TheoremBase

A diagonal argument on a dense sequence in the unit ball gives the weak-star convergent subsequence; the weighted supremum metric is shown to metrize weak-star convergence on the ball by an approximation estimate; sequential compactness then gives completeness and total boundedness, hence compactness.

Proof

Throughout, U={x∈X:∥x∥≤1}U=\{x\in X:\|x\|\le1\} is the closed unit ball of XX, and every F∈BM∗F\in B^{*}_{M} satisfies

∣F(x)∣≤∥F∥∗ ∥x∥≤M∥x∥(x∈X).|F(x)|\le\|F\|_{*}\,\|x\|\le M\|x\|\qquad(x\in X).

The proofs of (i) and (ii) use no choice principle; the proof of (iii) uses the axiom of dependent choice, a standing hypothesis by Radon Measures on Locally Compact Metric Spaces: Standing Notation (iv), both directly and through the compactness lemma it cites.

Proof of (i): such a sequence exists. The proof of (i) needs a sequence (xk)(x_{k}) in UU whose range is dense in UU; we construct one. By Separable Metric Space and Separable Real Normed Space (ii) and Separable Metric Space and Separable Real Normed Space (i) there is a countable subset D⊆XD\subseteq X that is dense in (X,dX)(X,d_{X}), dXd_{X} the norm metric. It is nonempty, since the open ball B(0,1)B(0,1) meets it, so by Countable Set (i) there is a surjective function N→D\mathbb{N}\to D, k↦ykk\mapsto y_{k}, whose range DD is dense in (X,dX)(X,d_{X}). Let

xk=1max⁡{1,∥yk∥} yk(k∈N).x_{k}=\frac{1}{\max\{1,\|y_{k}\|\}}\,y_{k}\qquad(k\in\mathbb{N}).

Then ∥xk∥=∥yk∥/max⁡{1,∥yk∥}≤1\|x_{k}\|=\|y_{k}\|/\max\{1,\|y_{k}\|\}\le1, so xk∈Ux_{k}\in U.

The range of (xk)(x_{k}) is dense in UU. Let x∈Ux\in U and δ>0\delta>0; we find kk with ∥xk−x∥<δ\|x_{k}-x\|<\delta. By density choose kk with ∥yk−x∥<δ/2\|y_{k}-x\|<\delta/2. If ∥yk∥≤1\|y_{k}\|\le1, then xk=ykx_{k}=y_{k}. Otherwise xk=yk/∥yk∥x_{k}=y_{k}/\|y_{k}\| and

∥yk−xk∥=(1−1∥yk∥)∥yk∥=∥yk∥−1≤∥yk∥−∥x∥≤∥yk−x∥<δ2.\|y_{k}-x_{k}\|=\Bigl(1-\frac{1}{\|y_{k}\|}\Bigr)\|y_{k}\|=\|y_{k}\|-1\le\|y_{k}\|-\|x\|\le\|y_{k}-x\|<\frac{\delta}{2}.

In both cases ∥xk−x∥≤∥xk−yk∥+∥yk−x∥<δ\|x_{k}-x\|\le\|x_{k}-y_{k}\|+\|y_{k}-x\|<\delta. So (xk)(x_{k}) has the properties the statement requires of the sequence in (ii); the proof of (i) uses this sequence, and the proof of (ii) the one of the statement.

An approximation estimate. Let (xk)(x_{k}) be a sequence in UU whose range is dense in UU. For F,G∈BM∗F,G\in B^{*}_{M}, x∈Ux\in U and k∈Nk\in\mathbb{N}, linearity gives F(x)−G(x)=(F(xk)−G(xk))+F(x−xk)−G(x−xk)F(x)-G(x)=\bigl(F(x_{k})-G(x_{k})\bigr)+F(x-x_{k})-G(x-x_{k}), and hence

∣F(x)−G(x)∣≤∣F(xk)−G(xk)∣+2M∥x−xk∥.(1)|F(x)-G(x)|\le|F(x_{k})-G(x_{k})|+2M\|x-x_{k}\|.\tag{1}

Moreover, for x∈Xx\in X with x≠0x\neq0 the vector x/∥x∥x/\|x\| lies in UU and F(x)=∥x∥ F(x/∥x∥)F(x)=\|x\|\,F(x/\|x\|); so a statement about the values at the points of UU that is preserved under multiplication by a constant passes to all of XX.

Proof of (i): the diagonal subsequence. Let (xk)(x_{k}) be the sequence constructed above, let (Fm)(F_{m}) be a sequence in BM∗B^{*}_{M}, and put am,k=Fm(xk)a_{m,k}=F_{m}(x_{k}). Since ∣am,k∣≤M∥xk∥≤M|a_{m,k}|\le M\|x_{k}\|\le M, every sequence (am,k)m∈N(a_{m,k})_{m\in\mathbb{N}} is bounded, and The Diagonal Subsequence Lemma for Bounded Real Arrays (i) gives a strictly increasing sequence (mj)j∈N(m_{j})_{j\in\mathbb{N}} in N\mathbb{N} such that (Fmj(xk))j∈N(F_{m_{j}}(x_{k}))_{j\in\mathbb{N}} converges for every kk. Write Gj=FmjG_{j}=F_{m_{j}}; then (Gj)(G_{j}) is a subsequence of (Fm)(F_{m}), and each (Gj(xk))j(G_{j}(x_{k}))_{j} is a Cauchy sequence by Completeness of the Real Numbers for Sequences: Monotone Convergence, the Bolzano-Weierstrass Theorem and Cauchy Sequences (v).

Convergence at every point. Let x∈Ux\in U and ε>0\varepsilon>0. First choose kk with ∥x−xk∥<ε/(4M+1)\|x-x_{k}\|<\varepsilon/(4M+1), so that 2M∥x−xk∥<ε/22M\|x-x_{k}\|<\varepsilon/2; then, with this kk fixed, choose N∈NN\in\mathbb{N} with ∣Gj(xk)−Gl(xk)∣<ε/2|G_{j}(x_{k})-G_{l}(x_{k})|<\varepsilon/2 for all j,l≥Nj,l\ge N. By (1), ∣Gj(x)−Gl(x)∣<ε|G_{j}(x)-G_{l}(x)|<\varepsilon for all j,l≥Nj,l\ge N. Thus (Gj(x))j(G_{j}(x))_{j} is a Cauchy sequence, and so convergent by Completeness of the Real Numbers for Sequences: Monotone Convergence, the Bolzano-Weierstrass Theorem and Cauchy Sequences (v), for every x∈Ux\in U; for x∉Ux\notin U the sequence Gj(x)=∥x∥ Gj(x/∥x∥)G_{j}(x)=\|x\|\,G_{j}(x/\|x\|) converges as well, by Limits of Sequences of Real Numbers: Uniqueness, Boundedness, Constants, Tails, Arithmetic, Quotients, Absolute Values, Finite Sums, Order, Squeezing, Domination and Subsequences (vi).

The limit functional. Define F:X→RF:X\to\mathbb{R} by F(x)=lim⁡j→∞Gj(x)F(x)=\lim_{j\to\infty}G_{j}(x). Passing to the limit in Gj(u+v)=Gj(u)+Gj(v)G_{j}(u+v)=G_{j}(u)+G_{j}(v) and Gj(λu)=λGj(u)G_{j}(\lambda u)=\lambda G_{j}(u) shows, by Limits of Sequences of Real Numbers: Uniqueness, Boundedness, Constants, Tails, Arithmetic, Quotients, Absolute Values, Finite Sums, Order, Squeezing, Domination and Subsequences (vi) and Limits of Sequences of Real Numbers: Uniqueness, Boundedness, Constants, Tails, Arithmetic, Quotients, Absolute Values, Finite Sums, Order, Squeezing, Domination and Subsequences (ii), that FF is a linear functional on XX; passing to the limit in ∣Gj(x)∣≤M∥x∥|G_{j}(x)|\le M\|x\| gives ∣F(x)∣≤M∥x∥|F(x)|\le M\|x\| by Limits of Sequences of Real Numbers: Uniqueness, Boundedness, Constants, Tails, Arithmetic, Quotients, Absolute Values, Finite Sums, Order, Squeezing, Domination and Subsequences (viii) and Limits of Sequences of Real Numbers: Uniqueness, Boundedness, Constants, Tails, Arithmetic, Quotients, Absolute Values, Finite Sums, Order, Squeezing, Domination and Subsequences (x). So F∈X∗F\in X^{*} and, ∥F∥∗\|F\|_{*} being the infimum of the bounds of FF by The Dual Norm (i), ∥F∥∗≤M\|F\|_{*}\le M. Hence F∈BM∗F\in B^{*}_{M}, and Gj(x)→F(x)G_{j}(x)\to F(x) for every x∈Xx\in X, which is Gj⇀∗FG_{j}\overset{*}{\rightharpoonup}F by Weak-Star Convergence in the Dual of a Real Normed Space (i).

Proof of (ii): dd is a metric. From here on, (xk)(x_{k}) and dd are as in the statement. Let F,G,H∈BM∗F,G,H\in B^{*}_{M}. For every kk, ∣F(xk)−G(xk)∣≤2M∥xk∥≤2M|F(x_{k})-G(x_{k})|\le2M\|x_{k}\|\le2M and 2−k≤1/22^{-k}\le1/2, so the numbers 2−k∣F(xk)−G(xk)∣2^{-k}|F(x_{k})-G(x_{k})| form a nonempty subset of [0,M][0,M]; its supremum d(F,G)d(F,G) exists and lies in [0,M][0,M]. Clearly d(F,G)=d(G,F)d(F,G)=d(G,F) and d(F,F)=0d(F,F)=0. For every kk,

2−k∣F(xk)−H(xk)∣≤2−k∣F(xk)−G(xk)∣+2−k∣G(xk)−H(xk)∣≤d(F,G)+d(G,H),2^{-k}|F(x_{k})-H(x_{k})|\le2^{-k}|F(x_{k})-G(x_{k})|+2^{-k}|G(x_{k})-H(x_{k})|\le d(F,G)+d(G,H),

and taking the supremum over kk gives the triangle inequality. Finally suppose d(F,G)=0d(F,G)=0, so that F(xk)=G(xk)F(x_{k})=G(x_{k}) for every kk. For x∈Ux\in U and ε>0\varepsilon>0 choose kk with 2M∥x−xk∥<ε2M\|x-x_{k}\|<\varepsilon; then (1) gives ∣F(x)−G(x)∣<ε|F(x)-G(x)|<\varepsilon. Hence F(x)=G(x)F(x)=G(x) for x∈Ux\in U, and by homogeneity for every x∈Xx\in X, so F=GF=G. By Metric Space (i), dd is a metric on BM∗B^{*}_{M}.

Convergence for dd implies weak-star convergence. Let (Gm)(G_{m}) be a sequence in BM∗B^{*}_{M} and G∈BM∗G\in B^{*}_{M}, and suppose Gm→GG_{m}\to G in (BM∗,d)(B^{*}_{M},d). For every kk, ∣Gm(xk)−G(xk)∣≤2kd(Gm,G)|G_{m}(x_{k})-G(x_{k})|\le2^{k}d(G_{m},G), so Gm(xk)→G(xk)G_{m}(x_{k})\to G(x_{k}). Let x∈Ux\in U and ε>0\varepsilon>0. First choose kk with 2M∥x−xk∥<ε/22M\|x-x_{k}\|<\varepsilon/2; then, with kk fixed, choose NN with ∣Gm(xk)−G(xk)∣<ε/2|G_{m}(x_{k})-G(x_{k})|<\varepsilon/2 for every m≥Nm\ge N. By (1), ∣Gm(x)−G(x)∣<ε|G_{m}(x)-G(x)|<\varepsilon for every m≥Nm\ge N. So Gm(x)→G(x)G_{m}(x)\to G(x) for every x∈Ux\in U, hence, by homogeneity and Limits of Sequences of Real Numbers: Uniqueness, Boundedness, Constants, Tails, Arithmetic, Quotients, Absolute Values, Finite Sums, Order, Squeezing, Domination and Subsequences (vi), for every x∈Xx\in X: Gm⇀∗GG_{m}\overset{*}{\rightharpoonup}G.

Weak-star convergence implies convergence for dd. Conversely, suppose Gm⇀∗GG_{m}\overset{*}{\rightharpoonup}G, and let ε>0\varepsilon>0. First choose K∈NK\in\mathbb{N} with 2−K(2M+1)<ε/22^{-K}(2M+1)<\varepsilon/2, which is possible since (1/2)n→0(1/2)^{n}\to0 by Completeness of the Real Numbers for Sequences: Monotone Convergence, the Bolzano-Weierstrass Theorem and Cauchy Sequences (ii). Then, with KK fixed, choose for each k≤Kk\le K an index NkN_{k} with ∣Gm(xk)−G(xk)∣<ε/2|G_{m}(x_{k})-G(x_{k})|<\varepsilon/2 for every m≥Nkm\ge N_{k}, and let N=max⁡{N1,…,NK}N=\max\{N_{1},\dots,N_{K}\}. For m≥Nm\ge N and every kk: if k≤Kk\le K, then 2−k∣Gm(xk)−G(xk)∣<ε/22^{-k}|G_{m}(x_{k})-G(x_{k})|<\varepsilon/2; if k>Kk>K, then 2−k∣Gm(xk)−G(xk)∣≤2−K⋅2M<ε/22^{-k}|G_{m}(x_{k})-G(x_{k})|\le2^{-K}\cdot2M<\varepsilon/2. Hence d(Gm,G)≤ε/2<εd(G_{m},G)\le\varepsilon/2<\varepsilon for every m≥Nm\ge N, that is, Gm→GG_{m}\to G in (BM∗,d)(B^{*}_{M},d) by Convergent Sequence in a Metric Space (i).

Proof of (iii): sequential compactness. Every sequence in BM∗B^{*}_{M} has, by (i), a subsequence converging weak-star to some G∈BM∗G\in B^{*}_{M}, and by (ii) this subsequence converges to GG in (BM∗,d)(B^{*}_{M},d). So BM∗B^{*}_{M} is sequentially compact in (BM∗,d)(B^{*}_{M},d).

Completeness. Let (Gm)(G_{m}) be a Cauchy sequence in (BM∗,d)(B^{*}_{M},d). By sequential compactness it has a subsequence (Gσ(j))j(G_{\sigma(j)})_{j} converging to some G∈BM∗G\in B^{*}_{M}. Given ε>0\varepsilon>0, choose NN with d(Gm,Gl)<ε/2d(G_{m},G_{l})<\varepsilon/2 for all m,l≥Nm,l\ge N, and then j≥Nj\ge N with d(Gσ(j),G)<ε/2d(G_{\sigma(j)},G)<\varepsilon/2; since σ\sigma is strictly increasing, σ(j)≥j≥N\sigma(j)\ge j\ge N. For every m≥Nm\ge N,

d(Gm,G)≤d(Gm,Gσ(j))+d(Gσ(j),G)<ε.d(G_{m},G)\le d(G_{m},G_{\sigma(j)})+d(G_{\sigma(j)},G)<\varepsilon .

So (Gm)(G_{m}) converges to GG, and (BM∗,d)(B^{*}_{M},d) is complete.

Total boundedness. Suppose that BM∗B^{*}_{M} is not totally bounded in (BM∗,d)(B^{*}_{M},d). Then there is ε>0\varepsilon>0 such that for every finite subset S⊆BM∗S\subseteq B^{*}_{M} some G∈BM∗G\in B^{*}_{M} satisfies d(G,H)≥εd(G,H)\ge\varepsilon for every H∈SH\in S (for S=∅S=\emptyset, any GG, such as 0∈BM∗0\in B^{*}_{M}). Applying dependent choice to the relation that extends a finite sequence (G1,…,Gn)(G_{1},\dots,G_{n}) in BM∗B^{*}_{M} by a Gn+1∈BM∗G_{n+1}\in B^{*}_{M} with d(Gn+1,Gl)≥εd(G_{n+1},G_{l})\ge\varepsilon for every l≤nl\le n, we obtain a sequence (Gm)(G_{m}) in BM∗B^{*}_{M} with d(Gm,Gl)≥εd(G_{m},G_{l})\ge\varepsilon whenever l<ml<m. No subsequence of (Gm)(G_{m}) converges in (BM∗,d)(B^{*}_{M},d): if Gσ(j)→GG_{\sigma(j)}\to G, then d(Gσ(j),Gσ(i))≤d(Gσ(j),G)+d(G,Gσ(i))<εd(G_{\sigma(j)},G_{\sigma(i)})\le d(G_{\sigma(j)},G)+d(G,G_{\sigma(i)})<\varepsilon for all large i<ji<j, a contradiction. This contradicts sequential compactness, so BM∗B^{*}_{M} is totally bounded in (BM∗,d)(B^{*}_{M},d).

Compactness. By The Open Subsets of a Metric Space Form a Topology (i), Td\mathcal{T}_{d} is a topology on BM∗B^{*}_{M}, so ∅=BM∗∖BM∗\emptyset=B^{*}_{M}\setminus B^{*}_{M} lies in Td\mathcal{T}_{d} and BM∗B^{*}_{M} is closed in (BM∗,Td)(B^{*}_{M},\mathcal{T}_{d}). Thus BM∗B^{*}_{M} is a closed, totally bounded subset of the complete metric space (BM∗,d)(B^{*}_{M},d), and A Closed Totally Bounded Subset of a Complete Metric Space is Compact (ii) shows that it is compact in (BM∗,Td)(B^{*}_{M},\mathcal{T}_{d}).

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…