Measure and integral
Ilkka Holopainen
August 29, 2017 2 Measure and integral
These are lecture notes of the course Measure and integral (Mitta ja integraali).
0 Some background
0.1 Basic operations on sets Let X be an arbitrary set. The power set of X is the set of all subsets of X,
(X)= A: A X , P { ⊂ } and any subset (X) is called a family (or collection) of subsets of X. The union of a family F ⊂ P is F A = x X : x A for some A { ∈ ∈ ∈ F} A[∈F and the intersection (of ) is F A = x X : x A for all A . { ∈ ∈ ∈ F} A\∈F Let be an index set (set of indices) and suppose that for every α there exists a unique A ∈ A subset V X. (In other words, α V is a mapping (X).) Then the collection α ⊂ 7→ α A → P = V : α F { α ∈ A} is an indexed family of X. The union of an indexed family is
V = x X : x V for some α α { ∈ ∈ α ∈ A} α[∈A and the intersection of an indexed family is
V = x X : x V for all α . α { ∈ ∈ α ∈ A} α\∈A We denote also V and V , if is clear from the context. α α A α α [ \ Example. 1. Let (X). We can interpret as an indexed family by using as the index F ⊂ P F F set. That is, if α (thus α is a subset of X), we write V = α. Then = V : α . ∈ F α F { α ∈ F} 2. X = x , x = a singleton. { } { } x[∈X If the index set is N = 1, 2, 3,... , we denote { } ∞ Vn or Vn or Vn, n n n[∈N [ [ and ∞ Vn or Vn or Vn. n n n\∈N \ \ Spring 2017 3
∞ Sequences (of sets) are denoted by (Vn), (Vn)n=1, (Vn)n∈N, or V1,V2,.... The difference of sets A, B X is ⊂ A B = x X : x A and x B . \ { ∈ ∈ 6∈ } The complement of a set B X (with respect to X) is ⊂ Bc = X B. \ Remark. A B = A Bc. \ ∩
X A Bc = A B ∩ \
A B
Bc
Theorem 0.2. Let V : α be a family of X. Then the following de Morgan’s laws hold: { α ∈ A} c c (0.3) Vα = Vα α α [ \ and
c c (0.4) Vα = Vα . α α \ [ Let B X. Then the following distributive laws for union and for intersection hold: ⊂ (0.5) B V = (B V ) ∩ α ∩ α α α [ [ and
(0.6) B V = (B V ). ∪ α ∪ α α α \ \ Proof. (0.3):
x V c x V α: x V α: x V c x V c. ∈ α ⇐⇒ 6∈ α ⇐⇒ ∀ 6∈ α ⇐⇒ ∀ ∈ α ⇐⇒ ∈ α α α α [ [ \ (0.4): Similarly. (0.5):
x B V x B and x V x B and x V for some α ∈ ∩ α ⇐⇒ ∈ ∈ α ⇐⇒ ∈ ∈ α ∈A α α [ [ x B V for some α x B V . ⇐⇒ ∈ ∩ α ∈A ⇐⇒ ∈ ∩ α α [ (0.6): Similarly. 4 Measure and integral
The images and preimages of the union/intersection of a family.
Let X and Y be non-empty sets and f : X Y a mapping. → The image of a set A X under the mapping f is ⊂ f(A)= f(x): x A . ( Y ) { ∈ } ⊂ We usually abbreviate fA. The preimage of a set B Y under the mapping f is ⊂ f −1(B)= x X : f(x) B . { ∈ ∈ } We also abbreviate f −1B and denote
f −1(y)= f −1( y ), { } if y Y. [Note: f need not have an inverse mapping.] ∈ Theorem 0.7. Let f : X Y be a mapping and let V : α be a family of X, and let → { α ∈ A} W : β be a family of Y . Then { β ∈ B}
(0.8) f Vα = fVα α α [ [
−1 −1 (0.9) f Wβ = f Wβ β β [ [
−1 −1 (0.10) f Wβ = f Wβ. β β \ \ Proof. (0.8):
y f V y = f(x) and x V y = f(x) and x V for some α ∈ α ⇐⇒ ∈ α ⇐⇒ ∈ α ∈A α α [ [ y fV for some α y fV . ⇐⇒ ∈ α ∈A ⇐⇒ ∈ α α [ (0.9) and (0.10): Similarly.
Remark. It is always true that f V fV , α ⊂ α α α \ \ but the inclusion can be strict. The equality f( V ) = fV holds, for example, if f os an ∩α α ∩α α injection. Spring 2017 5
Countable and uncountable sets
Countability is a very important notion is measure theory!
Definition. A set A is countable if A = or there exists an injection f : A N ( a ∅ → ⇐⇒ ∃ surjection g : N A). → A set A is uncountable if A is not countable.
Remark. 1. A countable A finite ¨a¨arellinen (including ) or countably infinite (when ⇐⇒ ∅ there exists a bijection f : A N). → 2. A countable A = x : n N (repetition allowed, so that A can be finite). ⇐⇒ { n ∈ } 3. A countable, B A B countable. ⊂ ⇒ Theorem 0.11. If the sets A are countable n N, then n ∀ ∈
An is countable. n[∈N (”countable union of countable sets is countable”.)
Proof. We may assume that A = n N. Since A is countable, we may write A = n 6 ∅ ∀ ∈ n n x (n): m N . Define a mapping { m ∈ } g : N N A , g(n,m)= x (n). × →∪n n m Then g is a surjection N N A . Hence it suffices to find a surjection h: N N N, because × →∪n n → × then g h: N A ◦ → n n[∈N is surjective and therefore A is countable. An example of a surjection h: N N N is: ∪n n → × (1, 1) (1, 2) (1, 3) (1, 4) (1, 5) =h(1) =h(3) =h(6) =h(10) =h(15) · · · ր ր ր ր (2, 1) (2, 2) (2, 3) (2, 4) =h(2) =h(5) =h(9) =h(14) ր ր ր (3, 1) (3, 2) (3, 3) =h(4) =h(8) =h(13) ր ր (4, 1) (4, 2) =h(7) =h(12) ր (5, 1) =h(11) . . 6 Measure and integral
Corollary. The set of all rational numbers m Q = n,m Z, n = 0 { n | ∈ 6 } is countable. Reason: The set m A = n,m Z, n = 0, m k, n k k { n | ∈ 6 | |≤ | |≤ } is finite (and hence countable) k N. Theorem 0.11 Q = A countable. ∀ ∈ ⇒ ∪k∈N k Example. (Uncountable set). The interval [0, 1] (and hence R) is uncountable. Idea: x [0, 1] x has a decimal expansion ∈ ⇒
x = 0, a1a2a3 ..., where a 0, 1, 2,..., 9 . j ∈ { } Contrapositive: [0, 1] is countable, so [0, 1] = x : n N . Points x have decimal expansions { n ∈ } n (1) (1) (1) x1 = 0, a1 a2 a3 . . . (2) (2) (2) x2 = 0, a1 a2 a3 . . . (3) (3) (3) x3 = 0, a1 a2 a3 . . . . . (n) (n) (n) (n) xn = 0, a1 a2 a3 . . . an . . . . .
(1) (2) (3) (n) (n) On the ”diagonal” there is a sequence a1 , a2 , a3 , . . . , an ,..., where an is the nth decimal of x . Let x [0, 1] be defined by x = 0, b b b ..., where n ∈ 1 2 3 (n) (n) an + 2, if an 0, 1, 2,..., 7 , (0.12) bn = (n) (n) ∈ { } (an 2, if an 8, 9 . − ∈ { } The nth decimal of x satisfies b an = 2 n N, and therefore x = x n N. This is a | n − n| ∀ ∈ 6 n ∀ ∈ contradiction, because [0, 1] = x : n N . Hence [0, 1] is uncountable. { n ∈ } [Note: A decimal expansion need not be unique: for instance, 0, 5999 . . . = 0, 6000 . . .. However, (n) this makes no harm, because in (0.12) b = an 2.] n ± Infinite sums. Let = be an arbitrary index set and a 0 α . Question: What does the sum A 6 ∅ α ≥ ∀ ∈A
aα αX∈A mean? Define a = sup a finite . α { α |A0 ⊂A } αX∈A αX∈A0 We will return to this a bit later. Spring 2017 7
0.13 Euclidean space Rn
n times Rn = R R Cartesian product ×···× The elements are called points or vectorsz }|. {
x Rn x = (x ,...,x ), x R, j = 1,...,n. ∈ ⇐⇒ 1 n j ∈ Algebraic structure. The sum of points x,y Rn is ∈ x + y = (x + y ,...,x + y ) Rn. 1 1 n x ∈ The product of a real number λ R and a point x Rn is ∈ ∈ λx = (λx , . . . , λx ) Rn. 1 n ∈ Zero vector 0= 0¯ = (0,..., 0). The inverse element (point) of x Rn is ∈ x = ( 1)x = ( x ,..., x ). − − − 1 − n The difference of x Rn and y Rn is ∈ ∈ x y = x + ( y). − − In Rn the addition and multiplication by a real number satisfy the axioms of a vector space, for example
x + y = y + x, x +0=0+ x = x, λ(x + y)= λx + λy, (λ + µ)x = λx + µx etc x,y Rn, λ, µ R. ∀ ∈ ∈ The inner product of x,y Rn is ∈ n x y = x y R. · i i ∈ Xi=1 Denote n 1/2 x = √x x = xixi norm of x. | | · ! Xi=1 The Euclidean distance in Rn. The distance between x,y Rn is ∈ n 1/2 x y = x y 2 . | − | i − i i=1 ! X 8 Measure and integral
Often we write d(x,y) = x y . Then d is a metric in Rn, i.e. the mapping d: Rn Rn R | − | × → satisfies the axioms of a metric:
d(x,y) 0 x,y Rn ≥ ∀ ∈ d(x,y) = 0 x = y ⇐⇒ d(x,y)= d(y,x) x,y Rn ∀ ∈ d(x,y) d(x, z)+ d(z,y) x,y,z Rn (triangle inequality, -ie). ≤ ∀ ∈ △ Open sets and closed sets in Rn. The Euclidean metric d determines open and closed sets of Rn (and hence the topology of Rn) as follows: Let x Rn and r> 0. The set ∈ B(x, r)= y Rn : y x < r { ∈ | − | } is an open ball with the center x and radius r and
S(x, r)= y Rn : y x = r { ∈ | − | } is the sphere (centered at x and with radius r. Similarly,
B¯(x, r)= y Rn : y x r { ∈ | − |≤ } is a closed ball (centered at x with radius r). A set V Rn is open if x V r = r(x) > 0 such that B(x, r) V . ⊂ ∀ ∈ ∃ ⊂ A set V Rn is closed is Rn V is open. ⊂ \ S(x,r) r −|x − y| > 0
r B(x,r) y x
Example. 1. B(x, r) is open x Rn,r > 0 ( -ie, see the picture above). ∀ ∈ △ 2. A closed ball B¯(x, r) is a closed set.
3. Rn and are both open and closed. ∅ 4. A half open interval, e.g. [0, 1), is neither open nor closed. Remark. The closure of a set A Rn is ⊂ A = x Rn : x A or x is an accumulation (or a cluster) point of A . { ∈ ∈ } Recall that x Rn is an accumulation point of A Rn if r> 0 B(x, r) (A x ) = . In Rn it ∈ ⊂ ∀ ∩ \ { } 6 ∅ holds that B¯(x, r)= B(x, r). Remark. If (X, d) is a metric space, i.e. d: X X R satisfies the axioms of a metric, we can × → define open and closed sets of X by using th emetric d as in the case of Rn by replacing y x | − | with the metric d(x,y). Spring 2017 9
The following result holds in general: Theorem 0.14.
(0.15) V Rn open α (arbitrary index set) V open; α ⊂ ∀ ∈A ⇒ α α[∈A
(0.16) V Rn closed α V closed; α ⊂ ∀ ∈A⇒ α α\∈A
k (0.17) V ,...,V Rn open V open; 1 k ⊂ ⇒ j j\=1
k (0.18) V ,...,V Rn closed V closed. 1 k ⊂ ⇒ j j[=1 Proof. (0.15):
x V α s.t. x V , ∈ α ⇒∃ 0 ∈A ∈ α0 α[∈A V open open ball B(x, r) V V . α0 ⇒∃ ⊂ α0 ⊂ α α[∈A (0.16):
V closed α V c open α α ∀ ⇒ α ∀ (0.15) de Morgan = V c = V c open ⇒ α α α α [ \ V closed. ⇒ α α \ (0.17) and (0.18): (Exerc.).
Remark. ∞ V open j N V open, j ∀ ∈ 6⇒ j j\=1 ∞ V closed j N V closed. (Exerc.) j ∀ ∈ 6⇒ j j[=1 1 Lebesgue measure in Rn
1.1 Introduction A geometric starting point: If I = [a, b] R is a bounded interval, its length is ⊂ ℓ(I)= b a. − 10 Measure and integral
(Similarly if I is an open or half open interval.) A set I Rn is an n-interval if it is of the form ⊂ I = I I , 1 ×···× n where each I R is an interval (either open, closed, or half open). j ⊂
b2
I
a2
a1 b1
An n-interval I is an open (respectively closed) n-interval if each Ij is open (resp. closed). Let Ij has the end points aj, bj; aj < bj. Then the geometric measure of I is n ℓ(I) = (b a )(b a ) (b a )= (b a ) 1 − 1 2 − 2 · · · n − n j − j jY=1 (n = 1 length, n = 2 area, n = 3 volume). Define ℓ( ) = 0. ∅ Our goal would be to define a ”measure” as a mapping m : (Rn) [0, + ], n P → ∞ such that it satisfies the conditions: (1) m (E) is defined E Rn and m (E) 0. n ∀ ⊂ n ≥ (2) If I is an n-interval, then mn(I)= ℓ(I). (3) If (E ) is a sequence of disjoint subsets of Rn (i.e. E E = if j = k), then k j ∩ k ∅ 6 ∞ m ∞ E = m (E ) countably additivity. n ∪k=1 k n k k=1 X (4) mn is translation invariant, i.e.
mn(E + x)= mn(E), where E Rn, x Rn, and E + x = y + x y E . ⊂ ∈ { | ∈ } It turns out that there exists no such mapping that would satisfy all the conditions (1) – (4) simultaneously. In the case of the (n-dimensional) Lebesgue measure mn we drop the condition (1). Hence m : Leb Rn [0, + ], n → ∞ will be a mapping that satisfies the conditions (2), (3) and (4), where Leb Rn ( (Rn) P is the family of Lebesgue measurable sets. The family Leb Rn contains, for instance, all open and closed subsets of Rn. Spring 2017 11
1.2 The Lebesgue outer measure in Rn
Convention.
a + = + a = , a = ∞ ∞ ∞ 6 −∞ a = + a = , a = −∞ −∞ −∞ 6 ∞ , + not defined ∞−∞ −∞ ∞ ( )= , ( )= − ∞ −∞ − −∞ ∞
, a> 0 ∞ a = a = , a< 0 ∞ · ·∞ −∞ 0, a = 0 Note! 0 = 0 ·∞ , a> 0 −∞ ( )a = a( )= + , a< 0 −∞ −∞ ∞ 0, a = 0
= ( )( )= ∞·∞ −∞ −∞ ∞ ( ) = ( )= −∞ ∞ ∞ −∞ −∞
, a> 0 a ∞ = , a< 0 0 −∞ not defined , a = 0 a a = = 0, a R ∈ ∞ −∞ ±∞ not defined ±∞
Recall: If (a ) N is a sequence such that a 0 j, then either j j∈ j ≥ ∀ ∞ k ∞ aj = lim aj R or aj =+ . k→∞ ∈ ∞ Xj=1 Xj=1 Xj=1 k Reason: partial sums j=1 aj form an increasing sequence. Let A Rn. Consider countable open covers of A (possibly finite) ⊂ P = I ,I ,... , F { 1 2 } where each I Rn is a bounded open n-interval (or ) and k ⊂ ∅ ∞ A I . ⊂ k k[=1 12 Measure and integral
A
Then we say that is a Lebesgue cover of A. We form a series F ∞ S( )= ℓ(I ), 0
m∗ (A) = inf S( ): is a Lebesgue cover of A . n { F F } (Later we will prove that closed n-intervals would work as well.) Remark. 1. Denote J = x Rn : x < k j (open n-interval). Clearly k { ∈ | j| ∀ } ∞ n R = Jk, k[=1 and therefore always there exist open covers ∞ I A (and hence inf exists). ∪k=1 k ⊃ 2. I Rn open n-interval 0 ℓ(I ) < the sum is well-defined and k ⊂ ⇒ ≤ k ∞ ⇒ ∞ 0 ℓ(I ) + . ≤ k ≤ ∞ Xk=1
3. The outer measure mn(A) depends (of course) on the dimension n. If n is clear from the ∗ ∗ context, we abbreviate m (A)= mn(A). 4. It follows directly from the definition that ε > 0 there exists a Lebesgue cover of A ∀ F (usually depending on ε) such that
S( ) m∗(A)+ ε. F ≤ (We allow m∗(A)=+ .) Note that it is usually not possible to find a Lebesgue cover of ∞ F A for which m∗ (A)= S( ). n F 5. Thus A m∗(A) is a mapping (Rn) [0, ], in particular, m∗ is defined in the whole 7→ P → ∞ (Rn). P Example. 1. Let n = 2 and let A = (x, 0): a x b R2 (a line segment in the plane). ∗ { ≤ ≤ }⊂ Claim: m2(A) = 0. Proof: Let ε> 0 and I =]a ε, b + ε[ ] ε, ε[ R2 an open 2-interval. ε − × − ⊂ A I 0 m∗(A) ℓ(I ) = 2ε(b a + 2ε) ε→0 0, ⊂ ε ⇒ ≤ 2 ≤ ε − −−−→ ∗ hence m2(A) = 0. Spring 2017 13
2. Let n = 1. Consider the set of rational numbers Q R. ∗ ⊂ Claim: m1(Q) = 0. Proof Since Q is countable, we may write Q = q : j N . Let ε> 0 be arbitrary. For each { j ∈ } j N let ∈ ε ε I = q ,q + R j j − 2j+1 j 2j+1 ⊂ j+1 j be an open interval. Its length is ℓ(Ij) = 2ε/2 = ε/2 .
q I j N Q I j ∈ j ∀ ∈ ⇒ ⊂ j ⇒ [j ∞ ∞ ∞ ε 1 ε→0 0 m∗(Q) ℓ(I )= = ε = ε 0, ≤ 1 ≤ j 2j 2j −−−→ Xj=1 Xj=1 Xj=1
∗ hence m1(Q) = 0. 3. Similarly, A Rn countable m∗ (A) = 0. ⊂ ⇒ n 4. Let A Rn be a bounded set, that is R> 0 such that A B(0,R). Then A I, where ⊂ ∃ ⊂ ⊂ n times I = ] R,R[ ] R,R[ open n-interval. − ×···× − R
R A
We get an estimate m∗(A) ℓ(I) = (2R)n. ≤ Basic properties of the (Lebesgue) outer measure.
Theorem 1.3. (1) m∗ ( ) = 0; n ∅ (2) ”monotonicity”: A B m∗ (A) m∗ (B); ⊂ ⇒ n ≤ n (3) ”subadditivity”: A , A ,... Rn 1 2 ⊂ ⇒ ∞ ∞ m∗ A m∗ (A ). n j ≤ n j j=1 j=1 [ X Remark. (3) holds also for finite unions k (A ) (choose A = = ). ∪j=1 j k+1 · · · ∅ Proof. (1): Clear. 14 Measure and integral
(2): Let be a Lebesgue cover of B. F A B is also a Lebesgue cover of A definition= m∗ (A) S( ). ⊂ ⇒ F ⇒ n ≤ F Take the inf over all Lebesgue covers of B m∗ (A) m∗ (B). ⇒ n ≤ n (3): Denote A = A . Let ε > 0. For each j choose a Lebesgue cover = I ,I . . . of A ∪j j Fj { j1 j2 } j such that S( ) m∗ (A )+ ε/2j . Fj ≤ n j Now = = I : j N, k N is a Lebesgue cover of A, hence (by definition) F j Fj { jk ∈ ∈ } S ∞ ∞ ∞ ∞ m∗ (A) S( )= S( ) m∗ (A )+ ε/2j = m∗ (A )+ ε. n ≤ F Fj ≤ n j n j Xj=1 Xj=1 Xj=1 Xj=1 Letting ε 0 we get the claim. →
Remark. Above we need some facts on ”summing” (more precisely, why S( ) = ∞ S( ))? F j=1 Fj See Lemma 1.7 and 1.8 below. P Theorem 1.4. Let A Rn. Then ⊂ ∗ ∗ (1.5) mn(A + x)= mn(A) for all x Rn, where A + x = y + x: y A ; ∈ { ∈ } ∗ n ∗ (1.6) mn(tA)= t mn(A), whenever t> 0 and tA = ty : y A . { ∈ } Proof (Exerc.) On summing. Let I be an (index) set and a 0 i I. If J I is finite, we denote i ≥ ∀ ∈ ⊂
SJ = ai , S∅ = 0. Xi∈J Definition. a = sup S : J I finite . i { J ⊂ } Xi∈I Lemma 1.7. n ai = lim ai. n→∞ Xi∈N Xi=1 That is, this ”new” definition coincide with the usual one (for countable sums).
Proof Denote J = 1,...,n , S = a (= sup S : J N finite ). n { } i∈N i { J ⊂ } P ′ SJ increasing sequence lim SJ = S n ⇒ ∃ n→∞ n S S S′ S. Jn ≤ ⇒ ≤ Spring 2017 15
On the other hand,
J N finite n N s.t. J J ⊂ ⇒ ∃ ∈ ⊂ n S S S′ ⇒ J ≤ Jn ≤ S S′ (taking sup over J). ⇒ ≤ ∀
Next both I and J are arbitrary index sets (i.e. they may be uncountable). (In addition, we abbreviate aij = a(i,j).) Lemma 1.8. aij = aij = aij. (i,jX)∈I×J Xi∈I Xj∈J Xj∈J Xi∈I Proof Denote by Svas the sum on the left hand side, by Skes the sum in the middle, and by Soik the sum on the right hand side. (a): If I J is finite, then finite I′ I, J ′ J s.t. I′ J ′ A⊂ × ∃ ⊂ ⊂ A⊂ × (∗) SA SI′×J′ = aij aij Skes ⇒ ≤ ′ ′ ≤ ′ ≤ Xi∈I jX∈J Xi∈I Xj∈J S S (taking sup over ). ⇒ vas ≤ kes ∀ A [( ): there is only finitely many terms in S ′ ′ , so the order of summing does not matter.] ∗ I ×J (b): Let I′ I be finite and J ′ J be finite i I′. Denote ⊂ i ⊂ ∀ ∈ = (i, j): i I′, j J ′ . A { ∈ ∈ i} Then S S = a . vas ≥ A ij i∈I′ ′ X jX∈Ji Take ( i I′) the sup over finite J ′ J ∀ ∈ i ⊂
Svas aij ≥ ′ Xi∈I Xj∈J sup over finite I′ I S S . ⊂ ⇒ vas ≥ kes
Similarly, Svas = Soik. Corollary 1.9. aij = aij = aij . (i,jX)∈N×N Xi∈N Xj∈N Xj∈N Xi∈N Remark. The subadditivity does not (in general) hold in the form
(1.10) m∗ A m∗ (A ), n i ≤ n i i∈I i∈I [ X where A Rn, i I, and I is an uncountable index set. Reason: i ⊂ ∈ n ∗ n R = x , mn( x ) = 0 x R . n{ } { } ∀ ∈ x[∈R 16 Measure and integral
If (1.10) would hold, then
(1.10) 0 m∗ (Rn)= m∗ x m∗ ( x ) = 0. ≤ n n { } ≤ n { } x∈Rn x∈Rn [ X On the other hand, we will prove later that m∗ (Rn)=+ . This is a contradiction, so (1.10) does n ∞ not hold!
1.11 (Lebesgue )measurable sets We will define the (Lebesgue) measurable sets of Rn, denoted by Leb Rn, by using so-called Carath´eodory’s condition. Recall the subadditivity (Theorem 1.3 (3)): A, B Rn ⊂ ⇒ m∗(A B) m∗(A)+ m∗(B). ∪ ≤ Later we will prove that A, B Rn s.t. A B = , but ∃ ⊂ ∩ ∅ m∗(A B) A E Definition. (Carath´eodory’s condition, 1914.) A set E Rn is (Lebesgue) measurable if ⊂ m∗(A)= m∗(A E)+ m∗(A E ) for all A Rn. ∩ \ ⊂ =A∩Ec Remark. E Rn measurable | {z } ⊂ ⇐⇒ m∗(A) m∗(A E)+ m∗(A E) for all A Rn, with m∗(A) < . ≥ ∩ \ ⊂ ∞ Reason: follows from the subadditivity and holds always if m∗(A)=+ . ≤ ≥ ∞ Definition. If E Rn is measurable, we denote ⊂ ∗ m(E)= m (E) or mn(E) if needed. m(E) is the (n-dimensional Lebesgue) measure of E. We write Leb Rn = E Rn : E Lebesgue measurable (Rn). { ⊂ } ⊂ P Spring 2017 17 Hence m = m∗ Leb Rn : Leb Rn [0, ], restriction of the outer measure. | → ∞ Later we will show that Leb Rn ( (Rn). P Theorem 1.12. m∗(E) = 0 E measurable. ⇒ Proof. Let A Rn be an arbitrary test set. ⊂ monotonicity A E E = m∗(A E) = 0 ∩ ⊂ ⇒ ∩ monotonocity A A E = m∗(A) m∗(A E)= m∗(A E) +m∗(A E) ⊃ \ ⇒ ≥ \ ∩ \ =0 E measurable. ⇒ | {z } Theorem 1.13. E measurable Ec measurable. ⇐⇒ Proof. It s enough to show : Let E be measurable and A Rn. Then ⇒ ⊂ m∗(A)= m∗(A E)+ m∗(A Ec) ∩ ∩ = m∗ A (Ec)c + m∗(A Ec) ∩ ∩ Ec measurable. ⇒ Example. Ex. 3 E Rn countable = m∗(E) = 0 ⊂ ⇒ Thm.= 1.12 E measurable Thm.= 1.13 Ec measurable. ⇒ ⇒ Special cases: Leb R, R Leb R, ∅∈ ∈ rational numbers Q Leb R, irrational numbers R Q Leb R. ∈ \ ∈ Let E1, E2,... be measurable. We will prove that ∞ ∞ Ei and Ei are measurable. i[=1 i\=1 To prove these statements we need some auxiliary lemmata. First the case of a finite union/intersection: Lemma 1.14. E ,...,E measurable k E and k E measurable. 1 k ⇒ i=1 i i=1 i S T 18 Measure and integral Proof. (a) union: k k−1 Ei = Ei Ek ! ∪ i[=1 i[=1 we may assume k = 2. ⇒ Suppose E and E are measurable. Let A Rn be a test set. 1 2 ⊂ E measurable 1 ⇒ m∗(A)= m∗(A E )+ m∗(A Ec) ∩ 1 ∩ 1 = c ⇒ E2 measurable, with test set A E1 ∗ c ∗ c ∗ ∩ c⇒ c m (A E1)= m (A E1 E2)+ m (A E1 E2) ∩ ∩ ∩ ∩ ∩ m∗(A)= m∗(A E )+ m∗(A Ec E ) +m∗(A Ec Ec), ∩ 1 ∩ 1 ∩ 2 ∩ 1 ∩ 2 (subadd. ⇒) ≥ m∗(B) where | {z } B = (A E ) (A Ec E )= A E (Ec E ) = A E (E E ) ∩ 1 ∪ ∩ 1 ∩ 2 ∩ 1 ∪ 1 ∩ 2 ∩ 1 ∪ 2 \ 1 = A (E E ). ∩ 1 ∪ 2 Hence m∗(A) m∗(B)+ m∗(A Ec Ec) ≥ ∩ 1 ∩ 2 = m∗ A (E E ) + m∗ A (E E )c ∩ 1 ∪ 2 ∩ 1 ∪ 2 E E measurable. ⇒ 1 ∪ 2 B E1 E2 A c A ∩ E1 (b) intersection: de Morgan, Theorem 1.13 (”measurability of the complement”) and part (a) ⇒ k k c c Ei = Ei measurable. ! i\=1 i[=1 Theorem 1.15. E , E measurable E E measurable. 1 2 ⇒ 1 \ 2 Proof. E E = E Ec. 1 \ 2 1 ∩ 2 Spring 2017 19 Lemma 1.16. Let E ,...,E be disjoint and measurable, and let A Rn be an arbitrary set. 1 k ⊂ Then k k m∗ A E = m∗(A E ). ∩ i ∩ i i=1 i=1 [ X E4 E1 E3 A E2 Proof. (a) The case k = 2 : E measurable, A (E E )= B as the test set 1 ∩ 1 ∪ 2 ⇒ m∗(B)= m∗(B E )+ m∗(B E ) ∩ 1 \ 1 =A∩E1 =A∩E2 | {z } | {z } = m∗(A E )+ m∗(A E ) i.e. the claim. ∩ 1 ∩ 2 (b) general case: By induction: Suppose that the claim holds for 2 k p, that is ≤ ≤ E1,...,Ep measurable p p E E = , i = j m∗ A E = m∗(A E ). i ∩ j ∅ 6 ⇒ ∩ i ∩ i A Rn i=1 i=1 ⊂ [ X Thus we get (for k = p + 1) p+1 p A i=1 Ei = A i=1 Ei Ep+1 ∩ ∩ ∪ = p S S ⇒ i=1 Ei, Ep+1 disjoint and measurable p+1 p S k=2 m∗ A E = m∗ A E + m∗(A E ) ∩ i ∩ i ∩ p+1 i=1 i=1 [ p [ k=p = m∗(A E )+ m∗(A E ) ∩ i ∩ p+1 Xi=1 p+1 = m∗(A E ). ∩ i Xi=1 ∞ Lemma 1.17. Let E = i=1 Ei, where the sets Ei are measurable. Then there exist disjoint and measurable sets Fi Ei s.t. ⊂ S ∞ E = Fi. i[=1 20 Measure and integral Proof. Choose F1 = E1, [measurable] F = E E , [measurable (Thm. 1.15)] 2 2 \ 1 . . k−1 F = E E , [measurable (Thm. 1.15 and L. 1.14)] k k \ i i[=1 . . F2 = E1 = F1 E2 E3 F3 = Then clearly ∞ F E i, E = F and F F = i = j. i ⊂ i ∀ i i ∩ j ∅ ∀ 6 i[=1 The main result of Lebesgue measurable sets Theorem 1.18. Let E1, E2,... be a sequence (possibly finite) of measurable sets. Then the sets Ei and Ei [i \i are measurable. If, in addition, the sets Ei are disjoint, then (1.19) m Ei = m(Ei). (”countably additivity”) i i [ X Proof. Denote 1.17 S = Ei = Fi, Fi measurable and disjoint, [i [i k S = F , S S. k i k ⊂ [i L. 1.14 (measurability of finite unions) S measurable. Let A be a test set. Then ⇒ k m∗(A)= m∗(A S )+ m∗(A S ) ∩ k \ k monot. m∗(A S )+ m∗(A S) ≥ ∩ k \ k 1=.16 m∗(A F )+ m∗(A S) k N. ∩ i \ ∀ ∈ Xi=1 Spring 2017 21 Letting k we get →∞ ∞ (1.20) m∗(A) m∗(A F )+ m∗(A S) ≥ ∩ i \ Xi=1 subadd. m∗ ∞ (A F ) + m∗(A S) ≥ ∪i=1 ∩ i \ = m∗(A S)+ m∗(A S) ∩ \ S = E measurable. ⇒ i [i Inequality (1.20), in the case A = S, and the subadditivity ⇒ =0 ∞ ∞ =Fi ∞ subadd. (1.20) m(F ) m(S) m∗(S F )+ m∗(S S)= m(F ). i ≥ ≥ ∩ i \ i Xi Xi=1 z }| { z }| { Xi=1 If Ei are disjoint, we may choose Fi = Ei, and therefore (1.19) holds. c c The first part of the proof and Thm. 1.13 imply that i Ei = i Ei is measurable. T S Example. Let A R2 s.t. ⊂ (1.21) m∗ A B(x, r) x r3 x R2, r> 0. ∩ ≤ | | ∀ ∈ ∀ Claim: m(A) = 0 Proof. (a) Suppose first that A is bounded, so A Q = [ a, a] [ a, a] (closed square) for ⊂ − × − some a. Let n N. Devide Q into closed (sub-)squares Q , with side length = 2a/n, j = 1,...,n2. ∈ j Let xj be the center of Qj. Then x 2a and Q B(x , 2a/n) (rough estimates) | j|≤ j ⊂ j monot. (1.21) m∗(A Q ) m∗ A B(x , 2a/n) x (2a/n)3 (2a)4n−3. ⇒ ∩ j ≤ ∩ j ≤ | j| ≤ n2 subadd. A = (A Q ) = ∩ j ⇒ j[=1 n2 n2 m∗(A)= m∗ (A Q ) m∗(A Q ) ∩ j ≤ ∩ j j=1 j=1 [ X n2(2a)4n−3 = (2a)4n−1 n ≤ ∀ n=→∞ m∗(A) = 0 m(A) = 0. ⇒ ⇒ (b) General case: A = A , where A = A B(0, j) bounded. j j ∩ j[∈N (a) A A A satisfies the assumption (1.21) m(A ) = 0 j j ⊂ ⇒ j ⇒ j ∀ subadd. = m(A) = 0. ⇒ 22 Measure and integral 1.22 Examples of measurable sets So far we know that: m∗(A) = 0 A and Ac measurable. ⇒ Now we will prove that, for example, open sets and closed sets are measurable. First: I Rn n-interval (open, closed, etc.) I is measurable and m(I)= ℓ(I). ⊂ ⇒ We use (Riemann) integration: Let I = I I Rn n-interval, where I R is an interval, with end points a < b , j = 1 ×···× n ⊂ j ⊂ j j 1,...,n. Let χ : Rn 0, 1 (the characteristic function of I) I → { } 1, x I χI (x)= ∈ (0, x I . 6∈ Choose an n-interval Q I and (Riemann) integrate ⊃ b1 bn χ = 1 dx dx = (b a ) (b a )= ℓ(I). I · · · 1 · · · n 1 − 1 · · · n − n ZQ Za1 Zan Lemma 1.23. Let I and I ,...,I be n-intervals s.t. I k I . Then ℓ(I) k ℓ(I ). If, 1 k ⊂ j=1 j ≤ j=1 j furthermore, the intersections Ii Ij, i = j, do not have interior points (i.e. no Ii Ij, i = j, ∩k 6 k S P∩ 6 contains an open ball) and I = j=1 Ij, then ℓ(I)= j=1 ℓ(Ij). Proof. Define χ,χ : Rn 0, 1S, P j → { } 1, x I 1, x Ij χ(x)= ∈ and χj(x)= ∈ 0, x I 0, x I . ( 6∈ ( 6∈ j Then it follows from the assumption I k I that χ(x) k χ (x) x Rn. Choose an ⊂ j=1 j ≤ j=1 j ∀ ∈ n-interval Q that contains all the n-intervals mentioned above and (Riemann) integrate over Q S P ℓ(I)= χ χ = χ = ℓ(I ). ≤ j j j ZQ ZQ j j ZQ j X X X k If the n-intervals Ij do not have common interior points, then χ(x)= j=1 χj(x) except possible on the boundaries of n-intervals that do not contribute to the integrals. P Lemma 1.24. If I is an n-interval, then m∗(I)= ℓ(I). Proof. (a): ε> 0 an open n-interval J I s.t. ℓ(J) <ℓ(I)+ ε. ∀ ∃ ⊃ J Leb. cover of I m∗(I) ℓ(I)+ ε { } ⇒ ≤ ε> 0 arbitr. m∗(I) ℓ(I). ⇒ ≤ Spring 2017 23 (b): Suppose first that I is closed. Let be a Lebesgue cover of I. Since I is closed and bounded, F I is compact. So a finite subcover = I ,...,I . Lemma 1.23 ∃ F0 { 1 k} ⊂ F ⇒ ℓ(I) S( ) S( ) ≤ F0 ≤ F inf over ℓ(I) m∗(I). ∀F ⇒ ≤ Hence: ℓ(I) = m∗(I) if I is closed. Suppose then that I need not be closed. Let ε > 0. Now a ∃ closed n-interval I I s.t. ℓ(I ) >ℓ(I) ε. Thus c ⊂ c − monot. m∗(I) m∗(I )= ℓ(I ) >ℓ(I) ε ≥ c c − ε> 0 arbitr. m∗(I) ℓ(I). ⇒ ≥ n Remark. The above holds also for degenerate n-intervals I = I1 In R , where at least def. ∗ ×···× ⊂ one Ij is a singleton. Then ℓ(I) = 0= mn(I). Let A Rn, ε> 0 and let J , J ,... Rn be arbitrary n-intervals s.t. A ∞ J . For each ⊂ 1 2 ⊂ ⊂ i=1 i i open n-interval I J s.t. ℓ(I ) <ℓ(J )+ ε/2i. Now I ,I . . . is a Lebesgue cover of A, and ∃ i ⊃ i i i { 1 2 } S therefore m∗(A) ∞ ℓ(I ) ∞ ℓ(J )+ ε. (Recall a geometric series.) It follows that ≤ i=1 i ≤ i=1 i P P∞ ∞ m∗(A) = inf ℓ(J ): A J , J arbitrary n-interval . i ⊂ i i i=1 i=1 X [ Theorem 1.25. If I is an n-interval, then I is measurable and m(I)= ℓ(I). Proof. L. 1.24 it suffices to prove that I is measurable. Let A Rn be a test set. Claim: ⇒ ⊂ m∗(A) m∗(A I)+ m∗(A I). ≥ ∩ \ Let ε> 0. Then a Lebesgue cover of A by open n-intervals = I ,I ,... s.t. ∃ F { 1 2 } S( ) m∗(A)+ ε. F ≤ I =∆ ∆ 1 ×···× n ⇒ I =]a , b [ ]a , b [ j 1 1 ×···× n n ′ n-interval Ij Ij I = ]a1, b1[ ∆1 ]an, bn[ ∆n = ∩ ∩ ×···× ∩ ( . ∅ I I is not necessarily an n-interval but j \ I I = I′′ j \ j,k [k is a finite union of n-intervals s.t. the intersections I′ I′′ and I′′ I′′ , k = i, do not have interior j ∩ j,k j,k ∩ j,i 6 points. 24 Measure and integral A ′′ Ij,1 ′′ Ij,2 Ij ′ Ij I Lemma 1.23 and 1.24 ⇒ 1.23 ′ ′′ 1.24 ∗ ′ ∗ ′′ ℓ(Ij) = ℓ(Ij)+ ℓ(Ij,k) = m (Ij)+ m (Ij,k). Xk Xk Taking the sum over j ⇒ m∗(A)+ ε S( )= ℓ(I )= m∗(I′ )+ m∗(I′′ ) ≥ F j j j,k Xj Xj Xj Xk subadd. m∗ I′ + m∗ I′′ ≥ j j,k [j [j,k ⊃A∩I ⊃A\I monot. m∗(A| {z I})+ m∗(A| {zI).} ≥ ∩ \ Letting ε 0 m∗(A) m∗(A I)+ m∗(A I). → ⇒ ≥ ∩ \ Theorem 1.26. (Lindel¨of’s theorem) Let A Rn be an arbitrary set and ⊂ V A, α ⊃ α[∈A where the sets V Rn, α are open. Then there exists a countable sub-cover α ⊂ ∈A V A. αj ⊃ j[∈N Proof. Exerc. Theorem 1.27. Open subsets and closed subsets of Rn are measurable. Proof. (a) Let A be open. If x A, an open n-interval I(x) s.t. x I(x) A ( an open ball ∈ ∃ ∈ ⊂ ∃ B(x, r ) A and it contains an open n-interval). x ⊂ I(x): x A is an open cover of A. { ∈ } Lindel¨of countable sub-cover I(x ): j N ⇒ ∃ { j ∈ } A = I(x ) is a countable union of measurable sets ⇒ j j[∈N A is measurable. ⇒ (b) If A is closed, its complement Ac is open and hence measurable A = (Ac)c is measurable. ⇒ Spring 2017 25 Example. Let f : R2 R2 be continuous. Claim: fR2 is measurable. → Proof. 2 R = Aj, where Aj = B¯(0, j) si compact j[∈N f continuous fA compact ⇒ j fA closed fA measurable ⇒ j ⇒ j fR2 = fA fR2 measurable. j ⇒ j[∈N Recall: Let n,m 1. A mapping f : Rn Rm is continuous f −1U Rn is open open ≥ → ⇐⇒ ⊂ ∀ U Rm. ⊂ Rn Rm f − U f 1U If f : Rn Rm is continuous and C Rn is compact, then fC Rm is compact. Reason: → ⊂ ⊂ fC U open cover ⊂ i i[∈I C f −1U open cover ⇒ ⊂ i∈I i finite sub-cover S ⇒ ∃ C compact k k C f −1U fC U . ⊂ ij ⇒ ⊂ ij j[=1 j[=1 More general measurable sets, σ-algebras. sets F , F closed (e.g. Q, [a, b), (a, b]) Fσ i i i[∈N sets G , G open (e.g. R Q, [a, b), (a, b]) Gδ i i \ i\∈N sets A , A Fσδ j j ∈ Fσ i\∈N sets B , B Gδσ j j ∈ Gδ i[∈N etc. Definition. Let X be an arbitrary set. A family Γ (X) is a σ-algebra (”sigma-algebra”) of X ⊂ P if 26 Measure and integral (a) Γ; ∅∈ (b) A Γ X A Γ; ∈ ⇒ \ ∈ (c) A Γ, i N ∞ A Γ. i ∈ ∈ ⇒ i=1 i ∈ Remark. (1) If Γ is aSσ-algebra and A Γ, i N, then also A Γ since i ∈ ∈ i i ∈ T A = Ac c = Ac c Γ. i i i ∈ i i i=1 \ \ [ (2) We have proved: The family of Lebesgue measurable sets Leb Rn is a σ-algebra of Rn (The- orems 1.12, 1.13, 1.18). (3) (X) is the largest σ-algebra of X; , X is the smallest σ-algebra of X; A X (fixed) P {∅ } ⊂ , X, A, Ac is a σ-algebra of X. ⇒ {∅ } Definition. The family of Borel sets Bor Rn is the smallest σ-algebra of Rn that contains all closed sets. Existence: Denote = Γ: Γ is a σ-algebra of Rn, Γ contains closed sets . B { } \ (For instance Γ = (Rn) is a σ-algebra of Rn that contains all closed sets.) P is a σ-algebra since: B (a) ; ∅∈B (b) A Ac Γ Γ Ac ; ∈B ⇒ ∈ ∀ ⇒ ∈B (c) A A Γ Γ A . i ∈B ⇒ i∈N i ∈ ∀ ⇒ i∈N i ∈B The construction S is the smallestS σ-algebra of Rn that contains closed sets, and so ⇒ B Bor Rn = . B Open sets, closed sets, sets, sets, etc. are Borel sets. Fσ Gδ Theorem 1.28. Every Borel sets is measurable. Proof. The family of measurable sets Leb Rn is a σ-algebra and contains closed sets, and therefore Bor Rn Leb Rn. ⊂ Spring 2017 27 1.29 General measure theory Definition. Let Γ be a σ-algebra in X. A function µ: Γ [0, + ] is a measure in X if → ∞ (i) µ( ) = 0; ∅ (ii) A Γ, i N, disjoint µ ∞ A = µ(A ). ”countably additivity” i ∈ ∈ ⇒ i=1 i i∈N i The triple (X, Γ,µ) is a measure space