Bourbaki

§ 5. Measurable functions and sets

1. Definition of measurable functions and sets

Definition 1

Let XX be a locally compact space, μ\mu a measure on XX. A mapping ff of XX into a topological space FF is said to be measurable for the measure μ\mu (or to be μ\mu-measurable) if, for every compact subset KK of XX, there exist a μ\mu-negligible set N⊂KN \subset K and a partition of K−NK - N formed by a sequence (finite or infinite) (Kn)(K_n) of compact sets, such that the restriction of ff to each KnK_n is continuous.

It is clear that every continuous mapping of XX into FF is measurable.

Note that if μ\mu and ν\nu are two measures on XX such that every μ\mu-negligible set is ν\nu-negligible, then every μ\mu-measurable function is also ν\nu-measurable (cf. Ch. V, §5, Nos. 5, 6).

Definition 1 may be transformed into the following criterion:

Proposition 1

For a mapping ff of XX into FF to be measurable, it is necessary and sufficient that, for every compact set K⊂XK \subset X and every number ε>0\varepsilon > 0, there exist a compact set K1⊂KK_1 \subset K such that ∣μ∣(K−K1)≤ε|\mu|(K - K_1) \leq \varepsilon and the restriction of ff to K1K_1 is continuous.

If this condition is fulfilled, we may define recursively a sequence of pairwise disjoint compact sets Kn⊂KK_n \subset K such that ∣μ∣(K−⋃i=1nKi)≤1/n|\mu|(K - \bigcup_{i=1}^n K_i) \leq 1/n and such that the restriction of ff to each KnK_n is continuous (§4§ 4, No. 6, Th. 4); the complement with respect to KK of the union of the KnK_n is then negligible (§4§ 4, No. 5, Cor. of Prop. 7), thus ff is measurable. Conversely, suppose there exist a negligible set N⊂KN \subset K and a partition (Kn)(K_n) of K−NK - N formed of compact sets such that the restriction of ff to each KnK_n is continuous; for every ε>0\varepsilon > 0 there exists an integer nn such that, if H=⋃i=1nKiH = \bigcup_{i=1}^n K_i, then ∣μ∣(K−H)≤ε|\mu|(K - H) \leq \varepsilon (§ 4, No. 5, Cor. of Prop. 7); the set HH is compact, the KiK_i (1≤i≤n1 \leq i \leq n) form a finite partition of HH into compact sets, and the restriction of ff to each KiK_i is continuous; therefore the restriction of ff to HH is continuous.

Proposition 2

Let* (Fn)(F_n) be a sequence of topological spaces and, for each nn, let fnf_n be a measurable mapping of XX into FnF_n. For every compact set K⊂XK \subset X and every ε>0\varepsilon > 0, there exists a compact set K0⊂KK_0 \subset K such that ∣μ∣(K−K0)≤ε|\mu|(K - K_0) \leq \varepsilon and the restriction to K0K_0 of each of the functions fnf_n is continuous.

For each integer n≥1n \geq 1 there exists a compact set Kn⊂KK_n \subset K such that ∣μ∣(K−Kn)≤ε/2n|\mu|(K - K_n) \leq \varepsilon / 2^n and the restriction of fnf_n to KnK_n is continuous. The set K0=⋂n=1∞KnK_0 = \bigcap_{n=1}^\infty K_n is compact, the restrictions to K0K_0 of all of the functions fnf_n are continuous and, since K−K0K - K_0 is contained in the union of the K−KnK - K_n, ∣μ∣(K−K0)≤∑n=1∞ε/2n=ε|\mu|(K - K_0) \leq \sum_{n=1}^\infty \varepsilon / 2^n = \varepsilon.

Definition 2

A subset* AA of XX is said to be measurable if its characteristic function φA\varphi_A is measurable.

In view of Def. 1, it comes to the same to say that a measurable set AA is a set such that, for every compact set KK, there exist a negligible set N⊂KN \subset K and a partition (Kn)(K_n) of K−NK - N formed by a sequence of compact sets each of which is contained either in K∩AK \cap A or in K∩∁AK \cap \complement A.

This definition yields at once the following criterion:

Proposition 3

For a set* AA to be measurable, it is necessary and sufficient that, for every compact set KK, A∩KA \cap K be integrable.

The condition is necessary, because the union of a sequence of integrable sets AnA_n is integrable when ∑n∣μ∣(An)\sum_n |\mu|(A_n) is finite (§ 4, No. 5, Cor. of Prop. 8). The condition is sufficient because, for every integrable set BB, there exist a negligible set N⊂BN \subset B and a partition of B−NB - N into a sequence of compact sets (§ 4, No. 6, Cor. 2 of Th. 4).

Corollary 1

The open sets and the closed sets are measurable*.

In particular, the entire space XX is measurable.

Corollary 2

— If X is metrizable, then every Souslin subset A of X (GT, IX, §6, No. 2) is μ-measurable for every measure μ on X.

By virtue of Prop. 3, it suffices to verify that every relatively compact Souslin set A is μ-integrable. Now, such a set A is capacitable for ∣μ∣∗|\mu|^* (GT, IX, §6, No. 9, Th. 5). Therefore, for every ε>0\varepsilon > 0 there exists a compact subset K of A such that ∣μ∣∗(A)≤∣μ∣∗(K)+ε=∣μ∣(K)+ε|\mu|^*(A) \leq |\mu|^*(K) + \varepsilon = |\mu|(K) + \varepsilon. Let U be a relatively compact open set in X containing A such that

∣μ∣(U)=∣μ∣∗(U)≤∣μ∣∗(A)+ε. |\mu|(U) = |\mu|^*(U) \leq |\mu|^*(A) + \varepsilon.

Then ∣μ∣∗(U−K)=∣μ∣(U)−∣μ∣(K)≤2ε|\mu|^*(U - K) = |\mu|(U) - |\mu|(K) \leq 2\varepsilon, therefore ∣μ∣∗(A−K)≤2ε|\mu|^*(A - K) \leq 2\varepsilon, which proves that A is μ-integrable (§4, No. 6, Cor. 1 of Th. 4).

2. Principle of localization. Locally negligible sets

Proposition 4 (Principle of localization)

— *Let f be a mapping of X into a topological space F. Suppose that for every x∈Xx \in X, there exist an integrable neighborhood VxV_x of x and a measurable mapping gxg_x of X into F such that f(y)=gx(y)f(y) = g_x(y) almost everywhere in VxV_x. Then f is measurable.

Let K be a compact set; there exists a finite number of points xi∈Kx_i \in K such that the VxiV_{x_i} form a covering of K. It follows at once (§4, No. 9, Lemma) that there exist a negligible set N⊂KN \subset K and a finite partition of K−NK - N formed of integrable sets MjM_j such that each of the sets K∩VxiK \cap V_{x_i} is the union of a subset of N and a certain number of the MjM_j, and such that on each of the MjM_j, ff is equal to one of the functions gxig_{x_i}. Now, for each MjM_j there exist a negligible set Nj⊂MjN_j \subset M_j and a partition of Mj−NjM_j - N_j formed by a sequence of compact sets Knj (n∈N)K_{nj}\ (\mathbf{n} \in \mathbf{N}); on the other hand, for each KnjK_{nj} there exist a negligible set Pnj⊂KnjP_{nj} \subset K_{nj} and a partition of Knj−PnjK_{nj} - P_{nj} formed by a sequence of compact sets Kmnj (m∈N)K_{mnj}\ (\mathbf{m} \in \mathbf{N}) such that the restriction of ff to each of the KmnjK_{mnj} is continuous. Since the union of N, the NjN_j and the PnjP_{nj} is negligible, ff is measurable.

The concept of measurable function is therefore a concept of local character.

Definition 3

— *A set A⊂XA \subset X is said to be locally negligible (for the measure μ\mu) if, for every x∈Xx \in X, there exists a neighborhood VV of xx such that V∩AV \cap A is negligible.

By the principle of localization, every locally negligible set is measurable. The properties of negligible sets (§2) show that every subset of a locally negligible set is locally negligible, and that every countable union of locally negligible sets is locally negligible.

Proposition 5

— *For a set A to be locally negligible, it is necessary and sufficient that, for every compact set K, A∩KA \cap K be negligible.

The condition is obviously sufficient since every point of X has a compact neighborhood. It is necessary, because if, for every x∈Kx \in K, there exists a neighborhood VxV_x of xx such that A∩VxA \cap V_x is negligible, then there exists a finite number of points xi∈Kx_i \in K such that the VxiV_{x_i} form a covering of K, and A∩KA \cap K is contained in the union of the negligible sets A∩VxiA \cap V_{x_i}.

Corollary 1

— *For a set A to be negligible, it is necessary and sufficient that it be locally negligible and of finite outer measure.

The condition is obviously necessary. Conversely, if it is satisfied then A is contained in an integrable open set G, which is the union of a negligible set N and a sequence (Kn)(K_n) of compact sets (§ 4, No. 6, Cor. 2 of Th. 4); since A∩NA \cap N and the sets A∩KnA \cap K_n are negligible, the same is true of their union A.

Corollary 2

— Every locally negligible open set is negligible (and is therefore contained in the complement of the support of μ\mu).

For, the outer measure of an open set G is the supremum of the measures ∣μ∣(K)|\mu|(K) of the compact sets K⊂GK \subset G (§ 4, No. 6, Cor. 4 of Th. 4); if G is locally negligible then ∣μ∣(K)=0|\mu|(K) = 0 for every compact set K contained in G, therefore ∣μ∣∗(G)=0|\mu|^*(G) = 0.

Corollary 3

— *In a locally compact space X that is countable at infinity, every locally negligible set is negligible.

Since X is the union of a sequence (Kn)(K_n) of compact sets, every locally negligible set A is the union of the negligible sets A∩KnA \cap K_n, hence is negligible.

One can give examples of locally compact spaces that are not countable at infinity, and of measures on such a space X such that there exist sets in X that are locally negligible but not negligible (§ 1, Exer. 5).

Corollary 4

— *Let f be a mapping of X into a topological space F. If the set N of points of discontinuity of f is locally negligible, then f is measurable.

For every compact set K⊂XK \subset X, K∩NK \cap N is negligible (Prop. 5), therefore, for every ε>0\varepsilon > 0, there exists a compact set K1⊂K−(K∩N)K_1 \subset K - (K \cap N) such that ∣μ∣(K−K1)≤ε|\mu|(K - K_1) \leq \varepsilon (§ 4, No. 6, Th. 4), and by hypothesis the restriction of f to K1K_1 is continuous, whence the conclusion.

If P{x}P\{x\} is a property, the property « P\{x\} locally almost everywhere (with respect to μ\mu)» is by definition equivalent to the property «the set of x such that* (x ∈\in X and not P{x}P\{x\}) is locally negligible (for the measure μ\mu)». If F is any set, the relation «f(x)=g(x)f(x) = g(x) locally almost everywhere» is an equivalence relation in the set of mappings of X into F. In particular, if F is a vector space, a mapping ff of X into F such that f(x) = 0\text{ locally almost everywhere is said to be } \textit{locally negligible}. We leave to the reader the task of establishing for these concepts most of the properties corresponding to those that have been enumerated in §2, Nos. 4, 5 and 6 for functions equal almost everywhere. We shall limit ourselves to observing that if two \emph{continuous} mappings f,gf, g of X into a Hausdorff topological space F are equal \emph{locally almost everywhere}, then they are equal \emph{almost everywhere} by virtue of Cor. 2 of Prop. 5 (hence are equal at every point of the support of μ\mu (§ 2, No. 4, Prop. 9)); however, we state explicitly the following proposition, which is an immediate consequence of the principle of localization:

Proposition 6

Let ff be a measurable mapping of X into a topological space F. Every mapping of X into F, equal to ff locally almost everywhere, is measurable.

3. Elementary properties of measurable functions

Theorem 1

Let X be a locally compact space, μ\mu a measure on X, (Fn)(F_n) a sequence of topological spaces, F=∏nFnF = \prod_n F_n their product. For each index n, let fnf_n be a measurable mapping of X into FnF_n, and let f(x)=(fn(x))∈Ff(x) = (f_n(x)) \in F; then, for every continuous mapping u of f(X)f(X) into a topological space G, the function x↦u(f(x))x \mapsto u(f(x)) is measurable.

For every compact subset K of X and every number ε>0\varepsilon > 0, there exists a compact set K1⊂KK_1 \subset K such that ∣μ∣(K−K1)≤ε|\mu|(K - K_1) \leq \varepsilon and such that the restrictions to K1K_1 of all the functions fnf_n are continuous (No. 1, Prop. 2); it is clear that u∘fu \circ f is continuous on K1K_1, whence the theorem.

Remark

— 1) The theorem does not extend to an \emph{arbitrary} product of topological spaces (Exer. 1). 2) If ff is a continuous mapping of X into itself, and gg is a measurable mapping of X into F, then g∘fg \circ f is not necessarily measurable (Exer. 2).

Corollary 1

The upper envelope and the lower envelope of a finite number of measurable numerical functions (finite or not) are measurable.* For, sup⁡(u,v)\sup(u, v) and inf⁡(u,v)\inf(u, v) are continuous on R‾×R‾\overline{\mathbf{R}} \times \overline{\mathbf{R}}.

Corollary 2

For a numerical function ff (finite or not) to be measurable, it is necessary and sufficient that f+f^+ and f−f^- be measurable.* The condition is necessary by Cor. 1; it is sufficient because the image A of X in R‾×R‾\overline{\mathbf{R}} \times \overline{\mathbf{R}} under the mapping x↦(f+(x),f−(x))x \mapsto (f^+(x), f^-(x)) does not contain the points (+∞,+∞)(+\infty, +\infty) and (−∞,−∞)(-\infty, -\infty), consequently the mapping (u,v)↦u−v(u, v) \mapsto u - v is continuous on A.

Corollary 3

— *If f and g are two measurable mappings of X into a topological vector space F, then f + g and αf are measurable (α any scalar).

The set of measurable mappings of X into a topological vector space F is thus a vector space.

Corollary 4

— Let F be a vector space of dimension n over R and let (ek)_{1 \leq k \leq n} be a basis of F. In order that a function f = \sum_{k=1}^{n} e_k f_k be measurable, it is necessary and sufficient that each of the numerical functions f_k be measurable.

Corollary 5

— *Let F, G, H be three topological vector spaces, and let (u, v) \to [u \cdot v] be a continuous bilinear mapping of F \times G into H. If f is a measurable mapping of X into F, and g is a measurable mapping of X into G, then [f \cdot g] is a measurable mapping of X into H.

In particular, if f is a measurable mapping of X into a real (resp. complex) normed space F, and g is a measurable mapping of X into R (resp. C), then gf is measurable. If F is a normed algebra and f, g are two measurable mappings of X into F, then fg is measurable.

Corollary 6

— *If f is a measurable mapping of X into a normed space F, then the numerical function |f| is measurable.

4. Limits of measurable functions

Theorem 2 (Egoroff)

— *Let X be a locally compact space, μ a measure on X, A a countable set, F\mathfrak{F} a filter on A having a countable base, and (fα)α∈A(f_\alpha)_{\alpha \in A} a family of measurable mappings of X into a metrizable space F. Assume that lim⁡Ffα(x)=f(x)\lim_{\mathfrak{F}} f_\alpha(x) = f(x) exists in the complement of a locally negligible subset N of X. Under these conditions, 1° the function f (extended to N in any manner whatsoever) is measurable; 2° for every compact subset K of X and every ε>0\varepsilon > 0, there exists a compact set K1⊂KK_1 \subset K such that ∣μ∣(K−K1)≤ε|\mu|(K - K_1) \leq \varepsilon and such that the restrictions of the fαf_\alpha to K1K_1 are continuous and converge uniformly to f on K1K_1.

The first assertion obviously follows from the second, which we are going to prove. There exists a compact set K0⊂KK_0 \subset K such that ∣μ∣(K−K0)≤ε/2|\mu|(K - K_0) \leq \varepsilon/2 and such that the restrictions to K0K_0 of all the functions fαf_\alpha are continuous (No. 1, Prop. 2). Let (An)(A_n) be a decreasing countable base for the filter F\mathfrak{F}; let d be a metric on F compatible with the topology. For every pair of integers n>0, r>0n > 0,\ r > 0, let Bn,rB_{n,r} be the set of points x∈K0x \in K_0 such that, for at least one pair of indices α,β\alpha, \beta belonging to AnA_n, d(fα(x),fβ(x))≥1/rd(f_\alpha(x), f_\beta(x)) \geq 1/r; for fixed α\alpha and β\beta, the set of x∈K0x \in K_0 such that d(fα(x),fβ(x))≥1/rd(f_\alpha(x), f_\beta(x)) \geq 1/r is closed in K0K_0, hence is compact; consequently, Bn,rB_{n,r} is a countable union of compact sets contained in K0K_0, hence is integrable (§ 4, No. 5, Props. 6 and 8). If rr is fixed, the intersection of the decreasing sequence of sets Bn,rB_{n,r} (n=1,2,…n = 1, 2, \ldots) has measure zero, since fα(x)f_\alpha(x) tends to f(x)f(x) almost everywhere in K0K_0 with respect to the filter F\mathfrak{F}; thus lim⁡n→∞∣μ∣(Bn,r)=0\lim_{n \to \infty} |\mu|(B_{n,r}) = 0 (§ 4, No. 5, Cor. of Prop. 7), consequently there exists an integer nrn_r such that ∣μ∣(Bnr,r)≤ε/2r+2|\mu|(B_{n_r,r}) \leq \varepsilon/2^{r+2}. Let BB be the union (for r=1,2,…r = 1, 2, \ldots) of the sets Bnr,rB_{n_r,r}; BB is integrable and

∣μ∣(B)≤∑r=1∞∣μ∣(Bnr,r)≤ε/4 |\mu|(B) \leq \sum_{r=1}^\infty |\mu|(B_{n_r,r}) \leq \varepsilon/4

(§ 4, No. 5, Cor. of Prop. 8). Let CC be the complement of BB in K0K_0; by construction, fα(x)f_\alpha(x) converges uniformly to f(x)f(x) in CC with respect to the filter F\mathfrak{F}, and since the restrictions of the fαf_\alpha to CC are continuous, so is the restriction of ff to CC. It then suffices to take a compact set K1⊂CK_1 \subset C such that ∣μ∣(C−K1)≤ε/4|\mu|(C - K_1) \leq \varepsilon/4 in order to satisfy the conditions of the statement, since ∣μ∣(K−K1)=∣μ∣(K−K0)+∣μ∣(B)+∣μ∣(C−K1)≤ε|\mu|(K - K_1) = |\mu|(K - K_0) + |\mu|(B) + |\mu|(C - K_1) \leq \varepsilon.

The conclusions of Th. 2 do not necessarily hold if FF is not metrizable (Exer. 1). If FF is metrizable, and the set AA is not countable but the filter F\mathfrak{F} has a countable base, then the first conclusion of Th. 2 is again valid; for, if (An)(A_n) is a countable base for F\mathfrak{F} and αn\alpha_n is an element of AnA_n, then ff is the limit of the sequence (fαn)(f_{\alpha_n}) locally almost everywhere, hence is measurable; however, the second conclusion of Th. 2 is no longer necessarily valid (cf. Exer. 4).

Corollary 1

— Let (fn)(f_n) be a sequence of numerical functions (finite or not). If the fnf_n are measurable, then the functions sup⁡nfn\sup_n f_n, inf⁡nfn\inf_n f_n, lim sup⁡n→∞fn\limsup_{n \to \infty} f_n and lim inf⁡n→∞fn\liminf_{n \to \infty} f_n are measurable.

For, the extended real line R‾\overline{\mathbf{R}}, being homeomorphic to a compact interval of R\mathbf{R}, is metrizable. The function sup⁡nfn\sup_n f_n is the pointwise limit of the increasing sequence of functions gn=sup⁡(f1,f2,…,fn)g_n = \sup(f_1, f_2, \ldots, f_n), which are measurable (No. 3, Cor. 1 of Th. 1); similarly, lim sup⁡n→∞fn\limsup_{n \to \infty} f_n is the pointwise limit of the decreasing sequence of functions hn=sup⁡p≥0fn+ph_n = \sup_{p \geq 0} f_{n+p}, each of which is measurable by the foregoing. Finally, since inf⁡nfn=−sup⁡(−fn)\inf_n f_n = -\sup(-f_n) and lim inf⁡n→∞fn=−lim sup⁡n→∞(−fn)\liminf_{n \to \infty} f_n = -\limsup_{n \to \infty} (-f_n), these functions are measurable.

Corollary 2

— The measurable sets in XX form a tribe (GT, IX, §6, No. 3).

For, if M and N are measurable then the functions

φM∪N=φM+φN−φMφNandφM∩CN=φM−φMφN \varphi_{M \cup N} = \varphi_M + \varphi_N - \varphi_M \varphi_N \quad \text{and} \quad \varphi_{M \cap C_N} = \varphi_M - \varphi_M \varphi_N

are measurable by No. 3, Cors. 3 and 5 of Th. 1. If (Mn)(M_n) is a sequence of measurable sets and M is their union, then φM=sup⁡nφMn\varphi_M = \sup_n \varphi_{M_n} is measurable by Cor. 1 of Th. 2, whence the corollary.

In particular, since the open sets are measurable:

Corollary 3

The Borel sets in X* (GT, IX, §6, No. 3, Def. 4) are μ\mu-measurable for every measure μ\mu on X.

5. Criteria for measurability

When F is a topological vector space, every step function over the measurable sets (§4, No. 9, Def. 4), with values in F, is obviously measurable (No. 3, Cor. 3 of Th. 1); such a function ff takes on only a finite number of values, and for every y∈Fy \in F, f−1(y)f^{-1}(y) is measurable. More generally, let F be any topological space, ff a mapping of X into F taking on only a finite number of distinct values aia_i (1≤i≤m1 \leq i \leq m); if the sets Ai=f−1(ai)A_i = f^{-1}(a_i) are measurable, then the function ff is measurable. For, for every compact set K and for each of the sets Ai∩KA_i \cap K, there exist a negligible set Ni⊂Ai∩KN_i \subset A_i \cap K and a partition of Ai∩K∩CNiA_i \cap K \cap C N_i formed by a sequence (Kin)(K_{in}) of compact sets; since K is the union of the sets Ai∩KA_i \cap K and the restriction of ff to each of the KinK_{in} is constant, therefore continuous, ff is measurable. By an abuse of language we shall say that a mapping ff of X into F is a measurable step function if it takes on only a finite number of values and if, for every y∈Fy \in F, f−1(y)f^{-1}(y) is measurable.

Theorem 3

For a mapping ff of X into a metrizable space F to be measurable, it is necessary and sufficient that, for every compact set K⊂XK \subset X, there exist a sequence (gn)(g_n) of measurable step functions, with values in F, such that gn(x)g_n(x) tends to f(x)f(x) for almost every x∈Kx \in K*.

The condition is sufficient by Egoroff’s theorem and the principle of localization. Let us show that it is necessary: by hypothesis, there exist a negligible set N⊂KN \subset K and a partition (Km)(K_m) of K−NK - N formed of compact sets such that the restriction of ff to each KmK_m is continuous. To define the sequence (gn)(g_n), it suffices to proceed in the following manner: let dd be a metric compatible with the topology of F; for each KiK_i with index i≤ni \leq n, there exists a finite partition of KiK_i into integrable sets AijA_{ij} (1≤j≤qi1 \leq j \leq q_i) sufficiently small that the oscillation of ff on each of the AijA_{ij} is ≤1/n\leq 1/n (§ 4, No. 9, Lemma); take gng_n to be constant on each AijA_{ij}, equal to one of the values of ff in this set (for 1≤i≤n, 1≤j≤qi1 \leq i \leq n,\ 1 \leq j \leq q_i), and equal to a fixed element a∈Fa \in \mathbf{F} for every point of XX that does not belong to any of the AijA_{ij}. It is clear that the sequence (gn(x))(g_n(x)) converges to f(x)f(x) at every point of KK not belonging to NN.

Corollary 1

Let ff be a measurable mapping of XX into a Banach space FF; for every compact set K⊂XK \subset X, there exists a sequence (gn)(g_n) of measurable step functions, with supports contained in KK, such that ∣gn(x)∣≤∣f(x)∣|g_n(x)| \leq |f(x)| for all x∈Xx \in X and such that gn(x)g_n(x) tends to f(x)f(x) for almost every x∈Kx \in K.

With notations as in the proof of Th. 3, and denoting by aija_{ij} one of the values of ff in AijA_{ij}, it suffices to take, as the value of gng_n in AijA_{ij}, the point 0 if ∣aij∣≤1/n|a_{ij}| \leq 1/n and the point aij(1−1/(n∣aij∣))a_{ij}(1 - 1/(n|a_{ij}|)) otherwise; finally, set gn(x)=0g_n(x) = 0 on the complement of the union of the AijA_{ij} (1≤i≤n, 1≤j≤qi1 \leq i \leq n,\ 1 \leq j \leq q_i).

Corollary 2

Let XX be a locally compact space that is countable at infinity. If ff is a measurable mapping of XX into a metrizable space FF, then there exists a sequence (gn)(g_n) of measurable step functions, with values in FF, such that gn(x)g_n(x) tends to f(x)f(x) for almost every x∈Xx \in X.

If XX is the union of an increasing sequence (An)(A_n) of compact sets, the sets An−An−1A_n - A_{n-1} that are nonempty form a partition of XX into integrable sets; consequently, there exist a negligible set N⊂XN \subset X and a partition of CN\mathbf{C}N formed by a sequence (Kn)(K_n) of compact sets such that the restriction of ff to each KnK_n is continuous; the proof can then be concluded as in Th. 3 without modification.

Proposition 7

Let ff be a measurable mapping of XX into a topological space FF; the inverse image under ff of every closed (resp. open) set in FF is measurable.

It suffices to carry out the proof for the inverse image −1f(A)^{-1}f(A) of a closed set AA in FF. Let KK be a compact subset of XX; there exist a negligible set N⊂KN \subset K and a partition (Kn)(K_n) of K−NK - N formed of compact sets such that the restriction of ff to each KnK_n is continuous. The intersection Kn∩−1f(A)K_n \cap ^{-1}f(A) is thus the inverse image of the closed set AA under the restriction of ff to KnK_n; it is therefore a closed set in KnK_n, hence is compact. The set K∩−1f(A)K \cap ^{-1}f(A) is therefore the union of the negligible set N∩−1f(A)N \cap ^{-1}f(A) and the compact sets Kn∩−1f(A)K_n \cap ^{-1}f(A), which proves that −1f(A)^{-1}f(A) is measurable.

Theorem 4

— Let F be a metrizable space and let d be a metric on F compatible with the topology. For a mapping f of X into F to be measurable, it is necessary and sufficient that it satisfy the following two conditions: a) the inverse image under f of every closed ball of F is measurable; b) for every compact set K ⊂ X, there exists a countable subset H of F such that f(x)∈H‾f(x) \in \overline{H} for almost every x∈Kx \in K.

The condition a) is necessary by Prop. 7; on the other hand, in the notations of Th. 3, condition b) is satisfied by taking H to be the countable set formed by the values of all of the functions gng_n.

Let us now show that the conditions a) and b) are sufficient. Let K be any compact subset of X; there exists a negligible subset N of K such that f(K−N)f(K - N) is contained in the closure of a countable set of points of F, which we arrange in a sequence (an)(a_n). Let An,pA_{n,p} be the set of x∈K−Nx \in K - N such that d(f(x),an)≤1/pd(f(x), a_n) \leq 1/p. It follows from condition a) that An,pA_{n,p} is measurable. For fixed pp, define a sequence of sets Bn,p⊂K−NB_{n,p} \subset K - N recursively by setting

B1,p=A1,pandBn+1,p=An+1,p∩C(⋃k≤nAk,p); B_{1,p} = A_{1,p} \quad \text{and} \quad B_{n+1,p} = A_{n+1,p} \cap C \left( \bigcup_{k \leq n} A_{k,p} \right);

the sets Bn,pB_{n,p} are measurable, and those that are nonempty form a partition of K−NK - N. Let gm,pg_{m,p} be the function equal to aia_i on the set Bi,pB_{i,p} for 1≤i≤m1 \leq i \leq m and equal to a constant b∈Fb \in F on the complement of the union of these sets; gm,pg_{m,p} is a measurable step function; as mm tends to infinity, gm,pg_{m,p} converges pointwise to the function fpf_p equal to ana_n on Bn,pB_{n,p} (n≥1n \geq 1) and to bb on N∪CKN \cup C K, therefore (Th. 2) fpf_p is measurable. As pp tends to infinity, fp(x)f_p(x) tends to f(x)f(x) for every x∈K−Nx \in K - N, and to bb for x∈N∪CKx \in N \cup C K; the limit of the fpf_p is therefore measurable, and the principle of localization proves that ff itself is measurable.

Remark 1

Condition a) alone is not sufficient for ff to be measurable (Exer. 7).

Remark 2

If the topology of F has a countable base then condition b) of Th. 4 is automatically satisfied for every mapping of X into F. The proof shows, moreover, that it suffices to assume that the inverse images under ff of the closed balls with rational radii, whose centers belong to a dense countable subset of F, are measurable sets.

Remark 3

The hypothesis a) can be replaced by the condition that the inverse image under ff of every open ball of F is measurable.

The case of numerical functions (finite or not) deserves special mention:

Proposition 8

— Let D be a countable dense subset of R\mathbf{R}. In order that a numerical function ff (finite or not) be measurable, it is sufficient (and necessary) that for every a∈Da \in D, the set of x∈Xx \in X such that f(x)≥af(x) \geq a be measurable.

For, if this is so, for every b∈R‾b \in \overline{\mathbf{R}} the set of xx such that f(x)⩾bf(x) \geqslant b is measurable, being the intersection of the sets (which form a countable family) of the xx such that f(x)⩾af(x) \geqslant a, as aa runs over the set of points of D that are ⩽b\leqslant b. The set of xx such that f(x)<bf(x) < b is measurable, being the complement of a measurable set. Next, if bb is finite then the set of xx such that f(x)⩽bf(x) \leqslant b is measurable, being the intersection of the sets of the xx such that f(x)<b+1/nf(x) < b + 1/n; and f(−∞)−1f(-\infty)^{-1} is measurable, being the intersection of the sets of the xx such that f(x)<nf(x) < n, as nn runs over Z\mathbf{Z}. Finally, the inverse image under ff of every closed interval of R‾\overline{\mathbf{R}} is measurable, being the intersection of two measurable sets, and Th. 4 may be applied.

One could show similarly that it suffices that for every a∈Da \in \mathrm{D}, the set of xx such that f(x)>af(x) > a be measurable.

Corollary

— Every lower (resp. upper) semi-continuous function is measurable.

For, if ff is lower semi-continuous, then the set of x∈Xx \in \mathrm{X} such that f(x)⩽af(x) \leqslant a is closed for every a∈R‾a \in \overline{\mathbf{R}}.

When F is metrizable and compact, Prop. 7 makes it possible to sharpen Th. 3 as follows:

Proposition 9

— If F is a metrizable compact space, then every measurable mapping f of X into F is the uniform limit (on all of X) of a sequence of measurable step functions.

Let dd be a metric compatible with the topology of F. For every positive integer nn there exists a finite number of points ak∈Fa_k \in \mathrm{F} such that the closed balls BkB_k with center aka_k and radius 1/n1/n form a covering of F; the sets Ak=f−1(Bk)A_k = f^{-1}(B_k) are therefore measurable (Prop. 7) and form a covering of X. Consequently (§ 4, No. 9, Lemma) there exists a partition (Ci)(C_i) of X into a finite number of measurable sets such that each AkA_k is the union of certain of the CiC_i. Let cic_i be a point of CiC_i and let gng_n be the measurable step function equal to f(ci)f(c_i) on CiC_i (for each index ii). It is clear that d(f(x),gn(x))⩽2/nd(f(x), g_n(x)) \leqslant 2/n for all x∈Xx \in \mathrm{X}.

Proposition 10

— Let F be a separable Banach space, F' its dual, and (an′)(\mathbf{a}'_n) a weakly dense sequence in the unit ball of F' (TVS, III, § 3, No. 4, Cor. 2 of Prop. 6).¹ For a mapping f of X into F to be measurable, it is necessary and sufficient that for every n the scalar function x↦⟨f(x),an′⟩x \mapsto \langle \mathbf{f}(x), \mathbf{a}'_n \rangle be measurable.

The condition being obviously necessary (No. 3, Th. 1), let us prove that it is sufficient; it suffices to verify condition a) of Th. 4 and, to this end,

¹ See the footnote at the end of § 2, No. 4.

it will suffice by the principle of localization to prove that, for every compact subset K of X and every closed ball B⊂FB \subset F, with center aa and radius rr, the set A=K∩f−1(B)A = K \cap f^{-1}(B) is measurable; now, for every z∈Fz \in F,

∣z∣=sup⁡n∣⟨z,an′⟩∣/∣an′∣; |z| = \sup_n |\langle z, a'_n \rangle| / |a'_n|;

AA is thus the intersection of KK and the sets defined by

∣⟨f(x),an′⟩−⟨a,an′⟩∣≤r∣an′∣; |\langle f(x), a'_n \rangle - \langle a, a'_n \rangle| \leq r |a'_n|;

as these sets are measurable by hypothesis, AA is measurable.

Corollary 1

Let F be a Banach space. In order that a mapping f of X into F be measurable, it is necessary and sufficient that it satisfy the following two conditions:

a) for every a′∈F′a' \in F', the scalar function x↦⟨f(x),a′⟩x \mapsto \langle f(x), a' \rangle is measurable;

b) for every compact set K⊂XK \subset X, there exists a countable subset H of F such that f(x)∈H‾f(x) \in \overline{H} for almost every x∈Kx \in K.

The necessity of the conditions follows from No. 3, Th. 1, and Th. 4. To prove that the conditions are sufficient, it again suffices to verify condition a) of Th. 4. With notations as in the proof of Prop. 10, we may (on account of b)) suppose, after modifying ff if necessary on a negligible set, that f(K)⊂H‾f(K) \subset \overline{H}, where HH is a countable subset of FF. If VV is the closed linear subspace of FF generated by the set H∪{a}H \cup \{a\}, then VV is a separable Banach space and every continuous linear form on VV is the restriction of a form a′∈F′a' \in F'; the same reasoning as in Prop. 10 then shows that K∩f−1(B)K \cap f^{-1}(B) is measurable.

Corollary 2

Let F be a locally convex space that is metrizable and separable, and let F′F' be its dual. For a mapping f of X into F to be measurable, it is necessary and sufficient that for every a′∈F′a' \in F', the scalar function x↦⟨f(x),a′⟩x \mapsto \langle f(x), a' \rangle be measurable.

We may regard F as a subspace of a countable product ∏nEn\prod_n E_n of Banach spaces (TVS, II, §4, No. 3), and we can suppose that pr⁡n(F)\operatorname{pr}_n(F) is dense in EnE_n, which is therefore separable. For every nn, the mapping pr⁡n∘f\operatorname{pr}_n \circ f is then measurable by Prop. 10, therefore ff is measurable by No. 3, Th. 1.

Proposition 11

Let F be a locally convex space that is the direct limit of a sequence of separable metrizable locally convex spaces FnF_n, F being the union of the FnF_n. Let F′F' be the dual of F, equipped with the weak topology σ(F′,F)\sigma(F', F). For a mapping f of X into F′F' to be measurable, it is necessary and sufficient that, for every a∈Fa \in F, the scalar function x↦⟨a,f(x)⟩x \mapsto \langle a, f(x) \rangle be measurable.

The condition is necessary by No. 3, Th. 1; let us prove that it is sufficient. Suppose first that F is metrizable and separable, and let D be a countable dense set in F. Let (Vn)(V_n) be a decreasing fundamental sequence of balanced convex open neighborhoods of 0 in F; the polar sets Vn∘V_n^\circ are equicontinuous and their union is all of F′F'. Let Xn=f−1(Vn∘)X_n = f^{-1}(V_n^\circ); the sequence (Xn)(X_n) is increasing and X=⋃nXnX = \bigcup_n X_n; let us show that each XnX_n is μ\mu-measurable. Indeed, D∩VnD \cap V_n is dense in VnV_n; for every y∈D∩Vny \in D \cap V_n let SyS_y be the set of x∈Xx \in X such that ∣⟨y,f(x)⟩∣≤1|\langle y, f(x) \rangle| \leq 1; the hypothesis implies that each of the SyS_y is measurable, and XnX_n is the intersection of the countable family of SyS_y for y∈D∩Vny \in D \cap V_n. This being so, for every compact subset K of X and every ε>0\varepsilon > 0, there exists an integer n such that ∣μ∣(K−(K∩Xn))≤ε/4|\mu|(K - (K \cap X_n)) \leq \varepsilon/4, and then a compact subset K1K_1 of K∩XnK \cap X_n such that ∣μ∣((K∩Xn)−K1)≤ε/4|\mu|((K \cap X_n) - K_1) \leq \varepsilon/4; finally, there exists a compact subset K2K_2 of K1K_1 such that ∣μ∣(K1−K2)≤ε/2|\mu|(K_1 - K_2) \leq \varepsilon/2 and such that the restrictions to K2K_2 of all of the functions ⟨y,f⟩\langle y, f \rangle, where y∈Dy \in D, are continuous (No. 1, Prop. 2). Since the set f(K2)⊂f(Xn)⊂Vn∘f(K_2) \subset f(X_n) \subset V_n^\circ is equicontinuous, the topology induced by σ(F′,F)\sigma(F', F) on f(K2)f(K_2) is identical to the topology of pointwise convergence in D (GT, X, §2, No. 4, Th. 1); the restriction of f to K2K_2 is therefore continuous, whence our assertion in this first case.

Let us pass to the general case. If z′z' is a continuous linear form on F, its restriction zn′z'_n to FnF_n is continuous; since F=⋃nFnF = \bigcup_n F_n, the dual F′F' of F may be identified (algebraically) with a linear subspace of the product ∏nFn′\prod_n F'_n, and then pr⁡n(z′)=zn′\operatorname{pr}_n(z') = z'_n. Moreover, since each finite subset of F is contained in one of the FnF_n, the topology σ(F′,F)\sigma(F', F) is none other than the topology induced by the product topology of the topologies σ(Fn′,Fn)\sigma(F'_n, F_n). This being so, if ⟨a,f⟩\langle a, f \rangle is measurable for every a∈Fa \in F, then so is ⟨an,pr⁡n∘f⟩\langle a_n, \operatorname{pr}_n \circ f \rangle for every n and every an∈Fna_n \in F_n, since ⟨an,pr⁡n∘f⟩=⟨an,f⟩\langle a_n, \operatorname{pr}_n \circ f \rangle = \langle a_n, f \rangle; the first part of the proof then shows that pr⁡n∘f\operatorname{pr}_n \circ f is measurable for every n, therefore so is f (No. 3, Th. 1).

6. Criteria for integrability

Theorem 5

— In order that a mapping f of X into a Banach space F be p-th power integrable (1≤p<+∞1 \leq p < +\infty), it is necessary and sufficient that f be measurable and that Np(f)N_p(f) be finite.

The condition is necessary: for, if f∈LFpf \in \mathscr{L}_F^p then there exists a sequence (gn)(g_n) of continuous functions with compact support that converges almost everywhere to f (§3§ 3, No. 4, Cor. 2 of Th. 3); by Th. 2 of No. 4, f is measurable.

To prove that the conditions are sufficient, we first establish a lemma:

Lemma 1

Let g be a function with values in F, such that N_p(g) < +\infty* (in other words, a function in FFp\mathscr{F}_F^p). *The set A of points x \in X such that g(x) \neq 0 is contained in the union of a negligible set and a sequence of compact sets.

Let AnA_n be the set of points x∈Xx \in X such that ∣g(x)∣≥1/n|g(x)| \geq 1/n; A is the union of the AnA_n, and φAn≤n∣g∣\varphi_{A_n} \leq n|g|, whence ∣μ∣∗(An)≤(nNp(g))p|\mu|^*(A_n) \leq (n N_p(g))^p; it follows that AnA_n is contained in the union of a negligible set and a sequence of compact sets (§ 4, No. 6, Cor. 3 of Th. 4), therefore so is A.

The lemma proved, consider first the case that f has compact support K. By Cor. 1 of Th. 3 of No. 5, there exists a sequence (gn)(g_n) of measurable step functions such that ∣gn(x)∣≤∣f(x)∣|g_n(x)| \leq |f(x)| at every point x∈Xx \in X and such that gn(x)g_n(x) tends to f(x)f(x) almost everywhere. Now, gng_n is a linear combination of characteristic functions of measurable sets contained in K; since these sets are integrable by Prop. 3 of No. 1, gng_n belongs to LFp\mathscr{L}_F^p. Since Np(f)<+∞N_p(f) < +\infty, Lebesgue’s theorem (§ 3, No. 7, Th. 6) shows that f belongs to LFp\mathscr{L}_F^p.

In the general case, it follows from Lemma 1 that there exists an increasing sequence (Kn)(K_n) of compact sets such that f(x)f(x) is zero almost everywhere in the complement of the union of the KnK_n. Let fnf_n be the function equal to f(x)f(x) on KnK_n and to 0 elsewhere; fnf_n is measurable by No. 3, Cor. 5 of Th. 1; since ∣fn∣≤∣f∣|f_n| \leq |f|, fnf_n belongs to LFp\mathscr{L}_F^p by the first part of the argument. Since f(x)f(x) is equal almost everywhere to the limit of the sequence fn(x)f_n(x), Lebesgue’s theorem again proves that f∈LFpf \in \mathscr{L}_F^p, which completes the proof.

One should take care to note that a function that is locally negligible but not negligible is not integrable; thus, a function equal locally almost everywhere to an integrable function is not necessarily integrable.

Corollary 1

For a set to be integrable, it is necessary and sufficient that it be measurable and have finite outer measure.

Corollary 2

Let (Fn)(F_n) be a sequence of topological spaces; for every index n, let fnf_n be a measurable mapping of X into FnF_n and let f(x)=(fn(x))∈F=∏nFnf(x) = (f_n(x)) \in F = \prod_n F_n; finally, let u be a continuous mapping of f(X)f(X) into a Banach space G. For the function g(x)=u(f(x))g(x) = u(f(x)) to be integrable, it is necessary and sufficient that N1(g)<+∞N_1(g) < +\infty.

For, g is measurable (No. 3, Th. 1).

Corollary 3

For every integrable function f and every measurable set A, the function fφAf \varphi_A is integrable.

For, it follows from Th. 5 and No. 3, Cor. 5 of Th. 1 that fφAf \varphi_A is measurable, and N1(fφA)≤N1(f)N_1(f \varphi_A) \leq N_1(f).

We write ∫Afdμ=∫fφAdμ\int_A f d\mu = \int f \varphi_A d\mu (or ∫Afμ\int_A f \mu) for every integrable function ff and every measurable set AA. We also write ∫∗fd∣μ∣\int^* f d|\mu| (or ∫A∗f∣μ∣\int^*_A f |\mu|) in place of ∫∗fφAd∣μ∣\int^* f \varphi_A d|\mu| for every numerical function (finite or not) f⩾0f \geqslant 0 (on setting f(x)φA(x)=0f(x)\varphi_A(x) = 0 if f(x)=+∞f(x) = +\infty and φA(x)=0\varphi_A(x) = 0).

Corollary 4

For every sequence* (fn)(f_n) of measurable functions ⩾0\geqslant 0 on XX,

∫∗(∑nfn)d∣μ∣=∑n∫∗fnd∣μ∣. \int^*\left( \sum_n f_n \right) d|\mu| = \sum_n \int^* f_n d|\mu|.

In view of §1, No. 3, Th. 3, we are reduced to proving that for any two measurable functions f⩾0,g⩾0f \geqslant 0, g \geqslant 0 on XX,

∫∗(f+g)d∣μ∣=∫∗fd∣μ∣+∫∗gd∣μ∣. \int^*(f + g) d|\mu| = \int^* f d|\mu| + \int^* g d|\mu|.

This is nothing more than the additivity of the integral when ff and gg are integrable. In the contrary case, if for example ff is not integrable, then ∫∗fd∣μ∣=+∞\int^* f d|\mu| = +\infty by Th. 5; a fortiori, ∫∗(f+g)d∣μ∣=+∞\int^*(f + g) d|\mu| = +\infty.

Corollary 5

For every sequence* (An)(A_n) of pairwise disjoint measurable sets,

∣μ∣∗(⋃nAn)=∑n∣μ∣∗(An). |\mu|^*\left( \bigcup_n A_n \right) = \sum_n |\mu|^*(A_n).

This follows from Cor. 4 applied to the φAn\varphi_{A_n}.

7. Measure induced on a locally compact subspace

Let XX be a locally compact space, μ\mu a measure on XX, and YY a locally compact subspace of XX. Since YY is the intersection of an open set and a closed set in XX (GT, I, §9, No. 7, Prop. 12), it is μ\mu-measurable (No. 1, Cor. of Prop. 3). For every function g∈H(Y;C)g \in \mathscr{H}(Y; \mathbf{C}) let g′g' be the function, defined on all of XX, equal to gg on YY and to 0 on X−YX - Y; let us show that g′g' is μ\mu-*integrable*. We can restrict ourselves to the case that gg is real and ⩾0\geqslant 0 (on writing gg as a linear combination of such functions); since g′g' is bounded and has compact support, it suffices to show that g′g' is μ\mu-measurable (No. 6, Th. 5); but this follows from the fact that g′g' is upper semi-continuous on XX (No. 5, Cor. of Prop. 8). We may therefore make the following definition:

Definition 4

— *Given a locally compact subspace Y of a locally compact space X, we call measure induced on Y by a measure μ\mu on X, and denote by μY\mu_Y or μ∣Y\mu|Y, the measure defined by the formula

∫g dμY=∫g′ dμ \int g \, d\mu_Y = \int g' \, d\mu

for every function g∈K(Y;C)g \in \mathscr{K}(Y; \mathbf{C}), where g′g' denotes the function equal to gg on YY and to 0 on X−YX - Y.

Example

— Let μ\mu be Lebesgue measure on R\mathbf{R}, I any interval in R\mathbf{R}; I is a locally compact subspace of R\mathbf{R}, and the measure induced by μ\mu on I is the linear form

g↦∫abg(x) dx g \mapsto \int_a^b g(x) \, dx

on K(I;C)\mathscr{K}(I; \mathbf{C}), where aa and bb are the endpoints (finite or not) of I (cf. § 4, No. 4, Example), in other words, what we have called Lebesgue measure on I.

When Y is an open subspace of X, Def. 4 coincides with the definition of the measure induced by μ\mu on Y (or the restriction of μ\mu to Y) given in Ch. III, §2, No. 1: indeed, for every function g∈K(Y;C)g \in \mathscr{K}(Y; \mathbf{C}) the function g′g' is then continuous on X.

We shall study integration with respect to an induced measure in detail in Ch. V, §7, and until then we shall need only the following results:

Lemma 2

— *Let μ\mu be a positive measure on X, and let K be a compact subset of X.

(i) For every compact (resp. open) subset H of K, μK(H)=μ(H)\mu_K(H) = \mu(H).

(ii) For a subset N of K to be μK\mu_K-negligible, it is necessary and sufficient that it be μ\mu-negligible.

(iii) If S is the support of μK\mu_K, then Supp(μS)=S\mathrm{Supp}(\mu_S) = S.

(i) We can restrict ourselves to the case that H is compact. Denote by ff the characteristic function of H in the space K; ff is upper semi-continuous, hence is the lower envelope of a decreasing directed family (gα)(g_\alpha) of functions in K+(K)\mathscr{K}_+(K); we have μK(H)=inf⁡α∫gα dμK\mu_K(H) = \inf_\alpha \int g_\alpha \, d\mu_K (§ 4, No. 4, Cor. 2 of Prop. 5). If gα′g'_\alpha is the function equal to gαg_\alpha on K and to 0 on X−KX - K, then gα′g'_\alpha is upper semi-continuous, and the lower envelope of the decreasing directed family (gα′)(g'_\alpha) is the characteristic function φH\varphi_H of H in the space X; therefore

μ(H)=inf⁡α∫gα′ dμ=inf⁡α∫gα dμK=μK(H) \mu(H) = \inf_\alpha \int g'_\alpha \, d\mu = \inf_\alpha \int g_\alpha \, d\mu_K = \mu_K(H)

by (1).

(ii) If N is μ\mu-negligible then, for every ε>0\varepsilon > 0, there exists a relatively compact open neighborhood U of N in X such that μ(U)≤ε\mu(U) \leq \varepsilon; since K−(U∩K)K - (U \cap K) is compact, it follows from (i) that μK(U∩K)≤μ(U)≤ε\mu_K(U \cap K) \leq \mu(U) \leq \varepsilon, thus N is μK\mu_K-negligible. Conversely, if N is μK\mu_K-negligible then there exists an open neighborhood V of N in X such that μK(V∩K)≤ε\mu_K(V \cap K) \leq \varepsilon; by (i), we have μ(V∩K)=μK(V∩K)\mu(V \cap K) = \mu_K(V \cap K), therefore μ(N)=0\mu(N) = 0 since ε\varepsilon is arbitrary.

(iii) For every open set U in K that intersects S, by hypothesis μK(U∩S)≠0\mu_K(U \cap S) \neq 0, therefore μ(U∩S)≠0\mu(U \cap S) \neq 0 by (i), and since U∩SU \cap S is open in S, μS(U∩S)≠0\mu_S(U \cap S) \neq 0 by (i); since every nonempty open set in S is of the form U∩SU \cap S, where U is open in K, this proves that Supp⁡(μS)=S\operatorname{Supp}(\mu_S) = S.

Lemma 3

Let Y be a locally compact subspace of X; for every measure μ\mu on X, ∣μY∣=∣μ∣Y|\mu_Y| = |\mu|_Y.

Let f be a function in K+(Y)\mathscr{K}_+(Y) and ε\varepsilon any number >0> 0; by definition, there exists a function g∈K(Y;C)g \in \mathscr{K}(Y; \mathbf{C}) such that ∣g∣≤f|g| \leq f and ∣μY∣(f)≤∣μY(g)∣+ε|\mu_Y|(f) \leq |\mu_Y(g)| + \varepsilon. Denoting by f′f' and g′g' the functions obtained by extending f and g, respectively, to be 0 on X−YX - Y, we have μY(g)=μ(g′)\mu_Y(g) = \mu(g') and, since ∣g′∣≤f′|g'| \leq f',

∣μ(g′)∣≤∣μ∣(∣g′∣)≤∣μ∣(f′)=∣μ∣Y(f), |\mu(g')| \leq |\mu|(|g'|) \leq |\mu|(f') = |\mu|_Y(f),

whence ∣μY∣(f)≤∣μ∣Y(f)+ε|\mu_Y|(f) \leq |\mu|_Y(f) + \varepsilon and, since ε\varepsilon is arbitrary,

∣μY∣(f)≤∣μ∣Y(f). |\mu_Y|(f) \leq |\mu|_Y(f).

On the other hand, let K be the support of f and let U be a compact neighborhood of K in X such that ∣μ∣(U−K)≤ε|\mu|(U - K) \leq \varepsilon; by Urysohn's theorem, there exists a function f1∈K+(X)f_1 \in \mathscr{K}_+(X), extending f, with support contained in U and such that ∥f1∥=∥f∥\|f_1\| = \|f\|. There exists a function h1∈K(X;C)h_1 \in \mathscr{K}(X; \mathbf{C}) such that ∣h1∣≤f1|h_1| \leq f_1 and ∣μ∣(f1)≤∣μ(h1)∣+ε|\mu|(f_1) \leq |\mu(h_1)| + \varepsilon. If h is the restriction of h1h_1 to Y, then h∈K(Y;C)h \in \mathscr{K}(Y; \mathbf{C}), ∣h∣≤f|h| \leq f and μ(h1)−μY(h)=μ(h1φU−K)\mu(h_1) - \mu_Y(h) = \mu(h_1 \varphi_{U-K}), therefore

∣μ(h1)−μY(h)∣≤∥f∥⋅∣μ∣(U−K)≤ε∥f∥; |\mu(h_1) - \mu_Y(h)| \leq \|f\| \cdot |\mu|(U - K) \leq \varepsilon \|f\|;

moreover, we have similarly

∣μ∣(f1)−∣μ∣Y(f)=∣μ∣(f1φU−K)and∣∣μ∣(f1)−∣μ∣Y(f)∣∣≤ε∥f∥. |\mu|(f_1) - |\mu|_Y(f) = |\mu|(f_1 \varphi_{U-K}) \quad \text{and} \quad ||\mu|(f_1) - |\mu|_Y(f)|| \leq \varepsilon \|f\|.

From this we infer that

∣μ∣Y(f)≤∣μY(h)∣+ε(1+2∥f∥)≤∣μY∣(f)+ε(1+2∥f∥) |\mu|_Y(f) \leq |\mu_Y(h)| + \varepsilon(1 + 2\|f\|) \leq |\mu_Y|(f) + \varepsilon(1 + 2\|f\|)

and since ε\varepsilon is arbitrary, ∣μ∣Y(f)≤∣μY∣(f)|\mu|_Y(f) \leq |\mu_Y|(f), which completes the proof.

8. μ\mu-dense families of compact sets

Proposition 12

— Let μ\mu be a measure on a locally compact space XX, AA a μ\mu-measurable subset of XX, and K\mathscr{K} a set of compact subsets of AA satisfying the following conditions: (PLI) Every closed (hence compact) subset of a set of K\mathscr{K} belongs to K\mathscr{K}. (PLII) Every finite union of sets of K\mathscr{K} belongs to K\mathscr{K}. The following four properties are then equivalent: a) For a subset BB of AA to be locally μ\mu-negligible, it is necessary and sufficient that ∣μ∣∗(B∩K)=0|\mu|^*(B \cap K) = 0 for all K∈KK \in \mathscr{K}. b) For every compact subset K0K_0 of AA and every ε>0\varepsilon > 0, there exists a set K∈KK \in \mathscr{K}, contained in K0K_0 and such that ∣μ∣(K0−K)≤ε|\mu|(K_0 - K) \leq \varepsilon. c) For every compact subset BB of AA, there exists a partition of BB formed by a μ\mu-negligible set NN and a sequence (Hn)(H_n) of compact sets belonging to K\mathscr{K}. d) For every compact subset BB of AA, there exists an increasing sequence (Kn)(K_n) of compact sets belonging to K\mathscr{K}, contained in BB and such that the set N=B−⋃nKnN = B - \bigcup_n K_n is μ\mu-negligible.

It is immediate (No. 2, Prop. 5) that d) implies a);c)a); c) implies d) on taking KnK_n to be the union of the HpH_p for p≤np \leq n and citing (PLII). To prove that b) implies c), one defines recursively a sequence (Hp)(H_p) of sets of K\mathscr{K} such that Hn+1⊂B−⋃p≤nHpH_{n+1} \subset B - \bigcup_{p \leq n} H_p and ∣μ∣(B−⋃p≤nHp)≤1/n|\mu|(B - \bigcup_{p \leq n} H_p) \leq 1/n (§ 4, No. 6, Th. 4).

It remains to prove that a) implies b). Let us argue by contradiction, and suppose that the supremum α\alpha of the numbers ∣μ∣(K)|\mu|(K), where KK runs over the set of subsets of K0K_0 belonging to K\mathscr{K}, is <∣μ∣(K0)< |\mu|(K_0). By (PLII), there exists an increasing sequence (Ln)(L_n) of compact subsets of K0K_0, belonging to K\mathscr{K} and such that sup⁡n∣μ∣(Ln)=α\sup_n |\mu|(L_n) = \alpha. Set B=⋃nLnB = \bigcup_n L_n; BB is integrable and ∣μ∣(B)=α|\mu|(B) = \alpha, therefore ∣μ∣(K0−B)=∣μ∣(K0)−α>0|\mu|(K_0 - B) = |\mu|(K_0) - \alpha > 0. On the other hand, we shall see that for every set K∈KK \in \mathscr{K}, ∣μ∣(K∩(K0−B))=0|\mu|(K \cap (K_0 - B)) = 0, which, by virtue of a), will imply a contradiction. Indeed, if there existed a set K∈KK \in \mathscr{K} such that ∣μ∣(K∩(K0−B))>0|\mu|(K \cap (K_0 - B)) > 0, then there would exist a compact subset HH of K∩(K0−B)K \cap (K_0 - B) such that ∣μ∣(H)>0|\mu|(H) > 0. By (PLI), we would have H∈KH \in \mathscr{K}, and, for nn sufficiently large,

∣μ∣(Ln∪H)=∣μ∣(Ln)+∣μ∣(H)>α. |\mu|(L_n \cup H) = |\mu|(L_n) + |\mu|(H) > \alpha .

But Ln∪HL_n \cup H belongs to K\mathscr{K} by (PLII), and this contradicts the definition of α\alpha.

Definition 6

— Let AA be a μ\mu-measurable subset of XX. A set K\mathscr{K} of compact subsets of AA is said to be μ\mu-dense in AA if it satisfies the conditions (PLI), (PLII), a),b),c),d)a), b), c), d) of Prop. 12.

The set of all compact subsets of AA is μ\mu-dense in AA.

When A=XA = X, we shall say simply ' μ\mu-dense set' in place of 'set μ\mu-dense in XX'. If X−AX - A is locally μ\mu-negligible, then every set of compact subsets of AA that is μ\mu-dense in AA is also μ\mu-dense in XX.

Remark

— Suppose that AA is the union of a sequence (Ln)(L_n) of compact sets and a μ\mu-negligible (resp. locally μ\mu-negligible) set, and let K\mathscr{K} be a set of compact subsets that is μ\mu-dense in AA. Applying to each LnL_n the property c) of the statement of Prop. 12, we see that AA is the union a sequence of compact sets belonging to K\mathscr{K} and a μ\mu-negligible (resp. locally μ\mu-negligible) set.

If KK is a compact subset of XX, it comes to the same to say that a set of compact subsets of KK is μ\mu-dense in KK or that it is μK\mu_K-*dense* in KK; this follows from Lemmas 2 and 3 of No. 7 and condition b) of Prop. 12.

Proposition 13

Let AA be a μ\mu-measurable subset of XX, K\mathscr{K} a set of compact subsets that is μ\mu-dense in AA. Let H\mathfrak{H} be a set of compact subsets of AA satisfying (PLI_I) and (PLII_{II}) and such that, for every K∈KK \in \mathscr{K}, the set of H∈HH \in \mathfrak{H} such that H⊂KH \subset K is μK\mu_K-*dense* (or, what amounts to the same, μ\mu-*dense*) in KK. Then H\mathfrak{H} is μ\mu-dense in AA.

For, let LL be a compact subset of AA. For every ε>0\varepsilon > 0 there exists a K∈KK \in \mathscr{K} such that K⊂LK \subset L and ∣μ∣(L−K)≤ε/2|\mu|(L - K) \leq \varepsilon/2, and then an H∈HH \in \mathfrak{H} such that H⊂KH \subset K and ∣μ∣(K−H)≤ε/2|\mu|(K - H) \leq \varepsilon/2; it follows that ∣μ∣(L−H)≤ε|\mu|(L - H) \leq \varepsilon, whence the proposition.

9. Locally countable partitions

Definition 7

A set A\mathfrak{A} of subsets of a topological space TT is said to be locally countable if, for every t∈Tt \in T, there exists a neighborhood VV of tt such that the set of A∈AA \in \mathfrak{A} that intersect VV is countable.

If the set A\mathfrak{A} of subsets of TT is locally countable then, for every compact subset KK of TT, the set of A∈AA \in \mathfrak{A} that intersect KK is countable, since KK can be covered by a finite number of open neighborhoods of points of KK, each of which intersects only a countable set of subsets belonging to A\mathfrak{A}.

Def. 7 shows that the union of a locally countable set of μ\mu-measurable (resp. locally μ\mu-negligible) subsets of a locally compact space is μ\mu-measurable (resp. locally μ\mu-negligible) (No. 1, Prop. 3 and No. 2, Prop. 5).

Proposition 14

Let XX be a locally compact space, μ\mu a measure on XX, AA a μ\mu-measurable subset of XX, and K\mathscr{K} a set of compact subsets of AA that is μ\mu-dense in AA. There exists a locally countable set H⊂K\mathfrak{H} \subset \mathscr{K}, formed of pairwise disjoint sets, such that A−⋃K∈HKA - \bigcup_{K \in \mathfrak{H}} K is locally μ\mu-negligible and such that, for every K∈HK \in \mathfrak{H}, the support of μK\mu_K is all of KK.

Consider sets L⊂K\mathscr{L} \subset \mathfrak{K} formed of pairwise disjoint sets such that, for every L∈LL \in \mathscr{L}, Supp⁡(μL)=L\operatorname{Supp}(\mu_L) = L. The sets L\mathscr{L} form a subset H\mathscr{H} of P(K)\mathscr{P}(\mathfrak{K}) that is nonempty (because it contains the element ∅\varnothing) and which we shall order by the relation of inclusion in P(K)\mathscr{P}(\mathfrak{K}). It is immediate that H\mathscr{H} is inductive; let H\mathfrak{H} be a maximal element of H\mathscr{H} (S, R, §6, No. 10). Let us first show that H\mathfrak{H} is locally countable. Indeed, for every x∈Xx \in X, let VV be a relatively compact open neighborhood of xx; if (Ki)1≤i≤n(K_i)_{1 \leq i \leq n} is a finite family of distinct sets of H\mathfrak{H} that intersect VV, then

∑i=1n∣μ∣(Ki∩V)=∣μ∣(V∩(⋃i=1nKi)) \sum_{i=1}^n |\mu|(K_i \cap V) = |\mu|\left(V \cap \left( \bigcup_{i=1}^n K_i \right)\right)

because the KiK_i are pairwise disjoint, whence ∑i=1n∣μ∣(Ki∩V)≤∣μ∣(V)\sum_{i=1}^n |\mu|(K_i \cap V) \leq |\mu|(V). It follows that if HV\mathfrak{H}_V is the set of K∈HK \in \mathfrak{H} that intersect VV, then

∑K∈HV∣μ∣(K∩V)<+∞, \sum_{K \in \mathfrak{H}_V} |\mu|(K \cap V) < +\infty,

and since ∣μ∣(K∩V)>0|\mu|(K \cap V) > 0 for every K∈HVK \in \mathfrak{H}_V, HV\mathfrak{H}_V is necessarily countable. Next, let us prove that N=A−⋃K∈HKN = A - \bigcup_{K \in \mathfrak{H}} K is locally μ\mu-negligible. We saw above that NN is μ\mu-measurable. If NN were not locally negligible, it would contain a non-negligible compact set L0L_0, hence (No. 8, Prop. 12) a non-negligible compact set L⊂L0L \subset L_0 belonging to K\mathfrak{K}. Since ∣μL∣(L)=∣μ∣(L)>0|\mu_L|(L) = |\mu|(L) > 0 (No. 7, Lemmas 2 and 3), the measure μL\mu_L induced on LL by μ\mu is nonzero; its support SS is therefore a nonempty compact set belonging to K\mathfrak{K} by (PL₁), and Supp⁡(μS)=S\operatorname{Supp}(\mu_S) = S (No. 7, Lemma 2, (iii)). It follows that the set H∪{S}\mathfrak{H} \cup \{S\} belongs to H\mathscr{H}, which contradicts the definition of H\mathfrak{H}; the set NN is therefore locally negligible, which completes the proof.

10. Measurable functions defined on a measurable subset

Proposition 15

Let XX be a locally compact space, μ\mu a measure on XX, AA a μ\mu-measurable subset of XX, and ff a mapping of AA into a topological space FF. The following conditions are equivalent:

a) The set H\mathfrak{H} of compact subsets KK of AA such that the restriction of ff to KK is continuous, is μ\mu-dense in AA.

b) There exists a set K\mathfrak{K} of compact subsets of AA, μ\mu-dense in AA, such that the restriction of ff to every K∈KK \in \mathfrak{K} is μK\mu_K-measurable.

c) There exist a homeomorphism jj of FF onto a subspace of a topological space GG and a μ\mu-measurable mapping gg of XX into GG, such that g∣A=j∘fg|A = j \circ f.

d) Every extension of ff to a mapping of XX into FF, constant on X−AX - A, is μ\mu-measurable.

It is clear that a) implies b) and that d) implies c). The fact that c) implies a) follows from condition c) of Prop. 12 of No. 8. On the other hand, b) implies a): for, Def. 1 shows that, for each K∈KK \in \mathscr{K}, the set of subsets H∈HH \in \mathscr{H} contained in KK is μK\mu_K-dense in KK (No. 8, Prop. 12, c)), and Prop. 13 of No. 8 shows that H\mathscr{H} is μ\mu-dense in AA. It remains to see that a) implies d). Let gg be an extension of ff to XX, constant on X−AX - A. For every compact subset LL of XX, L∩AL \cap A and L∩(X−A)L \cap (X - A) are μ\mu-integrable; therefore, for every ε>0\varepsilon > 0, there exist a compact subset P⊂L∩AP \subset L \cap A and a compact subset Q⊂L∩(X−A)Q \subset L \cap (X - A) such that

∣μ∣((L∩A)−P)≤ε/4and∣μ∣((L∩(X−A))−Q)≤ε/4. |\mu|((L \cap A) - P) \leq \varepsilon/4 \quad \text{and} \quad |\mu|\left((L \cap (X - A)) - Q\right) \leq \varepsilon/4.

On the other hand, there exists a set H∈HH \in \mathscr{H} contained in PP such that ∣μ∣(P−H)≤ε/2|\mu|(P - H) \leq \varepsilon/2; the restriction of gg to the compact set K=H∪QK = H \cup Q is then continuous (gg being constant on QQ) and ∣μ∣(L−K)≤ε|\mu|(L - K) \leq \varepsilon, which completes the proof.

Definition 8

Let XX be a locally compact space, μ\mu a measure on XX, and AA a μ\mu-measurable subset of XX. A mapping ff of AA into a topological space FF is said to be μ\mu-measurable if it satisfies the equivalent conditions of Prop. 15.

If AA is locally μ\mu-negligible, then every mapping of AA into FF is therefore μ\mu-measurable.

Corollary 1

Let XX be a locally compact space, μ\mu a measure on XX, AA a μ\mu-measurable subset of XX, and ff a μ\mu-measurable mapping of AA into a topological space FF. Let K\mathscr{K} be a set of compact subsets of XX, μ\mu-dense in XX. Then, there exists a partition of AA formed by a locally negligible set NN and a locally countable family (Kλ)λ∈L(K_\lambda)_{\lambda \in L} of sets Kλ∈KK_\lambda \in \mathscr{K}, such that f∣Kλf|K_\lambda is continuous for every λ∈L\lambda \in L.

In view of No. 9, Prop. 14, it suffices to show that the set H⊂K\mathscr{H} \subset \mathscr{K} of subsets K∈KK \in \mathscr{K} such that K⊂AK \subset A and f∣Kf|K is continuous, is μ\mu-dense in AA. Now, it follows at once from Prop. 1 of No. 1 and condition d) of Prop. 15 that, for every compact subset K0K_0 of AA and every ε>0\varepsilon > 0, there exists a subset K⊂K0K \subset K_0 belonging to K\mathscr{K} such that ∣μ∣(K0−K)≤ε|\mu|(K_0 - K) \leq \varepsilon and f∣Kf|K is continuous; the conclusion therefore follows from Prop. 12 of No. 8.

Corollary 2

Let KK be a compact subspace of XX; in order that a mapping ff of KK into a topological space FF be μ\mu-measurable, it is necessary and sufficient that it be μK\mu_K-measurable.

In view of Lemma 2 of No. 7, this follows at once from Prop. 1 of No. 1 and condition a) of Prop. 15.

Proposition 16

Let A\mathfrak{A} be a locally countable set of μ\mu-measurable subsets of XX and let B=⋃A∈AAB = \bigcup_{A \in \mathfrak{A}} A. For a mapping ff of BB into a topological space FF to be μ\mu-measurable, it is necessary and sufficient that the restriction of ff to every A∈AA \in \mathfrak{A} be μ\mu-measurable.

We have already observed (No. 9) that BB is μ\mu-measurable. The condition being obviously necessary, let us prove that it is sufficient. Thus, let KK be a compact subset of BB. By hypothesis, there exists a sequence (An)(A_n) of sets belonging to A\mathfrak{A} such that the K∩AnK \cap A_n form a covering of KK. Set C0=K∩A0C_0 = K \cap A_0 and

Cn=K∩An∩C(⋃i<nCi) C_n = K \cap A_n \cap \mathbf{C}\left( \bigcup_{i < n} C_i \right)
for n>0n > 0, so that the nonempty CnC_n form a partition of KK into μ\mu-integrable sets. Since the restriction of ff to CnC_n is μ\mu-measurable, there exists a partition of CnC_n formed by a μ\mu-negligible set NnN_n and a sequence (Lmn)m≥0(L_{mn})_{m \geq 0} of compact sets such that f∣Lmnf|L_{mn} is continuous. Since N=⋃nNnN = \bigcup_n N_n is μ\mu-negligible, we see that condition a) of Prop. 15 is satisfied, whence the proposition.

Property d) of Prop. 15 makes it possible to immediately generalize the properties of measurable functions defined on all of XX, observed in Nos. 2 to 5, to measurable functions defined on a measurable subset AA of XX; these generalizations are left to the reader. We only point out explicitly that the principle of localization (No. 2, Prop. 4) remains valid when it is assumed that each of the functions gxg_x is only defined in VxV_x (or almost everywhere in VxV_x) and is measurable.

11. Convergence in measure

Let XX be a locally compact space, μ\mu a measure on XX, AA a μ\mu-measurable subset of XX, and FF a uniform space; we shall denote by S(A,μ;F)\mathscr{S}(A, \mu; F), or SF(A,μ)\mathscr{S}_F(A, \mu) (or simply SF(μ)\mathscr{S}_F(\mu), or even SF\mathscr{S}_F, when A=XA = X) the set of μ\mu-*measurable mappings of AA into F∗F* (No. 10, Def. 8). For every entourage VV of the uniform structure of FF, every μ\mu-integrable set B⊂AB \subset A and every number δ>0\delta > 0, we shall denote by W(V,B,δ)W(V, B, \delta) the set of pairs (f,g)(f, g) of functions in S(A,μ;F)\mathscr{S}(A, \mu; F) having the following property: the set MM of all x∈Bx \in B for which (f(x),g(x))∉V(f(x), g(x)) \notin V is such that ∣μ∣∗(M)≤δ|\mu|^*(M) \leq \delta. Let us show that the sets W(V,B,δ)W(V, B, \delta) form a fundamental system of entourages for a uniform structure on S(A,μ;F)\mathscr{S}(A, \mu; F): it is clear that W(V,B,δ)W(V, B, \delta) is symmetric if VV is, and that if V′⊂V, B′⊃BV' \subset V,\ B' \supset B and δ′≤δ\delta' \leq \delta, then

W(V′,B′,δ′)⊂W(V,B,δ); W(V', B', \delta') \subset W(V, B, \delta);

it therefore suffices to verify the axiom (UIII′)(U'_{III}) (GT, II, §1, No. 1). Now, if V′V' is an entourage such that V′2⊂V{V'}^2 \subset V, then

W(V′,B,δ/2)∘W(V′,B,δ/2)⊂W(V,B,δ). W(V', B, \delta/2) \circ W(V', B, \delta/2) \subset W(V, B, \delta).

Note that as KK runs over a μ\mu-dense set K\mathfrak{K} of compact subsets of AA, the sets W(V,K,δ)W(V, K, \delta) also form a fundamental system of entourages for the preceding uniform structure: for, for every integrable set B⊂AB \subset A, there exists a compact set K∈KK \in \mathfrak{K} contained in BB such that ∣μ∣(B−K)≤δ|\mu|(B - K) \leq \delta, and therefore W(V,K,δ)⊂W(V,B,2δ)W(V, K, \delta) \subset W(V, B, 2\delta).

Definition 9

The uniform structure on S(A,μ;F)\mathscr{S}(A, \mu; F) of which the W(V,B,δ)W(V, B, \delta) form a fundamental system of entourages is called the uniform structure of convergence in measure in AA.*.

The corresponding topology is called the topology of convergence in measure in AA, and a filter (or a sequence) that converges for this topology is said to be convergent in measure in AA; the mention of AA is often suppressed when A=XA = X.

Suppose FF is Hausdorff; then, for every μ\mu-integrable set B⊂AB \subset A, the intersection of the entourages W(V,B,δ)W(V, B, \delta), where VV runs over a fundamental system of entourages of FF and δ\delta runs over the set of numbers >0> 0, is the set of pairs (f,g)(f, g) such that f(x)=g(x)f(x) = g(x) almost everywhere in BB (with respect to μ\mu). For, the set MM of x∈Bx \in B such that f(x)≠g(x)f(x) \neq g(x) is μ\mu-integrable, because it is the inverse image, under the μ\mu-measurable mapping x↦(f(x),g(x))x \mapsto (f(x), g(x)), of the complement of the diagonal in F×FF \times F, which is open (No. 5, Prop. 7); if ∣μ∣(M)=α>0|\mu|(M) = \alpha > 0, there exists a compact subset K⊂MK \subset M such that ∣μ∣(M−K)<α/2|\mu|(M - K) < \alpha/2 and such that the restrictions of ff and gg to KK are continuous; therefore, there exists an entourage V0V_0 of FF such that (f(x),g(x))∉V0(f(x), g(x)) \notin V_0 for all x∈Kx \in K, consequently (f,g)∉W(V0,B,α/2)(f, g) \notin W(V_0, B, \alpha/2).

From this it follows that if FF is Hausdorff, then the intersection of all the entourages of S(A,μ;F)\mathscr{S}(A, \mu; F) is the set of pairs (f,g)(f, g) such that f(x)=g(x)f(x) = g(x) locally almost everywhere in AA. The Hausdorff uniform space associated with S(A,μ;F)\mathscr{S}(A, \mu; F), which we shall denote S(A,μ;F)S(A, \mu; F) or SF(A,μ)S_F(A, \mu) (or even SF(μ)S_F(\mu) or SFS_F when A=XA = X), therefore consists of the equivalence classes for the relation « f(x)=g(x)f(x) = g(x) locally almost everywhere in AA » in the set S(A,μ;F)\mathscr{S}(A, \mu; F).

Proposition 17

— Let (Aλ)λ∈L(A_\lambda)_{\lambda \in L} be a locally countable family of μ\mu-measurable subsets of AA, pairwise disjoint and such that A−⋃λ∈LAλA - \bigcup_{\lambda \in L} A_\lambda is locally μ\mu-negligible. If, for every class f∈S(A,μ;F)f \in S(A, \mu; F) and every λ∈L\lambda \in L, fλf_\lambda denotes the class of the restriction to AλA_\lambda of any of the functions in the class ff, then the mapping ψ:f↦(fλ)λ∈L\psi : f \mapsto (f_\lambda)_{\lambda \in L} is an isomorphism of the uniform space S(A,μ;F)S(A, \mu; F) onto the product uniform space ∏λ∈LS(Aλ,μ;F)\prod_{\lambda \in L} S(A_\lambda, \mu; F).

It follows from No. 10, Prop. 16 that ψ\psi is bijective. Consider an entourage TT of S(A,μ;F)S(A, \mu; F) that is the canonical image of a W(V,B,δ)W(V, B, \delta), where BB is a compact subset of AA; we know that the set JJ of λ∈L\lambda \in L such that B∩Aλ≠∅B \cap A_\lambda \neq \varnothing is countable (No. 9), and ∣μ∣(B)=∑λ∈J∣μ∣(B∩Aλ)|\mu|(B) = \sum_{\lambda \in J} |\mu|(B \cap A_\lambda); therefore, there exists a finite subset HH of JJ such that

∑λ∈J−H∣μ∣(B∩Aλ)≤δ2. \sum_{\lambda \in J - H} |\mu|(B \cap A_\lambda) \leq \frac{\delta}{2}.

The image of TT under ψ×ψ\psi \times \psi is then contained in the canonical image of the product ∏λ∈HW(V,B∩Aλ,δ)\prod_{\lambda \in H} W(V, B \cap A_\lambda, \delta). On the other hand, if mm is the number of elements of HH, then the image of TT under ψ×ψ\psi \times \psi contains the canonical image of the entourage ∏λ∈HW(V,Aλ,δ/2m)\prod_{\lambda \in H} W(V, A_\lambda, \delta/2m), which proves the proposition.

Proposition 18

— *If FF is metrizable, and AA is the union of a locally μ\mu-negligible set and a sequence (An)(A_n) of μ\mu-integrable sets, then the space S(A,μ;F)S(A, \mu; F) is metrizable.

Since each AnA_n is the union of a negligible set and a sequence of compact sets, we can suppose that the AnA_n are already compact and pairwise disjoint. Prop. 17 then allows us to suppose that AA is compact. If (Vn)(V_n) is a countable fundamental system of entourages of FF, it is clear that the W(Vn,A,1/n)W(V_n, A, 1/n) form a fundamental system of entourages of S(A,μ;F)S(A, \mu; F) as nn runs over N\mathbf{N}, whence the proposition.

Lemma 4

— *Let FF be a metrizable uniform space, and let B⊂AB \subset A be a countable union of μ\mu-integrable sets. Then, for every Cauchy sequence (fn)(f_n) in S(A,μ;F)S(A, \mu; F), there exists a subsequence (fnk)(f_{n_k}) of (fn)(f_n) such that (fnk(x))(f_{n_k}(x)) is a Cauchy sequence in FF for almost every x∈Bx \in B.

Suppose first that BB is integrable, and denote by dd a metric compatible with the uniform structure of FF. We shall define recursively a double sequence (fmn)(f_{mn}) of functions in S(A,μ;F)S(A, \mu; F) such that f0n=fnf_{0n} = f_n for every nn, (fmn)n≥0(f_{mn})_{n \geq 0} is a subsequence of (fm−1,n)n≥0(f_{m-1,n})_{n \geq 0} for every m>0m > 0 and, finally, such that for every m>0m > 0 the set MmnM_{mn} of x∈Bx \in B for which d(fmn(x),fm,n+1(x))>1/2m+n+1d(f_{mn}(x), f_{m,n+1}(x)) > 1/2^{m+n+1} has measure ∣μ∣(Mmn)≤1/2m+n+1|\mu|(M_{mn}) \leq 1/2^{m+n+1};

the possibility of such a definition follows from the fact that (fn)(f_n) is a Cauchy sequence in S(A,μ;F)\mathscr{S}(A, \mu; F). Set Mm=⋃n≥0MmnM_m = \bigcup_{n \geq 0} M_{mn}; then

∣μ∣(Mm)≤∑n=0∞∣μ∣(Mmn)≤1/2m |\mu|(M_m) \leq \sum_{n=0}^{\infty} |\mu|(M_{mn}) \leq 1/2^m

and, for every x∈B−Mmx \in B - M_m, we have d(fmn(x),fm,n+p(x))≤1/2m+nd(f_{mn}(x), f_{m,n+p}(x)) \leq 1/2^{m+n} for all n≥0n \geq 0 and all p>0p > 0; the sequence (fmn(x))n≥0(f_{mn}(x))_{n \geq 0} is therefore a Cauchy sequence in FF. Now let N=⋂m=0∞MmN = \bigcap_{m=0}^{\infty} M_m; NN is negligible. Set gn=fnng_n = f_{nn} for every n≥0n \geq 0; for every mm, the sequence (gn)n≥m(g_n)_{n \geq m} is a subsequence of the sequence (fmn)n≥0(f_{mn})_{n \geq 0}; if x∈B−Nx \in B - N, there exists an index mm such that x∉Mmx \notin M_m, which proves that the sequence (gn(x))(g_n(x)) is a Cauchy sequence in FF.

If now BB is the union of a sequence (Bm)(B_m) of integrable sets, one can define recursively a double sequence (gmn)(g_{mn}) such that g0n=fng_{0n} = f_n, (gmn)n≥0(g_{mn})_{n \geq 0} is a subsequence of (gm−1,n)n≥0(g_{m-1,n})_{n \geq 0} for every m>0m > 0, and such that the sequence (gmn(x))n≥0(g_{mn}(x))_{n \geq 0} is a Cauchy sequence in Bm−PmB_m - P_m, where PmP_m is negligible. Set hn=gnnh_n = g_{nn} for every n≥0n \geq 0, so that, for every mm, the sequence (hn)n≥m(h_n)_{n \geq m} is a subsequence of (gmn)n≥0(g_{mn})_{n \geq 0}; the sequence (hn(x))n≥0(h_n(x))_{n \geq 0} is then a Cauchy sequence in FF for every x∈B−Px \in B - P, where P=⋃m=0∞PmP = \bigcup_{m=0}^{\infty} P_m is negligible.

Proposition 19

If the uniform space FF is metrizable and complete, then the uniform space S(A,μ;F)S(A, \mu; F) is complete.

There exists a locally countable family (Kλ)λ∈L(K_{\lambda})_{\lambda \in L} of compact subsets of AA such that the KλK_{\lambda} are pairwise disjoint and A−⋃λKλA - \bigcup_{\lambda} K_{\lambda} is locally negligible (No. 9, Prop. 14). By Prop. 17, S(A,μ;F)S(A, \mu; F) is isomorphic to the product ∏λ∈LS(Kλ,μ;F)\prod_{\lambda \in L} S(K_{\lambda}, \mu; F); we are thus reduced to proving the proposition when AA is integrable; S(A,μ;F)S(A, \mu; F) is then metrizable (Prop. 18) and, by Lemma 4, for every Cauchy sequence (fn)(f_n) in S(A,μ;F)\mathscr{S}(A, \mu; F) there exists a subsequence (fnk)(f_{n_k}) that is convergent in A−NA - N, where NN is negligible; the limit ff of (fnk)(f_{n_k}) (extended in any way whatsoever to all of AA) is then μ\mu-measurable, and it follows from the extension of Egoroff’s theorem mentioned in No. 10 that the sequence (fnk)(f_{n_k}) converges in measure to ff in AA. This implies that ff is a cluster point of the sequence (fn)(f_n) in S(A,μ;F)\mathscr{S}(A, \mu; F), and since the sequence (fn)(f_n) is by hypothesis a Cauchy sequence, it converges to ff.

Q.E.D.

Corollary

— Let F be a metrizable uniform space.

(i) Every sequence (fn)(f_n) of elements of S(A,μ;F)\mathscr{S}(A, \mu; F) that converges locally almost everywhere to a mapping ff (necessarily μ\mu-measurable) of A into F, converges in measure to ff in A.

(ii) Let (fn)(f_n) be a sequence of elements of S(A,μ;F)\mathscr{S}(A, \mu; F) that converges in measure to a mapping ff of A into F. For every set B⊂AB \subset A that is a countable union of integrable sets, there exists a subsequence (fnk)(f_{n_k}) of (fn)(f_n) such that the sequence (fnk(x))(f_{n_k}(x)) converges in F to f(x)f(x) for almost every x∈Bx \in B.

(i) The assertion follows at once from the extension of Egoroff’s theorem mentioned in No. 10.

(ii) By Lemma 4, there exists a subsequence (fnk)(f_{n_k}) of (fn)(f_n) such that (fnk(x))(f_{n_k}(x)) is a Cauchy sequence in F for every x∈B−Nx \in B - N, where N is negligible; let f′(x)∈F^f'(x) \in \widehat{F} be the limit of this sequence for x∈B−Nx \in B - N. It is clear that f′f' is a μ\mu-measurable mapping of B−NB - N into F^\widehat{F}, and the sequence (fn)(f_n) converges in measure to f′f' in B−NB - N by (i); f′f' is therefore equal to ff almost everywhere in B.

Proposition 20

— Let F be a Banach space, equipped with the uniform structure defined by its norm.

(i) For every μ\mu-measurable subset A of X, the topology of convergence in measure is compatible with the vector space structure of S(A,μ;F)\mathscr{S}(A, \mu; F).

(ii) The space K(X;F)\mathscr{K}(X; F) is dense in S(X,μ;F)\mathscr{S}(X, \mu; F).

(iii) For every real number p≥1p \geq 1, the topology induced on the space LFp(X,μ)\mathscr{L}_F^p(X, \mu) by the topology of convergence in measure is coarser than the topology of convergence in mean of order pp.

(i) For every μ\mu-integrable subset B of A and every δ>0\delta > 0, denote by T(B,δ)T(B, \delta) the set of f∈S(A,μ;F)f \in \mathscr{S}(A, \mu; F) for which the set C of x∈Bx \in B such that ∣f(x)∣≥δ|f(x)| \geq \delta satisfies the relation ∣μ∣(C)≤δ|\mu|(C) \leq \delta; it is clear that if VδV_\delta is the entourage of F formed by the pairs (y,z)(y, z) such that ∣y−z∣≤δ|y - z| \leq \delta, the entourage W(Vδ,B,δ)W(V_\delta, B, \delta) is the set of pairs (f,g)(f, g) of measurable mappings of A into F such that f−g∈T(B,δ)f - g \in T(B, \delta). It is clear that the sets T(B,δ)T(B, \delta) are symmetric, and that T(B,δ)+T(B,δ)⊂T(B,2δ)T(B, \delta) + T(B, \delta) \subset T(B, 2\delta) and T(B,∣α∣δ)⊂αT(B,δ)T(B, |\alpha|\delta) \subset \alpha T(B, \delta) for every nonzero scalar α\alpha such that ∣α∣≤1|\alpha| \leq 1; it therefore suffices to verify that the sets T(B,δ)T(B, \delta) are absorbent (TVS, I, §1, No. 5, Prop. 4). Now, if ff is a μ\mu-measurable mapping of A into F, the numerical function ∣f∣|f| is also μ\mu-measurable (No. 3, Cor. 6 of Th. 1). Let CnC_n be the set of x∈Bx \in B such that ∣f(x)∣≥n|f(x)| \geq n; the CnC_n form a decreasing sequence of integrable sets whose intersection is empty; therefore there exists an integer nn such that ∣μ∣(Cn)≤δ|\mu|(C_n) \leq \delta (§ 4, No. 5, Cor. of Prop. 7); we can moreover suppose that nn is taken sufficiently large that 1/n≤δ1/n \leq \delta; then f/n2∈T(B,δ)f/n^2 \in T(B, \delta), which completes the proof of assertion (i).

(iii) The relation ∫∣f∣pd∣μ∣≤δp+1\int |f|^p d|\mu| \leq \delta^{p+1} implies that if CC is the set of x∈Xx \in X such that ∣f(x)∣≥δ|f(x)| \geq \delta, then

δp∣μ∣∗(C)≤∫∣f∣pd∣μ∣≤δp+1, \delta^p |\mu|^*(C) \leq \int |f|^p d|\mu| \leq \delta^{p+1},

whence ∣μ∣∗(C)≤δ|\mu|^*(C) \leq \delta, which proves (iii).

(ii) In view of (iii), it suffices to show for example that LF1\mathscr{L}_F^1 is dense in SF\mathscr{S}_F, since by definition H(X;F)\mathscr{H}(X; F) is dense in LF1\mathscr{L}_F^1 for the topology of convergence in mean. Now, let ff be any element of SF\mathscr{S}_F and let T(B,δ)T(B, \delta) be a neighborhood of 0 in this space; we see as in (i) that there exists an integrable subset CC of BB such that ∣μ∣(C)≤δ|\mu|(C) \leq \delta and such that ff is bounded on B−CB - C; denoting then by gg the function equal to ff on B−CB - C and to 0 on X−(B−C)X - (B - C), it follows from No. 6, Th. 5 that gg is integrable, and obviously f−g∈T(B,δ)f - g \in T(B, \delta).

Remark

— 1) The topological vector space S(X,μ;F)\mathscr{S}(X, \mu; F) is not necessarily locally convex (Exer. 24). 2) The topology induced on the set of ff such that Np(f)≤1N_p(f) \leq 1 by the topology of convergence in measure may be strictly coarser than the topology induced on this set by the topology of convergence in mean of order pp (Exer. 22). However, see Prop. 21 below.

Definition 10

Let XX be a locally compact space, μ\mu a measure on XX, FF a Banach space, and p∈[1,+∞[p \in [1, +\infty[. A subset HH of LFp(X,μ)\mathscr{L}_F^p(X, \mu) is said to be equi-integrable of order pp (for μ\mu) if it satisfies the following conditions:

(i) For every ε>0\varepsilon > 0 there exists a δ>0\delta > 0 such that, for every integrable set AA of measure ∣μ∣(A)≤δ|\mu|(A) \leq \delta and every f∈Hf \in H,

∫∣f∣pφAd∣μ∣≤ε. \int |f|^p \varphi_A d|\mu| \leq \varepsilon.

(ii) For every ε>0\varepsilon > 0 there exists a compact subset KK of XX such that, for every f∈Hf \in H, ∫∣f∣pφX−Kd∣μ∣≤ε\int |f|^p \varphi_{X-K} d|\mu| \leq \varepsilon.

When p=1p = 1 we say 'equi-integrable' instead of 'equi-integrable of order 1'.

Remark

— Suppose μ\mu is bounded. For every a>0a > 0, the set of measurable mappings of XX into FF such that ∣f(x)∣≤a|f(x)| \leq a almost everywhere is equi-integrable of order pp, and this is true for any p∈[1,+∞[p \in [1, +\infty[.

Proposition 21

Let HH be a subset of LFp(X,μ)\mathscr{L}_F^p(X, \mu) that is equi-integrable of order pp. On HH, the uniform structure of convergence in measure is equal to the uniform structure induced by that of LFp(X,μ)\mathscr{L}_F^p(X, \mu).

Let ε>0\varepsilon > 0. There exist δ\delta and KK with the properties (i) and (ii) of Def. 10. Let f,gf, g in HH be such that

∣f(x)−g(x)∣≤(ε∣μ∣(K))1/p |f(x) - g(x)| \leq \left( \frac{\varepsilon}{|\mu|(K)} \right)^{1/p}

for x∈Kx \in K, except on a set MM of measure ≤δ\leq \delta. Then

(∫X−K∣f−g∣pd∣μ∣)1/p≤(∫X−K∣f∣pd∣μ∣)1/p+(∫X−K∣g∣pd∣μ∣)1/p \left( \int_{X-K} |f-g|^p d|\mu| \right)^{1/p} \leq \left( \int_{X-K} |f|^p d|\mu| \right)^{1/p} + \left( \int_{X-K} |g|^p d|\mu| \right)^{1/p}
≤2ε1/p \leq 2\varepsilon^{1/p}

and similarly

(∫M∣f−g∣pd∣μ∣)1/p≤2ε1/p, \left( \int_M |f-g|^p d|\mu| \right)^{1/p} \leq 2\varepsilon^{1/p},
therefore
∫∣f−g∣pd∣μ∣=∫X−K∣f−g∣pd∣μ∣+∫M∣f−g∣pd∣μ∣+∫K−M∣f−g∣pd∣μ∣ \int |f-g|^p d|\mu| = \int_{X-K} |f-g|^p d|\mu| + \int_M |f-g|^p d|\mu| + \int_{K-M} |f-g|^p d|\mu|
≤2pε+2pε+ε∣μ∣(K)∣μ∣(K−M)≤(2p+1+1)ε. \leq 2^p \varepsilon + 2^p \varepsilon + \frac{\varepsilon}{|\mu|(K)} |\mu|(K-M) \leq (2^{p+1} + 1)\varepsilon.

Thus, the uniform structure of convergence in measure on HH is finer than the uniform structure induced by that of LFp(X,μ)\mathscr{L}_F^p(X, \mu). It then suffices to apply Prop. 20.

12. A property of vague convergence

Lemma 5

Let XX be a locally compact space, μ\mu a bounded positive measure on XX, FF a Banach space, and ff a bounded function on XX with values in FF. The following conditions are equivalent:

(i) The set of points of discontinuity of ff is μ\mu-negligible.

(ii) For every ε>0\varepsilon > 0, there exist elements a1,…,ana_1, \ldots, a_n of FF, functions g1,…,gng_1, \ldots, g_n belonging to K(X)\mathscr{K}(X), and a bounded continuous function h≥0h \geq 0 on XX such that ∣f−g1a1−⋯−gnan∣≤h≤2sup⁡∣f∣|f - g_1 a_1 - \cdots - g_n a_n| \leq h \leq 2 \sup |f| on XX, and ∫hdμ≤ε\int h d\mu \leq \varepsilon.

Denote by NN the set of points of discontinuity of ff, and let M=sup⁡∣f∣M = \sup |f|.

(i) ⇒\Rightarrow (ii). Suppose that condition (i) is satisfied. Let ε>0\varepsilon > 0. The function ff is μ\mu-integrable (No. 2, Cor. 4 of Prop. 5, and No. 6, Th. 5), therefore there exist a1,…,ana_1, \ldots, a_n in FF and g1,…,gng_1, \ldots, g_n in K(X)\mathscr{K}(X) such that, on setting k=∣f−g1a1−⋯−gnan∣k = |f - g_1 a_1 - \cdots - g_n a_n|, we have ∫kdμ≤ε/2\int k d\mu \leq \varepsilon/2 (§ 3, No. 5,

Prop. 10). Multiplying g1,…,gng_1, \ldots, g_n by a suitable same element of K(X)\mathscr{K}(X), we can suppose in addition that

∣g1a1+⋯+gnan∣≤M=sup⁡∣f∣ |g_1 a_1 + \cdots + g_n a_n| \leq M = \sup |f|

on X, whence k≤2Mk \leq 2M. The set N′N' of points of discontinuity of kk is contained in NN, hence is negligible. For every x∈Xx \in X, set l(x)=lim sup⁡y→xk(y)l(x) = \limsup_{y \to x} k(y).

Then 2M≥l≥k2M \geq l \geq k on X, and l=kl = k on X−N′X - N', that is, almost everywhere for μ\mu, therefore ∫l dμ≤ε/2\int l \, d\mu \leq \varepsilon / 2. On the other hand, ll is bounded and upper semi-continuous, hence is the lower envelope of the set of bounded continuous functions ≥l\geq l. Therefore there exists a bounded continuous function h≥lh \geq l on X such that h≤2Mh \leq 2M and ∫h dμ≤∫l dμ+ε/2\int h \, d\mu \leq \int l \, d\mu + \varepsilon / 2 (§ 4, No. 4, Cor. 2 of Prop. 5). Then ∫h dμ≤ε\int h \, d\mu \leq \varepsilon and ∣f−g1a1−⋯−gnan∣≤h|f - g_1 a_1 - \cdots - g_n a_n| \leq h.

(ii) ⇒\Rightarrow (i). Suppose that condition (ii) is satisfied. For every x∈Xx \in X let ω(x)\omega(x) be the oscillation of f at x (GT, IX, § 2, No. 3). Let ε>0\varepsilon > 0. There exist a1,…,an,g1,…,gn,ha_1, \ldots, a_n, g_1, \ldots, g_n, h with the properties of (ii). For every x∈Xx \in X, ω(x)\omega(x) is the oscillation of f−g1a1−⋯−gnanf - g_1 a_1 - \cdots - g_n a_n at x, therefore ω(x)≤2h(x)\omega(x) \leq 2h(x). Thus ∫ω dμ≤2ε\int \omega \, d\mu \leq 2\varepsilon. Consequently, the set AεA_\varepsilon of x∈Xx \in X such that ω(x)≥ε\omega(x) \geq \sqrt{\varepsilon} satisfies μ(Aε)≤2ε\mu(A_\varepsilon) \leq 2\sqrt{\varepsilon}. This proves that μ(N)≤2ε\mu(N) \leq 2\sqrt{\varepsilon}, whence μ(N)=0\mu(N) = 0.

Proposition 22

— *Let F be a Banach space, X a locally compact space, E\mathscr{E} the set of bounded positive measures on X, μ\mu an element of E\mathscr{E}, and B\mathscr{B} a filter base on E\mathscr{E}. Assume that B\mathscr{B} converges vaguely to μ\mu and that ∥ν∥\| \nu \| tends to ∥μ∥\| \mu \| with respect to B\mathscr{B}. Let f be a mapping of X into F satisfying the following conditions:

(i) f is bounded, and is integrable for μ\mu and for every measure that belongs to some element of B\mathscr{B}; (ii) the set of points of discontinuity of f is μ\mu-negligible. Then ∫f dν\int f \, d\nu tends to ∫f dμ\int f \, d\mu with respect to B\mathscr{B}.

Let ε>0\varepsilon > 0. There exist elements a1,…,ana_1, \ldots, a_n in F, functions g1,…,gng_1, \ldots, g_n in K(X)\mathscr{K}(X), and a bounded continuous function h≥0h \geq 0 on X, such that

∣f−g1a1−⋯−gnan∣≤h≤2sup⁡∣f∣ |f - g_1 a_1 - \cdots - g_n a_n| \leq h \leq 2 \sup |f|

on X and ∫h dμ≤ε\int h \, d\mu \leq \varepsilon (Lemma 5). Let M=sup⁡∣f∣M = \sup |f|. There exist a compact subset K of X such that μ∗(X−K)≤ε\mu^*(X - K) \leq \varepsilon (§ 4, No. 7, Prop. 12 and No. 6, Th. 4), a compact neighborhood K′K' of K in X, and a continuous mapping h′h' of X into [0,2M][0, 2M] such that h′=hh' = h on K, h′=2Mh' = 2M on X−K′X - K'; replacing h′h' by sup⁡(h,h′)\sup(h, h'), we can suppose in addition that h′≥hh' \geq h. Then ∫(h′−h) dμ≤2Mμ∗(X−K)≤2Mε\int (h' - h) \, d\mu \leq 2M \mu^*(X - K) \leq 2M \varepsilon. On the other hand, h′=h1+2Mh' = h_1 + 2M, where h1∈K(X)h_1 \in \mathscr{K}(X). Taking into account § 4, No. 7, Prop. 12, the number

∫h′ dν=∫h1 dν+2M∥ν∥ tends to ∫h1 dμ+2M∥μ∥=∫h′ dμ with respect to B. There then exists an A∈B such that, for every ν∈A, \int h' \, d\nu = \int h_1 \, d\nu + 2M \|\nu\| \text{ tends to } \int h_1 \, d\mu + 2M \|\mu\| = \int h' \, d\mu \text{ with respect to } \mathscr{B}. \text{ There then exists an } A \in \mathscr{B} \text{ such that, for every } \nu \in A,
∣∫(g1a1+⋯+gnan) dν−∫(g1a1+⋯+gnan) dμ∣≤ε, \left| \int (g_1 a_1 + \cdots + g_n a_n) \, d\nu - \int (g_1 a_1 + \cdots + g_n a_n) \, d\mu \right| \leq \varepsilon,
∫h dν≤∫h′ dν≤∫h′ dμ+ε≤∫h dμ+2Mε+ε≤2(M+1)ε. \int h \, d\nu \leq \int h' \, d\nu \leq \int h' \, d\mu + \varepsilon \leq \int h \, d\mu + 2M \varepsilon + \varepsilon \leq 2(M + 1)\varepsilon.
These inequalities imply
∣∫f dν−∫f dμ∣≤ \left| \int f \, d\nu - \int f \, d\mu \right| \leq
∫h dν+∣∫(g1a1+⋯+gnan) dν−∫(g1a1+⋯+gnan) dμ∣+∫h dμ \int h \, d\nu + \left| \int (g_1 a_1 + \cdots + g_n a_n) \, d\nu - \int (g_1 a_1 + \cdots + g_n a_n) \, d\mu \right| + \int h \, d\mu
≤2(M+2)ε, \leq 2(M + 2)\varepsilon,
which proves the proposition.

Remark

The conditions (i) and (ii) of Proposition 22 are satisfied if ff is continuous and bounded.

Example

Let us take for XX the compact space U\mathbf{U} of complex numbers of absolute value 1. On setting μ(f)=∫01f(e2iπt) dt\mu(f) = \int_0^1 f(e^{2i\pi t}) \, dt for every f∈H(U)f \in \mathscr{H}(\mathbf{U}), one defines a positive measure of mass 1 on U\mathbf{U}. On the other hand, let θ\theta be a real number; for every integer n≥0n \geq 0, let νn\nu_n be the unit mass placed at the point e2iπnθe^{2i\pi n \theta} of U\mathbf{U}, and let

μn=1n+1(ν0+⋯+νn), \mu_n = \frac{1}{n+1} (\nu_0 + \cdots + \nu_n),
so that μn\mu_n is a positive measure of mass 1 on U\mathbf{U}. Then, if θ\theta is irrational, μn\mu_n tends vaguely to μ\mu. For, since the linear combinations of the functions z↦zkz \mapsto z^k (k∈Zk \in \mathbf{Z}) are dense in H(U)\mathscr{H}(\mathbf{U}) (GT, X, §4, No. 4, Prop. 8), it suffices to prove that μn(zk)\mu_n(z^k) tends to μ(zk)\mu(z^k) for k∈Zk \in \mathbf{Z}. Now, for k=0k = 0, μn(zk)=μ(zk)=1\mu_n(z^k) = \mu(z^k) = 1; for k≠0k \neq 0,
μn(zk)=1n+1(1+e2iπkθ+e4iπkθ+⋯+e2iπknθ). \mu_n(z^k) = \frac{1}{n+1} (1 + e^{2i\pi k \theta} + e^{4i\pi k \theta} + \cdots + e^{2i\pi k n \theta}).
Since e2iπkθ≠1e^{2i\pi k \theta} \neq 1 (because θ\theta is irrational), we infer that
∣μn(zk)∣=∣1n+1e2iπk(n+1)θ−1e2iπkθ−1∣≤1n+12∣e2iπkθ−1∣, |\mu_n(z^k)| = \left| \frac{1}{n+1} \frac{e^{2i\pi k(n+1)\theta} - 1}{e^{2i\pi k \theta} - 1} \right| \leq \frac{1}{n+1} \frac{2}{|e^{2i\pi k \theta} - 1|},

therefore μn(zk)\mu_n(z^k) tends to 0=μ(zk)0 = \mu(z^k). Under these conditions, Proposition 22 can be applied, and we see in particular that if AA is a subset of UU with negligible boundary with respect to μ\mu, then μn(A)\mu_n(A) tends to μ(A)\mu(A). In other words, if pnp_n denotes the number of integers k∈[0,n]k \in [0, n] such that e2iπkθ∈Ae^{2i\pi k \theta} \in A, then n−1pnn^{-1} p_n tends to μ(A)\mu(A) as nn tends to +∞+\infty.

Exercises

See the exercises for § 5.

Transcribed from the printed volume. Not translated.