Bourbaki

Theorem 5

Tag 08TN, permanent. Cite it as [Bourbaki, Tag 08TN].

Integration, IV, p. 54, § 4. Integrable functions and sets

Let Φ\Phi be a set of subsets of a locally compact space X, satisfying (PCI) and (PC_{II}), and let α\alpha be a real-valued function, defined on Φ\Phi, satisfying the conditions (PM_I), (PM_{II}) and (PM_{III}). In order that there exist a positive measure μ\mu on X such that the sets of Φ\Phi are μ\mu-integrable and μ(M)=α(M)\mu(M) = \alpha(M) for all M∈ΦM \in \Phi, it is necessary and sufficient that α\alpha satisfy in addition the following condition:*

(PM_{IV}) For every ε>0\varepsilon > 0 and every M∈ΦM \in \Phi, there exist a compact set K⊂MK \subset M and an open set U⊃MU \supset M such that, for every N∈ΦN \in \Phi satisfying the relation K⊂N⊂UK \subset N \subset U, one has ∣α(N)−α(M)∣≤ε|\alpha(N) - \alpha(M)| \leq \varepsilon.

Moreover, if the condition (PM{IV}) is satisfied, then the measure μ\mu is unique; for every compact set KK, μ(K)=inf⁡M∈Φ,M⊃Kα(M)\mu(K) = \inf_{M \in \Phi, M \supset K} \alpha(M); for every open set UU, μ∗(U)=sup⁡M∈Φ,M⊂Uα(M)\mu^*(U) = \sup_{M \in \Phi, M \subset U} \alpha(M).*

Note that the condition (PM_{IV}) is equivalent to the conjunction of the following two:

(PM'_{IV}) For every ε>0\varepsilon > 0 and every M∈ΦM \in \Phi, there exists an open set U⊃MU \supset M such that, for every N∈ΦN \in \Phi contained in UU, α(N)≤α(M)+ε\alpha(N) \leq \alpha(M) + \varepsilon. (PM''_{IV}) For every ε>0\varepsilon > 0 and every M∈ΦM \in \Phi, there exists a compact set K⊂MK \subset M such that, for every N∈ΦN \in \Phi containing KK, α(N)≥α(M)−ε\alpha(N) \geq \alpha(M) - \varepsilon.

For, it is obvious that (PM'_{IV}) and (PM''_{IV}) imply (PM_{IV}). Conversely, let us show for example that (PM_{IV}) implies (PM'_{IV}): let K be a compact set and U an open set such that K⊂M⊂UK \subset M \subset U and ∣α(P)−α(M)∣≤ε|\alpha(P) - \alpha(M)| \leq \varepsilon for every P∈ΦP \in \Phi satisfying K⊂P⊂UK \subset P \subset U. Then, if N∈ΦN \in \Phi and N⊂UN \subset U, M∪NM \cup N belongs to Φ\Phi and K⊂M∪N⊂UK \subset M \cup N \subset U, whence α(M∪N)≤α(M)+ε\alpha(M \cup N) \leq \alpha(M) + \varepsilon and a fortiori α(N)≤α(M)+ε\alpha(N) \leq \alpha(M) + \varepsilon.

When the set Φ\Phi, satisfying (PC_I) and (PC_{II}), consists of compact sets, then the condition (PM''_{IV}) is trivially verified, and (PM_{IV}) is then equivalent to (PM'_{IV}).

The condition (PM_{IV}) is necessary: this follows at once from Th. 4 of No. 6 on the 'approximation' of an integrable set by a compact set and an open set. To prove the other assertions of the theorem, we proceed in several steps.

1° Definition of a topology on P(X)\mathfrak{P}(X).

For every pair (K,U)(K, U) consisting of a compact set KK and an open set UU in XX, we denote by I(K,U)I(K, U) the set of subsets M⊂XM \subset X such that K⊂M⊂UK \subset M \subset U; in order that I(K,U)I(K, U) be nonempty, it is necessary and sufficient that K⊂UK \subset U. If (K′,U′)(K', U') is a second pair, formed by a compact set K′K' and an open set U′U', we have

I(K,U)∩I(K′,U′)=I(K∪K′,U∩U′). I(K, U) \cap I(K', U') = I(K \cup K', U \cap U').
Let TT be the topology on P(X)\mathfrak{P}(X) generated by the set of subsets I(K,U)I(K, U) as KK runs over the set of compact subsets of XX, and UU over the set of open subsets of XX; by the foregoing, the I(K,U)I(K, U) form a base for the topology TT (GT, I, §1, No. 3).

We observe that the definition of TT implies that, in P(X)\mathfrak{P}(X), the set of compact subsets of XX is dense. The condition (PC_{II}) expresses that Φ\Phi is dense in P(X)\mathfrak{P}(X), and condition (PM_{IV}) expresses that the function α\alpha is continuous on Φ\Phi for the topology induced by TT. Finally, Th. 4 of No. 6 expresses that the function M↦μ(M)M \mapsto \mu(M) is continuous on the clan of μ\mu-integrable sets, for the topology induced by TT.

2° Uniqueness of μ\mu.

We denote by Φ‾\overline{\Phi} the set of subsets M⊂XM \subset X such that α(N)\alpha(N) tends to a finite limit as NN tends to MM (for the topology TT) while remaining in Φ\Phi; we may then extend α\alpha in only one way to a continuous mapping α‾\overline{\alpha} of Φ‾\overline{\Phi} into R\mathbf{R} (GT, I, §8, No. 5, Th. 1). If there exists a measure μ\mu meeting the requirements, the above remarks prove that the clan Ψ\Psi of μ\mu-integrable sets is contained in Φ‾\overline{\Phi} and that μ(M)=α‾(M)\mu(M) = \overline{\alpha}(M) for every M∈ΨM \in \Psi; this relation holds in particular for every compact subset MM of XX, which proves the uniqueness of μ\mu (No. 10, Cor. 3 of Prop. 19).

3° Extension of α\alpha to the compact sets.

Without assuming the existence of μ\mu, we are now going to study the set Φ‾\overline{\Phi} and the extension α‾\overline{\alpha} of α\alpha to Φ‾\overline{\Phi}. We first show that every compact set KK belongs to Φ‾\overline{\Phi} and that α‾(K)=inf⁡P∈Φ,P⊃Kα(P)\overline{\alpha}(K) = \inf_{P \in \Phi, P \supset K} \alpha(P). Set a=inf⁡P∈Φ,P⊃Kα(P)a = \inf_{P \in \Phi, P \supset K} \alpha(P); for every ε>0\varepsilon > 0, there exists an M∈ΦM \in \Phi such that K⊂MK \subset M and α(M)≤a+ε\alpha(M) \leq a + \varepsilon. By (PM'_{IV}), there exists an open set U⊃MU \supset M such that, for every N∈ΦN \in \Phi contained in UU, we have α(N)≤α(M)+ε≤a+2ε\alpha(N) \leq \alpha(M) + \varepsilon \leq a + 2\varepsilon; for every N∈ΦN \in \Phi such that K⊂N⊂UK \subset N \subset U, we therefore have a≤α(N)≤a+2εa \leq \alpha(N) \leq a + 2\varepsilon, which, by the definitions, shows that K∈Φ‾K \in \overline{\Phi} and α‾(K)=a\overline{\alpha}(K) = a.

This result proves at once that if K1K_1 and K2K_2 are two compact sets such that K1⊂K2K_1 \subset K_2, then α‾(K1)≤α‾(K2)\overline{\alpha}(K_1) \leq \overline{\alpha}(K_2). If K1K_1 and K2K_2 are any two compact sets, we have α‾(K1∪K2)≤α‾(K1)+α‾(K2)\overline{\alpha}(K_1 \cup K_2) \leq \overline{\alpha}(K_1) + \overline{\alpha}(K_2) by (PM_{II}). We shall see, moreover, that if K1K_1 and K2K_2 are disjoint then α‾(K1∪K2)=α‾(K1)+α‾(K2)\overline{\alpha}(K_1 \cup K_2) = \overline{\alpha}(K_1) + \overline{\alpha}(K_2). For, there then exist two disjoint open sets U1,U2U_1, U_2 such that K1⊂U1K_1 \subset U_1, K2⊂U2K_2 \subset U_2 (GT, II, §4, Prop. 4). Therefore, by (PC_{II}), there also exist two sets M1∈ΦM_1 \in \Phi, M2∈ΦM_2 \in \Phi such that K1⊂M1⊂U1K_1 \subset M_1 \subset U_1 and K2⊂M2⊂U2K_2 \subset M_2 \subset U_2. Now let PP be any set of Φ\Phi containing K1∪K2K_1 \cup K_2; the union of the two sets P∩M1P \cap M_1 and P∩M2P \cap M_2 belongs to Φ\Phi by (PC_{I}), and since these two sets are disjoint, application of (PM_I) and (PM_{III}) yields

α(P)≥α(P∩M1)+α(P∩M2)≥α‾(K1)+α‾(K2), \alpha(P) \geq \alpha(P \cap M_1) + \alpha(P \cap M_2) \geq \overline{\alpha}(K_1) + \overline{\alpha}(K_2),

which establishes our assertion.

4° Extension of α\alpha to the open sets.

We shall now see that, for an open set UU to belong to Φ‾\overline{\Phi}, it is necessary and sufficient that, as KK runs over the set of compact subsets of UU, the supremum of the numbers α‾(K)\overline{\alpha}(K) is finite; moreover, α‾(U)\overline{\alpha}(U) is then equal to this supremum.

For, let UU be an open set belonging to Φ‾\overline{\Phi}; for every ε>0\varepsilon > 0 there exists a compact set K⊂UK \subset U such that, for every set M∈ΦM \in \Phi satisfying K⊂M⊂UK \subset M \subset U, one has ∣α‾(U)−α(M)∣≤ε|\overline{\alpha}(U) - \alpha(M)| \leq \varepsilon, whence ∣α‾(U)−α‾(K)∣≤ε|\overline{\alpha}(U) - \overline{\alpha}(K)| \leq \varepsilon; on the other hand, if K′K' is any compact set contained in UU, then K⊂K∪K′⊂UK \subset K \cup K' \subset U, whence ∣α‾(U)−α‾(K∪K′)∣≤ε|\overline{\alpha}(U) - \overline{\alpha}(K \cup K')| \leq \varepsilon and so α‾(U)≥α‾(K∪K′)−ε≥α‾(K′)−ε\overline{\alpha}(U) \geq \overline{\alpha}(K \cup K') - \varepsilon \geq \overline{\alpha}(K') - \varepsilon; α‾(U)\overline{\alpha}(U) is therefore indeed equal to the supremum of the numbers α‾(K)\overline{\alpha}(K) as KK runs over the set of compact subsets of UU.

Conversely, let UU be an open set such that b=sup⁡K⊂Uα‾(K)<+∞b = \sup_{K \subset U} \overline{\alpha}(K) < +\infty (KK running over the set of compact subsets of UU), and let us show that U∈Φ‾U \in \overline{\Phi}. For every ε>0\varepsilon > 0, there exists a compact set K⊂UK \subset U such that b−ε≤α‾(K)≤bb - \varepsilon \leq \overline{\alpha}(K) \leq b; by (PM''_{IV}), for every set M∈ΦM \in \Phi such that K⊂M⊂UK \subset M \subset U, there exists a compact set K′⊂MK' \subset M such that

α(M)≤α‾(K′)+ε≤b+ε; \alpha(M) \leq \overline{\alpha}(K') + \varepsilon \leq b + \varepsilon;

therefore b−ε≤α(M)≤b+εb - \varepsilon \leq \alpha(M) \leq b + \varepsilon, which proves that U∈Φ‾U \in \overline{\Phi}.

From this characterization of the open sets U∈Φ‾U \in \overline{\Phi}, and of α‾(U)\overline{\alpha}(U), it follows first of all that if U1U_1 and U2U_2 are two open sets such that U1⊂U2U_1 \subset U_2 and U2∈Φ‾U_2 \in \overline{\Phi}, then U1∈Φ‾U_1 \in \overline{\Phi} and α‾(U1)≤α‾(U2)\overline{\alpha}(U_1) \leq \overline{\alpha}(U_2). On the other hand, if U1U_1 and U2U_2 are two open sets belonging to Φ‾\overline{\Phi}, then the same is true of U1∪U2U_1 \cup U_2, and α‾(U1∪U2)≤α‾(U1)+α‾(U2)\overline{\alpha}(U_1 \cup U_2) \leq \overline{\alpha}(U_1) + \overline{\alpha}(U_2). For, let KK be any compact set contained in U1∪U2U_1 \cup U_2; for every point x∈Kx \in K, there exists a compact neighborhood of xx contained in either U1U_1 or U2U_2; one can therefore cover KK by a finite number of these neighborhoods; if K1K_1 (resp. K2K_2) is the union of those that are contained in U1U_1 (resp. U2U_2), then K⊂K1∪K2K \subset K_1 \cup K_2, whence

α‾(K)≤α‾(K1∪K2)≤α‾(K1)+α‾(K2)≤α‾(U1)+α‾(U2), \overline{\alpha}(K) \leq \overline{\alpha}(K_1 \cup K_2) \leq \overline{\alpha}(K_1) + \overline{\alpha}(K_2) \leq \overline{\alpha}(U_1) + \overline{\alpha}(U_2),

which establishes the asserted property.

5° Properties of Φ‾\overline{\Phi} and α‾\overline{\alpha}.

The definition of Φ‾\overline{\Phi} and α‾\overline{\alpha} can now be transformed as follows (taking into account (PC_{II})): in order that M∈Φ‾M \in \overline{\Phi}, it is necessary and sufficient that, for every ε>0\varepsilon > 0, there exist a compact set KK and an open set U∈Φ‾U \in \overline{\Phi} such that K⊂M⊂UK \subset M \subset U and α‾(U)−α‾(K)≤ε\overline{\alpha}(U) - \overline{\alpha}(K) \leq \varepsilon; α‾(M)\overline{\alpha}(M) is, moreover, the infimum of the α‾(U)\overline{\alpha}(U) for the open sets U∈Φ‾U \in \overline{\Phi} containing MM, and the supremum of the α‾(K)\overline{\alpha}(K) for the compact sets K⊂MK \subset M.

From this, we shall first deduce that if M1,M2M_1, M_2 and M1∪M2M_1 \cup M_2 belong to Φ‾\overline{\Phi}, then α‾(M1∪M2)≤α‾(M1)+α‾(M2)\overline{\alpha}(M_1 \cup M_2) \leq \overline{\alpha}(M_1) + \overline{\alpha}(M_2). Indeed, if U1U_1 and U2U_2 are two open sets of Φ‾\overline{\Phi} containing M1M_1 and M2M_2, respectively, and such that α‾(U1)≤α‾(M1)+ε\overline{\alpha}(U_1) \leq \overline{\alpha}(M_1) + \varepsilon and α‾(U2)≤α‾(M2)+ε\overline{\alpha}(U_2) \leq \overline{\alpha}(M_2) + \varepsilon, then U1∪U2U_1 \cup U_2 belongs to Φ‾\overline{\Phi}, contains M1∪M2M_1 \cup M_2, and consequently

α‾(M1∪M2)≤α‾(U1∪U2)≤α‾(U1)+α‾(U2)≤α‾(M1)+α‾(M2)+2ε, \overline{\alpha}(M_1 \cup M_2) \leq \overline{\alpha}(U_1 \cup U_2) \leq \overline{\alpha}(U_1) + \overline{\alpha}(U_2) \leq \overline{\alpha}(M_1) + \overline{\alpha}(M_2) + 2\varepsilon,

whence our assertion.

Next, let us show that if KK is a compact set and UU is an open set of Φ‾\overline{\Phi} such that K⊂UK \subset U, then α‾(U−K)=α‾(U)−α‾(K)\overline{\alpha}(U - K) = \overline{\alpha}(U) - \overline{\alpha}(K). By the foregoing, we have α‾(U)≤α‾(K)+α‾(U−K)\overline{\alpha}(U) \leq \overline{\alpha}(K) + \overline{\alpha}(U - K). On the other hand, for every compact set K′⊂U−KK' \subset U - K,

α‾(K∪K′)=α‾(K)+α‾(K′)≤α‾(U); \overline{\alpha}(K \cup K') = \overline{\alpha}(K) + \overline{\alpha}(K') \leq \overline{\alpha}(U);

since U−KU - K is open and belongs to Φ‾\overline{\Phi}, α‾(U−K)\overline{\alpha}(U - K) is the supremum of the α‾(K′)\overline{\alpha}(K'), which shows that α‾(K)+α‾(U−K)≤α‾(U)\overline{\alpha}(K) + \overline{\alpha}(U - K) \leq \overline{\alpha}(U).

The definition of Φ‾\overline{\Phi} may therefore now be expressed in the following way: in order that M∈Φ‾M \in \overline{\Phi}, it is necessary and sufficient that, for every ε>0\varepsilon > 0, there exist a compact set KK and an open set U∈Φ‾U \in \overline{\Phi} such that K⊂M⊂UK \subset M \subset U and α‾(U−K)≤ε\overline{\alpha}(U - K) \leq \varepsilon.

We are now in a position to prove that Φ‾\overline{\Phi} is a clan and α‾\overline{\alpha} an additive set function on Φ‾\overline{\Phi}. We first show that if MM and NN belong to Φ‾\overline{\Phi} then so do M∩CNM \cap \mathbf{C}N and M∪NM \cup N. By hypothesis, for every ε>0\varepsilon > 0 there exist two compact sets K,K′K, K' and two open sets U,U′U, U' of Φ‾\overline{\Phi} such that

K⊂M⊂U,K′⊂N⊂U′,α‾(U−K)≤ε,α‾(U′−K′)≤ε. K \subset M \subset U,\quad K' \subset N \subset U',\quad \overline{\alpha}(U - K) \leq \varepsilon,\quad \overline{\alpha}(U' - K') \leq \varepsilon.

The set K′′=K∩CU′K'' = K \cap \mathbf{C}U' is compact, the set U′′=U∩CK′U'' = U \cap \mathbf{C}K' is open and belongs to Φ‾\overline{\Phi}, and K′′⊂M∩CN⊂U′′K'' \subset M \cap \mathbf{C}N \subset U''; on the other hand, U′′−K′′U'' - K'' is contained in the union of U∩CKU \cap \mathbf{C}K and U′∩CK′U' \cap \mathbf{C}K', whence α‾(U′′−K′′)≤2ε\overline{\alpha}(U'' - K'') \leq 2\varepsilon, which proves that M∩CN∈Φ‾M \cap \mathbf{C}N \in \overline{\Phi}. Similarly, U1=U∪U′U_1 = U \cup U' is open and belongs to Φ‾\overline{\Phi}, K1=K∪K′K_1 = K \cup K' is compact, and K1⊂M∪N⊂U1K_1 \subset M \cup N \subset U_1; on the other hand, U1−K1U_1 - K_1 is contained in the union of U−KU - K and U′−K′U' - K', whence again α‾(U1−K1)≤2ε\overline{\alpha}(U_1 - K_1) \leq 2\varepsilon, and M∪NM \cup N belongs to Φ‾\overline{\Phi}. Finally, if MM and NN are disjoint, then

α‾(K1)=α‾(K)+α‾(K′)≥α‾(M)+α‾(N)−2ε, \overline{\alpha}(K_1) = \overline{\alpha}(K) + \overline{\alpha}(K') \geq \overline{\alpha}(M) + \overline{\alpha}(N) - 2\varepsilon,

consequently α‾(M∪N)≥α‾(M)+α‾(N)−2ε\overline{\alpha}(M \cup N) \geq \overline{\alpha}(M) + \overline{\alpha}(N) - 2\varepsilon; since ε\varepsilon is arbitrary, we have α‾(M∪N)=α‾(M)+α‾(N)\overline{\alpha}(M \cup N) = \overline{\alpha}(M) + \overline{\alpha}(N).

6° Existence of the measure μ\mu.

By Prop. 18 of No. 9, there exists one and only one positive linear form β\beta on the vector space E(Φ‾)\mathscr{E}(\overline{\Phi}) of Φ‾\overline{\Phi}-step functions, such that β(φM)=α‾(M)\beta(\varphi_M) = \overline{\alpha}(M) for all M∈Φ‾M \in \overline{\Phi}. For every compact subset KK of XX, let us denote by G(K)\mathscr{G}(K) the space of uniform limits of functions of E(Φ‾)\mathscr{E}(\overline{\Phi}) whose support is contained in KK. Since β\beta is positive, ∣β(f)∣≤α‾(K)⋅∥f∥|\beta(f)| \leq \overline{\alpha}(K) \cdot \|f\| for every function f∈E(Φ‾)f \in \mathscr{E}(\overline{\Phi}) whose support is contained in KK; the restriction of β\beta to the space of these functions is a continuous linear form for the topology of uniform convergence; it may therefore be extended to a positive continuous linear form β‾K\overline{\beta}_K on G(K)\mathscr{G}(K). Moreover, if K⊂K1K \subset K_1, then the restriction of β‾K1\overline{\beta}_{K_1} to G(K)\mathscr{G}(K) is identical to β‾K\overline{\beta}_K, therefore there exists a positive linear form β‾\overline{\beta} on the union G\mathscr{G} of the G(K)\mathscr{G}(K), that extends each of the forms β‾K\overline{\beta}_K.

Now, since every compact set belongs to Φ‾\overline{\Phi}, the space H\mathscr{H} of continuous real-valued functions with compact support is a subspace of G\mathscr{G} (No. 10, Prop. 19); the restriction to H\mathscr{H} of the positive linear form β‾\overline{\beta} is therefore a positive measure μ\mu. Let us show that for every compact set KK, μ(K)=α‾(K)\mu(K) = \overline{\alpha}(K). For every ε>0\varepsilon > 0, there exists an open set U∈Φ‾U \in \overline{\Phi} such that K⊂UK \subset U, μ(U)≤μ(K)+ε\mu(U) \leq \mu(K) + \varepsilon and α‾(U)≤α‾(K)+ε\overline{\alpha}(U) \leq \overline{\alpha}(K) + \varepsilon. Let ff be a continuous mapping of XX into [0,1][0, 1] whose support is contained in UU and such that f(x)=1f(x) = 1 on KK (Ch. III, §1, No. 2, Lemma 1). Then μ(K)≤μ(f)≤μ(U)≤μ(K)+ε\mu(K) \leq \mu(f) \leq \mu(U) \leq \mu(K) + \varepsilon, and, on the other hand,

α‾(K)=β(φK)≤β‾(f)≤β(φU)=α‾(U)≤α‾(K)+ε; \overline{\alpha}(K) = \beta(\varphi_K) \leq \overline{\beta}(f) \leq \beta(\varphi_U) = \overline{\alpha}(U) \leq \overline{\alpha}(K) + \varepsilon;

since μ(f)=β‾(f)\mu(f) = \overline{\beta}(f), we see that ∣μ(K)−α‾(K)∣≤ε|\mu(K) - \overline{\alpha}(K)| \leq \varepsilon, and since ε\varepsilon is arbitrary, μ(K)=α‾(K)\mu(K) = \overline{\alpha}(K).

The characterization of the open sets belonging to Φ‾\overline{\Phi}, combined with Cor. 4 of Th. 4 of No. 6, then shows that the open sets belonging to Φ‾\overline{\Phi} are none other than the μ\mu-integrable open sets, and that, for such a set UU, we have μ(U)=α‾(U)\mu(U) = \overline{\alpha}(U). Th. 4 of No. 6 and the characterization of the sets of Φ‾\overline{\Phi} given in 5° then show that the μ\mu-integrable sets are the sets of Φ‾\overline{\Phi} and that, for such a set MM, μ(M)=α‾(M)\mu(M) = \overline{\alpha}(M). Finally, the fact that μ∗(U)=sup⁡M∈Φ,M⊂Uα(M)\mu^*(U) = \sup_{M \in \Phi, M \subset U} \alpha(M) for every open set UU follows at once from (PCII_{II}) and Cor. 4 of Th. 4 of No. 6.

Theorem 5 is thus completely proved.

Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.

Cited by

Cites

Transcribed from the printed volume. Not translated.