Appendix A. Measure and integration
We suppose the reader is familiar with the basic facts concerning set theory and integration as they are presented in the introductory course of analysis. In this appendix, we review them briefly, and add some more which we shall need in the text. Basic references for proofs and a detailed exposition are, e.g., [[ H a l 1 ]] , [[ J a r 1 , 2 ]] , [[ K F 1 , 2 ]] , [[ L i L ]] , [[ R u 1 ]] , or any other textbook on analysis you might prefer.
A.1 Sets, mappings, relations
A set is a collection of objects called elements. The symbol card X denotes the cardi- nality of the set X. The subset M consisting of the elements of X which satisfy the conditions P1(x),...,Pn(x) is usually written as M = { x ∈ X : P1(x),...,Pn(x) }.A set whose elements are certain sets is called a system or family of these sets; the family of all subsystems of a given X is denoted as 2X . The operations of union, intersection, and set difference are introduced in the standard way; the first two of these are commutative, associative, and mutually distributive. In a { } system Mα of any cardinality, the de Morgan relations , X \ Mα = (X \ Mα)and X \ Mα = (X \ Mα), α α α α are valid. Another elementary property is the following: for any family {Mn} ,whichis { } at most countable, there is a disjoint family Nn of the same cardinality such that ⊂ \ ∪ \ Nn Mn and n Nn = n Mn.Theset(M N) (N M) is called the symmetric difference of the sets M,N and denoted as M #N. It is commutative, M #N = N #M, and furthermore, we have M #N =(M ∪N)\(M ∩N)andM #N =(X\M)#(X\N)for any X ⊃ M∪N. The symmetric difference is also associative, M#(N#P )=(M#N)#P , and distributive with respect to the intersection, (M # N) ∩ P =(M ∩ P ) # (N ∩ P ). A family R is called a set ring if M # N ∈R and M ∩ N ∈R holds for any pair M,N ∈R. The relation M \ N =(M # N) ∩ M also gives M \ N ∈R, and this in turn implies ∅∈R and M ∪ N ∈R. If the symmetric difference and intersection are understood as a sum and product, respectively, then a set ring is a ring in the sense of the general algebraic definition of Appendix B.1. A.1.1 Example: Let J d be the family of all bounded intervals in Rd,d≥ 1. The family Rd, which consists of all finite unions of intervals J ⊂Jd together with the empty set, is a set ring, and moreover, it is the smallest set ring containing J d. As mentioned above, any R ∈Rd can be expressed as a finite union of disjoint bounded intervals.
595 596 Appendix A Measure and integration
A set ring R⊂2X is called a set field if it contains the set X (notice that in the terminology of Appendix B.1 it is a ring with a unit element but not an algebra). A set field A⊂ X ∞ ∈A { }∞ ⊂A 2 is called a σ–field if n=1 M n holds for any countable system Mn n=1 . ∞ ∈A A⊂ X De Morgan relations show that also n=1 Mn , and furthermore that a family 2 \ ∈A ∈A ∈A containing the set X is a σ–field iff X M for all M and n Mn for any X at most countable subsystem {Mn}⊂A. Given a family S⊂2 we consider all σ–fields A⊂2X containing S (there is at least one, A =2X ). Their intersection is again a σ–field containing S;wecallittheσ–field generated by S and denote it as A(S). A.1.2 Example: The elements of Bd := A(J d) are called Borel sets in Rd. In particular, all the open and closed sets, and thus also the compact sets, are Borel. The σ–field Bd is also generated by other systems, e.g., by the system of all open sets in Rd. In general, Borel sets in a topological space (X, τ) are defined as the elements of the σ–field A(τ). { }∞ ⊂ ⊃ A sequence Mn n=1 is nondecreasing or nonincreasing if Mn Mn+1 or Mn M Mn+1, respectively, holds for n =1, 2,.... A set family is monotonic if it contains { } the set n Mn together with any nondecreasing sequence Mn ,and n Mn together with any nonincreasing sequence {Mn}.Anyσ–field represents an example of a monotonic system. To any S there is the smallest monotonic system M(S) containing S and we M S ⊂AS R M R ∈ have ( ) ( ). If is a ring, the same is true for ( ); in addition, if M∈R M M(R), then M(R)isaσ–field and M(R)=A(R). A mapping (or map) f from a set X to Y is a rule, which associates with any x ∈ X a unique element y ≡ f(x) of the set Y ;wewrite f : X → Y and also x → f(x). If Y = R or Y = C the map f is usually called a real or a complex function, respectively. It is also often useful to consider maps which are defined on a subset Df ⊂ X only. The symbol f : X → Y must then be completed by specifying the set Df which is called the domain of f;wedenoteitalsoas D(f). If Df is not specified, it is supposed to coincide with X. The sets Ran f := { y ∈ Y : y = f(x),x∈ Df } and Ker f := { x ∈ Df : f(x)=0} are the range and kernel of the map f, respectively. A map f : X → Y is injective if f(x)=f(x) holds for any x, x ∈ X only if x = x; it is surjective if Ran f = Y . A map which is simultaneously injective and surjective is called bijective or a bijection. The sets X and Y have the same cardinality if there is a bijection f : X → Y with Df = X. The relation f = g between f : X → Y and g : X → Y means by definition Df = Dg and f(x)=g(x) for all x ∈ Df .If Df ⊃ Dg and f(x)=g(x) holds for all x ∈ Dg we say that f is an extension of g while g is a \ restriction of f to the set Dg;wewrite f ⊃ g and g = f | Dg.
A.1.3 Example: For any X ⊂ M we define a real function χM : χM (x)=1ifx ∈ M, χM (x)=0ifx ∈ X \ M; it is called the characteristic (or indicator) function of the X set M. The map M → χM is a bijection of the system 2 to the set of all functions f : X → R such that Ran f = {0, 1}. (−1) { ∈ ∈ } → ⊂ The set f (N):= x Df : f(x) N for given f : X Y and N Y is called (−1) (−1) the pull–back of the set N by the map f. One has f α∈I Nα = α∈I f (Nα)for any family {Nα}⊂Y , and the analogous relation is valid for intersections. Furthermore, (−1) (−1) (−1) (−1) f (N1 \ N2)=f (N1) \ f (N2)andf(f (N)=Ranf ∩ N. On the other hand, f (−1)(f(M)) ⊃ M; the inclusion turns to identity if f is injective. Y A.1.4 Example: Let f : X → Y with Df = X.Toaσ–field B⊂2 we can construct the family f (−1) := { f (−1)(N): N ∈B}, which is obviously again a σ–field. Similarly, if A.1 Sets, mappings, relations 597
A⊂2X is a σ–field, then the same is true for { N ⊂ Y : f (−1)(N) ∈A}. Hence for any family S⊂2Y we can construct the σ–field f (−1)(A(S)) ⊂ 2X and the latter coincides with the σ–field generated by f (−1)(S), i.e., f (−1)(A(S)) = A(f (−1)(S)). Given f : X → Y and g : Y → Z we can define the composite map g◦f : X → Z (−1) (−1) with the domain D(g◦f):=f (Dg)=f (Dg ∩ Ran f)by(g◦f)(x):=g(f(x)). We have (g◦f)(P )(−1) = f (−1)(g(−1)(P )) for any P ⊂ Z. If f : X → Y is injective, then for any y ∈ Ran f there is just one xy ∈ Df such that y = f(xy); the prescription y → g(y):=xy defines a map g : Y → X,whichis −1 −1 −1 called the inverse of f and denoted as f .Wehave D(f )=Ranf, Ran f = Df −1 −1 and f (f(x)) = x, f(f (y)) = y for any x ∈ Df and y ∈ Ran f, respectively. These relations further imply f (−1)(N)=f −1(N) for any N ⊂ Ran f.Oftenwehaveapairof mappings f : X → Y and g : Y → X and we want to know whether f is invertible and f −1 = g; this is true if one of the following conditions is valid:
(i) Dg =Ranf and g(f(x)) = x for all x ∈ Df (ii) Ran f ⊂ Dg, g(f(x)) = x for all x ∈ Df and Ran g ⊂ Df , f(g(y)) = y for all y ∈ Dg
If f : X → Y is injective, then f −1 is also injective and (f −1)−1 = f.If g : Y → Z is also injective, then the composite map g◦f is invertible and (g◦f)−1 = f −1◦g−1. The Cartesian product M × N is the set of ordered pairs [x, y]withx ∈ M and y ∈ N; the Cartesian product of the families S and S is defined by S×S := { M × N : M ∈S,N ∈S }. For instance, the systems of bounded intervals of Example 1 satisfy J m+n = J m ×Jn.If M × N is empty, then either M = ∅ or N = ∅. On the other hand, if M × N is nonempty, then the inclusion M × N ⊂ P × R implies M ⊂ P and N ⊂ R.Wehave(M ∪ P ) × N =(M × N) ∪ (P × N) and similar simple relations for the intersection and set difference. Notice, however, that (M × N)∪(P × R) can be expressed in the form S × T only if M = P or N = R. The definition of the Cartesian product extends easily to any finite family of sets. ×···× { }→ n Alternatively, we can interpret M1 Mn as the set of maps f : 1,...,n j=1 Mj ∈ { ∈ } such that f(j) Mj. This allows us to define Xα∈I Mα for a system Mα : α I of → ∈ any cardinality as the set of maps f : I α∈I Mα which fulfil f(α) Mα for any α ∈ I. The existence of such maps is related to the axiom of choice (see below). Given f : X → C and g : Y → C, we define the function f × g on X × Y by (f × g)(x, y):=f(x)g(y). Let M ⊂ X × Y ; then to any x ∈ X we define the x–cut of the set M by Mx := { y ∈ Y :[x, y] ∈ M}; we define the y–cuts analogously. Let A⊂2X , B⊂2Y be σ–fields; then the σ–field A(A×B) is called the direct product of the fields A and B and is denoted as A⊗B. A.1.5 Example: The Borel sets in Rm and Rn in this way generate all Borel sets in Rm+n, i.e.,wehave Bm ⊗Bn = Bm+n. On the other hand, the cuts of a set M ∈A⊗B belong to the original fields: we have Mx ∈B and My ∈A for any x ∈ X and y ∈ Y , respectively. A subset Rϕ ⊂ X × X defines a relation ϕ on X:if[x, y] ∈ Rϕ we say the element x is in relation with y and write xϕy. A common example is an equivalence,whichis a relation ∼ on X that is reflexive ( x ∼ x for any x ∈ X ), symmetric ( x ∼ y implies y ∼ x ), and transitive ( x ∼ y and y ∼ z imply x ∼ z ). For any x ∈ X we define the 598 Appendix A Measure and integration
equivalence class of x as the set Tx := { y ∈ X : y ∼ x }.Wehave Tx = Ty iff x ∼ y,so the set X decomposes into a disjoint union of the equivalence classes. Another important example is a partial ordering on X, which means any relation ≺ that is reflexive, transitive, and antisymmetric, i.e., such that the conditions x ≺ y and y ≺ x imply x = y.If X is partially ordered, then a subset M ⊂ X is said to be (fully) ordered if any elements x, y ∈ M satisfy either x ≺ y or x y. An element x ∈ X is an upper bound of a set M ⊂ X if y ≺ x holds for all y ∈ M;itisamaximal element of M if for any y ∈ M the condition y x implies y = x. A.1.6 Theorem (Zorn’s lemma): Let any ordered subset of a partially ordered set X have an upper bound; then X contains a maximal element. Zorn’s lemma is equivalent to the so–called axiom of choice, which postulates for a system { Mα : α ∈ I } of any cardinality the existence of a map α → xα such that xα ∈ Mα for all α ∈ I — see, e.g., [[ DS 1 ]], Sec.I.2, [[ Ku ]], Sec.I.6. Notice that the maximal element in a partially ordered set is generally far from unique.
A.2 Measures and measurable functions
Let us have a pair (X, A), where X is a set and A⊂2X a σ–field. A function f : X → R is called measurable (with respect to A )if f (−1)(J) ∈A holds for any bounded interval J ⊂ R, i.e., f (−1)(J ) ⊂A. This is equivalent to any of the following statements: (i) f (−1)((c, ∞)) ∈A for all c ∈ R, (ii) f (−1)(G) ∈A for any open G ⊂ R, (iii) f (−1)(B) ⊂ A.If X is a topological space, a function f : X → R is called Borel if it is measurable w.r.t. the σ–field B of Borel sets in X. A.2.1 Example: Any continuous function f : Rd → R is Borel. Furthermore, let f : X → R be measurable (w.r.t. some A )and g : R → R be Borel; then the composite function g◦f is measurable w.r.t. A. If functions f,g : X → R are measurable, then the same is true for their linear combinations af + bg and product fg as well as for the function x → (f(x))−1 provided f(x) =0 forall x ∈ X. Even if the last condition is not valid, the function h, defined by h(x):=(f(x))−1 if f(x) =0 and h(x) := 0 otherwise, is measurable. Furthermore, if a sequence {fn} converges pointwise, then the function x → limn→∞ fn(x) is again measurable. The notion of measurability extends to complex functions: a function ϕ : X → C is measurable (w.r.t. A ) if the functions Re ϕ(·)andImϕ(·) are measurable; this is true iff ϕ(−1)(G) ∈A holds for any open set G ⊂ C. A complex linear combination of measurable functions is again measurable. Furthermore, if ϕ is measurable, then |ϕ(·)| is also measurable. In particular, the modulus of a measurable f : X → R is measurable, ± 1 | |± as are the functions f := 2 ( f f). → C ∈ C A function ϕ : X is simple (σ–simple)if ϕ = n ynχMn , where yn and the ∈A sets Mn form a finite (respectively, at most countable) disjoint system with n Mn = X. By definition, any such function is measurable; the sets of (σ–)simple functions are closed with respect to the pointwise defined operations of summation, multiplication, and scalar multiplication. The expression ϕ = n ynχMn is not unique, however, unless the numbers yn are mutually different. A.2 Measures and measurable functions 599
A.2.2 Proposition: A function f : X → R is measurable iff there is a sequence {fn} of σ–simple functions, which converges to f uniformly on X.If f is bounded, there is a sequence of simple functions with the stated property.
In fact, the approximating sequence {fn} canbechoseneventobenondecreasing.If f is not bounded it can still be approximated pointwise by a sequence of simple functions, but not uniformly. Given (X, A)and(Y,B) we can construct the pair (X × Y,A⊗B). Let ϕ : M → C be a function on M ∈A⊗B; then its x–cut is the function ϕx defined on Mx by ϕx(y):=ϕ(x, y); we define the y–cut similarly. Cuts of a measurable functions may not be measurable in general, however, it is usually important to ensure measurability a.e. – cf. Theorem A.3.13 below. A mapping λ defined on a set family S and such that λ(M) is either non–negative or λ(M)=+∞ for any M ∈S is called (a non–negative) set function.Itismonotonic if M ⊂ N implies λ(M) ≤ λ(N), additive if λ(M ∪ N)=λ(M)+λ(N) for any pair of sets such that M ∪ N ∈S and M ∩ N = ∅,andσ–additive if the last property generalizes, { }⊂S λ ( n Mn)= n λ(Mn), to any disjoint at most countable system Mn such that ∈S n Mn . A set function µ, which is defined on a certain A⊂2X ,isσ–additive, and satisfies µ(∅) = 0 is called a (non–negative) measure on X. If at least one M ∈A has µ(M) < ∞, then µ(∅) = 0 is a consequence of the σ–additivity. The triplet (X, A,µ) is called a measure space; the sets and functions measurable w.r.t. A are in this case often specified as µ–measurable.Aset M ∈A is said to be µ–zero if µ(M) = 0, a proposition–valued function defined on M ∈A is valid µ–almost everywhere if the set N ⊂ M,onwhichit is not valid, is µ–zero. A measure µ is complete if N ⊂ M implies N ∈A for any µ–zero set M; below we shall show that any measure can be extended in a standard way to a complete one. Additivity implies that any measure is monotonic, and µ(M ∪ N)=µ(M)+µ(N) − µ(M ∩ N) ≤ µ(M)+µ(N) for any sets M,N ∈Awhich satisfy µ(M ∩ N) < ∞. ∞ Using the σ–additivity, one can check that limk→∞ µ(Mk)=µ ( n=1 Mn) holds for any nondecreasing sequence {Mn}⊂A, and a similar relation with the union replaced by intersection is valid for nonincreasing sequences. A measure µ is said be finite if µ(X) < ∞ ∞ ∈A ∞ and σ–finite if X = n=1 Mn, where Mn and µ(Mn) < for n =1, 2,.... Let (X, τ) be a topological space, in which any open set can be expressed as a count- able union of compact sets (as, for instance, the space Rd; recall that an open ball there is a countable union of closed balls). Suppose that µ is a measure on X with the domain A⊃τ; then the following is true: if any point of an open set G has a µ–zero neighborhood, then µ(G)=0. Given a measure µ we can define the function µ : A×A → [0, ∞)byµ(M ×N):= µ(M # N). The condition µ(M # N) = 0 defines an equivalence relation on A and µ is a metric on the corresponding set of equivalence classes. Apoint x ∈ X such that the one–point set {x} belongs to A and µ({x}) =0 is called a discrete point of µ; the set of all such points is denoted as Pµ.If µ is σ–finite the set Pµ is at most countable. A measure µ is discrete if Pµ ∈Aand µ(M)=µ(M ∩ Pµ) for any M ∈A. 600 Appendix A Measure and integration
A measure µ is said to be concentrated on a set S ∈A if µ(M)=µ(M ∩ S) for any M ∈A. For instance, a discrete measure is concentrated on the set of its discrete points. If (X, τ) is a topological space and τ ⊂A, then the support of µ denoted as supp µ is the smallest closed set on which µ is concentrated. Next we are going to discuss some ways in which measures can be constructed. First we shall describe a construction, which starts from a given non–negative σ–additive set X functionµ ˙ defined on a ring R⊂2 ; we assume that there exists an at most countable { }⊂R ∞ disjoint system Bn such that n Bn = X andµ ˙ (Bn) < for n =1, 2,.... Let S be the system of all at most countable unions of the elements of R; it is closed w.r.t. countable unions and finite intersections, and M \R ∈Sholds for all M ∈Sand R ∈ R ∈S { }⊂R .Any M can be expressed as M = j Rj, where Rj is an at most countable S disjoint system; using it we can define the set function µ ¨ on by µ ¨(M):= j µ˙ (Rj). It is ≤ monotonic and σ–additive. Furthermore, we haveµ ¨ ( n Mn) n µ¨(Mn); this property is called countable semiadditivity. Together with the monotonicity, it is equivalent to the ⊂ ∞ ≤ ∞ fact that M k=1 Mk impliesµ ¨(M) k=1 µ¨(Mk). The next step is to extend the functionµ ¨ to the whole system 2X by defining the outer measure by µ∗(A):=inf{ µ¨(M): M ∈S,M ⊃ A }. The outer measure is again monotonic and countably semiadditive; however, it is not additive so it is not a measure. Its importance lies in the fact that the system
∗ Aµ := {A⊂X :infµ (A # M)=0} M∈S
∗ \ is a σ–field. This finally allows us to define µ := µ | Aµ;itisacompleteσ–additive measure on the σ–field Aµ ⊃A(R), which is determined uniquely by the set functionµ ˙ in the sense that any measure ν on A(R), which is an extension ofµ ˙ , satisfies ν = µ |\ A(R). The measure µ is called the Lebesgue extension ofµ ˙ . A measure µ on a topological space (X, τ) is called Borel if it is defined on B≡A(τ) and µ(C) < ∞ holds for any compact set C. We are particularly interested in Borel measures on Rd, where the last condition is equivalent to the requirement µ(K) < ∞ for any compact interval K ⊂ Rd. Any Borel measure on Rd is therefore σ–additive and corresponds to a unique σ– additive set functionµ ˙ on Rd. The space Rd, however, has the special property that d for any bounded interval J ∈J we can find a nonincreasing sequence of open intervals ⊃ ⊂ In J and a nondecreasing sequence of compact intervals Kn J such that n In = n Kn = J. This allows us to replace the requirement of σ–additivity by the condition d d µ˙ (J)=inf{ µ˙ (I): I ∈GJ } =sup{ µ˙ (K): K ∈FJ } for any J ∈J , where GJ ⊂J is d the system of all open intervals containing J,and FJ ⊂J is the system of all compact intervals contained in J. A set functionµ ˜ on J d which is finite, additive, and fulfils the last condition is called regular. A.2.3 Theorem: There is a one–to–one correspondence between regular set functionsµ ˜ and µ := µ∗ |\ Bd on Rd. In particular, Borel measures µ and ν coincide if µ(J)=ν(J) holds for any J ∈Jd. A.2.4 Example: Let f : R → R be a nondecreasing right–continuous function. For any a, b ∈ R,a
1 J ; the corresponding Borel measure µf is called the Lebesgue–Stieltjes measure generated by the function f. In particular, if f is the identical function, f(x)=x,wespeak about the Lebesgue measure on R. Let us remark that the Lebesgue–Stieltjes measure is sometimes understood as a Lebesgue extension with a domain which is generally dependent on f; however, it contains B in any case. A.2.5 Example: Letµ ˜ andν ˜ be regular set functions on J m ⊂ Rm and J n ⊂ Rn, respectively; then the function ˜ on J m+n defined by ˜(J × L):=˜µ(J)˜ν(L) is again regular; the corresponding Borel measure is called the direct product of the measures µ and ν which correspond toµ ˜ andν ˜, respectively, and is denoted as µ ⊗ ν. In particular, repeating the procedure d times, we can in this way construct the Lebesgue measure on Rd which associates its volume with every parallelepiped. A.2.6 Proposition: Any Borel measure on Rd is regular, i.e., µ(B)=inf{µ(G): G ⊃ B, G open} =sup{µ(C): C ⊂ B, C compact}. As a consequence of this result, we can find to any B ∈Bd a nonincreasing sequence ⊃ ⊂ of open sets Gn B and a nondecreasing sequence of compact sets Cn B (both dependent generally on the measure µ ) such that µ(B) = limn→∞ µ(Gn)=µ ( n Gn)= limn→∞ µ(Cn)=µ ( n Cn). Proposition A.2.6 generalizes to Borel measures on a locally compact Hausdorff space, in which any open set is a countable union of compact sets – see [[Ru 1 ]], Sec.2.18. Let us finally remark that there are alternative ways to construct Borel measures. One can use, e.g., the Riesz representation theorem, according to which Borel measures correspond bijectively to positive linear functionals on the vector space of continuous functions with a compact support — cf. [[Ru 1 ]], Sec.2.14; [[RS 1 ]], Sec.IV.4.
A.3 Integration
Now we shall briefly review the Lebesgue integral theory on a measure space (X, A,µ). It is useful from the beginning to consider functions which may assume infinite values; this requires to define the algebraic operations a + ∞ := ∞, a ·∞:= ∞ for a>0and a ·∞ := 0 for a =0,etc., to add the requirement f (−1)(∞) ∈A to the definition of measurability, and several other simple modifications. Given a simple non–negative function s := n ynχMn on X, we define its integral by X sdµ:= n ynµ(Mn); correctness of the definition follows from the additivity of µ. In the next step, we extend it to all measurable functions f : X → [0, ∞] putting
fdµ ≡ f(x) dµ(x):=sup sdµ : s ∈ Sf , X X X → ∞ ≤ where Sf is the set of all simple functions s : X [0, ) such that s f.Wealso ∈A define M fdµ:= X fχM dµ for any M ; in this way we associate with the function f and the set M anumberfrom[0, ∞], which is called the (Lebesgue) integral of f over M w.r.t. the measure µ.
A.3.1 Proposition: Let f,g be measurable functions X → [0, ∞]andM ∈A; then ∈ ∞ ≤ M (kf) dµ = k M fdµ holds for any k [0, ), and moreover, the inequality f g ≤ implies M fdµ M gdµ. 602 Appendix A Measure and integration
Notice that the integral of f = 0 is zero even if µ(X)=∞. On the other hand, the | | → C { ∈ relation X ϕ dµ = 0 for any measurable function ϕ : X implies that µ( x X : ϕ(x) =0 })=0,i.e., that the function ϕ is zero µ–a.e. Let us turn to limits which play the central role in the theory of integration. { } A.3.2 Theorem (monotone convergence): Let fn be a nondecreasing sequence of non- negative measurable functions; then limn→∞ X fn dµ = X (limn→∞ fn) dµ. The right side of the last relation makes sense since the limit function is measurable. However, we often need some conditions under which both sides are finite. The correspond- ing modification is also called the monotone–convergence (or Levi’s) theorem: if {fn} is a nondecreasing sequence of non–negative measurable functions and there is a k>0such ≤ → that X fn dµ k for n =1, 2,..., then the function x f(x) := limn→∞ fn(x)is ≤ µ–a.e. finite and limn→∞ X fn dµ = X fdµ k. The monotone–convergence theorem implies, in particular, that the integral of a measurable function can be approximated by a nondecreasing sequence of integrals of simple functions. ≤ A.3.3 Corollary (Fatou’s lemma): X (liminf n→∞fn) dµ liminf n→∞ X fn dµ holds for any sequence of measurable functions fn : X → [0, ∞]. { } This result has the following easy consequence: let a sequence fn of non–negative ≤ measurable functions have a limit everywhere, limn→∞ fn(x)=f(x), and X fn dµ k ≤ for n =1, 2,...; then X fdµ k. Applying the monotone–convergence theorem to a se- { } ∞ quence fn of non–negative measurable functions we get the relation X ( n=1 fn) dµ = ∞ ∞ fn dµ. In particular, if f : X → [0, ∞] is measurable and {Mn} ⊂A is n=1 X ∞ n=1 a disjoint family with Mn = M, then putting fn := fχM we get fdµ = ∞ n=1 n M fdµ. This relation is called σ–additivity of the integral; it expresses the fact n=1 Mn that the function f together with the measure µ generates another measure.
A.3.4 Proposition: Let f : X → [0, ∞] be a measurable function; then the map M → A ν(M):= M fdµ is a measure with the domain ,and X gdν = X gf dµ holds for any measurable g : X → [0, ∞].
Let us pass to integration of complex functions. A measurable function ϕ : X → C | | ∞ is integrable (over X w.r.t. µ )if X ϕ dµ < (recall that if ϕ is measurable so is |ϕ| ). The set of all integrable functions is denoted as L(X, dµ); in the same way we define L(M,dµ) for any M ∈A. Given ϕ ∈L(X, dµ)wedenotef := Re ϕ and g := Im ϕ; then f ± and g± are non–negative measurable functions belonging to L(X, dµ). This allows us to define the integral of complex functions through the positive and negative parts of the functions f,g as the mapping ϕ −→ ϕdµ := f + dµ − f − dµ + i g+ dµ − i g− dµ X X X X X
d of L(X, dµ)toC. If, in particular, µ is the Lebesgue measure on R we often use the d d d symbol L(R )orL (R ,dx) instead of L(R ,dµ), and the integral is written as ϕ(x) dx, or occasionally as ϕ(x) dx. The above definition has the following easy consequence: if M ϕdµ =0 holdsfor ∈A ≥ ∈A all M , then ϕ(x)=0 µ–a.e. in X. Similarly M ϕdµ 0 for all M implies ϕ(x) ≥ 0 µ–a.e. in X; further generalizations can be found in [[ Ru 1 ]], Sec.1.40. The integral has the following basic properties: A.3 Integration 603