arbitrary it follows that (Ek ) k (Ek ) as required. (ii) That (A) (A) follows from the denition of (one can take A1 = A and the other Ak empty). For the converse suppose A Ak where Ak A. We have to show (Ak ) (A). To c do this let Bk = Ak (A1 Ak1 ) so that A Bk and the Bk are disjoint. Then by the countable additivity of we have (A) (Bk ) (Ak ) as required. (iii) From (i) we have (A E ) + (Ac E ) (E ). To show the reverse inequality, let > 0 and nd A1 , A2 , in A such that E Ak and (Ak ) < (E ) + . Then the sets Ak A cover E A, so (Ak A) (E A) and similarly (Ak Ac ) (E Ac ) and since (Ak A) + (Ak Ac ) = (Ak ) we deduce that (Ak ) (A E ) + (Ac E ). Hence c (E ) + (A E ) + (A E ) and as is arbitrary (E ) (A E ) + (Ac E ) as required. Part (i) of the proposition says that is countably subadditive. In general it will not be additive on arbitrary sets. However as we shall see it is countably additive on a suitable collection of sets, and this is how we obtain our measure. We now state the main result: Theorem. (Carath eodory extension theorem) Let A be an algebra of subsets of and let F be the -algebra generated by A. Let be a -nite countably additive set function on A. Then there is a unique measure on F such that (A) = (A) for all A A. Proof. We dene (E ), for arbitrary E , as the outer measure as introduced above. We shall show that it is in fact a measure on F . To do this we introduce a collection of sets C , consisting of all sets B such that (B E ) + (B c E ) = (E ) (1) for every set E . Note rst that by part (ii) of the proposition this equation is satised by every B A, so A C . Next we show that C is an algebra. It is clear from (1) that B C implies B c C . So we suppose A, B C and aim to prove A B C . To do this, rst note that as A C we get from (1) applied with B c E in place of E that (Ac B c E ) + (A B c E ) = (B c E ) (2)

Essentials of Analysis and Probability, 1st Semester 2009-2010 and as B C , applying (1) with (A B ) E in place of E gives ((A B ) E ) = (A B c E ) + (B E ) Combining (2) and (3) gives ((A B )c E ) + ((A B ) E ) = (B c E ) + (B E ) = (E )

(3)

showing that C is an algebra. Note that (3) shows that, if A and B are disjoints sets in C , then ((A B ) E ) = (A E ) + (B E ). By iterating this we obtain, for any disjoint B1 , , Bn C and any set E , that
n

((B1 Bn ) E ) =
k=1

(Bk E )

(4)

Now we nish the proof by showing that C is a -algebra and is countably additive on C . To do this suppose B1 , B2 , are disjoint sets in C . Write An = n k=1 Bk and A = k=1 Bk . Now for any set E , by (4) we have (An E ) = n k=1 (Bk E ). On the other hand by part (i) of the proposition (A E ) k=1 (Bk E ). Combining these we deduce (A E ) = ( B E ) and ( A E ) ( A E ). Now as An C we have k n k=1 c c (An E ) + (An E ) = (E ) so (An E ) + (A E ) (E ) and letting n gives (A E ) + (Ac E ) (E ). The reverse inequality follows from (i) of the proposition so we have (1) holding for A and A C . Also taking E = we have (A) = k=1 (Bk ). Hence we conclude that C is a -algebra containing A and that is a measure on it. Then C contains F so is a measure on F . To prove uniqueness suppose is a measure on F such that (A) = (A) for all A A. First note that (E ) (E ) for any E F , since if A1 , A2 , A are such that E Ak then (E ) (Ak ) = (Ak ) using the countable additivity of . Next, if E F and A A with (A) < we have (E A) = (E A). To see this, suppose on the contrary that (E A) < (E A). Then (A) = (A) = (E A) + (E c A) < (E A) + (E c A) = (A) = (A) which is a contradiction. So (E A) = (E A) when (A) > 0, as claimed. We apply this to get (E Bk ) = (E Bk ) for each k and any E F , and we deduce (E ) = (E Bk ) = (E Bk ) = (E ), so = as required.