Bourbaki

§ 2. Support of a measure

1. Restriction of a measure to an open set. Definition of a measure by means of local data

Let X be a locally compact space, Y an open set in X. The subspace Y of X is locally compact, and every continuous function with values in a topological vector space E, defined on Y and with compact support, may be extended by continuity to all of X, by giving it the value 0 on CY\mathbf{C}Y; one can therefore identify in this way the space K(Y;E)\mathscr{K}(Y;E) with the linear subspace of K(X;E)\mathscr{K}(X;E) formed by the continuous functions with compact support contained in Y. If μ\mu is a measure on X, it is clear that the restriction of μ\mu to K(Y;C)\mathscr{K}(Y;\mathbf{C}) is a measure on Y, which is called the restriction of μ\mu to the open subspace Y, or the measure induced on Y by μ\mu, and is denoted μ∣Y\mu|Y. The restrictions to Y of ∣μ∣|\mu|, Rμ\mathscr{R}\mu and Iμ\mathscr{I}\mu are, respectively, ∣μ∣Y∣|\mu|Y|, R(μ∣Y)\mathscr{R}(\mu|Y) and I(μ∣Y)\mathscr{I}(\mu|Y), by virtue of §1, Nos. 5 and 6. If μ\mu is real then the restrictions of μ+\mu^+ and μ−\mu^- to Y are, respectively, (μ∣Y)+(\mu|Y)^+ and (μ∣Y)−(\mu|Y)^-, by virtue of formula (8) of §1, No. 5.

One sees immediately that if Y and Z are two open sets in X such that Y⊃ZY \supset Z, and if μ∣Y\mu|Y and μ∣Z\mu|Z are the restrictions of μ\mu to Y and Z, then μ∣Z\mu|Z is also the restriction of μ∣Y\mu|Y to the open subspace Z of the locally compact space Y.

In Ch. IV, §5, No. 7 we shall generalize this definition to the case that Y is a locally compact subspace of X.

Note that a measure on Y is not necessarily the restriction of some measure on X (cf. Ch. V, §7, No. 2, Prop. 3).

For example, let Y be the open interval ]0,1[ of X=RX = \mathbf{R}; the mapping

f↦∫01f(x)xdx f \mapsto \int_0^1 \frac{f(x)}{x} dx

is a measure on YY, because every function in K(Y;C)\mathscr{K}(Y; \mathbf{C}) is zero on a neighborhood of 0 in R\mathbf{R}. However, this measure cannot be extended to a measure on R\mathbf{R} because, in the contrary case, its restriction to the set of functions f∈K(Y;C)f \in \mathscr{K}(Y; \mathbf{C}) such that ∥f∥≤1\|f\| \leq 1 would be bounded; but this is false.

However, we have the following proposition:

Proposition 1

Let* (Yα)α∈A(Y_\alpha)_{\alpha \in A} be an open covering of XX and suppose given, on each subspace YαY_\alpha, a measure μα\mu_\alpha, in such a way that for every pair (α,β)(\alpha, \beta), the restrictions of μα\mu_\alpha and μβ\mu_\beta to Yα∩YβY_\alpha \cap Y_\beta are identical. Under these conditions, there exists one and only one measure μ\mu on XX whose restriction to YαY_\alpha is equal to μα\mu_\alpha for every index α\alpha.

Let us first show that every function f∈K(X;C)f \in \mathscr{K}(X; \mathbf{C}) may be written in the form of a finite sum f=∑ifif = \sum_i f_i where, for each of the functions fi∈K(X;C)f_i \in \mathscr{K}(X; \mathbf{C}), there exists an index αi\alpha_i such that Supp⁡(fi)⊂Yαi\operatorname{Supp}(f_i) \subset Y_{\alpha_i}. If K=Supp⁡(f)K = \operatorname{Supp}(f), there exists a finite number of indices αi\alpha_i (1≤i≤n1 \leq i \leq n) such that the YαiY_{\alpha_i} form a covering of KK; let hih_i (1≤i≤n1 \leq i \leq n) be continuous mappings of XX into [0,1][0, 1] such that the support of hih_i is compact and is contained in YαiY_{\alpha_i} for 1≤i≤n1 \leq i \leq n, and such that ∑i=1nhi(x)=1\sum_{i=1}^n h_i(x) = 1 on KK (§ 1, No. 2, Lemma 1); the functions fi=fhif_i = f h_i meet the requirements. This shows in the first place that if there exists a measure μ\mu meeting the requirements then it is unique, because for every finite sum f=∑i=1nfif = \sum_{i=1}^n f_i, where fi∈K(Yαi;C)f_i \in \mathscr{K}(Y_{\alpha_i}; \mathbf{C}), necessarily μ(f)=∑i=1nμαi(fi)\mu(f) = \sum_{i=1}^n \mu_{\alpha_i}(f_i). Moreover, we will have shown the existence of a linear form μ\mu on K(X;C)\mathscr{K}(X; \mathbf{C}) whose restriction to each subspace K(Yα;C)\mathscr{K}(Y_\alpha; \mathbf{C}) is μα\mu_\alpha, provided we demonstrate the following property: if (gi)1≤i≤m(g_i)_{1 \leq i \leq m} and (hj)1≤j≤n(h_j)_{1 \leq j \leq n} are two finite sequences of functions in K(X;C)\mathscr{K}(X; \mathbf{C}) such that gi∈K(Yαi;C)g_i \in \mathscr{K}(Y_{\alpha_i}; \mathbf{C}) for 1≤i≤m1 \leq i \leq m, hj∈K(Yβj;C)h_j \in \mathscr{K}(Y_{\beta_j}; \mathbf{C}) for 1≤j≤n1 \leq j \leq n and

∑i=1mgi(x)=∑j=1nhj(x)=1 \sum_{i=1}^m g_i(x) = \sum_{j=1}^n h_j(x) = 1
on KK, then
∑i=1mμαi(fgi)=∑j=1nμβj(fhj). \sum_{i=1}^m \mu_{\alpha_i}(f g_i) = \sum_{j=1}^n \mu_{\beta_j}(f h_j).
Now,
fgi=∑j=1nfgihj, f g_i = \sum_{j=1}^n f g_i h_j,

whence

∑i=1mμαi(fgi)=∑i=1m∑j=1nμαi(fgihj). \sum_{i=1}^{m} \mu_{\alpha_i}(f g_i) = \sum_{i=1}^{m} \sum_{j=1}^{n} \mu_{\alpha_i}(f g_i h_j).
Similarly,
∑j=1nμβj(fhj)=∑j=1n∑i=1mμβj(fgihj). \sum_{j=1}^{n} \mu_{\beta_j}(f h_j) = \sum_{j=1}^{n} \sum_{i=1}^{m} \mu_{\beta_j}(f g_i h_j).
But since the support of fgihjf g_i h_j is contained in Yαi∩YβjY_{\alpha_i} \cap Y_{\beta_j}, we have μαi(fgihj)=μβj(fgihj)\mu_{\alpha_i}(f g_i h_j) = \mu_{\beta_j}(f g_i h_j), which establishes our assertion.

It remains to see that μ\mu is a measure on XX; now, every point of XX admits a compact neighborhood contained in one of the YαY_\alpha; the conclusion therefore follows at once from the definition of μ\mu and from Prop. 6 of §1, No. 3.

Corollary (Principle of localization). — Let μ\mu and ν\nu be two measures on XX, and let (Yα)(Y_\alpha) be a family of open sets of XX such that, for every α\alpha, the restrictions to YαY_\alpha of μ\mu and ν\nu are equal; then the restrictions of μ\mu and ν\nu to Y=⋃αYαY = \bigcup_\alpha Y_\alpha are equal.

2. Support of a measure

Let μ\mu be a measure on a locally compact space XX, and let G\mathscr{G} be the set of open sets U⊂XU \subset X such that the restriction of μ\mu to UU is zero; it follows at once from the principle of localization (No. 1, Cor. of Prop. 1) that if U0U_0 is the union of the sets U∈GU \in \mathscr{G}, then U0U_0 itself belongs to G\mathscr{G} and is therefore the largest of the sets of G\mathscr{G}.

Definition 1

If μ\mu is a measure on a locally compact space XX, one defines the support of μ\mu, denoted Supp⁡(μ)\operatorname{Supp}(\mu), to be the closed set complementary to the largest of the open sets in XX on which the restriction of μ\mu is zero.

To say that a point x∈Xx \in X does not belong to the support of μ\mu means that there exists an open neighborhood VV of xx such that the restriction of μ\mu to VV is zero; to say that xx belongs to the support of μ\mu therefore means that for every neighborhood VV of xx, there exists a function f∈K(X;C)f \in \mathscr{K}(X; \mathbf{C}), whose support is contained in VV, such that μ(f)≠0\mu(f) \neq 0.

Example

— 1) For a measure on XX to be zero, it is necessary and sufficient that its support be empty. 2) The support of Lebesgue measure on R\mathbf{R} is the entire line R\mathbf{R}; for, it is nonempty and is invariant under every translation.

3) On the interval X=[0,1]X = [0,1] of R\mathbf{R} consider a countable dense subset, arranged into a sequence (an)(a_n), and let μ\mu be the measure defined by placing the mass 2−n2^{-n} at the point ana_n for every n≥0n \geq 0 (§ 1, No. 3, Example I). The support of μ\mu is all of XX; for, let xx be any point of XX, VV a neighborhood of xx, and ff a continuous real-valued function ≥0\geq 0 on XX, equal to 1 at the point xx, whose support is contained in VV (§ 1, No. 2, Lemma 1); the set of y∈Vy \in V such that f(y)>0f(y) > 0 is open in XX, therefore contains a point ana_n, consequently μ(f)≥f(an)2−n>0\mu(f) \geq f(a_n)2^{-n} > 0.

Proposition 2

The support of a measure μ\mu is identical to the support of the measure ∣μ∣|\mu|; if μ\mu is real, its support is the union of the supports of the measures μ+\mu^+ and μ−\mu^-.

For, if the restriction of μ\mu to an open set UU is zero, then so is that of ∣μ∣|\mu| (resp. of μ+\mu^+ and μ−\mu^- when μ\mu is real), and conversely.

Note that the supports of μ+\mu^+ and μ−\mu^- can be nonempty and identical (cf. Ch. V, § 5, Exer. 5).

Proposition 3

If μ\mu and ν\nu are two measures on a locally compact space XX such that ∣μ∣≤∣ν∣|\mu| \leq |\nu|, then Supp(μ)⊂Supp(ν)\mathrm{Supp}(\mu) \subset \mathrm{Supp}(\nu).

For, if the restriction of ν\nu to an open set is zero, then so is that of μ\mu.

Proposition 4

The support of the sum of two measures is contained in the union of their supports.

For, if the restrictions of two measures to an open set are zero, then the same is true of the restriction of their sum.

If μ\mu and ν\nu are two positive measures, then the support of λ=μ+ν\lambda = \mu + \nu is equal to the union of the supports of μ\mu and ν\nu; for, if x0x_0 is a point of the union and VV is any neighborhood of x0x_0, then there exists a continuous function f≥0f \geq 0 with support contained in VV and such that one of the numbers μ(f)\mu(f), ν(f)\nu(f) is >0> 0; a fortiori, λ(f)=μ(f)+ν(f)>0\lambda(f) = \mu(f) + \nu(f) > 0.

Proposition 5

The support of the restriction of a measure μ\mu to an open set UU is the trace on UU of the support of μ\mu.

The proposition is obvious from the definitions.

Proposition 6

The set of measures on a locally compact space XX, whose support is contained in a closed set FF, is a vaguely closed linear subspace of M(X;C)\mathscr{M}(X; \mathbf{C}).

For, it is the intersection of the vaguely closed hyperplanes with equation μ(f)=0\mu(f) = 0, where ff runs over the set of functions in H(X;C)\mathscr{H}(X; \mathbf{C}) whose support does not intersect FF.

Suppose XX is not compact: given a filter Φ\Phi on the space M(X;C)\mathscr{M}(X; \mathbf{C}) of measures on XX, we shall say that the support of a measure μ\mu recedes indefinitely along Φ\Phi if, for every compact subset KK of XX, there exists a set M∈ΦM \in \Phi such that Supp(μ)∩K=∅\mathrm{Supp}(\mu) \cap K = \varnothing for every measure μ∈M\mu \in M.

Proposition 7

— *If Φ\Phi is a filter on M(X;C)\mathscr{M}(X; \mathbf{C}) such that the support of μ\mu recedes indefinitely along Φ\Phi, then μ\mu converges vaguely to 0 with respect to Φ\Phi.

Let ff be any function in K(X;C)\mathscr{K}(X; \mathbf{C}) and let KK be its support. By hypothesis, there exists a set M∈ΦM \in \Phi such that for every measure μ∈M\mu \in M, Supp⁡(μ)∩K=∅\operatorname{Supp}(\mu) \cap K = \varnothing; it follows that μ(f)=0\mu(f) = 0 for all μ∈M\mu \in M, which proves the proposition.

3. Characterization of the support of a measure

By definition, if the support of a function f∈K(X;C)f \in \mathscr{K}(X; \mathbf{C}) does not intersect the support of a measure μ\mu, then μ(f)=0\mu(f) = 0; but the following more precise result is true:

Proposition 8

— *Let μ\mu be a measure on a locally compact space XX. For every function f∈K(X;C)f \in \mathscr{K}(X; \mathbf{C}) that is zero on Supp⁡(μ)\operatorname{Supp}(\mu), μ(f)=0\mu(f) = 0.

Set K=Supp⁡(f)K = \operatorname{Supp}(f), S=Supp⁡(μ)S = \operatorname{Supp}(\mu). Given a number ε>0\varepsilon > 0, let VV be the set of x∈Xx \in X such that ∣f(x)∣<ε|f(x)| < \varepsilon; VV is an open set containing SS by hypothesis, therefore CSCS is a neighborhood of the compact set CVCV. It follows that there exists a continuous mapping hh of XX into [0,1][0, 1], equal to 1 on CVCV and with support contained in CSCS (§ 1, No. 2, Lemma 1). Since the support of fhfh does not intersect SS, μ(fh)=0\mu(fh) = 0. On the other hand, f=fhf = fh on K∩CVK \cap CV, and ∣fh∣≤∣f∣|fh| \leq |f| on XX, therefore ∣f−fh∣≤2ε|f - fh| \leq 2\varepsilon on XX, by the choice of VV. Observe finally that there exists a number MKM_K such that ∣μ(g)∣≤MK∥g∥|\mu(g)| \leq M_K \|g\| for every function g∈K(X;C)g \in \mathscr{K}(X; \mathbf{C}) whose support is contained in KK; since the support of f−fhf - fh is contained in KK, ∣μ(f−fh)∣≤2MKε|\mu(f - fh)| \leq 2M_K \varepsilon and consequently ∣μ(f)∣=∣μ(f−fh)∣≤2MKε|\mu(f)| = |\mu(f - fh)| \leq 2M_K \varepsilon; since ε\varepsilon is arbitrary, μ(f)=0\mu(f) = 0.

Corollary 1

— *If two functions f,gf, g in K(X;C)\mathscr{K}(X; \mathbf{C}) are equal on Supp⁡(μ)\operatorname{Supp}(\mu), then μ(f)=μ(g)\mu(f) = \mu(g).

Corollary 2

— *Let μ\mu be a positive measure on XX; if f∈K(X;C)f \in \mathscr{K}(X; \mathbf{C}) is such that f(x)≥0f(x) \geq 0 on Supp⁡(μ)\operatorname{Supp}(\mu), then μ(f)≥0\mu(f) \geq 0.

For, f=∣f∣f = |f| on Supp⁡(μ)\operatorname{Supp}(\mu), therefore μ(f)=μ(∣f∣)≥0\mu(f) = \mu(|f|) \geq 0 by Corollary 1.

Corollary 3

— *Let μ\mu be a bounded measure on XX; if f∈K(X;C)f \in \mathscr{K}(X; \mathbf{C}) is such that ∣f(x)∣≤a|f(x)| \leq a on Supp⁡(μ)\operatorname{Supp}(\mu), then ∣μ(f)∣≤a∥μ∥|\mu(f)| \leq a \|\mu\|.

For, Supp⁡(∣μ∣)=Supp⁡(μ)\operatorname{Supp}(|\mu|) = \operatorname{Supp}(\mu), and if hh is a continuous mapping of XX into [0,1][0, 1], equal to 1 on Supp⁡(f)\operatorname{Supp}(f) and with compact support, then ∣f(x)∣≤ah(x)|f(x)| \leq ah(x) on Supp⁡(μ)\operatorname{Supp}(\mu), therefore

∣μ∣(∣f∣)≤a∣μ∣(h)≤a∥μ∥ |\mu|(|f|) \leq a|\mu|(h) \leq a\|\mu\|

by Corollary 2; the conclusion then follows from formula (13) of § 1, No. 6.

Proposition 9

— Let μ\mu be a positive measure on XX; if ff is a function in K+(X)\mathscr{K}_+(X) such that μ(f)=0\mu(f) = 0, then ff is zero on Supp(μ)\mathrm{Supp}(\mu).

Let xx be a point of XX such that f(x)>0f(x) > 0; let us show that xx does not belong to Supp(μ)\mathrm{Supp}(\mu). For, there exist a compact neighborhood VV of xx and a number a>0a > 0 such that f(y)≥af(y) \geq a on VV. If gg is any continuous function ≥0\geq 0 with support contained in VV, let us show that μ(g)=0\mu(g) = 0; indeed, if one sets b=∥g∥b = \|g\| then g≤bf/ag \leq bf/a, whence μ(g)≤bμ(f)/a=0\mu(g) \leq b\mu(f)/a = 0.

Proposition 10

— Let μ\mu be a measure on a locally compact space XX; for every function g∈C(X;C)g \in \mathscr{C}(X; \mathbf{C}), the support of the measure g⋅μg \cdot \mu is the closure TT of the set of points x∈Supp(μ)x \in \mathrm{Supp}(\mu) such that g(x)≠0g(x) \neq 0.

Set S=Supp(μ)S = \mathrm{Supp}(\mu); let x0x_0 be a point not belonging to TT; there exists an open neighborhood VV of x0x_0 such that at every point of V∩SV \cap S, gg is zero; if f∈K(X;C)f \in \mathscr{K}(X; \mathbf{C}) has support contained in VV, then fgfg is zero on SS, therefore (Prop. 8) μ(gf)=0\mu(gf) = 0; in other words, the restriction of g⋅μg \cdot \mu to VV is zero.

Conversely, assuming that the restriction of g⋅μg \cdot \mu to an open neighborhood WW of a point x0∈Xx_0 \in X is zero, let us show that there does not exist a point of W∩SW \cap S at which gg is ≠0\neq 0. Indeed, if there were such a point yy, there would exist a compact neighborhood UU of yy, contained in WW, at every point xx of which g(x)≠0g(x) \neq 0; but then every function f∈K(X;C)f \in \mathscr{K}(X; \mathbf{C}) with support contained in UU could be written f=ghf = gh, where h∈K(X;C)h \in \mathscr{K}(X; \mathbf{C}) has support contained in U⊂WU \subset W; it would then follow that μ(f)=μ(gh)=0\mu(f) = \mu(gh) = 0, contrary to the hypothesis y∈Sy \in S.

Note that TT is contained in the intersection of the support SS of μ\mu and the support of gg, but it is not necessarily equal to this intersection. For example, if X=RX = \mathbf{R}, μ\mu is the Dirac measure at the point 0, and g(x)=xg(x) = x, then g⋅μ=0g \cdot \mu = 0 even though the intersection of the supports of gg and μ\mu reduces to the point 0, thus is nonempty.

Corollary

— In order that the measure g⋅μg \cdot \mu be zero, it is necessary and sufficient that gg be zero on the support of μ\mu.

Proposition 11

— Every measure with compact support is bounded.

For, ∣μ∣|\mu| is also a measure with compact support, thus we can restrict attention to the case that μ≥0\mu \geq 0; if hh is a continuous mapping of XX into [0,1][0, 1], with compact support and equal to 1 on Supp(μ)\mathrm{Supp}(\mu), then for every function f∈K(X;C)f \in \mathscr{K}(X; \mathbf{C}) one has ∣f(x)∣≤∥f∥h(x)|f(x)| \leq \|f\| h(x) on Supp(μ)\mathrm{Supp}(\mu), therefore (Cor. 2 of Prop. 8) μ(∣f∣)≤μ(h)∥f∥\mu(|f|) \leq \mu(h)\|f\|, which proves the proposition (§ 1, No. 8).

4. Point measures. Measures with finite support

Proposition 12

— Let aia_i (1≤i≤n1 \leq i \leq n) be distinct points in a locally compact space XX. Every measure on XX whose support is contained in the set of the aia_i is a linear combination of the measures εai\varepsilon_{a_i} (1≤i≤n1 \leq i \leq n).

For, such a measure μ\mu is zero for every function f∈K(X;C)f \in \mathscr{K}(X; \mathbf{C}) satisfying the nn relations f(ai)=0f(a_i) = 0 (No. 3, Prop. 8); since these relations may be written εai(f)=0\varepsilon_{a_i}(f) = 0, μ\mu is a linear combination of the εai\varepsilon_{a_i} (A, II, §7, No. 5, Cor. 1 of Th. 7).

In particular, every measure whose support is either empty or reduced to a single point xx is of the form αεx\alpha \varepsilon_x, where α\alpha is a complex number; such a measure is said to be a point measure; thus every measure whose support is finite is a sum of point measures.

Theorem 1

— Every measure μ\mu on a locally compact space XX is in the vague closure of the vector space VV of measures whose support is finite and contained in Supp(μ)\mathrm{Supp}(\mu).

It suffices to prove that μ\mu is orthogonal to the subspace V∘V^\circ of K(X;C)\mathscr{K}(X; \mathbf{C}) orthogonal to VV (TVS, II, §6, No. 3, Cor. 2 of Th. 1), that is, that the relations ⟨f,εa⟩=0\langle f, \varepsilon_a \rangle = 0, where aa runs over the support of μ\mu, imply ⟨f,μ⟩=0\langle f, \mu \rangle = 0; but this is just Prop. 8 of No. 3.

Corollary 1

— Every bounded measure μ\mu on XX is in the vague closure of the convex set AA of measures whose support is finite and is contained in that of μ\mu, and whose norm is ≤∥μ∥\leq \| \mu \|. Moreover, if ν\nu tends vaguely to μ\mu while remaining in AA, then ∥ν∥\| \nu \| tends to ∥μ∥\| \mu \|.

To prove the first assertion, it suffices to establish that the measure μ\mu belongs to the polar set of the polar set A∘A^\circ of AA in K(X;C)\mathscr{K}(X; \mathbf{C}) (TVS, II, §6, No. 3, Th. 1 and §8, No. 4); this means that for f∈K(X;C)f \in \mathscr{K}(X; \mathbf{C}), the relations ∣⟨f,εa⟩∣≤1/∥μ∥|\langle f, \varepsilon_a \rangle| \leq 1/\| \mu \| for all a∈Supp(μ)a \in \mathrm{Supp}(\mu) imply that ∣⟨f,μ⟩∣≤1|\langle f, \mu \rangle| \leq 1; but this is a consequence of Cor. 3 of Prop. 8 of No. 3.

To prove the second assertion, we note that

lim inf⁡ν→μ,ν∈A∥ν∥≥∥μ∥ \liminf_{\nu \to \mu, \nu \in A} \| \nu \| \geq \| \mu \|

since the function ν↦∥ν∥\nu \mapsto \| \nu \| is lower semi-continuous for the vague topology (§1, No. 9, Cor. 4 of Prop. 15), and the conclusion follows from the fact that ∥ν∥≤∥μ∥\| \nu \| \leq \| \mu \| for ν∈A\nu \in A by definition.

Corollary 2

— Every bounded measure μ\mu on XX is in the vague closure of the set of measures whose support is finite and contained in that of μ\mu and whose norm is equal to ∥μ∥\| \mu \|.

We can suppose that μ≠0\mu \neq 0. Let VV be an open neighborhood of 0 for the vague topology; for every ε\varepsilon such that 0<ε<10 < \varepsilon < 1, there exists, by virtue of Cor. 1, a measure ν0\nu_0 whose support is finite and contained in Supp(μ)\mathrm{Supp}(\mu) and for which ν0−μ∈V\nu_0 - \mu \in V and ∥μ∥≥∥ν0∥≥(1−ε)∥μ∥\| \mu \| \geq \| \nu_0 \| \geq (1 - \varepsilon) \| \mu \|. Setting ν=(∥μ∥/∥ν0∥)ν0\nu = (\| \mu \| / \| \nu_0 \|) \nu_0, we have ∥ν∥=∥μ∥\| \nu \| = \| \mu \| and ∥ν−ν0∥≤∥μ∥\| \nu - \nu_0 \| \leq \| \mu \|; for ε\varepsilon sufficiently small we therefore have ν−μ∈V+V\nu - \mu \in V + V, whence the conclusion.

Corollary 3

Every bounded positive measure μ\mu on XX is in the vague closure of the convex set of positive measures whose support is finite and contained in that of μ\mu and whose norm is equal to ∥μ∥\| \mu \|.

The same reasoning as in Cor. 2 shows that we can limit ourselves to proving that μ\mu is in the vague closure of the convex set BB formed by the positive measures with finite support contained in Supp(μ)\mathrm{Supp}(\mu) and with norm ≤∥μ∥\leq \| \mu \|. Again, it suffices to establish that μ\mu belongs to the polar set of B∘B^\circ, the polar set of BB in the space K(X;R)\mathscr{K}(X; \mathbf{R}) (TVS, II, §6, No. 3, Th. 1); but this means that for f∈K(X;R)f \in \mathscr{K}(X; \mathbf{R}) the relations ⟨f,εa⟩≥−1/∥μ∥\langle f, \varepsilon_a \rangle \geq -1 / \| \mu \| for all a∈Supp(μ)a \in \mathrm{Supp}(\mu) imply ⟨f,μ⟩≥−1\langle f, \mu \rangle \geq -1, which is a consequence of No. 3, Cor. 2 of Prop. 8.

Corollary 4

In the space M(X;C)\mathscr{M}(X; \mathbf{C}), the set of point measures is total for the topology of strictly compact convergence* (§1, No. 10).

On the cone M+(X)\mathscr{M}_+(X), the topology of strictly compact convergence is identical to the vague topology (§1, No. 10, Prop. 18), and every measure on XX may be written μ1−μ2+iμ3−iμ4\mu_1 - \mu_2 + i \mu_3 - i \mu_4, where the μj\mu_j (1≤j≤41 \leq j \leq 4) are positive measures; the conclusion therefore follows from Th. 1.

Proposition 13

Let μ\mu be a measure on a locally compact space XX. For a point x0x_0 to belong to Supp(μ)\mathrm{Supp}(\mu), it is necessary and sufficient that the point measure εx0\varepsilon_{x_0} be in the vague closure of the set of measures g⋅μg \cdot \mu, where gg runs over the set of continuous functions with compact support such that ∥g⋅μ∥≤1\| g \cdot \mu \| \leq 1.

The condition is obviously sufficient by virtue of Prop. 6 of No. 2. To see that it is necessary, suppose x0∈Supp(μ)x_0 \in \mathrm{Supp}(\mu); consider a finite number of functions fkf_k (1≤k≤n1 \leq k \leq n) in K(X;C)\mathscr{K}(X; \mathbf{C}), and an arbitrary number δ>0\delta > 0; we are to prove that there exists a function g∈K(X;C)g \in \mathscr{K}(X; \mathbf{C}) such that ∥g⋅μ∥≤1\| g \cdot \mu \| \leq 1 and such that

∣fk(x0)−μ(gfk)∣≤δ |f_k(x_0) - \mu(g f_k)| \leq \delta
for 1≤k≤n1 \leq k \leq n. Let UU be a relatively compact open neighborhood of x0x_0 such that the oscillation of each of the fkf_k (1≤k≤n1 \leq k \leq n) on UU is ≤δ/2\leq \delta / 2. By hypothesis, since x0∈Supp(μ)x_0 \in \mathrm{Supp}(\mu), there exists a function g0∈K(X;C)g_0 \in \mathscr{K}(X; \mathbf{C}) whose support is contained in UU and such that μ(g0)≠0\mu(g_0) \neq 0; the measure ν=g0⋅μ\nu = g_0 \cdot \mu is not zero, since for every function f∈K(X;C)f \in \mathscr{K}(X; \mathbf{C}) equal to 1 on UU, ν(f)=μ(g0)≠0\nu(f) = \mu(g_0) \neq 0. Moreover, ν\nu is bounded (No. 3, Prop. 11);

multiplying g0g_0 by a scalar, we can suppose that ∥ν∥=1\| \nu \| = 1. This being so, setting αk=fk(x0)\alpha_k = f_k(x_0) we can write, for 1≤k≤n1 \leq k \leq n and for every function h∈K(X;C)h \in \mathscr{K}(X; \mathbf{C}),

fk(x0)−ν(fkh)=αk(1−ν(h))+ν((αk−fk)h). f_k(x_0) - \nu(f_k h) = \alpha_k (1 - \nu(h)) + \nu((\alpha_k - f_k)h) .

Since ν\nu has its support in UU, we may identify it with its restriction to UU; the hypothesis ∥ν∥=1\| \nu \| = 1 then implies that there exists a function h∈K(X;C)h \in \mathscr{K}(X; \mathbf{C}), with support contained in UU, such that ∥h∥≤1\| h \| \leq 1 and such that ∣αk(1−ν(h))∣≤δ/2| \alpha_k (1 - \nu(h)) | \leq \delta / 2 for 1≤k≤n1 \leq k \leq n. The definition of UU moreover shows that ∣(αk−fk(x))h(x)∣≤δ/2| (\alpha_k - f_k(x)) h(x) | \leq \delta / 2 for all x∈Ux \in U; since ∥ν∥=1\| \nu \| = 1 and Supp⁡(ν)⊂U\operatorname{Supp}(\nu) \subset U we therefore have ∣ν((αk−fk)h)∣≤δ/2| \nu((\alpha_k - f_k)h) | \leq \delta / 2 and so, setting g=g0hg = g_0 h,

∣fk(x0)−μ(gfk)∣≤δfor 1≤k≤n. | f_k(x_0) - \mu(g f_k) | \leq \delta \quad \text{for } 1 \leq k \leq n .

This proves the proposition, since ∥g⋅μ∥=∥(g0h)⋅μ∥≤∥g0⋅μ∥=1\| g \cdot \mu \| = \| (g_0 h) \cdot \mu \| \leq \| g_0 \cdot \mu \| = 1.

Corollary

Let μ\mu be a measure on XX. For a measure ν\nu on XX to be in the vague closure of the set of measures g⋅μg \cdot \mu, where gg runs over K(X;C)\mathscr{K}(X; \mathbf{C}), it is necessary and sufficient that Supp⁡(ν)⊂Supp⁡(μ)\operatorname{Supp}(\nu) \subset \operatorname{Supp}(\mu).

For, Supp⁡(g⋅μ)⊂Supp⁡(μ)\operatorname{Supp}(g \cdot \mu) \subset \operatorname{Supp}(\mu) by No. 3, Prop. 10; therefore the support of every vague limit of measures of the form g⋅μg \cdot \mu is also contained in Supp⁡(μ)\operatorname{Supp}(\mu) (No. 2, Prop. 6). Conversely, if Supp⁡(ν)⊂Supp⁡(μ)\operatorname{Supp}(\nu) \subset \operatorname{Supp}(\mu) then ν\nu is the vague limit of measures with finite support contained in Supp⁡(μ)\operatorname{Supp}(\mu) (Th. 1), hence is in the vague closure of the set of measures g⋅μg \cdot \mu by Prop. 13.

5. Discrete measures

Proposition 14

For a measure μ\mu on a locally compact space XX to be a discrete measure (§ 1, No. 3, Example I), it is necessary and sufficient that Supp⁡(μ)\operatorname{Supp}(\mu) be a discrete closed subspace of XX.

Let μ\mu be a discrete measure on XX, defined by the masses h(x)≠0h(x) \neq 0 placed at the points xx of a discrete closed subspace NN of XX; let us show that Supp⁡(μ)=N\operatorname{Supp}(\mu) = N. For every a∈Na \in N and every neighborhood VV of aa, there exists a function f∈K(X;C)f \in \mathscr{K}(X; \mathbf{C}) with support contained in VV, equal to 1 at the point aa and to 0 at the other points of NN, whence μ(f)=h(a)≠0\mu(f) = h(a) \neq 0. On the other hand if b∉Nb \notin N then there exists a neighborhood WW of bb not intersecting NN; for every function g∈K(X;C)g \in \mathscr{K}(X; \mathbf{C}) with support contained in WW, we therefore have μ(g)=0\mu(g) = 0, which proves that b∉Supp⁡(μ)b \notin \operatorname{Supp}(\mu).

Conversely, let μ\mu be a measure such that Supp⁡(μ)\operatorname{Supp}(\mu) is a discrete closed subspace NN of XX. For every a∈Na \in N, there exists an open neighborhood VaV_a of aa that contains no point of NN distinct from aa; the restriction of μ\mu to VaV_a is therefore a point measure with support {a}\{a\} (No. 2, Prop. 5), hence (No. 4, Prop. 12) is of the form h(a)εah(a)\varepsilon_a, where h(a)≠0h(a) \neq 0. Setting h(x)=0h(x) = 0 at the points of CN\mathbf{C}\mathbf{N}, and denoting by ν\nu the measure defined by the masses h(x)h(x), the principle of localization shows that ν=μ\nu = \mu.

We thus see that on a discrete space XX, every measure is discrete.

Exercises

See the exercises for § 2.

Transcribed from the printed volume. Not translated.