Bourbaki

§ 3. Measures and additive set functions

In this section, we shall denote by K(T)\mathscr{K}(T) and B(T)\mathscr{B}(T), respectively, the set of compact subsets of a Hausdorff topological space T and the Borel tribe of T.

1. Measures and additive set functions of compact sets

Theorem 1

— Let T be a topological space, and I a mapping of K(T)\mathscr{K}(T) into R+\mathbf{R}_+. In order that there exist a measure μ\mu on T such that I(K)=μ∙(K)I(K) = \mu^\bullet(K) for all K∈K(T)K \in \mathscr{K}(T), it is necessary and sufficient that I satisfy the following conditions:

1) If K and L are compact subsets of T such that K⊂LK \subset L, then I(K)≤I(L)I(K) \leq I(L) ('I is increasing').

2) If K and L are compact subsets of T, then I(K∪L)≤I(K)+I(L)I(K \cup L) \leq I(K) + I(L).

3) If K and L are disjoint compact subsets of T, then I(K∪L)=I(K)+I(L)I(K \cup L) = I(K) + I(L) ('I is additive').

4) For every decreasing directed family (Kα)α∈A(K_\alpha)_{\alpha \in A} of compact subsets of T, one has I(⋂α∈AKα)=inf⁡α∈AI(Kα)I(\bigcap_{\alpha \in A} K_\alpha) = \inf_{\alpha \in A} I(K_\alpha).

5) For every x∈Tx \in T, there exists a neighborhood V of x such that

sup⁡K∈K(T)K⊂VI(K)<+∞ \sup_{\substack{K \in \mathscr{K}(T) \\ K \subset V}} I(K) < +\infty
('I is locally bounded').

The measure μ\mu is then unique.

The uniqueness of μ\mu follows from the Cor. of Prop. 2 of § 1, No. 2. The above conditions are necessary, the first three in obvious fashion, the last from the fact that μ∙\mu^\bullet is a locally bounded encumbrance, and condition 4) by the Cor. of Prop. 5 of § 1, No. 6.

To show that these conditions are sufficient, we begin by treating the case that T is compact.

Lemma 1

Assume that T is compact, and set l = I(T). For every A⊂TA \subset T, set

J(A)=sup⁡K∈K(T)K⊂AI(K) J(A) = \sup_{\substack{K \in \mathscr{K}(T) \\ K \subset A}} I(K)

and let Φ\Phi be the set of A⊂TA \subset T such that J(A)+J(CA)=lJ(A) + J(\mathbf{C}A) = l. The set Φ\Phi is then a clan that contains K(T)\mathscr{K}(T), and the function JJ on Φ\Phi is increasing and additive.

It is clear that JJ is an increasing set function, extending II, and that J(A)+J(CA)≤lJ(A) + J(\mathbf{C}A) \leq l for every A⊂TA \subset T.

Let KK and SS be two compact sets in TT; we are first going to show that

J(K∩S)+J(CK∩S)=J(S). J(K \cap S) + J(\mathbf{C}K \cap S) = J(S).

On considering the restrictions of II to K(S)\mathscr{K}(S) and of JJ to P(S)\mathfrak{P}(S), one reduces immediately to the case that S=TS = T. Since TT is normal, KK is the intersection of the decreasing directed family of its compact neighborhoods, and condition 4) implies the existence, for every ε>0\varepsilon > 0, of a compact neighborhood HH of KK such that I(H)≤I(K)+εI(H) \leq I(K) + \varepsilon. Let LL be the closure of T−HT - H; LL is compact, L∩K=∅L \cap K = \varnothing and H∪L=TH \cup L = T, therefore l=I(H∪L)≤I(H)+I(L)≤I(K)+I(L)+εl = I(H \cup L) \leq I(H) + I(L) \leq I(K) + I(L) + \varepsilon (condition 2)), whence the relation J(K)+J(CK)≥I(K)+I(L)≥l−εJ(K) + J(\mathbf{C}K) \geq I(K) + I(L) \geq l - \varepsilon. Since ε\varepsilon is arbitrary, J(K)+J(CK)=lJ(K) + J(\mathbf{C}K) = l. This proves the formula (2), as well as the inclusion K(T)⊂Φ\mathscr{K}(T) \subset \Phi.

Let us now prove that Φ\Phi is a clan. Since Φ\Phi is obviously stable under passage to the complement, it suffices to show that if A1A_1 and A2A_2 denote elements of Φ\Phi, then A1∪A2∈ΦA_1 \cup A_2 \in \Phi, or again that

J(A1∪A2)+J(C(A1∪A2))≥l. J(A_1 \cup A_2) + J(\mathbf{C}(A_1 \cup A_2)) \geq l.

Denote by ε\varepsilon a number >0> 0, and, for i=1,2i = 1, 2, let KiK_i be a compact set contained in AiA_i, and LiL_i a compact set contained in CAi\mathbf{C}A_i, such that

I(Ki)≥J(Ai)−ε,I(Li)≥J(CAi)−ε. I(K_i) \geq J(A_i) - \varepsilon, \quad I(L_i) \geq J(\mathbf{C}A_i) - \varepsilon.

Set M1=K1∪L1M_1 = K_1 \cup L_1; the relations l=J(M1)+J(CM1)l = J(M_1) + J(\mathbf{C}M_1) and

J(M1)=I(K1)+I(L1)≥J(A1)+J(CA1)−2ε=l−2ε J(M_1) = I(K_1) + I(L_1) \geq J(A_1) + J(\mathbf{C}A_1) - 2\varepsilon = l - 2\varepsilon

imply that J(CM1)≤2εJ(\mathbf{C}M_1) \leq 2\varepsilon. Then if SS is a compact subset of TT, the relation (2) (applied to K=M1K = M_1) implies J(S)≤J(M1∩S)+2εJ(S) \leq J(M_1 \cap S) + 2\varepsilon, whence

J(S)≤J(K1∩S)+J(L1∩S)+2ε. J(S) \leq J(K_1 \cap S) + J(L_1 \cap S) + 2\varepsilon.

Let us add the inequalities obtained by making S=K2S = K_2 and S=L2S = L_2, and take into account the inequality J(K2)+J(L2)≥l−2εJ(K_2) + J(L_2) \geq l - 2\varepsilon and the fact that K1∩K2K_1 \cap K_2, L1∩K2L_1 \cap K_2 and K1∩L2K_1 \cap L_2 are three disjoint compact sets contained in A1∪A2A_1 \cup A_2. Denoting by CC the union of these three compact sets, it follows that

l−2ε≤J(K2)+J(L2)≤J(C)+J(L1∩L2)+4ε≤J(A1∪A2)+J(C(A1∪A2))+4ε, l - 2\varepsilon \leq J(K_2) + J(L_2) \leq J(C) + J(L_1 \cap L_2) + 4\varepsilon \\ \leq J(A_1 \cup A_2) + J(\mathscr{C}(A_1 \cup A_2)) + 4\varepsilon,

whence immediately the desired formula (3), in view of the arbitrariness of ε\varepsilon. This having been established, the preceding inequalities imply that J(C)≥J(A1∪A2)−6εJ(C) \geq J(A_1 \cup A_2) - 6\varepsilon; if A1A_1 and A2A_2 are disjoint, then CC is the union of K1∩L2⊂A1K_1 \cap L_2 \subset A_1 and K2∩L1⊂A2K_2 \cap L_1 \subset A_2, from which one deduces that J(A1∪A2)≤J(A1)+J(A2)J(A_1 \cup A_2) \leq J(A_1) + J(A_2). The reverse inequality being obvious, JJ is indeed additive on Φ\Phi, and the lemma is established.

Let us complete the proof of the theorem for the case that TT is compact. Let E(Φ)\mathscr{E}(\Phi) be the vector space of Φ\Phi-step functions on TT, equipped with the topology of uniform convergence (Ch. IV, §4, No. 9, Def. 4); we shall again denote by JJ the positive linear form on E(Φ)\mathscr{E}(\Phi) associated with the additive function JJ (loc. cit., Prop. 18). Since J(T)=lJ(T) = l, JJ is continuous and has norm ll. Let then H\mathscr{H} be the closure of E(Φ)\mathscr{E}(\Phi) for the topology of uniform convergence; one verifies at once that JJ may be extended by continuity to a positive linear form on H\mathscr{H}, again denoted JJ. Since H\mathscr{H} contains C(T)\mathscr{C}(T) (loc. cit., No. 10, Prop. 19) the restriction of JJ to C(T)\mathscr{C}(T) is a positive measure μ\mu. It remains to show that μ∙(K)=I(K)\mu^\bullet(K) = I(K) for every compact subset KK of TT. Now, we have μ∙(K)=inf⁡f∈SKμ∙(f)\mu^\bullet(K) = \inf_{f \in S_K} \mu^\bullet(f), where SKS_K denotes the set of elements of C(T)\mathscr{C}(T) that are ≥φK\geq \varphi_K (§ 1, No. 6, Prop. 5). Since J(f)=μ∙(f)J(f) = \mu^\bullet(f) for f∈C(T)f \in \mathscr{C}(T), it clearly suffices to show that J(K)≥inf⁡f∈SKJ(f)J(K) \geq \inf_{f \in S_K} J(f). As in the proof of Lemma 1, let HH be a compact neighborhood of KK such that J(H)≤J(K)+εJ(H) \leq J(K) + \varepsilon, and let ff be a continuous function on TT, between 0 and 1, equal to 1 on KK and to 0 outside HH (GT, IX, §4, No. 1, Prop. 1). Then

J(f)≤J(H)≤J(K)+ε; J(f) \leq J(H) \leq J(K) + \varepsilon;

ε\varepsilon being arbitrary, the desired inequality is proved, and the theorem is thus established when TT is compact.

Let us now pass to the general case. For every compact set LL in TT, let ILI_L be the restriction of II to K(L)\mathfrak{K}(L). By the special case just treated, there exists a measure μL\mu_L on LL, unique, such that μL(K)=IL(K)\mu_L(K) = I_L(K) for every compact set K⊂LK \subset L. Let then L′L' be a compact set contained in LL; we have (μL)L′∙(K)=μL′∙(K)=μL′∙(K)(\mu_L)_{L'}^\bullet(K) = \mu_{L'}^\bullet(K) = \mu_{L'}^\bullet(K) for every compact set K⊂L′K \subset L', therefore μL′=(μL)L′\mu_{L'} = (\mu_L)_{L'}; the mapping μ:L↦μL\mu : L \mapsto \mu_L is a premeasure. The condition 5) expresses that μ\mu is a measure, and the relation I(K)=μ∗(K)I(K) = \mu^*(K) for all compact K⊂TK \subset T is obvious.

Remark 1

The condition 4) may be replaced, in the statement of Theorem 1, by the following condition ('right-continuity'): 4′)4') For every K∈K(T)K \in \mathscr{K}(T) and every ε>0\varepsilon > 0, there exists an open set UU containing KK, such that I(H)≤I(K)+εI(H) \leq I(K) + \varepsilon for every compact set H⊂UH \subset U. For, if μ\mu is a measure, the function I:K↦μ∗(K)I : K \mapsto \mu^*(K) satisfies 4′4' (§ 1, No. 9, Prop. 13). Conversely, suppose that II satisfies 1) and 4′4'; let us show that II then satisfies 4). With notations as in the statement of Theorem 1, choose an ε>0\varepsilon > 0 and an open set UU containing the compact set K=⋂α∈AKαK = \bigcap_{\alpha \in A} K_\alpha and such that 4′4') is satisfied. There then exists an index β∈A\beta \in A such that Kβ⊂UK_\beta \subset U, and this implies

inf⁡α∈AI(Kα)≤I(Kβ)≤I(K)+ε \inf_{\alpha \in A} I(K_\alpha) \leq I(K_\beta) \leq I(K) + \varepsilon
and 4) is indeed verified.

Remark 2

The set of conditions 2) and 3) may be replaced, in the statement of Theorem 1, by the following condition: If KK and LL are compact subsets of TT, then

I(K∪L)+I(K∩L)=I(K)+I(L). I(K \cup L) + I(K \cap L) = I(K) + I(L).
Indeed, this condition implies 2) and 3), and on the other hand
μ∗(K∪L)=μ∗(K∩L)=μ∗(K)+μ∗(L) \mu^*(K \cup L) = \mu^*(K \cap L) = \mu^*(K) + \mu^*(L)
for every measure μ\mu, by virtue of the relation φK∪L+φK∩L=φK+φL\varphi_{K \cup L} + \varphi_{K \cap L} = \varphi_K + \varphi_L between the characteristic functions.

2. Inner regular set functions

Definition 1

Let TT be a topological space, and let B(T)\mathscr{B}(T) be the Borel tribe of TT; let II be a mapping of B(T)\mathscr{B}(T) into R‾+\overline{\mathbf{R}}_+. a) II is said to be countably additive if, for every sequence (An)(A_n) of pairwise disjoint elements of B(T)\mathscr{B}(T),

I(⋃nAn)=∑nI(An). I\left( \bigcup_n A_n \right) = \sum_n I(A_n).
b) II is said to be inner regular if, for every set A∈B(T)A \in \mathscr{B}(T),
I(A)=sup⁡KI(K), I(A) = \sup_K I(K),
where KK runs over the set of compact subsets of AA.

c) I is said to be bounded (resp. locally bounded) if I(T)<+∞I(T) < +\infty (resp. if every point x∈Tx \in T admits an open neighborhood VV such that I(V)<+∞I(V) < +\infty).

Remark 1

The condition a) clearly implies that I is an increasing mapping of B(T)\mathscr{B}(T) (ordered by inclusion) into R‾+\overline{\mathbf{R}}_+.

Remark 2

Suppose that I is countably additive; let (An)n∈N(A_n)_{n \in \mathbf{N}} be an increasing sequence of Borel sets, and let A=⋃n∈NAnA = \bigcup_{n \in \mathbf{N}} A_n. The sets D0=A0D_0 = A_0, Dn=An−An−1D_n = A_n - A_{n-1} being pairwise disjoint, and their union being A, we have I(A)=∑nI(Dn)=lim⁡n→∞I(An)I(A) = \sum_n I(D_n) = \lim_{n \to \infty} I(A_n). Similarly, if (Bn)(B_n) is a decreasing sequence of Borel sets, and if I(B0)<+∞I(B_0) < +\infty, then I(⋂nBn)=lim⁡n→∞I(Bn)I(\bigcap_n B_n) = \lim_{n \to \infty} I(B_n): it suffices to apply the preceding to the sets An=B0−BnA_n = B_0 - B_n.

Remark 3

Let (An)(A_n) be any sequence of Borel sets of T. If I is countably additive, then I(⋃nAn)≤∑nI(An)I(\bigcup_n A_n) \leq \sum_n I(A_n). By the preceding remark, it suffices to establish this inequality for a finite sequence. One immediately reduces to the case of two sets A1A_1 and A2A_2; but the relation (4) implies that

I(A1∪A2)=I(A1)+I(A2−(A1∩A2))≤I(A1)+I(A2). I(A_1 \cup A_2) = I(A_1) + I(A_2 - (A_1 \cap A_2)) \leq I(A_1) + I(A_2).

The desired inequality then follows immediately.

Remark 4

If I is a countably additive and locally bounded function, the preceding remark implies at once that I(K)<+∞I(K) < +\infty for every compact set K⊂TK \subset T.

Remark 5

One can show that if I is additive, that is, satisfies (4) for finite sequences, and if I is inner regular, then I is countably additive (Exer. 7). The reader can also ascertain that only additivity and inner regularity are used in the proof of Th. 2 below.

Theorem 2

— Let T be a topological space, and let I be a function defined on B(T)\mathscr{B}(T), with values in R‾+\overline{\mathbf{R}}_+. In order that there exist a measure μ\mu on T such that μ∗(A)=I(A)\mu^*(A) = I(A) for every A∈B(T)A \in \mathscr{B}(T), it is necessary and sufficient that I be countably additive, locally bounded and inner regular. The measure μ\mu is then unique.

These three conditions are necessary: for, the mapping A↦μ∗(A)A \mapsto \mu^*(A) on B(T)\mathscr{B}(T) is countably additive (§ 1, No. 5, Cor. of Prop. 4), locally bounded by the definition of measures (§ 1, No. 2, Def. 5), and inner regular by Remark 3 of § 1, No. 2.

We pass to existence. It is clear that the restriction of I to K(T)\mathscr{K}(T) satisfies conditions 1), 2), 3) and 5) of the statement of Th. 1; let us show that 4) is satisfied as well. Let K be a compact subset of T, the intersection of a decreasing directed family (Kα)α∈A(K_\alpha)_{\alpha \in A} of compact sets, and let ε\varepsilon be a number >0> 0; I being locally bounded, there exists an open (hence Borel) neighborhood V of K such that I(V)<+∞I(V) < +\infty, and then there exists an index α\alpha such that Kα⊂VK_\alpha \subset V; changing notation if necessary, we can suppose that Kα⊂VK_\alpha \subset V for all α∈A\alpha \in A. By the inner regularity of I, there exists a compact set L⊂V−KL \subset V - K such that I(L)≥I(V−K)−εI(L) \geq I(V - K) - \varepsilon; since L does not intersect K, there exists an index α\alpha such that L∩Kα=∅L \cap K_\alpha = \varnothing, and one then has I(V−Kα)≥I(L)≥I(V−K)−εI(V - K_\alpha) \geq I(L) \geq I(V - K) - \varepsilon. Since Kα⊂VK_\alpha \subset V, it follows that I(Kα)≤I(K)+εI(K_\alpha) \leq I(K) + \varepsilon and the condition 4) is verified.

By Th. 1, there exists a measure μ\mu such that μ∙(K)=I(K)\mu^\bullet(K) = I(K) for all K∈K(T)K \in \mathscr{K}(T). The inner regularity of the set functions μ∙\mu^\bullet and II on B(T)\mathscr{B}(T) then implies that μ∙(A)=I(A)\mu^\bullet(A) = I(A) for all A∈B(T)A \in \mathscr{B}(T), and existence is proved. The uniqueness of μ\mu follows from the uniqueness assertion of Th. 1.

3. Radon spaces

Definition 2

Let T be a topological space. One says that T is a Radon (resp. strongly Radon) space if T is Hausdorff and if every function defined on the Borel tribe B(T)\mathscr{B}(T) of T, with values in R‾+\overline{\mathbf{R}}_+, that is countably additive and bounded (resp. locally bounded) is inner regular.

For example, we shall see later on (Prop. 3) that every Polish space is strongly Radon. In particular, every locally compact space with a countable base is strongly Radon.

There exist Radon spaces that are not strongly Radon.

Proposition 1

Every Lindelöf*(1)^{(1)}* Radon space is strongly Radon.

Let T be a Lindelöf space that is Radon, and let I be a set function on the tribe B(T)\mathscr{B}(T) that is positive, countably additive and locally bounded. The open sets V such that I(V)<+∞I(V) < +\infty form a covering of T, from which one can extract a countable covering (Vn)n∈N(V_n)_{n \in \mathbf{N}}. Set Gn=V0∪V1∪⋯∪VnG_n = V_0 \cup V_1 \cup \cdots \cup V_n for every n∈Nn \in \mathbf{N}; set H0=G0H_0 = G_0 and Hn=Gn−Gn−1H_n = G_n - G_{n-1} for n≥1n \geq 1; finally, denote by InI_n the set function A↦I(A∩Hn)A \mapsto I(A \cap H_n) on B(T)\mathscr{B}(T), which is obviously countably additive and bounded. Since the sets HnH_n form a partition of T, we have I=∑nInI = \sum_n I_n. The space T being Radon, for every n∈Nn \in \mathbf{N} there exists a bounded measure μn\mu_n on T such that μn∙(A)=In(A)\mu_n^\bullet(A) = I_n(A) for all A∈B(T)A \in \mathscr{B}(T); therefore also ∑nμn∙(A)=I(A)\sum_n \mu_n^\bullet(A) = I(A). Since I is locally bounded, the family (μn)(\mu_n) is summable (§ 1, No. 7, Prop. 7); if μ\mu denotes ∑nμn\sum_n \mu_n, we have μ∙(A)=I(A)\mu^\bullet(A) = I(A) for all A∈B(T)A \in \mathscr{B}(T), and it follows that I is inner regular. In other words, T is strongly Radon.

Recall that a subset A of a topological space T is said to be universally measurable if A is μ\mu-measurable for every measure μ\mu on T. This is equivalent to saying that A is μ\mu-measurable for every measure μ\mu on T with compact support (§ 1, No. 8, Prop. 9).

(1) Recall (GT, I, §9, Exer. 14; TG, IX, Appendix I) that a Lindelöf space is a topological space T such that every open covering of T contains a countable covering.

Proposition 2

— Let X be a topological space and T a subspace of X.

a) Suppose that T is a Radon space. Then, for every function I defined on B(X)\mathscr{B}(X) that is positive, countably additive and bounded, one has

sup⁡K compactK⊂TI(K)=inf⁡B∈B(X)B⊃TI(B). \sup_{\substack{K \text{ compact} \\ K \subset T}} I(K) = \inf_{\substack{B \in \mathscr{B}(X) \\ B \supset T}} I(B).

Moreover, T is universally measurable in X.

b) Conversely, suppose that X is a Radon space and that T is universally measurable in X; then T is a Radon space.

Let us prove a). We denote by α\alpha the right side of (6); for every n∈Nn \in \mathbf{N}, there exists a set Cn∈B(X)C_n \in \mathscr{B}(X) containing T, such that I(Cn)≤α+2−nI(C_n) \leq \alpha + 2^{-n}. Setting C=⋂nCnC = \bigcap_n C_n, we then have T⊂CT \subset C, I(C)=αI(C) = \alpha. If A∈B(T)A \in \mathscr{B}(T), let us choose a Borel set B of X such that A=B∩TA = B \cap T (GT, IX, §6, No. 3) and set J(A)=I(B∩C)J(A) = I(B \cap C). This number does not depend on the choice of B, for if B′B' is a second Borel set in X such that A=B′∩TA = B' \cap T, then B∩CB \cap C and B′∩CB' \cap C differ only by a Borel set M contained in C−TC - T, and I(M)=0I(M) = 0 by the construction of C. Clearly J(K)=I(K)J(K) = I(K) for every compact set K⊂TK \subset T. Let (An)(A_n) be a sequence of Borel sets of T, pairwise disjoint, and, for each nn, let BnB_n be a Borel set of X such that Bn∩T=AnB_n \cap T = A_n. Replacing BnB_n by Bn−(⋃k<nBk)B_n - \left( \bigcup_{k < n} B_k \right) if necessary, we can suppose that the sets BnB_n are pairwise disjoint. Set A=⋃nAnA = \bigcup_n A_n and B=⋃nBnB = \bigcup_n B_n; then

J(A)=I(B∩C)=∑nI(Bn∩C)=∑nJ(An); J(A) = I(B \cap C) = \sum_n I(B_n \cap C) = \sum_n J(A_n);

J is thus a countably additive and bounded function on B(T)\mathscr{B}(T). Since T is by hypothesis a Radon space, there exists a bounded measure μ\mu on T such that J(A)=μ∙(A)J(A) = \mu^\bullet(A) for every A∈B(T)A \in \mathscr{B}(T); consequently

α=J(T)=μ∙(T)=sup⁡Kμ∙(K)=sup⁡KJ(K), \alpha = J(T) = \mu^\bullet(T) = \sup_K \mu^\bullet(K) = \sup_K J(K),

by the definition of μ∙\mu^\bullet. Formula (6) is thus established.

Let us now show that T is universally measurable. Let λ\lambda be a bounded measure on X; the preceding argument may be applied to the set function I:A↦λ∙(A)I : A \mapsto \lambda^\bullet(A) on B(X)\mathscr{B}(X), thus there exists a sequence (Kn)(K_n) of compact subsets of T such that (with the above notations)

sup⁡nλ∙(Kn)=J(T)=λ∙(C). \sup_n \lambda^\bullet(K_n) = J(T) = \lambda^\bullet(C).

Set K′=⋃n∈NKnK' = \bigcup_{n \in \mathbf{N}} K_n; K′K' is Borel in XX, K′⊂T⊂CK' \subset T \subset C, λ∗(K′)=λ∗(C)\lambda^*(K') = \lambda^*(C), therefore these three sets differ only by λ\lambda-negligible sets, and so TT is λ\lambda-measurable. This completes the proof of a).

Let us pass to b). Suppose that XX is a Radon space, and that TT is universally measurable in XX. Let II be a positive function on B(T)\mathscr{B}(T) that is countably additive and bounded; the function A↦I(A∩T)A \mapsto I(A \cap T) on B(X)\mathscr{B}(X) is then positive, countably additive and bounded, therefore there exists a bounded measure ν\nu on XX such that I(A∩T)=ν∗(A)I(A \cap T) = \nu^*(A) for all A∈B(X)A \in \mathscr{B}(X). Now, TT is ν\nu-measurable; the preceding relation shows that ν∗(K)=0\nu^*(K) = 0 for every compact subset KK of XX that is disjoint from TT, therefore ν\nu is concentrated on TT. Consequently, for every Borel set AA of XX, we have I(A∩T)=ν∗(A∩T)=μ∗(A∩T)I(A \cap T) = \nu^*(A \cap T) = \mu^*(A \cap T), where μ\mu is the measure induced by ν\nu on TT. Finally, it follows that I(B)=μ∗(B)I(B) = \mu^*(B) for every set B∈B(T)B \in \mathscr{B}(T) (GT, IX, §6, No. 3, Remark 2), and II is indeed inner regular.

Corollary

— *If XX is a Radon space, then every Borel subset TT of XX is Radon.

For, TT is universally measurable in XX.

Proposition 3

— *Every Souslin space (in particular, every Polish or Lusin space) is strongly Radon.

Let TT be a Souslin space; since TT is a Lindelöf space (TG, IX, Appendix I, Cor. of Prop. 1),(2) it suffices to show that TT is Radon (Prop. 1). Let II be a function defined on B(T)\mathscr{B}(T), positive, countably additive and bounded. We extend II to P(T)\mathscr{P}(T) by setting, for every subset AA of TT,

I(A)=inf⁡B∈B(T)B⊇AI(B). I(A) = \inf_{\substack{B \in \mathscr{B}(T) \\ B \supseteq A}} I(B).

Let us show that this extension is a capacity on TT (GT, IX, §6, No. 9). It is clear that the relation A⊂A′A \subset A' implies I(A)≤I(A′)I(A) \leq I(A'). Let (An)(A_n) be an increasing sequence of subsets of TT, and let A=⋃nAnA = \bigcup_n A_n. The set of Borel sets that contain AnA_n being stable for countable intersections, there exists for each nn a Borel set BnB_n such that An⊂BnA_n \subset B_n and I(An)=I(Bn)I(A_n) = I(B_n) (cf. the proof of Prop. 2). Set Cn=⋂p≥nBpC_n = \bigcap_{p \geq n} B_p; CnC_n is Borel, and An⊂Cn⊂BnA_n \subset C_n \subset B_n, therefore I(An)=I(Cn)I(A_n) = I(C_n). On the other hand, the sequence (Cn)(C_n) is increasing. Let C=⋃nCnC = \bigcup_n C_n: the relation A⊂CA \subset C implies that

I(A)≤I(C)=lim⁡nI(Cn)=lim⁡nI(An), I(A) \leq I(C) = \lim_n I(C_n) = \lim_n I(A_n),

(2) Every Souslin space has a countable base for open sets (GT, IX, §6, No. 2, Prop. 4), hence is Lindelöf (GT, I, §9, Exer. 14).

whence the equality I(A)=lim⁡nI(An)I(A) = \lim_n I(A_n) is immediate. Consequently, I is a capacity.

If (Hn)(H_n) is a decreasing sequence of closed sets in T, obviously I(⋂nHn)=inf⁡nI(Hn)I\left( \bigcap_n H_n \right) = \inf_n I(H_n). It follows that every Souslin subset F of T is capacitable for I (TG, IX, §6, No. 10, Prop. 15). In particular, every Borel set A of T is capacitable (loc. cit., §6, No. 3, Prop. 10).(3) In other words,

I(A)=sup⁡KI(K), I(A) = \sup_K I(K),

where K runs over the set of compact sets contained in A; we have proved that I is inner regular.

Remark

— Let X be a Lusin space (in particular, any Polish space), and f a bijective continuous mapping of X onto a (Lusin) regular space Y. One knows (TG, IX, §6, No. 7, Prop. 14) that the mapping B↦f−1(B)B \mapsto f^{-1}(B) is a bijection of the Borel tribe of Y onto the Borel tribe of X. The spaces X and Y are Lusin, hence strongly Radon (Prop. 3). It follows immediately that the mapping μ↦f(μ)\mu \mapsto f(\mu) is a bijection of the set of bounded measures on X onto the set of bounded measures on Y.(4)

Exercises

See the exercises for § 3.

Transcribed from the printed volume. Not translated.