MAT4300 - Mandatory Assignment Sindre Froyn (a) We have A1 , . . . , An ∈ A. We use (i),(ii) and (iii) as the axioms of algebras. (iii) A1 , A2 ∈ A =⇒ A1 ∪ A2 ∈ A We define the set B2 := A1 ∪ A2 , which we just showed was in A. (iii) B2 , A3 ∈ A =⇒ B2 ∪ A3 ∈ A =⇒ A1 ∪ A2 ∪ A3 ∈ A We define B3 := A1 ∪A2 ∪A3 and as we have just shown, B3 ∈ A. We assume this is true for Bk := A1 ∪ . . . ∪ Ak , so Bk ∈ A. We check the next step. (iii) Bk , Ak+1 ∈ A =⇒ Bk ∪ Ak+1 ∈ A =⇒ A1 ∪ . . . ∪ Ak ∪ Ak+1 ∈ A By induction, and repeated use of property (iii) for algebras, we have verified that the algebra A is closed under a countable and finite number of unions, so (a) A1 , . . . , An ∈ A =⇒ A1 ∪ A2 ∪ . . . ∪ An ∈ A. (b) We assume A, B ∈ A. (ii) (iii) A, B ∈ A =⇒ Ac , B c ∈ A =⇒ Ac ∪ B c ∈ A We apply DeMorgans identity. (ii) Ac ∪ B c ∈ A ⇐⇒ (A ∩ B)c ∈ A =⇒ A ∩ B ∈ A We have now shown that for A, B ∈ A, we have A ∩ B ∈ A. We use this in the next part. (ii) (b) A, B ∈ A =⇒ A, B c ∈ A =⇒ A ∩ B c ∈ A ⇐⇒ A \ B ∈ A (c) We are going to determine if the set B is an algebra. To do this we must verify the three axioms of algebras. ∅ ⊂ N, ∅ is finite =⇒ ∅ ∈ B 1 (i) We assume we have a set B ∈ B, and show that this implies B c ∈ B. Since B ∈ B, we have B or B c finite, which we can use directly as properties for the set B c (since (B c )c = B). B ∈ B =⇒ B ⊂ N =⇒ B c = N \ B ⊂ N =⇒ B c ∈ B (ii) For the last property we assume A, B ∈ B. Since A ⊂ N and B ⊂ N we obviously have A ∪ B ⊂ N, since the “worst case scenario” is N ⊂ N. We check if (A ∪ B) or (A ∪ B)c are finite, and using DeMorgan’s law on the complemented union we can rewrite it as Ac ∩B c . There are four possibilities: so if A and B are finite, then A ∪ B is finite. In any other case (A ∪ B)c is finite, since the intersection Ac ∩ B c restricts us to a finite set. A, B ∈ B =⇒ A ∪ B ∈ B (iii) We have checked the three axioms of algebras for B and verified that it is an algebra. (d) The σ-algebra generated by B is the smallest σ-algebra containing it, so B ⊂ σ(B). This means the first two properties we verified for the algebra B also apply for σ(B) for all sets B ∈ B. The σ-algebra allows countable unions, which the algebra did not, and these unions provide new sets. For sets A1 , A2 , A3 , . . . there are two possibilities. First if all are finite: then their union is also finite. [ A1 , A2 , . . . ∈ σ(B) =⇒ Ai ∈ σ(B) i∈N If at least one of the sets Ak is not finite, then the complement to the union is finite. [ c \ [ Ai ∈ σ(B) =⇒ Ai ∈ σ(B) =⇒ Aci ∈ σ(B) i∈N i∈N i∈N As described in (c), the one, or more, finite set(s) Ack restricts this intersection to a finite set, so (∪i∈N Ai )c is finite. All the sets in the σ-algebra generated 2 by B satisfy the same conditions as the sets in the underlying algebra: either it is finite, or it has a finite complement. σ(B) = B ⊂ N B or B c finite (e) We assume A1 , A2 , . . . , An ∈ A are all mutually disjoint. We use (i) and (ii) as the axioms of the finitely additive measure. We define B2 = A1 ∪ A2 . (ii) µ(B2 ) = µ A1 ∪ A2 = µ(A1 ) + µ(A2 ) We define B3 := A1 ∪ A2 ∪ A3 = B2 ∪ A3 . (ii) µ(B3 ) = µ(B2 ∪ A3 ) = µ(B2 ) + µ(A3 ) = µ(A1 ) + µ(A2 ) + µ(A3 ) P We assume this is true for Bk , so µ(Bk ) = µ(A1 ∪ . . . ∪ Ak ) = ki µ(Ak ). We check the next step, Bk+1 = Bk ∪ Ak+1 . (ii) µ(Bk+1 ) = µ(Bk ∪ Ak+1 ) = µ(Bk ) + µ(Ak+1 ) = k+1 X µ(Ai ) i=1 By induction, this is true for a countable, finite amount of sets, so µ n [ i=1 Ai = n X µ(Ai ). i=1 (f) To check that ν : B 7→ [0, ∞] is a finitely additive measure, we must verify the two defining properties. Property (i) is verified directly. ν(∅) = 0, since ∅ is finite. To check property (ii) we must consider different scenarios. We recall that A and B are disjoint. When A and B both are finite, then so is A ∪ B. ν(A ∪ B) = 0 = ν(A) + ν(B) When Ac is finite and B is finite, (A ∪ B)c = Ac ∩ B c is finite. By symmetry the same argument applies for the reverse case. ν(A ∪ B) = 1 = 1 + 0 = ν(A) + ν(B) 3 For the last scenario, Ac and B c finite, it is impossible to choose disjoint sets A and B. To illustrate this, we can set Ac = {1, . . . , 10} in which case A = {11, 12, . . .}, and now we cannot choose a finite B c such that B is disjoint from A, so the function ν does not apply to this case. Both properties are met, and ν is a finitely additive measure. (g) To extend the measure ν to the σ-algebra σ(B) we need to use Caratheodory’s theorem. To do this we must verify if the algebra B is a semi-ring, and check if ν satisfies the conditions of a pre-measure on B. To verify that B is a semi-ring, we must check the three defining properties. Obviously, ∅ ∈ B. As we showed in (b), the intersection of two sets in an algebra, is again in the algebra: S, T ∈ B ⇒ S ∩ T ∈ B. For the last property, we must check that for two sets S, T ∈ B, there are finitely many disjoint sets S1 , . . . , Sn such that n [ S\T = Sj . j=1 However, as we showed in (b), if S, T ∈ B then S \ T ∈ B, since B is an algebra, so this difference is always a single set in B. With all the three properties of semi-rings met, B is a semi-ring. We verify that ν satisfies the two necessary properties of pre-measures on the semi-ring. The first is ν(∅) = 0 which is satisfied since ν(∅) = 0 by (f). For the second property, we assume P (Sj )j∈N ⊂ B are disjoint, set S = ∪j∈N Sj ∈ BPand check that ν(S) = j∈N ν(Sj ). If Sj is finite for all P j, then the sum j∈N ν(Sj ) = j∈N 0 = 0, but on the left side we combine all natural numbers and get an infinite set, so ν(S) = 1. This property is not met for the finitely additive measure ν, so we cannot use Caratheodory’s theorem. In conclusion we cannot extend ν to be a measure on the σ-algebra σ(B). (h) Due to laziness, procrastination, excessive workload and confusion I have chosen not to answer this question. My apologies! (i) Pn Let f and g be step functions, f ≤ g, with representations f = i=1 ai 1Ai Pm and g = j=1 bj 1Aj , where we have chosen the standard representation over the same partition {Ai } of X. Using the representation of f and g: n X i=1 ai 1Ai ≤ n X i=1 bi 1Ai =⇒ 0 ≤ n X i=1 4 bi 1Ai − n X i=1 ai 1Ai n X = (bi − ai )1Ai i=1 We have a step function representation of a function that is greater or equal to 0. Using the preliminary integral, we replace the indicator function 1Ai with the non-negative measure µ(Ai ), and we preserve the inequality. It is permissible with a negative measure, since we know µ(X) < ∞. 0≤ n X (bi − ai )µ(Ai) = i=1 n X bi µ(Ai ) − i=1 n X ai µ(Ai ) ≤ i=1 n X n X ai µ(Ai ) =⇒ i=1 bi µ(Ai ) =⇒ I(f ) ≤ I(g) i=1 (j) We have a non-negative, simple function f . Z f dµ = inf I(h) h a simple function, f ≤ h = I(f ) Z f dµ = sup I(h) h a simple function, 0 ≤ h ≤ f = I(f ) In both cases we get I(f ) since f itself is among the simple functions h in the sets. Since the upper and lower integral are the same, f is integrable. Z Z Z f dµ = f dµ = f dµ = I(f ) (k) We have a function h : N 7→ R, given by h(n) = n1 . This function has a simple function representation, and thus we have an expression for the preliminary integral. n n X X ai ν(Ai ) ai 1Ai =⇒ I(h) = lim h = lim n→∞ n→∞ i=1 i=1 1 1 The sets {A1 } = 1, {A2 } = 2, etc and a1 = = 1, a2 = 12 and so on. Since h has a simple function representation, both the upper and lower integral will be I(h) as in part (j). Z Z Z n X hdν = hdν = hdν = I(h) = lim ai ν(Ai ) n→∞ i=1 All the sets Ai are finite so we get ν(Ai ) = 0 for all i. Z n X ai ν(Ai ) = 0. hdν = I(h) = lim n→∞ 5 i=1 (l) For the measure space (Y, C, λ) we have λ(Y ) 6= 0. By definition of measures λ: Y 7→ [0, ∞], but the inequality means that there is at least one set A ⊂ Y such that λ(A) > 0. A strictly positive, measurable function f : Y 7→ R means f = 0 on a null set: λ({f = 0}) = 0. By definition of the integral of measurable functions: Z Z Z Z + − f dλ = f dλ − f dλ = f + dλ We only consider the positive part since f is strictly positive and f − = 0. By definition of the integral of positive functions: Z f dλ = sup I(h) h a simple, strictly positive function, h ≤ f We can choose some h from this set, which can be a crude approximation of f , but still strictly positive and over the sets Ai ⊂ Y . h= n X ai 1Ai i=1 For this simple function we know that there exists at least one set Ak such that λ(Ak ) > 0, and since this set is not a null set h is positive on Ak so ak > 0. To sum up we know there exists som k ∈ N such that ak λ(Ak ) > 0. With this we can prove the desired inequality. Z n X 0 < ak λ(Ak ) ≤ ai λ(Ai ) = I(h) ≤ f dλ i=1 We have assumed that we integrate over the entire set Y . Even if the measure of Y isn’t 0 doesn’t mean there aren’t subsets with measure 0. We can construct a simple example to illustrate this. Y ∗ = [1], [2], (3, 4) We define a strictly positive function f ∗ on this set, where g is some strictly positive function. x=1 1 ∗ f (x) = 1 x=2 g(x) x ∈ (3, 4) The Lebesgue measure λ∗ gives us λ∗ (Y ∗ ) 6= 0, the function f ∗ : Y ∗ 7→ R is strictly positive, but the integral over a subset of Y ∗ is zero. Z f ∗ dλ∗ = 0. [1]∪[2] 6