Bourbaki

Theorem 2

Tag 09QS, permanent. Cite it as [Bourbaki, Tag 09QS].

Integration, chapter IX, § 4. Inverse limits of measures

— Assume that the directed preordered set I has a countable cofinal subset. Let T=(Ti,pij)\mathscr{T} = (T_i, p_{ij}) be an inverse system of topological spaces, T=lim⁡←TiT = \lim_{\leftarrow} T_i and pip_i the canonical mapping of TT into TiT_i. Then every inverse system (μi)i∈I(\mu_i)_{i \in I} of measures on T\mathscr{T} admits an inverse limit.

We shall first treat the case that I=NI = \mathbf{N} and set qn=pn,n+1q_n = p_{n,n+1}. Let ε>0\varepsilon > 0. Define recursively a sequence of compact sets Ln⊂TnL_n \subset T_n as follows: L0L_0 is a compact subset of T0T_0 such that μ0∙(T0−L0)≤ε/2\mu_0^\bullet(T_0 - L_0) \leq \varepsilon/2, and for n≥0n \geq 0 the compact set Ln+1L_{n+1} is contained in qn−1(Ln)‾\overline{q_n^{-1}(L_n)} and satisfies

μn+1∙(qn−1(Ln)‾−Ln+1)≤ε/2n+2. \mu_{n+1}^\bullet(\overline{q_n^{-1}(L_n)} - L_{n+1}) \leq \varepsilon/2^{n+2}.
This construction is possible by virtue of Remark 3 of §1, No. 2. We have
μn+1∙(Tn+1−Ln+1)=μn+1∙(Tn+1−qn−1(Ln)‾)+μn+1∙(qn−1(Ln)‾−Ln+1)≤μn+1∙(Tn+1−qn−1(Ln)‾)+ε/2n+2=μn∙(Tn−Ln)+ε/2n+2 \begin{align*} \mu_{n+1}^\bullet(T_{n+1} - L_{n+1}) &= \mu_{n+1}^\bullet(T_{n+1} - \overline{q_n^{-1}(L_n)}) + \mu_{n+1}^\bullet(\overline{q_n^{-1}(L_n)} - L_{n+1}) \\ &\leq \mu_{n+1}^\bullet(T_{n+1} - \overline{q_n^{-1}(L_n)}) + \varepsilon/2^{n+2} \\ &= \mu_n^\bullet(T_n - L_n) + \varepsilon/2^{n+2} \end{align*}
because μn=qn(μn+1)\mu_n = q_n(\mu_{n+1}); by induction on pp, one deduces that
μp∙(Tp−Lp)≤ε(1−1/2p+1)≤ε. \mu_p^\bullet(T_p - L_p) \leq \varepsilon(1 - 1/2^{p+1}) \leq \varepsilon.
Since TT is a closed subspace of ∏n∈NTn\prod_{n \in \mathbf{N}} T_n and the product space ∏n∈NLn\prod_{n \in \mathbf{N}} L_n is compact, the subset L=T∩∏n∈NLn=⋂n∈Npn−1(Ln)‾L = T \cap \prod_{n \in \mathbf{N}} L_n = \bigcap_{n \in \mathbf{N}} \overline{p_n^{-1}(L_n)} of TT is compact. Let n∈Nn \in \mathbf{N}; we have pn(L)=⋂m≥npnm(Lm)p_n(L) = \bigcap_{m \geq n} p_{nm}(L_m) (GT, I, §9, No. 6, Prop. 8) and pnm(Lm)⊃pnm′(Lm′)p_{nm}(L_m) \supset p_{nm'}(L_{m'}) for m′≥m≥nm' \geq m \geq n, whence
μn∙(Tn−pn(L))=lim⁡m→∞μn∙(Tn−pnm(Lm)). \mu_n^\bullet(T_n - p_n(L)) = \lim_{m \to \infty} \mu_n^\bullet(T_n - p_{nm}(L_m)).

But, for m≥nm \geq n, the measure μn\mu_n is the image of μm\mu_m under pnmp_{nm}, whence

μn∙(Tn−pnm(Lm))=μm∙(Tm−p‾nm(pnm(Lm)))≤μm∙(Tm−Lm)≤ε; \mu_n^\bullet(T_n - p_{nm}(L_m)) = \mu_m^\bullet(T_m - \overline{p}_{nm}(p_{nm}(L_m))) \leq \mu_m^\bullet(T_m - L_m) \leq \varepsilon;

passing to the limit with respect to mm, we obtain μn∙(Tn−pn(L))≤ε\mu_n^\bullet(T_n - p_n(L)) \leq \varepsilon. In other words, Prokhorov’s condition (P) is satisfied, and there exists a bounded measure μ\mu on TT such that μn=pn(μ)\mu_n = p_n(\mu) for all n∈Nn \in \mathbf{N} (No. 2, Th. 1).

Let us pass to the general case: there exists in II an increasing cofinal sequence (in)n∈N(i_n)_{n \in \mathbf{N}}. The mapping t↦(pin(t))n∈Nt \mapsto (p_{i_n}(t))_{n \in \mathbf{N}} is a homeomorphism of TT onto the inverse limit of the inverse system (Tin,pinim)(T_{i_n}, p_{i_n i_m}) (GT, I, §4, No. 4). By the first part of the proof, there exists therefore a bounded measure μ\mu on TT such that μin=pin(μ)\mu_{i_n} = p_{i_n}(\mu) for all n∈Nn \in \mathbf{N}. Let i∈Ii \in I; there exists an n∈Nn \in \mathbf{N} with i≤ini \leq i_n, whence

pi(μ)=piin(pin(μ))=piin(μin)=μi. p_i(\mu) = p_{i i_n}(p_{i_n}(\mu)) = p_{i i_n}(\mu_{i_n}) = \mu_i.

Q.E.D.

Theorem 2 is often used in the following situation: let DD be a countable set and (Xt)t∈D(X_t)_{t \in D} a family of topological spaces. Let F\mathfrak{F} be the set of finite subsets of DD, ordered by inclusion. For JJ in F\mathfrak{F}, set XJ=∏t∈JXtX_J = \prod_{t \in J} X_t, and for J⊂J′J \subset J' let pJJ′p_{JJ'} be the canonical projection of XJ′X_{J'} onto the partial product XJX_J. Also set X=∏t∈DXtX = \prod_{t \in D} X_t and denote by pJp_J the canonical projection of XX onto the partial product XJX_J. One shows easily (cf. S, III, §7, No. 2, Remark 3) that the family (pJ)J∈F(p_J)_{J \in \mathfrak{F}} defines a homeomorphism of XX onto lim⁡←XJ\lim_{\leftarrow} X_J. An inverse system of measures is then a family of bounded measures μJ\mu_J on XJX_J such that μJ=pJJ′(μJ′)\mu_J = p_{JJ'}(\mu_{J'}) for J⊂J′J \subset J'. There exists one and only one bounded measure μ\mu on XX such that μJ=pJ(μ)\mu_J = p_J(\mu) for every finite subset JJ of DD (Kolmogoroff’s theorem). One sometimes says that μ\mu is the measure on ∏t∈DXt\prod_{t \in D} X_t having margins μJ\mu_J.

In particular, suppose given, for every t∈Dt \in D, a measure νt\nu_t on XtX_t of total mass 1. Set μJ=⨂t∈Jνt\mu_J = \bigotimes_{t \in J} \nu_t for every finite subset JJ of DD. Let J⊂J′J \subset J' be two finite subsets of DD and let K=J′−JK = J' - J; identifying XJ′X_{J'} with XJ×XKX_J \times X_K, one has μJ′=μJ⊗μK\mu_{J'} = \mu_J \otimes \mu_K, and since the measure μK\mu_K has total mass 1, the projection of μJ⊗μK\mu_J \otimes \mu_K on XJX_J is equal to μJ\mu_J. The measure on XX admitting the margins μJ\mu_J is denoted ⨂t∈Dνt\bigotimes_{t \in D} \nu_t and is called the product of the family (νt)t∈D(\nu_t)_{t \in D}. When the spaces XtX_t are compact, we recover the construction of Ch. III, §4, No. 6.

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.