Measure Theory First Week Robert Simon e-mail: [email protected] We follow the book by Donald Cohn, Mea- sure Theory. 1 Algebras and Measures: A collection A of subsets of X is called an algebra if (a) all of X is a member of A, (b) for all A 2 A, the set XnA is in A, n (c) for every A1;:::;An 2 A, [i=1Ai 2 A, n (d) for every A1;:::;An 2 A, \i=1Ai 2 A. The condition (d) is superfluous, n n as \i=1Ai = Xn([i=1(XnAi). The empty set ; is always in the collection. 2 The collection A is called a sigma-algebra (or σ-algebra) if (a) and (b) hold and additionally for every infinite sequence A1;A2;::: of sets in A: 1 1 [i=1Ai 2 A and \i=1Ai 2 A. In general we work with sigma-algebras. A set in the sigma-algebra A is called a A- measurable set. A space X with a sigma algebra A is ex- pressed as (X; A) and is called a measurable space. 3 Examples: (a) A is the collection of all subsets of some set X. { sigma-algebra (b) A = fX; ;g { sigma-algebra (c) X is infinite and A is the collection of sets A such that A is finite or XnA is finite. { algebra, but not a sigma-algebra. (d) X is infinite and A is the collection of sets A such that A is finite or countable or XnA is finite or countable. { sigma-algebra. 4 (e) X = R and A is the collection of all intervals of R. { not an algebra (f) X = R and A is the collection of all finite unions of intervals of R. { algebra, but not sigma-algebra. If A were a sigma-algebra, all points of R would be in A, and so the rational numbers Q should be in A. 5 Given a collection of sets A that may not be a sigma-algebra, we want to way to construct a collection σ(A) that is a sigma-algebra and is the sigma-algebra that is generated by A{ meaning that all sets which must be in the sigma algebra are there and all that need not be there are not there. We mean that it is the smallest sigma-algebra containing A. 6 Lemma: Let X be a set and let (Ai j i 2 I) be an arbitrary collection of sigma alge- bras. The collection A := fA j 8i 2 IA 2 Aig is a sigma-algebra. Proof: (a) X 2 Ai for every i 2 I ) X 2 A. (b) A 2 A ) 8 i 2 IA 2 Ai ) 8 i 2 IXnA 2 Ai ) XnA 2 A. (c) A1; · · · 2 A ) 8 i 2 IA1; · · · 2 Ai 1 1 ) 8 i 2 I [j=1 Aj 2 Ai )[j=1Aj 2 A. 2 7 Lemma: For any collection A of subsets of X there is a smallest sigma-algebra, called σ(A), containing A. This means that any other sigma algebra containing A also con- tains σ(A). Proof: Let (Bi j i 2 I) be all the sigma algebras of X that contain A. There is at least one member of the family, namely all the subsets of X. By the above lemma, the collection B := fB j 8i 2 IB 2 Big is a sigma-algebra. Fix any sigma algebra Bi containing A: any B 2 B is also contained in Bi. 2 8 Determining the smallest sigma algebra con- taining a collection of sets can be a difficult task. Consider the collection A of subsets A1;A2;::: of the integers such that Ai = fni jn is an integerg. What is σ(A)? 9 Topology: A topology for a set X is a collection A of open subsets of X such that: (a) X 2 A, (b) 8A1;A2 2 A A1 \ A2 2 A, (c) if (Ai j i 2 I) is any collection in A then [iAi 2 A. As the empty intersection is included, the empty set is an open set. In Rn a set A is open if for every x 2 A there is some δ such that the open ball Bδ(x) = fy j jjy − xjj < δg is contained in A. It is easy to check that this definition satis- fies the three conditions (a), (b), (c). 10 A base for a topology is a special collection B of open sets such that every open set A of the topology is a union of sets of the base. Lemma: A base for the topology of Rn is the collection of sets Bδ(x) where δ is a rational number and each coordinate of x is rational. 11 Proof: Let x be a member of A, an open set. With δ small enough so that Bδ(x) ⊆ A, let x be a point with rational coefficients within δ=3 of x and let δ be a rational num- ber with δ=2 < δ < 2δ=3. The ball Bδ(x) both contains x and is con- tained within A. Unioning such open balls for every x 2 A will recreate the set A. 2. 12 The above base of open sets of Rn is a count- able collection, meaning that every open set is the union of countably many open balls of this base. For any space X with a topology B(X) is defined to be the smallest sigma- algebra containing all the open sets of X. This special sigma-algebra is called the col- lection of Borel sets. B(Rn) include most of the sets one could imagine. 13 Let F be the collection of closed sets of Rn and G the collection of open sets of Rn. Let Fσ be the collection of sets of the form 1 [i=1Ai for some sequence A1;A2;::: of closed sets (in F). Let Gδ be the collection of sets of the form 1 \i=1Ai for some sequence A1;A2;::: of open sets (in G). Members of Gδ are called Gδ sets and mem- bers of Fσ are called Fδ sets. 14 n Lemma: Every closed set of R is in Gδ n and every open set of R is in Fσ. Proof: Let C be a closed set of Rn. For every > 0 define C := fy j jjy − xjj < for some x 2 C. C is open: if jjy − xjj < for some x 2 C then B−||y−xjj (y) is also in C. 2 We claim that \iC1 is equal to C. i Any points y of \iC1 has a sequence x1; x2;::: i 1 of points in C such that jjy − xijj < i . Therefore the sequence xi converges to y. As C is closed, this means that y is in C. 15 Now take any open set A and consider its complement B = RnnA. As B = \Ai for a sequence of open sets, 1 we can write XnB = A = [i=1XnAi for a sequence XnA1;XnA2;::: of closed sets. 2 16 Corollary: The Borel collection of Rn are the sigma-algebra generated by (a) the open sets of Rn, (b) the closed sets of Rn, (c) a countable base for the topology of Rn. 17 Any countable process of taking unions and intersections, starting with the open sets, will result in a Borel set. Indeed, it is consistent logically to assume that all subsets of Rn are Borel sets, and it requires some axiom of choice to prove that a non-Borel set exists. 18 Measure: Let X be a set with an algebra A. A func- tion µ : A! [0; 1] is called finitely ad- ditive if for every finite collection A1;A2;:::;An of mutually disjoint sets in A, n Pn µ([i=1Ai) = i=1 µ(Ai). It is called a measure if additionally µ(;) = 0. Let X be a set with a sigma algebra A.A function µ : A! [0; 1] is called count- ably additive if for every infinite collec- tion A1;A2;::: of mutually disjoint sets in A, 1 P1 µ([i=1Ai) = i=1 µ(Ai). It is called a measure if additionally µ(;) = 0. 19 Countably additive implies finitely additive, by equating Aj = ; for all j > n such that n is large enough. But does finite additivity on a sigma algebra imply countable additivity? People thought so, until about 1900. The theory of finite additive measures is deep, related to duality in functional analysis, and not handled in this course. For us, measure means countably additive. 20 Examples: (a) µ(A) = jAj, the cardinality of A. (b) x 2 X chosen, δx(A) = 1 if x 2 A and δx(A) = 0 if x 62 A. 21 (c) Z is the set of integers and A is the collection of subsets such that A is in A if and only if A is finite or ZnA is finite. µ(A) = 1 if ZnA is finite and µ(A) = 0 if A is finite. Now try to extend µ to the sigma algebra σ(A), which are all the subsets of Z. Problem: what should be the weight given to the set of even numbers and the set of odd numbers? And what of the other ways to partition the integers into finitely many disjoint infinite subsets? It can be done, but never in a sigma additive way. If so, µ(fig) = 0 for all i 2 Z, but then µ(Z) = 0, a contradiction.