TheoremBase

The metric axioms and the three inequalities come from the operator-norm rules; entrywise convergence gives convergence through the entry-sum bound and the limit of a finite sum, and the entrywise limit of a Cauchy sequence is its limit. The limit laws follow entrywise, and the closed sets are handled with explicit balls.

Proof

Each result cited is universally quantified over the data in its own statement.

Let r,s∈Nr,s\in\mathbb{N}; every clause is proved for these arbitrary sizes, using no others. Write M=Mr×s(R)\mathcal{M}=\mathcal{M}_{r\times s}(\mathbb{R}) and d=dr×sd=d_{r\times s}, so that d(A,B)=∥A−B∥d(A,B)=\lVert A-B\rVert by The Distance between Real Matrices §distance. By Metric Spaces: Balls, the Metric Topology, and the Real and Complex Numbers as Metric Spaces §real-line, dRd_{\mathbb{R}} is a metric on R\mathbb{R} with dR(a,b)=∣a−b∣d_{\mathbb{R}}(a,b)=|a-b| for a,b∈Ra,b\in\mathbb{R}.

Entrywise identities. For A,B,C∈MA,B,C\in\mathcal{M}, comparing entries by Real Matrices: the Set of Real m x n Matrices with Their Entrywise Operations §operations, Real Matrices: the Set of Real m x n Matrices with Their Entrywise Operations §zero and Real Matrices: the Set of Real m x n Matrices with Their Entrywise Operations §equality gives B−A=(−1)(A−B)B-A=(-1)(A-B), A−C=(A−B)+(B−C)A-C=(A-B)+(B-C) and A=(A−B)+BA=(A-B)+B, and shows that A−B=0A-B=0 if and only if A=BA=B; the entries of A−BA-B are Aij−BijA_{ij}-B_{ij}.

Clause metric. dd is the map M×M→R\mathcal{M}\times\mathcal{M}\to\mathbb{R} of The Operator-Norm Distance between Real Matrices Is a Map §map, by The Distance between Real Matrices §distance. Let A,B,C∈MA,B,C\in\mathcal{M}. Nonnegative: d(A,B)=∥A−B∥≥0d(A,B)=\lVert A-B\rVert\ge0 by Rules for the Operator Norm: Bound, Least Constant, Definiteness, Homogeneity, Triangle Inequality and Products §nonnegative. Zero: by Rules for the Operator Norm: Bound, Least Constant, Definiteness, Homogeneity, Triangle Inequality and Products §definite, d(A,B)=0d(A,B)=0 if and only if A−B=0A-B=0, that is, if and only if A=BA=B. Symmetric: d(B,A)=∥(−1)(A−B)∥=∣−1∣ ∥A−B∥=d(A,B)d(B,A)=\lVert(-1)(A-B)\rVert=|-1|\,\lVert A-B\rVert=d(A,B) by Rules for the Operator Norm: Bound, Least Constant, Definiteness, Homogeneity, Triangle Inequality and Products §homogeneity. Triangle: d(A,C)=∥(A−B)+(B−C)∥≤d(A,B)+d(B,C)d(A,C)=\lVert(A-B)+(B-C)\rVert\le d(A,B)+d(B,C) by Rules for the Operator Norm: Bound, Least Constant, Definiteness, Homogeneity, Triangle Inequality and Products §triangle. Hence dd is a metric on M\mathcal{M} by Metric Space §metric, and (M,d)(\mathcal{M},d) is a metric space by Metric Space §metric-space.

Clause entry-bound. Let A,B∈MA,B\in\mathcal{M}, i∈[r]i\in[r] and j∈[s]j\in[s]. Since the (i,j)(i,j) entry of A−BA-B is Aij−BijA_{ij}-B_{ij}, The Operator Norm of Entries, Transposes, Multiples of the Identity, Orthogonal Factors and Diagonal Matrices §entry-bound gives ∣Aij−Bij∣≤∥A−B∥=d(A,B)|A_{ij}-B_{ij}|\le\lVert A-B\rVert=d(A,B).

Clause norm-lipschitz. Let A,B∈MA,B\in\mathcal{M}. From A=(A−B)+BA=(A-B)+B and Rules for the Operator Norm: Bound, Least Constant, Definiteness, Homogeneity, Triangle Inequality and Products §triangle, ∥A∥≤d(A,B)+∥B∥\lVert A\rVert\le d(A,B)+\lVert B\rVert; in the same way ∥B∥≤d(B,A)+∥A∥=d(A,B)+∥A∥\lVert B\rVert\le d(B,A)+\lVert A\rVert=d(A,B)+\lVert A\rVert, by the clause metric (Metric Space §symmetric); so ∣∥A∥−∥B∥∣≤d(A,B)\bigl|\lVert A\rVert-\lVert B\rVert\bigr|\le d(A,B).

Clause vector-bound. Let A,B∈MA,B\in\mathcal{M} and x∈Rsx\in\mathbb{R}^{s}. From A=(A−B)+BA=(A-B)+B and Algebraic Rules for Matrix Products, Matrix-Vector Products and Transposes §matrix-vector, Ax=(A−B)x+BxAx=(A-B)x+Bx, so Ax−Bx=(A−B)xAx-Bx=(A-B)x, comparing components by Euclidean Space Rn\mathbb{R}^n: n-Tuples of Real Numbers with Their Componentwise Operations §operations and Euclidean Space Rn\mathbb{R}^n: n-Tuples of Real Numbers with Their Componentwise Operations §equality; hence ∥Ax−Bx∥≤∥A−B∥ ∥x∥=d(A,B) ∥x∥\lVert Ax-Bx\rVert\le\lVert A-B\rVert\,\lVert x\rVert=d(A,B)\,\lVert x\rVert by Rules for the Operator Norm: Bound, Least Constant, Definiteness, Homogeneity, Triangle Inequality and Products §bound.

Entry sequences. A sequence (A(k))(A^{(k)}) in M\mathcal{M} is a sequence in R[r]×[s]\mathbb{R}^{[r]\times[s]}, as M=R[r]×[s]\mathcal{M}=\mathbb{R}^{[r]\times[s]} by Real Matrices: the Set of Real m x n Matrices with Their Entrywise Operations §matrices; for i∈[r]i\in[r] and j∈[s]j\in[s], its entry sequence (Aij(k))(A^{(k)}_{ij}), with Aij(k)=A(k)(i,j)A^{(k)}_{ij}=A^{(k)}(i,j) by Real Matrices: the Set of Real m x n Matrices with Their Entrywise Operations §entries, is the sequence k↦A(k)(i,j)k\mapsto A^{(k)}(i,j), a sequence in R\mathbb{R} (The Sequences k ↦ f_k(p) of a Sequence in ℝ^I, and the Map of Their Limits §sections). The set [r]×[s][r]\times[s] is finite: [r]={1,…,r}[r]=\{1,\dots,r\} and [s]={1,…,s}[s]=\{1,\dots,s\} are finite intervals (Finite Sets: the Pigeonhole Principle, Uniqueness of the Length, Subsets, Unions, Products, Images, Bounded Sets of Natural Numbers, Extreme Elements, Sets of Maps, Finite Unions and Finite Choice §naturals), and a product of finite sets is finite (Finite Sets: the Pigeonhole Principle, Uniqueness of the Length, Subsets, Unions, Products, Images, Bounded Sets of Natural Numbers, Extreme Elements, Sets of Maps, Finite Unions and Finite Choice §union).

Clause convergence. Let (A(k))(A^{(k)}) be a sequence in M\mathcal{M} and A∈MA\in\mathcal{M}.

Suppose (A(k))(A^{(k)}) converges to AA in (M,d)(\mathcal{M},d), and let i∈[r]i\in[r], j∈[s]j\in[s] and ε>0\varepsilon>0. The choices are made in this order: ii, jj and ε\varepsilon, then NN depending only on ε\varepsilon, then k≥Nk\ge N. By Convergent Sequence in a Metric Space §converges there is N∈NN\in\mathbb{N} with d(A(k),A)<εd(A^{(k)},A)<\varepsilon for every k≥Nk\ge N; for these kk, dR(Aij(k),Aij)=∣Aij(k)−Aij∣≤d(A(k),A)<εd_{\mathbb{R}}(A^{(k)}_{ij},A_{ij})=|A^{(k)}_{ij}-A_{ij}|\le d(A^{(k)},A)<\varepsilon by the clause entry-bound. So (Aij(k))(A^{(k)}_{ij}) converges to AijA_{ij} in (R,dR)(\mathbb{R},d_{\mathbb{R}}), by the same definition.

Conversely, suppose (Aij(k))(A^{(k)}_{ij}) converges to AijA_{ij} in (R,dR)(\mathbb{R},d_{\mathbb{R}}) for all i∈[r]i\in[r] and j∈[s]j\in[s], and let ε>0\varepsilon>0; put η=ε/(rs+1)>0\eta=\varepsilon/(rs+1)>0, where rsrs and rs+1rs+1 are read in R\mathbb{R}. The choices are made in this order: ε\varepsilon, then η\eta, then one NN depending only on ε\varepsilon, then k≥Nk\ge N.

The condition. Let F:N→MF:\mathbb{N}\to\mathcal{M}, k↦A(k)k\mapsto A^{(k)}, be the given sequence (Sequences §sequence). As M=R[r]×[s]\mathcal{M}=\mathbb{R}^{[r]\times[s]} (entry sequences), the evaluation map of Projection, Evaluation and Currying Maps §evaluation, with R\mathbb{R} for its YY and [r]×[s][r]\times[s] for its II, is a map e:M×([r]×[s])→Re:\mathcal{M}\times([r]\times[s])\to\mathbb{R} with e((B,q))=B(q)e((B,q))=B(q) for B∈MB\in\mathcal{M} and q∈[r]×[s]q\in[r]\times[s]. As a−b∈Ra-b\in\mathbb{R} for a,b∈Ra,b\in\mathbb{R}, Maps and Relations Given by Formulas §binary gives a map δ:R×R→R\delta:\mathbb{R}\times\mathbb{R}\to\mathbb{R} with δ((a,b))=a−b\delta((a,b))=a-b, the difference map; let α:R→R\alpha:\mathbb{R}\to\mathbb{R} be the absolute-value map a↦∣a∣a\mapsto|a| of Absolute Value and Squaring as Maps on the Reals, and Entrywise Absolute Values and Squares of a Real Matrix §absolute-value. The maps FF, AA, ee, δ\delta and α\alpha, the number η\eta, the sets [r]×[s][r]\times[s] and N\mathbb{N}, and the strict relation << of R\mathbb{R}, for which a<ba<b means (a,b)∈<(a,b)\in{<}, enter the formulas below as parameters: set variables, or for << a class variable, to which they are assigned. Let pp and kk be distinct set variables, and let γ\gamma be the guard p∈[r]×[s]∧k∈Np\in[r]\times[s]\wedge k\in\mathbb{N} for them. Under γ\gamma: pp and kk are terms of type [r]×[s][r]\times[s] and N\mathbb{N} (Typed Terms under a Predicative Guard: Variables, Parameters, Values of Maps and Ordered Pairs, and Predicative Equations and Memberships between Them §variable); A(k)=F(k)A^{(k)}=F(k) is of type M\mathcal{M}, and A(p)A(p) is of type R\mathbb{R} (Typed Terms under a Predicative Guard: Variables, Parameters, Values of Maps and Ordered Pairs, and Predicative Equations and Memberships between Them §value); A(k)(p)=e((F(k),p))A^{(k)}(p)=e((F(k),p)) is of type R\mathbb{R} (Typed Terms under a Predicative Guard: Variables, Parameters, Values of Maps and Ordered Pairs, and Predicative Equations and Memberships between Them §pair, then the clause value); the difference A(k)(p)−A(p)=δ((A(k)(p),A(p)))A^{(k)}(p)-A(p)=\delta((A^{(k)}(p),A(p))) and its absolute value ∣A(k)(p)−A(p)∣=α(δ((A(k)(p),A(p))))|A^{(k)}(p)-A(p)|=\alpha(\delta((A^{(k)}(p),A(p)))) are of type R\mathbb{R}, by the clauses pair and value; and, η\eta being of type R\mathbb{R} (Typed Terms under a Predicative Guard: Variables, Parameters, Values of Maps and Ordered Pairs, and Predicative Equations and Memberships between Them §parameter), the pair (∣A(k)(p)−A(p)∣,η)(|A^{(k)}(p)-A(p)|,\eta) is of type R×R\mathbb{R}\times\mathbb{R}. Let Q(p,k)Q(p,k) be the condition "p∈[r]×[s]p\in[r]\times[s], k∈Nk\in\mathbb{N} and ∣A(k)(p)−A(p)∣<η|A^{(k)}(p)-A(p)|<\eta", that is, γ∧(∣A(k)(p)−A(p)∣,η)∈<\gamma\wedge(|A^{(k)}(p)-A(p)|,\eta)\in{<}. By Typed Terms under a Predicative Guard: Variables, Parameters, Values of Maps and Ordered Pairs, and Predicative Equations and Memberships between Them §membership, with << as its class CC, QQ is expressed by a predicative formula.

One threshold. Let p∈[r]×[s]p\in[r]\times[s], so p=(i,j)p=(i,j) with i∈[r]i\in[r] and j∈[s]j\in[s], and A(k)(p)=Aij(k)A^{(k)}(p)=A^{(k)}_{ij} and A(p)=AijA(p)=A_{ij} by Real Matrices: the Set of Real m x n Matrices with Their Entrywise Operations §entries. By the same definition of convergence, with η\eta in place of ε\varepsilon, there is NpN_{p} with Np∈NN_{p}\in\mathbb{N} and dR(Aij(k),Aij)=∣A(k)(p)−A(p)∣<ηd_{\mathbb{R}}(A^{(k)}_{ij},A_{ij})=|A^{(k)}(p)-A(p)|<\eta, that is, Q(p,k)Q(p,k), for every k∈Nk\in\mathbb{N} with Np≤kN_{p}\le k. As [r]×[s][r]\times[s] is finite, One Threshold for Finitely Many Monotone Conditions and for Tails §tails, with [r]×[s][r]\times[s] for its II and kk for its nn, gives one common N∈NN\in\mathbb{N} such that Q(p,k)Q(p,k) holds for every p∈[r]×[s]p\in[r]\times[s] and every k∈Nk\in\mathbb{N} with N≤kN\le k; that is, ∣Aij(k)−Aij∣<η|A^{(k)}_{ij}-A_{ij}|<\eta for all k≥Nk\ge N, i∈[r]i\in[r] and j∈[s]j\in[s]. Let k≥Nk\ge N. By The Operator Norm of Entries, Transposes, Multiples of the Identity, Orthogonal Factors and Diagonal Matrices §entry-sum applied to A(k)−AA^{(k)}-A, whose entries are Aij(k)−AijA^{(k)}_{ij}-A_{ij}, and by Finite Sums in a Commutative Ring and in an Ordered Field: Distributivity, Differences, Telescoping, Constant Terms, Comparison and the Triangle Inequality §comparison and Finite Sums in a Commutative Ring and in an Ordered Field: Distributivity, Differences, Telescoping, Constant Terms, Comparison and the Triangle Inequality §constant in the ordered field R\mathbb{R}, applied first to the sum over j∈[s]j\in[s] for each i∈[r]i\in[r] and then to the sum over i∈[r]i\in[r],

d(A(k),A)≤∑i=1r∑j=1s∣Aij(k)−Aij∣≤∑i=1rs η=rs η<ε,d(A^{(k)},A)\le\sum_{i=1}^{r}\sum_{j=1}^{s}|A^{(k)}_{ij}-A_{ij}|\le\sum_{i=1}^{r}s\,\eta=rs\,\eta<\varepsilon,

the last step as rs<rs+1rs<rs+1 and ε>0\varepsilon>0. So (A(k))(A^{(k)}) converges to AA in (M,d)(\mathcal{M},d), by the same definition of convergence.

Claim (the matrix of limits). Let (A(k))(A^{(k)}) be a sequence in M\mathcal{M} such that (Aij(k))(A^{(k)}_{ij}) is a Cauchy sequence in (R,dR)(\mathbb{R},d_{\mathbb{R}}) for all i∈[r]i\in[r] and j∈[s]j\in[s]. Then there is L∈ML\in\mathcal{M} with Lij=lim⁡k→∞Aij(k)L_{ij}=\lim_{k\to\infty}A^{(k)}_{ij} for all i∈[r]i\in[r] and j∈[s]j\in[s], and (A(k))(A^{(k)}) converges to LL in (M,d)(\mathcal{M},d).

Proof of the claim. As (R,dR)(\mathbb{R},d_{\mathbb{R}}) is complete by Convergence, Limits and the Cauchy Condition for Real Sequences Agree with Those in the Real Line as a Metric Space, which is Complete §complete, each (Aij(k))(A^{(k)}_{ij}) is convergent in (R,dR)(\mathbb{R},d_{\mathbb{R}}) by Complete Metric Space §complete. By The Sequences k ↦ f_k(p) of a Sequence in ℝ^I, and the Map of Their Limits §limit-map, applied with I=[r]×[s]I=[r]\times[s] to (A(k))(A^{(k)}), a sequence in R[r]×[s]\mathbb{R}^{[r]\times[s]} whose sequences k↦A(k)(p)k\mapsto A^{(k)}(p) at the points pp are the entry sequences, there is L∈R[r]×[s]=ML\in\mathbb{R}^{[r]\times[s]}=\mathcal{M} such that, for all i∈[r]i\in[r] and j∈[s]j\in[s], (Aij(k))(A^{(k)}_{ij}) converges to L(i,j)=LijL(i,j)=L_{ij} in (R,dR)(\mathbb{R},d_{\mathbb{R}}) and Lij=lim⁡k→∞Aij(k)L_{ij}=\lim_{k\to\infty}A^{(k)}_{ij}. By the clause convergence, (A(k))(A^{(k)}) converges to LL in (M,d)(\mathcal{M},d).

Clause cauchy. Let (A(k))(A^{(k)}) be a sequence in M\mathcal{M}.

Suppose (A(k))(A^{(k)}) is Cauchy in (M,d)(\mathcal{M},d), and let i∈[r]i\in[r], j∈[s]j\in[s] and ε>0\varepsilon>0. The choices are made in this order: ii, jj and ε\varepsilon, then NN depending only on ε\varepsilon, then k,ℓ≥Nk,\ell\ge N. By Cauchy Sequence in a Metric Space §cauchy there is N∈NN\in\mathbb{N} with d(A(k),A(ℓ))<εd(A^{(k)},A^{(\ell)})<\varepsilon for all k,ℓ≥Nk,\ell\ge N, and then dR(Aij(k),Aij(ℓ))≤d(A(k),A(ℓ))<εd_{\mathbb{R}}(A^{(k)}_{ij},A^{(\ell)}_{ij})\le d(A^{(k)},A^{(\ell)})<\varepsilon by the clause entry-bound. So (Aij(k))(A^{(k)}_{ij}) is Cauchy in (R,dR)(\mathbb{R},d_{\mathbb{R}}), by Cauchy Sequence in a Metric Space §cauchy.

Conversely, suppose (Aij(k))(A^{(k)}_{ij}) is Cauchy in (R,dR)(\mathbb{R},d_{\mathbb{R}}) for all i∈[r]i\in[r] and j∈[s]j\in[s]. By the claim, (A(k))(A^{(k)}) converges in (M,d)(\mathcal{M},d) to some L∈ML\in\mathcal{M}. Let ε>0\varepsilon>0; the choices are made in this order: ε\varepsilon, then NN depending only on ε\varepsilon, then k,ℓ≥Nk,\ell\ge N. By the definition of convergence cited in the clause convergence, there is N∈NN\in\mathbb{N} with d(A(k),L)<ε/2d(A^{(k)},L)<\varepsilon/2 for every k≥Nk\ge N, and for k,ℓ≥Nk,\ell\ge N the clause metric (Metric Space §triangle and Metric Space §symmetric) gives d(A(k),A(ℓ))≤d(A(k),L)+d(A(ℓ),L)<εd(A^{(k)},A^{(\ell)})\le d(A^{(k)},L)+d(A^{(\ell)},L)<\varepsilon. So (A(k))(A^{(k)}) is Cauchy in (M,d)(\mathcal{M},d), by Cauchy Sequence in a Metric Space §cauchy.

Clause complete. Let (A(k))(A^{(k)}) be a Cauchy sequence in (M,d)(\mathcal{M},d). By the clause cauchy, (Aij(k))(A^{(k)}_{ij}) is Cauchy in (R,dR)(\mathbb{R},d_{\mathbb{R}}) for all i∈[r]i\in[r] and j∈[s]j\in[s], so by the claim (A(k))(A^{(k)}) converges in (M,d)(\mathcal{M},d) to some L∈ML\in\mathcal{M}. So (A(k))(A^{(k)}) is convergent in (M,d)(\mathcal{M},d) by Convergent Sequence in a Metric Space §convergent, and (M,d)(\mathcal{M},d) is complete by Complete Metric Space §complete.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…