TheoremBase

Proof

Preliminaries. Since A\mathcal{A} is compact it is closed by Compact Subset of Rn\mathbb{R}^n is Closed and bounded by Compact Subset of Rn\mathbb{R}^n is Bounded, so there are x∈Rmx\in\mathbb{R}^{m} and a real R′≥0R'\ge0 with dE(x,a)≤R′d_{E}(x,a)\le R' for every a∈Aa\in\mathcal{A}, where dEd_{E} is the Euclidean distance. Putting R=∣x∣+R′R=|x|+R' and using the triangle inequality for the Euclidean norm (Elementary Properties of the Euclidean Norm on Rn\mathbb{R}^n) gives ∣a∣≤∣a−x∣+∣x∣=dE(x,a)+∣x∣≤R|a|\le|a-x|+|x|=d_{E}(x,a)+|x|\le R for every a∈Aa\in\mathcal{A}. By The Set of Controls with Values in a Closed Bounded Set is Nonempty, Bounded, Convex and Closed, UA\mathcal{U}_{\mathcal{A}} is nonempty, satisfies ∥ξ∥L2≤RT1/2\lVert\xi\rVert_{L^{2}}\le RT^{1/2} for every ξ∈UA\xi\in\mathcal{U}_{\mathcal{A}}, is convex, and is closed for the topology determined by dL2d_{L^{2}}. Claim numbers for HH refer to The Lebesgue Space of Square-Integrable Vector-Valued Functions is a Real Inner Product Space: the pairing is symmetric and linear in each argument, ⟨v,v⟩L2=∥v∥L22\langle v,v\rangle_{L^{2}}=\lVert v\rVert^{2}_{L^{2}}, the Cauchy-Schwarz inequality holds, and the norm vanishes only at 00 (claims 4 and 5).

We also record two elementary facts about the natural powers of 22 in R\mathbb{R}, where ι\iota denotes the canonical map of R\mathbb{R}. First, every 2r2^{r} is positive, and 2s≤2r2^{s}\le2^{r} whenever s≤rs\le r. Indeed, by the recursion of the natural powers one has 21=22^{1}=2 and 2r+1=2⋅2r2^{r+1}=2\cdot2^{r}, so positivity follows by Principle of Induction for the Natural Numbers, and then 2r≤2⋅2r=2r+12^{r}\le2\cdot2^{r}=2^{r+1} for every rr; for s≤rs\le r either s=rs=r, or r=s+ir=s+i for some i∈Ni\in\mathbb{N} by claim 7 of Properties of the Order on the Natural Numbers, and induction on ii gives 2s≤2s+i2^{s}\le2^{s+i}. Since all these numbers are positive, 2−r≤2−s2^{-r}\le2^{-s} for s≤rs\le r. Second, for every real ε>0\varepsilon>0 there is J∈NJ\in\mathbb{N} with 2−J<ε2^{-J}<\varepsilon: by Principle of Induction for the Natural Numbers one has ι(J)≤2J\iota(J)\le2^{J} for every J∈NJ\in\mathbb{N} — for J=1J=1 this reads 1≤21\le2, and if ι(J)≤2J\iota(J)\le2^{J} then ι(J+1)=ι(J)+1≤2J+2J=2J+1\iota(J+1)=\iota(J)+1\le2^{J}+2^{J}=2^{J+1}, using 1≤2J1\le2^{J} and claim 1 of Properties of the Canonical Map from the Natural Numbers to an Ordered Field — hence 2−J≤ι(J)−12^{-J}\le\iota(J)^{-1}, and claim 3 of The Archimedean Property of the Real Numbers provides JJ with ι(J)−1<ε\iota(J)^{-1}<\varepsilon.

Claim 1. Fix ξ,ζ∈H\xi,\zeta\in H and put ar=min⁡(2−r,∣⟨ξ−ζ,wr⟩L2∣)a_{r}=\min\bigl(2^{-r},|\langle\xi-\zeta,w_{r}\rangle_{L^{2}}|\bigr). Each ara_{r} satisfies 0≤ar≤2−r≤2−10\le a_{r}\le2^{-r}\le2^{-1}, so the set {ar:r∈N}\{a_{r}:r\in\mathbb{N}\} is nonempty and bounded above; its supremum exists because R\mathbb{R} is Dedekind complete. Thus ρ(ξ,ζ)\rho(\xi,\zeta) is a well-defined nonnegative real number.

Symmetry. ⟨ζ−ξ,wr⟩L2=−⟨ξ−ζ,wr⟩L2\langle\zeta-\xi,w_{r}\rangle_{L^{2}}=-\langle\xi-\zeta,w_{r}\rangle_{L^{2}} by linearity, and the absolute value is unchanged, so the two defining sets coincide.

Vanishing on the diagonal. ρ(ξ,ξ)=0\rho(\xi,\xi)=0, since every ara_{r} is then min⁡(2−r,0)=0\min(2^{-r},0)=0.

Triangle inequality. Let ξ,η,ζ∈H\xi,\eta,\zeta\in H and r∈Nr\in\mathbb{N}. Write a=∣⟨ξ−η,wr⟩L2∣a=|\langle\xi-\eta,w_{r}\rangle_{L^{2}}|, b=∣⟨η−ζ,wr⟩L2∣b=|\langle\eta-\zeta,w_{r}\rangle_{L^{2}}| and c=2−rc=2^{-r}; by linearity of the pairing and the triangle inequality for the absolute value, ∣⟨ξ−ζ,wr⟩L2∣≤a+b|\langle\xi-\zeta,w_{r}\rangle_{L^{2}}|\le a+b. We check min⁡(c,a+b)≤min⁡(c,a)+min⁡(c,b)\min(c,a+b)\le\min(c,a)+\min(c,b): if a+b≤ca+b\le c then a≤ca\le c and b≤cb\le c, so the right-hand side is a+ba+b; if a+b>ca+b>c and a≥ca\ge c then the right-hand side is at least cc; symmetrically if b≥cb\ge c; and if a<ca<c and b<cb<c the right-hand side is a+b>ca+b>c. In all cases the inequality holds, and since min⁡(c,a)≤ρ(ξ,η)\min(c,a)\le\rho(\xi,\eta) and min⁡(c,b)≤ρ(η,ζ)\min(c,b)\le\rho(\eta,\zeta) we get min⁡(c,∣⟨ξ−ζ,wr⟩L2∣)≤ρ(ξ,η)+ρ(η,ζ)\min\bigl(c,|\langle\xi-\zeta,w_{r}\rangle_{L^{2}}|\bigr)\le\rho(\xi,\eta)+\rho(\eta,\zeta) for every rr. The right-hand side is therefore an upper bound of the defining set, so ρ(ξ,ζ)≤ρ(ξ,η)+ρ(η,ζ)\rho(\xi,\zeta)\le\rho(\xi,\eta)+\rho(\eta,\zeta).

Separation. Suppose ρ(ξ,ζ)=0\rho(\xi,\zeta)=0 and put v=ξ−ζv=\xi-\zeta. For every rr, 0≤min⁡(2−r,∣⟨v,wr⟩L2∣)≤00\le\min(2^{-r},|\langle v,w_{r}\rangle_{L^{2}}|)\le0, and 2−r>02^{-r}>0, so ⟨v,wr⟩L2=0\langle v,w_{r}\rangle_{L^{2}}=0. Let ε>0\varepsilon>0 be real. The set of terms of (wr)r(w_{r})_{r} is dense, so by Characterization of the Closure in a Metric Space by Open Balls there is rr with ∥v−wr∥L2<ε\lVert v-w_{r}\rVert_{L^{2}}<\varepsilon, whence by linearity and Cauchy-Schwarz

∥v∥L22=⟨v,v−wr⟩L2+⟨v,wr⟩L2=⟨v,v−wr⟩L2≤∥v∥L2 ε.\lVert v\rVert^{2}_{L^{2}}=\langle v,v-w_{r}\rangle_{L^{2}}+\langle v,w_{r}\rangle_{L^{2}}=\langle v,v-w_{r}\rangle_{L^{2}}\le\lVert v\rVert_{L^{2}}\,\varepsilon .

If ∥v∥L2>0\lVert v\rVert_{L^{2}}>0, dividing by it gives ∥v∥L2≤ε\lVert v\rVert_{L^{2}}\le\varepsilon for every real ε>0\varepsilon>0, which fails for ε=∥v∥L2/2\varepsilon=\lVert v\rVert_{L^{2}}/2. Hence ∥v∥L2=0\lVert v\rVert_{L^{2}}=0 and ξ=ζ\xi=\zeta by claim 5.

So ρ\rho is a metric on HH, and its restriction to UA\mathcal{U}_{\mathcal{A}} is a metric on UA\mathcal{U}_{\mathcal{A}}.

Claim 2. Suppose first that ξn⇀ξ\xi_{n}\rightharpoonup\xi, and let ε>0\varepsilon>0 be real. Choose J∈NJ\in\mathbb{N} with 2−J<ε2^{-J}<\varepsilon. For each rr with r≤Jr\le J, weak convergence and linearity give that the real sequence (⟨ξn−ξ,wr⟩L2)n\bigl(\langle\xi_{n}-\xi,w_{r}\rangle_{L^{2}}\bigr)_{n} has limit 00, so there is Nr∈NN_{r}\in\mathbb{N} with ∣⟨ξn−ξ,wr⟩L2∣<ε|\langle\xi_{n}-\xi,w_{r}\rangle_{L^{2}}|<\varepsilon for n≥Nrn\ge N_{r}. Let NN be the greatest element of the finite family (Nr)r≤J(N_{r})_{r\le J}, which exists by Greatest Element of a Finite Family in a Totally Ordered Set. Let n≥Nn\ge N and r∈Nr\in\mathbb{N}. If r≤Jr\le J then min⁡(2−r,∣⟨ξn−ξ,wr⟩L2∣)≤∣⟨ξn−ξ,wr⟩L2∣<ε\min\bigl(2^{-r},|\langle\xi_{n}-\xi,w_{r}\rangle_{L^{2}}|\bigr)\le|\langle\xi_{n}-\xi,w_{r}\rangle_{L^{2}}|<\varepsilon; if r>Jr>J then the minimum is at most 2−r≤2−J<ε2^{-r}\le2^{-J}<\varepsilon. So ε\varepsilon is an upper bound of the defining set and ρ(ξn,ξ)≤ε\rho(\xi_{n},\xi)\le\varepsilon for every n≥Nn\ge N. Hence the real sequence (ρ(ξn,ξ))n\bigl(\rho(\xi_{n},\xi)\bigr)_{n} has limit 00.

Conversely suppose that sequence has limit 00. Fix r∈Nr\in\mathbb{N}. Since 2−r>02^{-r}>0 there is NN with ρ(ξn,ξ)<2−r\rho(\xi_{n},\xi)<2^{-r} for n≥Nn\ge N; for such nn the minimum min⁡(2−r,∣⟨ξn−ξ,wr⟩L2∣)\min\bigl(2^{-r},|\langle\xi_{n}-\xi,w_{r}\rangle_{L^{2}}|\bigr) is at most ρ(ξn,ξ)<2−r\rho(\xi_{n},\xi)<2^{-r}, so it cannot equal 2−r2^{-r} and therefore equals ∣⟨ξn−ξ,wr⟩L2∣≤ρ(ξn,ξ)|\langle\xi_{n}-\xi,w_{r}\rangle_{L^{2}}|\le\rho(\xi_{n},\xi). Consequently, given a real ε>0\varepsilon>0 and choosing N′≥NN'\ge N with ρ(ξn,ξ)<min⁡(ε,2−r)\rho(\xi_{n},\xi)<\min(\varepsilon,2^{-r}) for every n≥N′n\ge N', we obtain ∣⟨ξn−ξ,wr⟩L2∣≤ρ(ξn,ξ)<ε|\langle\xi_{n}-\xi,w_{r}\rangle_{L^{2}}|\le\rho(\xi_{n},\xi)<\varepsilon for every n≥N′n\ge N'. By linearity of the pairing the real sequence (⟨ξn,wr⟩L2)n\bigl(\langle\xi_{n},w_{r}\rangle_{L^{2}}\bigr)_{n} therefore has limit ⟨ξ,wr⟩L2\langle\xi,w_{r}\rangle_{L^{2}}. Since ∥ξn∥L2≤RT1/2\lVert\xi_{n}\rVert_{L^{2}}\le RT^{1/2} for every nn and the set of terms of (wr)r(w_{r})_{r} is dense in (H,dL2)(H,d_{L^{2}}), claim 5 of Basic Properties of Weak Convergence in the Lebesgue Space of Square-Integrable Vector-Valued Functions gives ξn⇀ξ\xi_{n}\rightharpoonup\xi.

Claim 3. Let (ξn)n∈N(\xi_{n})_{n\in\mathbb{N}} be a sequence in UA\mathcal{U}_{\mathcal{A}}. Its terms satisfy ∥ξn∥L2≤RT1/2\lVert\xi_{n}\rVert_{L^{2}}\le RT^{1/2}, so by Bounded Sequences in the Lebesgue Space of Square-Integrable Vector-Valued Functions Have Weakly Convergent Subsequences there are natural numbers n1<n2<…n_{1}<n_{2}<\dots and ξ∈H\xi\in H with ξnj⇀ξ\xi_{n_{j}}\rightharpoonup\xi. The set UA\mathcal{U}_{\mathcal{A}} is convex and closed, so ξ∈UA\xi\in\mathcal{U}_{\mathcal{A}} by Closed Convex Subsets of the Lebesgue Space of Square-Integrable Vector-Valued Functions are Weakly Sequentially Closed. By claim 2 the real sequence (ρ(ξnj,ξ))j\bigl(\rho(\xi_{n_{j}},\xi)\bigr)_{j} has limit 00, that is, the subsequence (ξnj)j(\xi_{n_{j}})_{j} converges to ξ\xi in the metric space (UA,ρ)(\mathcal{U}_{\mathcal{A}},\rho), and ξ∈UA\xi\in\mathcal{U}_{\mathcal{A}}.

Thus every sequence in UA\mathcal{U}_{\mathcal{A}} has a subsequence converging in (UA,ρ)(\mathcal{U}_{\mathcal{A}},\rho) to a point of UA\mathcal{U}_{\mathcal{A}}, which is sequential compactness of UA\mathcal{U}_{\mathcal{A}} as a subset of that metric space. By A Sequentially Compact Subset of a Metric Space is Compact it is therefore a compact subset of (UA,ρ)(\mathcal{U}_{\mathcal{A}},\rho).

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…