Bourbaki

Proposition 3

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

Integration, chapter IV, § 7. Barycenters

— Let K be a compact convex subset of a Hausdorff locally convex space E, x a point of K. Every measure μ\mu on K, positive, of total mass 1, and admitting x as barycenter, is in the vague closure of the set of discrete positive measures of total mass 1 that admit x as barycenter.

Let U be a neighborhood of μ\mu for the vague topology; we can suppose that U consists of the measures ν\nu on K such that

∣μ(fi)−ν(fi)∣≤δ |\mu(f_i) - \nu(f_i)| \leq \delta
for a finite number of functions fi∈C(K;C)f_i \in \mathscr{C}(K; \mathbf{C}) (1≤i≤p1 \leq i \leq p) and a number δ>0\delta > 0. For every point a∈Ka \in K, there exists a closed convex neighborhood VaV_a of 0 in E such that
∣fi(y)−fi(a)∣≤δ/2 |f_i(y) - f_i(a)| \leq \delta/2

for 1≤i≤p1 \leq i \leq p and for every y∈Wa=K∩(a+Va)y \in W_a = K \cap (a + V_a). Since KK is compact, there exists a finite number of points aja_j (1≤j≤r1 \leq j \leq r) of KK such that the WajW_{a_j} form a covering of KK (1≤j≤r1 \leq j \leq r). Consider a continuous partition of unity (gj)1≤j≤r(g_j)_{1 \leq j \leq r} on KK, subordinate to the covering (Waj)(W_{a_j}), and set αj=μ(gj)\alpha_j = \mu(g_j) for all jj; if αj≠0\alpha_j \neq 0 set μj=αj−1gj⋅μ\mu_j = \alpha_j^{-1} g_j \cdot \mu, and if αj=0\alpha_j = 0 set μj=εaj\mu_j = \varepsilon_{a_j}. Each of the measures μj\mu_j is positive, of total mass 1, and its support is contained in the compact convex set WajW_{a_j}; moreover, by definition,

μ=∑j=1rαjμj \mu = \sum_{j=1}^r \alpha_j \mu_j

since gj⋅μ=0g_j \cdot \mu = 0 if μ(gj)=0\mu(g_j) = 0; the αj\alpha_j are ≥0\geq 0 and

∑j=1rαj=∑j=1rμ(gj)=μ(∑j=1rgj)=μ(1)=1. \sum_{j=1}^r \alpha_j = \sum_{j=1}^r \mu(g_j) = \mu \left( \sum_{j=1}^r g_j \right) = \mu(1) = 1 .

Let xjx_j be the barycenter of μj\mu_j, which belongs to WajW_{a_j} (No. 1, Prop. 1), and consider the discrete measure ν=∑j=1rαjεxj\nu = \sum_{j=1}^r \alpha_j \varepsilon_{x_j}; it is positive and of total mass 1, and its barycenter is ∑j=1rαjxj\sum_{j=1}^r \alpha_j x_j, which is also the barycenter of μ\mu by virtue of (3), thus is equal to xx. Moreover, by (2), ∣fi(y)−fi(aj)∣≤δ/2|f_i(y) - f_i(a_j)| \leq \delta/2 for all y∈Wajy \in W_{a_j} and for all ii, whence, since Supp⁡(μj)⊂Waj\operatorname{Supp}(\mu_j) \subset W_{a_j}, ∣μj(fi)−fi(aj)∣≤δ/2|\mu_j(f_i) - f_i(a_j)| \leq \delta/2 for 1≤i≤p1 \leq i \leq p. On the other hand, since xj∈Wajx_j \in W_{a_j}, one also has

∣εxj(fi)−fi(aj)∣≤δ/2 |\varepsilon_{x_j}(f_i) - f_i(a_j)| \leq \delta/2

for 1≤i≤p1 \leq i \leq p, whence ∣μj(fi)−εxj(fi)∣≤δ|\mu_j(f_i) - \varepsilon_{x_j}(f_i)| \leq \delta for all ii and jj. Since the αj\alpha_j are ≥0\geq 0 and have sum 1, it follows from (3) and the definition of ν\nu that ν\nu satisfies the inequality (1).

Q.E.D.

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.