Chapter 1 Sigma-Algebras

1.1 Denition

Consider a set X . A algebra F of subsets of X is a collection F of subsets of X satisfying the following conditions: (a) F (b) if B F then its complement B c is also in F (c) if B1 , B2 , ... is a countable collection of sets in F then their union n=1 Bn Sometimes we will just write sigma-algebra instead of sigma-algebra of subsets of X . There are two extreme examples of sigma-algebras: the collection {, X } is a sigma-algebra of subsets of X the set P (X ) of all subsets of X is a sigma-algebra Any sigma-algebra F of subsets of X lies between these two extremes: {, X } F P (X ) An atom of F is a set A F such that the only subsets of A which are also in F are the empty set and A itself. 1

CHAPTER 1. SIGMA-ALGEBRAS

A partition of X is a collection of disjoint subsets of X whose union is all of X . For simplicity, consider a partition consisting of a nite number of sets A1 , ..., AN . Thus Ai Aj = and A1 AN = X Then the collect F consisting of all unions of the sets Aj forms a algebra. Here are a few simple observations: Proposition 1 Let F be a sigma-algebra of subsets of X . (i) X F (ii) If A1 , ..., An F then A1 An F (iii) If A1 , ..., An F then A1 An F (iv) If A1 , A2 , ... is a countable collection of sets in F then n=1 An F (v) If A, B F then A B F . Proof Since F and X = c it follows that X F . For (ii) we have A1 A n = A 1 A n F Then (iii) follows by complementation:
c c A1 An = (Ac 1 An )

which is in F because each Ac i F and, by (i), F is closed under nite unions. Similarly, (iv) follows by taking complements :
c c n=1 An = [n=1 An ]

which belongs to F because F is closed under complements and countable unions. Finally, A B = A Bc is in F , because A, B c F . QED

1.2. GENERATED SIGMA-ALGEBRA (B )

1.2

Generated Sigma-algebra (B)

Let X be a set and B a non-empty collection of subsets of X . The smallest algebra containing all the sets of B is denoted (B ) and is called the sigma-algebra generated by the collection B . The term smallest here means that any sigma-algebra containing the sets of B would have to contain all the sets of (B ) as well. We need to check that such a smalled sigma-algebra exists. To this end observe rst the following fact: If G is any non-empty collection of sigma-algebras of subsets of X then the intersection G is also a sigma-algebra of subsets of X . Here G = {A X | A F for every F G}

consists of all sets A which belong to each sigma-algebra F of G. The verication of this statement is left as an (easy) exercise. Given a collection B of subsets of X , let GB be the collection of all sigmaalgebras containing all the sets of B . Note that P (X ) GB and so GB is not empty. Then GB is a sigma-algebra, contains all the sets of B , and is minimal among such sigma-algebras. Minimality here means that if F is a sigma-algebra such that BF then GB F Thus GB is the sigma-algebra generated by B : (B ) = GB If B is itself a sigma-algebra then of course (B ) = B .

CHAPTER 1. SIGMA-ALGEBRAS

1.3

The Dynkin Theorem

Let X be a set. A collection P of subsets of X is a -system if ( ) P is closed under nite intersections: if A, B P then A B P Note that by the usual induction argument, this condition implies that if A1 , ..., An are a nite number of sets in P then their intersection A1 An is also in P . A collection L of subsets of X is called a system if (1) L contains the empty set (2) L is closed under complements: if A L then Ac L (3) L is closed under countable disjoint union: if A1 , A2 , ... L and Ai Aj = for every i = j , then n=1 An L Unlike a algebra, the notions of system and system are not in themselves fundamental. Their signicance is contained in the following theorem which will be of great use later in proving uniqueness of measures: Theorem 1 The Dynkin theorem If P is a system and L a system of subsets of X then (P ) L, i.e. the sigma-algebra generated by P is contained in L. The proof of this result is long but can be broken up into simple little pieces. As a rst step, we have Lemma 1 A system is closed under proper dierences, i.e. if A, B L, where L is a system, and A B then the dierence B A is also in L. Proof. It is best to draw a little diagram illustrating the fact that A B . From this you can see that B A is the complement of the set A B c , and the latter, being the disjoint union of A L and B c L, is in L; thus B A L. More formally, B A = B Ac = (B c A)c

1.3. THE DYNKIN THEOREM

is in L because it is the complement of the set B c A which is in L because it is the union of two disjoint sets A and B c both of which are in L. QED The next step is more substantial: Lemma 2 A family which is both a system and a system is a algebra. Proof. Let S be a collection of subsets of X which is both a system and a system. To prove that S is a algebra it will be enough to show that S is closed under countable unions (not just disjoint countable unions). Let A1 , A2 , ... S . We have to show that their union n=1 An is in S . The trick (and it is a very useful trick) is to rewrite A n=1 n as a countable union of disjoint sets: n=1 An = n=1 Bn where B1 = A1 and, for n 1,
c c Bn = An (A1 A2 An1 ) = An Ac 1 A2 A n

(1.1)

Thus Bn consists of all elements of An which do not appear in any previous Ai . It is clear that the sets B1 , B2 , ... are mutually disjoint. Since S is a system, each complement Ac i is in S , and since S is a system it follows then that Bn , which is a nite intersection of sets in S , is also in S . QED As further preparation for the proof of the main theorem let us make one more observation, though its signicance will only become clear later: Lemma 3 Suppose L is a system of subsets of X . For any set A L , let SA be the set of all B X for which A B L . Then SA is a system. Proof. First note that SA , because A = L . It is also clear that SA is closed under countable disjoint unions. The last thing we have to show is that SA is closed under complements. To this end, let B SA and observe that A B c = A B = A (A B ) The utility in writing the dierence A B as the proper dierence A (A B ) lies in the fact that A B A and we can appeal to Lemma 1, along with the facts that A and A B are both in L , to conclude that A (A B ) is in L . QED

CHAPTER 1. SIGMA-ALGEBRAS

Now we return to the proof of the main theorem. As before, P is a system and L a system, with P L. Our objective is to show that the sigma-algebra (P ) generated by P is contained in L. The strategy will be to produce a sigma-algebra which lies between P and L, i.e. contains P and is contained in L. This will imply that (P ), which is the smallest sigmaalgebra containing P , is contained in L. We look at l(P ), the intersection of all systems containing P . Clearing l(P ) is itself also a system and contains P , and is thus the minimal system containing P . This means that any system which contains P must also contain l(P ). The objective will be to show that the system l(P ) is also a system. This would imply that l(P ) is a sigma-algebra. It contains P and, being the minimal system containing P , is a subset of L. This would provide our sigma-algebra lying between P and L. So the last piece of the argument is: Lemma 4 l(P ) is a system. The proof of this uses a bootstrap argument which is often useful in measure theory. We start with a set A P and show that A B is in l(P ) for every B in l(P ); then we turn around and use this to show that if A and B are in l(P ) then so is their intersection. Proof. Let A P , and let SA be the set of all sets B X for which A B is in l(P ). We have already proven that SA is a system. Moreover, it is clear that every element of P is in SA . Thus SA is a system with P SA . Therefore, l(P ) SA . Which means that we have proven that for any A P and any B l(P ) the intersection A B is in l(P ). So now consider a B l(P ), and look at SB . The preceding paragraph proves that P SB . On the other hand, by Lemma 3, SB is a system. Therefore, l(P ) SB . Which means: for any A l(P ), the intersection A B is in l(P ). Thus, l(P ) is a system. QED Putting all of the strands together, we have: Proof of Dynkins theorem. We have proven that the system l(P ) is also a system, and is therefore a algebra. On the other hand, we also know that P l(P ) L

1.3. THE DYNKIN THEOREM

because l(P ) is the intersection of all systems containing P , and L is just one system containing P . Thus we have produced a sigma-algebra l(P ) lying between P and L. Therefore, P (P ) l(P ) L since (P ) is the intersection of all sigma-algebras which contain P . QED There are several other similar results which can substitute for the Dynkin theorem. The best known alternative is the monotone class lemma, but we shall not go into this.