Each result cited below is universally quantified over the data in its own statement.
Preliminaries. (i) Conventions. The four conditions of Metric Space are referred to as conditions 1 to 4. For a real ε > 0 \varepsilon>0 ε > 0 write ε / 2 = ε ⋅ 2 − 1 \varepsilon/2=\varepsilon\cdot2^{-1} ε /2 = ε ⋅ 2 − 1 and ε / 4 = ( ε / 2 ) / 2 \varepsilon/4=(\varepsilon/2)/2 ε /4 = ( ε /2 ) /2 ; by clause 8 of Elementary Order Arithmetic in an Ordered Field these are positive, ε / 2 + ε / 2 = ε \varepsilon/2+\varepsilon/2=\varepsilon ε /2 + ε /2 = ε and ε / 4 + ε / 4 = ε / 2 \varepsilon/4+\varepsilon/4=\varepsilon/2 ε /4 + ε /4 = ε /2 . A sum of inequalities of which one is strict is strict by clause 3 of Elementary Order Arithmetic in an Ordered Field , and a chain mixing ≤ \le ≤ and < < < is strict by clause 2 there. Given N 1 , N 2 ∈ N N_{1},N_{2}\in\mathbb{N} N 1 , N 2 ∈ N , one of N 1 < N 2 N_{1}<N_{2} N 1 < N 2 , N 1 = N 2 N_{1}=N_{2} N 1 = N 2 , N 2 < N 1 N_{2}<N_{1} N 2 < N 1 holds by clause 3 of Properties of the Order on the Natural Numbers ; letting N N N be the larger of the two, every k ≥ N k\ge N k ≥ N satisfies k ≥ N 1 k\ge N_{1} k ≥ N 1 and k ≥ N 2 k\ge N_{2} k ≥ N 2 by clause 1 there. We write ξ m → ξ \xi_{m}\to\xi ξ m → ξ for convergence in a metric space.
(ii) Real limits. A real sequence ( a k ) k ∈ N (a_{k})_{k\in\mathbb{N}} ( a k ) k ∈ N converges to L L L in the sense of Limit of a Sequence of Real Numbers if and only if it converges to L L L in the real line ( R , d R ) (\mathbb{R},d_{\mathbb{R}}) ( R , d R ) of The Absolute Value Metric on the Real Line , since both conditions require ∣ a k − L ∣ < ε |a_{k}-L|<\varepsilon ∣ a k − L ∣ < ε for all large k k k and d R ( a k , L ) = ∣ a k − L ∣ d_{\mathbb{R}}(a_{k},L)=|a_{k}-L| d R ( a k , L ) = ∣ a k − L ∣ . By Uniqueness of Limits in a Metric Space , limits in any metric space, and hence limits of real sequences, are unique.
(iii) Constant sequences. In a metric space ( Z , ρ ) (Z,\rho) ( Z , ρ ) a constant sequence with value c c c converges to c c c , since ρ ( c , c ) = 0 < ε \rho(c,c)=0<\varepsilon ρ ( c , c ) = 0 < ε for every ε > 0 \varepsilon>0 ε > 0 by condition 2.
(iv) Continuity of the distance. Let ( Z , ρ ) (Z,\rho) ( Z , ρ ) be a metric space. For p , q , p ′ , q ′ ∈ Z p,q,p',q'\in Z p , q , p ′ , q ′ ∈ Z put c = ρ ( p , p ′ ) + ρ ( q , q ′ ) c=\rho(p,p')+\rho(q,q') c = ρ ( p , p ′ ) + ρ ( q , q ′ ) . Conditions 4 and 3 give ρ ( p , q ) ≤ ρ ( p , p ′ ) + ρ ( p ′ , q ′ ) + ρ ( q ′ , q ) = ρ ( p ′ , q ′ ) + c \rho(p,q)\le\rho(p,p')+\rho(p',q')+\rho(q',q)=\rho(p',q')+c ρ ( p , q ) ≤ ρ ( p , p ′ ) + ρ ( p ′ , q ′ ) + ρ ( q ′ , q ) = ρ ( p ′ , q ′ ) + c , that is ρ ( p , q ) − ρ ( p ′ , q ′ ) ≤ c \rho(p,q)-\rho(p',q')\le c ρ ( p , q ) − ρ ( p ′ , q ′ ) ≤ c by clause 3 of Elementary Arithmetic in an Ordered Field ; exchanging ( p , q ) (p,q) ( p , q ) with ( p ′ , q ′ ) (p',q') ( p ′ , q ′ ) , using condition 3 to rewrite ρ ( p ′ , p ) + ρ ( q ′ , q ) \rho(p',p)+\rho(q',q) ρ ( p ′ , p ) + ρ ( q ′ , q ) as c c c , and applying clause 4 of Elementary Order Arithmetic in an Ordered Field gives − c ≤ ρ ( p , q ) − ρ ( p ′ , q ′ ) -c\le\rho(p,q)-\rho(p',q') − c ≤ ρ ( p , q ) − ρ ( p ′ , q ′ ) . By clause 6 of Properties of the Absolute Value in an Ordered Field ,
∣ ρ ( p , q ) − ρ ( p ′ , q ′ ) ∣ ≤ ρ ( p , p ′ ) + ρ ( q , q ′ ) . (P) \bigl|\rho(p,q)-\rho(p',q')\bigr|\le\rho(p,p')+\rho(q,q').\tag{P} ρ ( p , q ) − ρ ( p ′ , q ′ ) ≤ ρ ( p , p ′ ) + ρ ( q , q ′ ) . ( P )
Consequently, if p k → p p_{k}\to p p k → p and q k → q q_{k}\to q q k → q in ( Z , ρ ) (Z,\rho) ( Z , ρ ) , then the real sequence ( ρ ( p k , q k ) ) k ∈ N (\rho(p_{k},q_{k}))_{k\in\mathbb{N}} ( ρ ( p k , q k ) ) k ∈ N converges to ρ ( p , q ) \rho(p,q) ρ ( p , q ) : given ε > 0 \varepsilon>0 ε > 0 , choose K 1 K_{1} K 1 with ρ ( p k , p ) < ε / 2 \rho(p_{k},p)<\varepsilon/2 ρ ( p k , p ) < ε /2 for k ≥ K 1 k\ge K_{1} k ≥ K 1 , then K 2 K_{2} K 2 with ρ ( q k , q ) < ε / 2 \rho(q_{k},q)<\varepsilon/2 ρ ( q k , q ) < ε /2 for k ≥ K 2 k\ge K_{2} k ≥ K 2 , and let K K K be the larger; for k ≥ K k\ge K k ≥ K , (P) gives ∣ ρ ( p k , q k ) − ρ ( p , q ) ∣ < ε |\rho(p_{k},q_{k})-\rho(p,q)|<\varepsilon ∣ ρ ( p k , q k ) − ρ ( p , q ) ∣ < ε .
(v) A null sequence of radii. Let ι \iota ι be the canonical map of R \mathbb{R} R and r m = ι ( m ) − 1 r_{m}=\iota(m)^{-1} r m = ι ( m ) − 1 for m ∈ N m\in\mathbb{N} m ∈ N . By clause 3 of Properties of the Canonical Map from the Natural Numbers to an Ordered Field , r m r_{m} r m exists and 0 < r m 0<r_{m} 0 < r m . For every ε > 0 \varepsilon>0 ε > 0 there is M ∈ N M\in\mathbb{N} M ∈ N with r m < ε r_{m}<\varepsilon r m < ε for all m ≥ M m\ge M m ≥ M : clause 3 of The Archimedean Property of the Real Numbers gives M M M with ι ( M ) − 1 < ε \iota(M)^{-1}<\varepsilon ι ( M ) − 1 < ε ; for m ≥ M m\ge M m ≥ M we have ι ( M ) ≤ ι ( m ) \iota(M)\le\iota(m) ι ( M ) ≤ ι ( m ) (equality if m = M m=M m = M , clause 6 of Properties of the Canonical Map from the Natural Numbers to an Ordered Field if M < m M<m M < m ), so r m ≤ ι ( M ) − 1 < ε r_{m}\le\iota(M)^{-1}<\varepsilon r m ≤ ι ( M ) − 1 < ε by Order Reversal under Reciprocals, and Summability of the Reciprocals of the Squares §reciprocal .
(vi) Elements of the completion. By The Metric Completion of a Metric Space §completion , every element of X ^ \widehat{X} X equals [ x ] [x] [ x ] for some x ∈ C ( X ) x\in\mathcal{C}(X) x ∈ C ( X ) , and d ^ ( [ x ] , [ y ] ) = δ ( x , y ) \widehat{d}([x],[y])=\delta(x,y) d ([ x ] , [ y ]) = δ ( x , y ) for x , y ∈ C ( X ) x,y\in\mathcal{C}(X) x , y ∈ C ( X ) , where δ \delta δ is the limit distance of Cauchy Sequences in a Metric Space: Constant Sequences, Convergence of Termwise Distances, the Null Relation, and Independence of Representatives §distances . By The Metric Completion of a Metric Space §canonical-map , κ X ( a ) = [ a ˉ ] \kappa_{X}(a)=[\bar{a}] κ X ( a ) = [ a ˉ ] , where a ˉ ∈ C ( X ) \bar{a}\in\mathcal{C}(X) a ˉ ∈ C ( X ) by Cauchy Sequences in a Metric Space: Constant Sequences, Convergence of Termwise Distances, the Null Relation, and Independence of Representatives §constant .
(vii) Sequential continuity. Once clause 1 is proved, for a metric space ( Z , ρ ) (Z,\rho) ( Z , ρ ) and a map φ : X ^ → Z \varphi:\widehat{X}\to Z φ : X → Z we apply Continuity Between Metric Spaces is Equivalent to Sequential Continuity §on-subset with A = X ^ A=\widehat{X} A = X , whose restricted metric is d ^ \widehat{d} d itself: φ \varphi φ is continuous if and only if φ ( ξ m ) → φ ( ξ ) \varphi(\xi_{m})\to\varphi(\xi) φ ( ξ m ) → φ ( ξ ) in ( Z , ρ ) (Z,\rho) ( Z , ρ ) whenever ξ m → ξ \xi_{m}\to\xi ξ m → ξ in ( X ^ , d ^ ) (\widehat{X},\widehat{d}) ( X , d ) .
Proof of clause 1 (Metric). By The Metric Completion of a Metric Space §completion , d ^ \widehat{d} d is a well-defined map X ^ × X ^ → R \widehat{X}\times\widehat{X}\to\mathbb{R} X × X → R . Let α , β , γ ∈ X ^ \alpha,\beta,\gamma\in\widehat{X} α , β , γ ∈ X and by (vi) write α = [ x ] \alpha=[x] α = [ x ] , β = [ y ] \beta=[y] β = [ y ] , γ = [ z ] \gamma=[z] γ = [ z ] with x , y , z ∈ C ( X ) x,y,z\in\mathcal{C}(X) x , y , z ∈ C ( X ) . By Cauchy Sequences in a Metric Space: Constant Sequences, Convergence of Termwise Distances, the Null Relation, and Independence of Representatives §distances , d ^ ( α , β ) = δ ( x , y ) ≥ 0 \widehat{d}(\alpha,\beta)=\delta(x,y)\ge0 d ( α , β ) = δ ( x , y ) ≥ 0 , d ^ ( α , β ) = δ ( y , x ) = d ^ ( β , α ) \widehat{d}(\alpha,\beta)=\delta(y,x)=\widehat{d}(\beta,\alpha) d ( α , β ) = δ ( y , x ) = d ( β , α ) and d ^ ( α , γ ) = δ ( x , z ) ≤ δ ( x , y ) + δ ( y , z ) = d ^ ( α , β ) + d ^ ( β , γ ) \widehat{d}(\alpha,\gamma)=\delta(x,z)\le\delta(x,y)+\delta(y,z)=\widehat{d}(\alpha,\beta)+\widehat{d}(\beta,\gamma) d ( α , γ ) = δ ( x , z ) ≤ δ ( x , y ) + δ ( y , z ) = d ( α , β ) + d ( β , γ ) ; these are conditions 1, 3 and 4. For condition 2: d ^ ( α , β ) = 0 \widehat{d}(\alpha,\beta)=0 d ( α , β ) = 0 means δ ( x , y ) = 0 \delta(x,y)=0 δ ( x , y ) = 0 , that is x ≈ y x\approx y x ≈ y , which by Cauchy Sequences in a Metric Space: Constant Sequences, Convergence of Termwise Distances, the Null Relation, and Independence of Representatives §null holds if and only if [ x ] = [ y ] [x]=[y] [ x ] = [ y ] , that is α = β \alpha=\beta α = β .
Proof of clause 2 (Isometry). Let a , b ∈ X a,b\in X a , b ∈ X . By (vi) and the last identity of Cauchy Sequences in a Metric Space: Constant Sequences, Convergence of Termwise Distances, the Null Relation, and Independence of Representatives §distances , d ^ ( κ X ( a ) , κ X ( b ) ) = d ^ ( [ a ˉ ] , [ b ˉ ] ) = δ ( a ˉ , b ˉ ) = d ( a , b ) \widehat{d}(\kappa_{X}(a),\kappa_{X}(b))=\widehat{d}([\bar{a}],[\bar{b}])=\delta(\bar{a},\bar{b})=d(a,b) d ( κ X ( a ) , κ X ( b )) = d ([ a ˉ ] , [ b ˉ ]) = δ ( a ˉ , b ˉ ) = d ( a , b ) . If κ X ( a ) = κ X ( b ) \kappa_{X}(a)=\kappa_{X}(b) κ X ( a ) = κ X ( b ) , then d ( a , b ) = d ^ ( κ X ( a ) , κ X ( a ) ) = 0 d(a,b)=\widehat{d}(\kappa_{X}(a),\kappa_{X}(a))=0 d ( a , b ) = d ( κ X ( a ) , κ X ( a )) = 0 by condition 2 for d ^ \widehat{d} d (clause 1), so a = b a=b a = b by condition 2 for d d d . Thus κ X \kappa_{X} κ X is injective.
Proof of clause 3 (Density). Let x ∈ C ( X ) x\in\mathcal{C}(X) x ∈ C ( X ) and ε > 0 \varepsilon>0 ε > 0 . First choose N N N by Cauchy Sequence in a Metric Space with d ( x k , x l ) < ε / 2 d(x_{k},x_{l})<\varepsilon/2 d ( x k , x l ) < ε /2 for all k , l ≥ N k,l\ge N k , l ≥ N . Now fix k ≥ N k\ge N k ≥ N . By (vi), κ X ( x k ) = [ x k ‾ ] \kappa_{X}(x_{k})=[\overline{x_{k}}] κ X ( x k ) = [ x k ] , so d ^ ( κ X ( x k ) , [ x ] ) = δ ( x k ‾ , x ) = : L k \widehat{d}(\kappa_{X}(x_{k}),[x])=\delta(\overline{x_{k}},x)=:L_{k} d ( κ X ( x k ) , [ x ]) = δ ( x k , x ) =: L k , which by Cauchy Sequences in a Metric Space: Constant Sequences, Convergence of Termwise Distances, the Null Relation, and Independence of Representatives §distances is the limit of the real sequence ( d ( x k , x l ) ) l ∈ N (d(x_{k},x_{l}))_{l\in\mathbb{N}} ( d ( x k , x l ) ) l ∈ N . Next choose N ′ N' N ′ with ∣ d ( x k , x l ) − L k ∣ < ε / 2 |d(x_{k},x_{l})-L_{k}|<\varepsilon/2 ∣ d ( x k , x l ) − L k ∣ < ε /2 for all l ≥ N ′ l\ge N' l ≥ N ′ , and let l l l be the larger of N N N and N ′ N' N ′ . By clauses 2 and 3 of Properties of the Absolute Value in an Ordered Field , L k − d ( x k , x l ) ≤ ∣ d ( x k , x l ) − L k ∣ < ε / 2 L_{k}-d(x_{k},x_{l})\le|d(x_{k},x_{l})-L_{k}|<\varepsilon/2 L k − d ( x k , x l ) ≤ ∣ d ( x k , x l ) − L k ∣ < ε /2 , hence L k < d ( x k , x l ) + ε / 2 < ε L_{k}<d(x_{k},x_{l})+\varepsilon/2<\varepsilon L k < d ( x k , x l ) + ε /2 < ε . Thus d ^ ( κ X ( x k ) , [ x ] ) < ε \widehat{d}(\kappa_{X}(x_{k}),[x])<\varepsilon d ( κ X ( x k ) , [ x ]) < ε for every k ≥ N k\ge N k ≥ N , and ( κ X ( x k ) ) k ∈ N (\kappa_{X}(x_{k}))_{k\in\mathbb{N}} ( κ X ( x k ) ) k ∈ N converges to [ x ] [x] [ x ] by Convergent Sequence in a Metric Space .
For density, let T \mathcal{T} T be the collection of open subsets of ( X ^ , d ^ ) (\widehat{X},\widehat{d}) ( X , d ) , a topology by Metric Open Sets Form a Topology . The closure cl X ^ ( κ X ( X ) ) \operatorname{cl}_{\widehat{X}}(\kappa_{X}(X)) cl X ( κ X ( X )) is by definition a subset of X ^ \widehat{X} X . Conversely, let ξ ∈ X ^ \xi\in\widehat{X} ξ ∈ X and write ξ = [ x ] \xi=[x] ξ = [ x ] by (vi). The sequence ( κ X ( x k ) ) k ∈ N (\kappa_{X}(x_{k}))_{k\in\mathbb{N}} ( κ X ( x k ) ) k ∈ N has all its terms in κ X ( X ) \kappa_{X}(X) κ X ( X ) and converges to ξ \xi ξ by the first part, so ξ ∈ cl X ^ ( κ X ( X ) ) \xi\in\operatorname{cl}_{\widehat{X}}(\kappa_{X}(X)) ξ ∈ cl X ( κ X ( X )) by Sequential Characterization of the Closure in a Metric Space . Hence cl X ^ ( κ X ( X ) ) = X ^ \operatorname{cl}_{\widehat{X}}(\kappa_{X}(X))=\widehat{X} cl X ( κ X ( X )) = X , which is density in the sense of Dense Subset of a Topological Space .
Proof of clause 4 (Completeness). Let ( ξ m ) m ∈ N (\xi_{m})_{m\in\mathbb{N}} ( ξ m ) m ∈ N be a Cauchy sequence in ( X ^ , d ^ ) (\widehat{X},\widehat{d}) ( X , d ) . For m ∈ N m\in\mathbb{N} m ∈ N let A m = { a ∈ X : d ^ ( κ X ( a ) , ξ m ) < r m } A_{m}=\{a\in X:\ \widehat{d}(\kappa_{X}(a),\xi_{m})<r_{m}\} A m = { a ∈ X : d ( κ X ( a ) , ξ m ) < r m } , with r m r_{m} r m from (v). Each A m A_{m} A m is nonempty: writing ξ m = [ x ] \xi_{m}=[x] ξ m = [ x ] by (vi), clause 3 gives κ X ( x k ) → ξ m \kappa_{X}(x_{k})\to\xi_{m} κ X ( x k ) → ξ m , and since 0 < r m 0<r_{m} 0 < r m some k k k satisfies d ^ ( κ X ( x k ) , ξ m ) < r m \widehat{d}(\kappa_{X}(x_{k}),\xi_{m})<r_{m} d ( κ X ( x k ) , ξ m ) < r m , so x k ∈ A m x_{k}\in A_{m} x k ∈ A m . By the axiom of countable choice , applied to the family ( A m ) m ∈ N (A_{m})_{m\in\mathbb{N}} ( A m ) m ∈ N of subsets of X X X , there is a sequence a = ( a m ) m ∈ N a=(a_{m})_{m\in\mathbb{N}} a = ( a m ) m ∈ N in X X X with a m ∈ A m a_{m}\in A_{m} a m ∈ A m for every m m m .
The sequence a a a is Cauchy in ( X , d ) (X,d) ( X , d ) . For m , l ∈ N m,l\in\mathbb{N} m , l ∈ N , clause 2 and conditions 3 and 4 for d ^ \widehat{d} d give
d ( a m , a l ) = d ^ ( κ X ( a m ) , κ X ( a l ) ) ≤ d ^ ( κ X ( a m ) , ξ m ) + d ^ ( ξ m , ξ l ) + d ^ ( ξ l , κ X ( a l ) ) < r m + d ^ ( ξ m , ξ l ) + r l . d(a_{m},a_{l})=\widehat{d}(\kappa_{X}(a_{m}),\kappa_{X}(a_{l}))\le\widehat{d}(\kappa_{X}(a_{m}),\xi_{m})+\widehat{d}(\xi_{m},\xi_{l})+\widehat{d}(\xi_{l},\kappa_{X}(a_{l}))<r_{m}+\widehat{d}(\xi_{m},\xi_{l})+r_{l}. d ( a m , a l ) = d ( κ X ( a m ) , κ X ( a l )) ≤ d ( κ X ( a m ) , ξ m ) + d ( ξ m , ξ l ) + d ( ξ l , κ X ( a l )) < r m + d ( ξ m , ξ l ) + r l .
Given ε > 0 \varepsilon>0 ε > 0 , first choose M 1 M_{1} M 1 by (v) with r m < ε / 4 r_{m}<\varepsilon/4 r m < ε /4 for m ≥ M 1 m\ge M_{1} m ≥ M 1 , then M 2 M_{2} M 2 with d ^ ( ξ m , ξ l ) < ε / 2 \widehat{d}(\xi_{m},\xi_{l})<\varepsilon/2 d ( ξ m , ξ l ) < ε /2 for m , l ≥ M 2 m,l\ge M_{2} m , l ≥ M 2 , and let M M M be the larger. For m , l ≥ M m,l\ge M m , l ≥ M we get d ( a m , a l ) < ε / 4 + ε / 2 + ε / 4 = ε d(a_{m},a_{l})<\varepsilon/4+\varepsilon/2+\varepsilon/4=\varepsilon d ( a m , a l ) < ε /4 + ε /2 + ε /4 = ε . So a ∈ C ( X ) a\in\mathcal{C}(X) a ∈ C ( X ) .
By clause 3, κ X ( a m ) → [ a ] \kappa_{X}(a_{m})\to[a] κ X ( a m ) → [ a ] . For every m m m , d ^ ( ξ m , [ a ] ) ≤ d ^ ( ξ m , κ X ( a m ) ) + d ^ ( κ X ( a m ) , [ a ] ) < r m + d ^ ( κ X ( a m ) , [ a ] ) \widehat{d}(\xi_{m},[a])\le\widehat{d}(\xi_{m},\kappa_{X}(a_{m}))+\widehat{d}(\kappa_{X}(a_{m}),[a])<r_{m}+\widehat{d}(\kappa_{X}(a_{m}),[a]) d ( ξ m , [ a ]) ≤ d ( ξ m , κ X ( a m )) + d ( κ X ( a m ) , [ a ]) < r m + d ( κ X ( a m ) , [ a ]) . Given ε > 0 \varepsilon>0 ε > 0 , choose M 1 M_{1} M 1 with r m < ε / 2 r_{m}<\varepsilon/2 r m < ε /2 for m ≥ M 1 m\ge M_{1} m ≥ M 1 , then M 2 M_{2} M 2 with d ^ ( κ X ( a m ) , [ a ] ) < ε / 2 \widehat{d}(\kappa_{X}(a_{m}),[a])<\varepsilon/2 d ( κ X ( a m ) , [ a ]) < ε /2 for m ≥ M 2 m\ge M_{2} m ≥ M 2 , and let M M M be the larger; then d ^ ( ξ m , [ a ] ) < ε \widehat{d}(\xi_{m},[a])<\varepsilon d ( ξ m , [ a ]) < ε for m ≥ M m\ge M m ≥ M . Thus ( ξ m ) (\xi_{m}) ( ξ m ) converges to [ a ] ∈ X ^ [a]\in\widehat{X} [ a ] ∈ X , and ( X ^ , d ^ ) (\widehat{X},\widehat{d}) ( X , d ) is complete by Complete Metric Space .
Proof of clause 5 (Extension). Step 1: limits along Cauchy sequences. For x ∈ C ( X ) x\in\mathcal{C}(X) x ∈ C ( X ) the sequence ( f ( x k ) ) k ∈ N (f(x_{k}))_{k\in\mathbb{N}} ( f ( x k ) ) k ∈ N is Cauchy in ( Y , e ) (Y,e) ( Y , e ) by hypothesis, so it converges by Complete Metric Space , to a unique limit by (ii). Write L ( x ) = lim k → ∞ f ( x k ) L(x)=\lim_{k\to\infty}f(x_{k}) L ( x ) = lim k → ∞ f ( x k ) .
Step 2: independence of the representative. Let x , y ∈ C ( X ) x,y\in\mathcal{C}(X) x , y ∈ C ( X ) with x ≈ y x\approx y x ≈ y ; we show L ( x ) = L ( y ) L(x)=L(y) L ( x ) = L ( y ) . Let T = { 0 , 1 } T=\{0,1\} T = { 0 , 1 } and let t : N × T → T t:\mathbb{N}\times T\to T t : N × T → T be given by t ( n , 0 ) = 1 t(n,0)=1 t ( n , 0 ) = 1 and t ( n , 1 ) = 0 t(n,1)=0 t ( n , 1 ) = 0 . By Definition of Sequences by Recursion on the Natural Numbers §recursion there is a sequence τ : N → T \tau:\mathbb{N}\to T τ : N → T with τ ( 1 ) = 0 \tau(1)=0 τ ( 1 ) = 0 and τ ( n + 1 ) = t ( n , τ ( n ) ) ≠ τ ( n ) \tau(n+1)=t(n,\tau(n))\ne\tau(n) τ ( n + 1 ) = t ( n , τ ( n )) = τ ( n ) for every n n n (recall S ( n ) = n + 1 S(n)=n+1 S ( n ) = n + 1 in Natural Numbers ). Hence for every N ∈ N N\in\mathbb{N} N ∈ N one of N N N , N + 1 N+1 N + 1 is an index m m m with τ ( m ) = 0 \tau(m)=0 τ ( m ) = 0 and the other an index with τ ( m ) = 1 \tau(m)=1 τ ( m ) = 1 , and both are ≥ N \ge N ≥ N by clauses 1 and 5 of Properties of the Order on the Natural Numbers . ( † ) (\dagger) ( † )
Define the sequence z z z in X X X by z m = x m z_{m}=x_{m} z m = x m if τ ( m ) = 0 \tau(m)=0 τ ( m ) = 0 and z m = y m z_{m}=y_{m} z m = y m if τ ( m ) = 1 \tau(m)=1 τ ( m ) = 1 . For every m m m we have d ( z m , x m ) ≤ d ( x m , y m ) d(z_{m},x_{m})\le d(x_{m},y_{m}) d ( z m , x m ) ≤ d ( x m , y m ) : if τ ( m ) = 0 \tau(m)=0 τ ( m ) = 0 the left side is 0 0 0 (condition 2) and the right side is nonnegative (condition 1); if τ ( m ) = 1 \tau(m)=1 τ ( m ) = 1 the two sides are equal by condition 3. Hence, by conditions 3 and 4,
d ( z m , z l ) ≤ d ( z m , x m ) + d ( x m , x l ) + d ( x l , z l ) ≤ d ( x m , y m ) + d ( x m , x l ) + d ( x l , y l ) . d(z_{m},z_{l})\le d(z_{m},x_{m})+d(x_{m},x_{l})+d(x_{l},z_{l})\le d(x_{m},y_{m})+d(x_{m},x_{l})+d(x_{l},y_{l}). d ( z m , z l ) ≤ d ( z m , x m ) + d ( x m , x l ) + d ( x l , z l ) ≤ d ( x m , y m ) + d ( x m , x l ) + d ( x l , y l ) .
Since δ ( x , y ) = 0 \delta(x,y)=0 δ ( x , y ) = 0 , the real sequence ( d ( x m , y m ) ) m (d(x_{m},y_{m}))_{m} ( d ( x m , y m ) ) m converges to 0 0 0 (Cauchy Sequences in a Metric Space: Constant Sequences, Convergence of Termwise Distances, the Null Relation, and Independence of Representatives §distances ), and by clause 3 of Properties of the Absolute Value in an Ordered Field d ( x m , y m ) ≤ ∣ d ( x m , y m ) − 0 ∣ d(x_{m},y_{m})\le|d(x_{m},y_{m})-0| d ( x m , y m ) ≤ ∣ d ( x m , y m ) − 0∣ . Given ε > 0 \varepsilon>0 ε > 0 , first choose N 1 N_{1} N 1 with d ( x m , y m ) < ε / 4 d(x_{m},y_{m})<\varepsilon/4 d ( x m , y m ) < ε /4 for m ≥ N 1 m\ge N_{1} m ≥ N 1 , then N 2 N_{2} N 2 with d ( x m , x l ) < ε / 2 d(x_{m},x_{l})<\varepsilon/2 d ( x m , x l ) < ε /2 for m , l ≥ N 2 m,l\ge N_{2} m , l ≥ N 2 , and let N N N be the larger; for m , l ≥ N m,l\ge N m , l ≥ N the display gives d ( z m , z l ) < ε d(z_{m},z_{l})<\varepsilon d ( z m , z l ) < ε . So z ∈ C ( X ) z\in\mathcal{C}(X) z ∈ C ( X ) and f ( z m ) → L ( z ) f(z_{m})\to L(z) f ( z m ) → L ( z ) by Step 1.
We show L ( x ) = L ( z ) L(x)=L(z) L ( x ) = L ( z ) . Let ε > 0 \varepsilon>0 ε > 0 . First choose K 1 K_{1} K 1 with e ( f ( x k ) , L ( x ) ) < ε / 2 e(f(x_{k}),L(x))<\varepsilon/2 e ( f ( x k ) , L ( x )) < ε /2 for k ≥ K 1 k\ge K_{1} k ≥ K 1 , then K 2 K_{2} K 2 with e ( f ( z k ) , L ( z ) ) < ε / 2 e(f(z_{k}),L(z))<\varepsilon/2 e ( f ( z k ) , L ( z )) < ε /2 for k ≥ K 2 k\ge K_{2} k ≥ K 2 , let K K K be the larger, and by ( † ) (\dagger) ( † ) choose m ≥ K m\ge K m ≥ K with τ ( m ) = 0 \tau(m)=0 τ ( m ) = 0 , so that z m = x m z_{m}=x_{m} z m = x m . Then e ( L ( x ) , L ( z ) ) ≤ e ( L ( x ) , f ( x m ) ) + e ( f ( z m ) , L ( z ) ) < ε e(L(x),L(z))\le e(L(x),f(x_{m}))+e(f(z_{m}),L(z))<\varepsilon e ( L ( x ) , L ( z )) ≤ e ( L ( x ) , f ( x m )) + e ( f ( z m ) , L ( z )) < ε . As ε > 0 \varepsilon>0 ε > 0 was arbitrary and 0 ≤ e ( L ( x ) , L ( z ) ) 0\le e(L(x),L(z)) 0 ≤ e ( L ( x ) , L ( z )) , Comparison of Real Numbers with Arbitrary Positive Slack §vanishing gives e ( L ( x ) , L ( z ) ) = 0 e(L(x),L(z))=0 e ( L ( x ) , L ( z )) = 0 , so L ( x ) = L ( z ) L(x)=L(z) L ( x ) = L ( z ) by condition 2. The same argument with an index m ≥ K m\ge K m ≥ K satisfying τ ( m ) = 1 \tau(m)=1 τ ( m ) = 1 , for which z m = y m z_{m}=y_{m} z m = y m , gives L ( y ) = L ( z ) L(y)=L(z) L ( y ) = L ( z ) . Hence L ( x ) = L ( y ) L(x)=L(y) L ( x ) = L ( y ) .
Step 3: existence and uniqueness of f ^ \widehat{f} f . Let f ^ = { ( ξ , L ( x ) ) : x ∈ C ( X ) , ξ = [ x ] } ⊆ X ^ × Y \widehat{f}=\{(\xi,L(x)):\ x\in\mathcal{C}(X),\ \xi=[x]\}\subseteq\widehat{X}\times Y f = {( ξ , L ( x )) : x ∈ C ( X ) , ξ = [ x ]} ⊆ X × Y . Every ξ ∈ X ^ \xi\in\widehat{X} ξ ∈ X occurs as a first coordinate by (vi), and if [ x ] = [ y ] [x]=[y] [ x ] = [ y ] then x ≈ y x\approx y x ≈ y by Cauchy Sequences in a Metric Space: Constant Sequences, Convergence of Termwise Distances, the Null Relation, and Independence of Representatives §null , so L ( x ) = L ( y ) L(x)=L(y) L ( x ) = L ( y ) by Step 2. Thus f ^ \widehat{f} f is a map X ^ → Y \widehat{X}\to Y X → Y (no choice is involved) and f ^ ( [ x ] ) = L ( x ) = lim k f ( x k ) \widehat{f}([x])=L(x)=\lim_{k}f(x_{k}) f ([ x ]) = L ( x ) = lim k f ( x k ) for every x ∈ C ( X ) x\in\mathcal{C}(X) x ∈ C ( X ) . If g : X ^ → Y g:\widehat{X}\to Y g : X → Y also satisfies this formula, then for ξ = [ x ] ∈ X ^ \xi=[x]\in\widehat{X} ξ = [ x ] ∈ X we get g ( ξ ) = L ( x ) = f ^ ( ξ ) g(\xi)=L(x)=\widehat{f}(\xi) g ( ξ ) = L ( x ) = f ( ξ ) ; so f ^ \widehat{f} f is the only such map.
Step 4: f ^ ∘ κ X = f \widehat{f}\circ\kappa_{X}=f f ∘ κ X = f . For a ∈ X a\in X a ∈ X , (vi) gives f ^ ( κ X ( a ) ) = f ^ ( [ a ˉ ] ) = L ( a ˉ ) \widehat{f}(\kappa_{X}(a))=\widehat{f}([\bar{a}])=L(\bar{a}) f ( κ X ( a )) = f ([ a ˉ ]) = L ( a ˉ ) , the limit of the constant sequence with value f ( a ) f(a) f ( a ) , which is f ( a ) f(a) f ( a ) by (iii) and (ii).
Step 5: continuity. By (vii) it suffices to show that ξ m → ξ \xi_{m}\to\xi ξ m → ξ in ( X ^ , d ^ ) (\widehat{X},\widehat{d}) ( X , d ) implies f ^ ( ξ m ) → f ^ ( ξ ) \widehat{f}(\xi_{m})\to\widehat{f}(\xi) f ( ξ m ) → f ( ξ ) in ( Y , e ) (Y,e) ( Y , e ) . For m ∈ N m\in\mathbb{N} m ∈ N let
A m = { a ∈ X : d ^ ( κ X ( a ) , ξ m ) < r m and e ( f ( a ) , f ^ ( ξ m ) ) < r m } . A_{m}=\{a\in X:\ \widehat{d}(\kappa_{X}(a),\xi_{m})<r_{m}\ \text{and}\ e(f(a),\widehat{f}(\xi_{m}))<r_{m}\}. A m = { a ∈ X : d ( κ X ( a ) , ξ m ) < r m and e ( f ( a ) , f ( ξ m )) < r m } .
Each A m A_{m} A m is nonempty: writing ξ m = [ x ] \xi_{m}=[x] ξ m = [ x ] by (vi), clause 3 gives κ X ( x k ) → ξ m \kappa_{X}(x_{k})\to\xi_{m} κ X ( x k ) → ξ m and Step 3 gives f ( x k ) → f ^ ( ξ m ) f(x_{k})\to\widehat{f}(\xi_{m}) f ( x k ) → f ( ξ m ) ; choose K 1 K_{1} K 1 with d ^ ( κ X ( x k ) , ξ m ) < r m \widehat{d}(\kappa_{X}(x_{k}),\xi_{m})<r_{m} d ( κ X ( x k ) , ξ m ) < r m for k ≥ K 1 k\ge K_{1} k ≥ K 1 , then K 2 K_{2} K 2 with e ( f ( x k ) , f ^ ( ξ m ) ) < r m e(f(x_{k}),\widehat{f}(\xi_{m}))<r_{m} e ( f ( x k ) , f ( ξ m )) < r m for k ≥ K 2 k\ge K_{2} k ≥ K 2 , and let k k k be the larger; then x k ∈ A m x_{k}\in A_{m} x k ∈ A m . By the axiom of countable choice there is a sequence a = ( a m ) m ∈ N a=(a_{m})_{m\in\mathbb{N}} a = ( a m ) m ∈ N in X X X with a m ∈ A m a_{m}\in A_{m} a m ∈ A m for every m m m .
First, κ X ( a m ) → ξ \kappa_{X}(a_{m})\to\xi κ X ( a m ) → ξ : for every m m m , d ^ ( κ X ( a m ) , ξ ) ≤ d ^ ( κ X ( a m ) , ξ m ) + d ^ ( ξ m , ξ ) < r m + d ^ ( ξ m , ξ ) \widehat{d}(\kappa_{X}(a_{m}),\xi)\le\widehat{d}(\kappa_{X}(a_{m}),\xi_{m})+\widehat{d}(\xi_{m},\xi)<r_{m}+\widehat{d}(\xi_{m},\xi) d ( κ X ( a m ) , ξ ) ≤ d ( κ X ( a m ) , ξ m ) + d ( ξ m , ξ ) < r m + d ( ξ m , ξ ) ; given ε > 0 \varepsilon>0 ε > 0 choose M 1 M_{1} M 1 with r m < ε / 2 r_{m}<\varepsilon/2 r m < ε /2 for m ≥ M 1 m\ge M_{1} m ≥ M 1 by (v), then M 2 M_{2} M 2 with d ^ ( ξ m , ξ ) < ε / 2 \widehat{d}(\xi_{m},\xi)<\varepsilon/2 d ( ξ m , ξ ) < ε /2 for m ≥ M 2 m\ge M_{2} m ≥ M 2 , and take m m m at least the larger. Second, a ∈ C ( X ) a\in\mathcal{C}(X) a ∈ C ( X ) : by clause 2, d ( a m , a l ) = d ^ ( κ X ( a m ) , κ X ( a l ) ) ≤ d ^ ( κ X ( a m ) , ξ ) + d ^ ( ξ , κ X ( a l ) ) d(a_{m},a_{l})=\widehat{d}(\kappa_{X}(a_{m}),\kappa_{X}(a_{l}))\le\widehat{d}(\kappa_{X}(a_{m}),\xi)+\widehat{d}(\xi,\kappa_{X}(a_{l})) d ( a m , a l ) = d ( κ X ( a m ) , κ X ( a l )) ≤ d ( κ X ( a m ) , ξ ) + d ( ξ , κ X ( a l )) ; given ε > 0 \varepsilon>0 ε > 0 choose M M M with d ^ ( κ X ( a m ) , ξ ) < ε / 2 \widehat{d}(\kappa_{X}(a_{m}),\xi)<\varepsilon/2 d ( κ X ( a m ) , ξ ) < ε /2 for m ≥ M m\ge M m ≥ M , so d ( a m , a l ) < ε d(a_{m},a_{l})<\varepsilon d ( a m , a l ) < ε for m , l ≥ M m,l\ge M m , l ≥ M . Third, by clause 3, κ X ( a m ) → [ a ] \kappa_{X}(a_{m})\to[a] κ X ( a m ) → [ a ] , so [ a ] = ξ [a]=\xi [ a ] = ξ by Uniqueness of Limits in a Metric Space ; hence f ( a m ) → L ( a ) = f ^ ( [ a ] ) = f ^ ( ξ ) f(a_{m})\to L(a)=\widehat{f}([a])=\widehat{f}(\xi) f ( a m ) → L ( a ) = f ([ a ]) = f ( ξ ) by Steps 1 and 3. Finally, for every m m m ,
e ( f ^ ( ξ m ) , f ^ ( ξ ) ) ≤ e ( f ^ ( ξ m ) , f ( a m ) ) + e ( f ( a m ) , f ^ ( ξ ) ) < r m + e ( f ( a m ) , f ^ ( ξ ) ) . e(\widehat{f}(\xi_{m}),\widehat{f}(\xi))\le e(\widehat{f}(\xi_{m}),f(a_{m}))+e(f(a_{m}),\widehat{f}(\xi))<r_{m}+e(f(a_{m}),\widehat{f}(\xi)). e ( f ( ξ m ) , f ( ξ )) ≤ e ( f ( ξ m ) , f ( a m )) + e ( f ( a m ) , f ( ξ )) < r m + e ( f ( a m ) , f ( ξ )) .
Given ε > 0 \varepsilon>0 ε > 0 choose M 1 M_{1} M 1 with r m < ε / 2 r_{m}<\varepsilon/2 r m < ε /2 for m ≥ M 1 m\ge M_{1} m ≥ M 1 , then M 2 M_{2} M 2 with e ( f ( a m ) , f ^ ( ξ ) ) < ε / 2 e(f(a_{m}),\widehat{f}(\xi))<\varepsilon/2 e ( f ( a m ) , f ( ξ )) < ε /2 for m ≥ M 2 m\ge M_{2} m ≥ M 2 ; for m m m at least the larger, e ( f ^ ( ξ m ) , f ^ ( ξ ) ) < ε e(\widehat{f}(\xi_{m}),\widehat{f}(\xi))<\varepsilon e ( f ( ξ m ) , f ( ξ )) < ε . So f ^ ( ξ m ) → f ^ ( ξ ) \widehat{f}(\xi_{m})\to\widehat{f}(\xi) f ( ξ m ) → f ( ξ ) , and f ^ \widehat{f} f is continuous.
Step 6: the Lipschitz property. Suppose f f f is Lipschitz with constant Λ ≥ 0 \Lambda\ge0 Λ ≥ 0 . Let α , β ∈ X ^ \alpha,\beta\in\widehat{X} α , β ∈ X and write α = [ x ] \alpha=[x] α = [ x ] , β = [ y ] \beta=[y] β = [ y ] by (vi). By Step 3, f ( x k ) → f ^ ( α ) f(x_{k})\to\widehat{f}(\alpha) f ( x k ) → f ( α ) and f ( y k ) → f ^ ( β ) f(y_{k})\to\widehat{f}(\beta) f ( y k ) → f ( β ) , so by (iv) the real sequence ( e ( f ( x k ) , f ( y k ) ) ) k (e(f(x_{k}),f(y_{k})))_{k} ( e ( f ( x k ) , f ( y k )) ) k converges to e ( f ^ ( α ) , f ^ ( β ) ) e(\widehat{f}(\alpha),\widehat{f}(\beta)) e ( f ( α ) , f ( β )) . By Cauchy Sequences in a Metric Space: Constant Sequences, Convergence of Termwise Distances, the Null Relation, and Independence of Representatives §distances , ( d ( x k , y k ) ) k (d(x_{k},y_{k}))_{k} ( d ( x k , y k ) ) k converges to δ ( x , y ) = d ^ ( α , β ) \delta(x,y)=\widehat{d}(\alpha,\beta) δ ( x , y ) = d ( α , β ) , so ( Λ d ( x k , y k ) ) k (\Lambda\,d(x_{k},y_{k}))_{k} ( Λ d ( x k , y k ) ) k converges to Λ d ^ ( α , β ) \Lambda\,\widehat{d}(\alpha,\beta) Λ d ( α , β ) by clause 3 of Arithmetic of Limits of Real Sequences . Since e ( f ( x k ) , f ( y k ) ) ≤ Λ d ( x k , y k ) e(f(x_{k}),f(y_{k}))\le\Lambda\,d(x_{k},y_{k}) e ( f ( x k ) , f ( y k )) ≤ Λ d ( x k , y k ) for every k k k , clause 1 (comparison) of Order Properties of Limits of Real Sequences gives e ( f ^ ( α ) , f ^ ( β ) ) ≤ Λ d ^ ( α , β ) e(\widehat{f}(\alpha),\widehat{f}(\beta))\le\Lambda\,\widehat{d}(\alpha,\beta) e ( f ( α ) , f ( β )) ≤ Λ d ( α , β ) . Thus f ^ \widehat{f} f is Lipschitz with constant Λ \Lambda Λ .
Proof of clause 6 (Uniqueness of continuous extensions). Let ξ ∈ X ^ \xi\in\widehat{X} ξ ∈ X and write ξ = [ x ] \xi=[x] ξ = [ x ] by (vi). By clause 3, κ X ( x k ) → ξ \kappa_{X}(x_{k})\to\xi κ X ( x k ) → ξ , so by (vii) g ( κ X ( x k ) ) → g ( ξ ) g(\kappa_{X}(x_{k}))\to g(\xi) g ( κ X ( x k )) → g ( ξ ) and h ( κ X ( x k ) ) → h ( ξ ) h(\kappa_{X}(x_{k}))\to h(\xi) h ( κ X ( x k )) → h ( ξ ) in ( Y , e ) (Y,e) ( Y , e ) . Since g ( κ X ( x k ) ) = h ( κ X ( x k ) ) g(\kappa_{X}(x_{k}))=h(\kappa_{X}(x_{k})) g ( κ X ( x k )) = h ( κ X ( x k )) for every k k k , one sequence converges to both g ( ξ ) g(\xi) g ( ξ ) and h ( ξ ) h(\xi) h ( ξ ) , and Uniqueness of Limits in a Metric Space gives g ( ξ ) = h ( ξ ) g(\xi)=h(\xi) g ( ξ ) = h ( ξ ) .
Proof of clause 7 (Inequalities). Let ξ ∈ X ^ \xi\in\widehat{X} ξ ∈ X and write ξ = [ x ] \xi=[x] ξ = [ x ] by (vi). By clause 3, κ X ( x k ) → ξ \kappa_{X}(x_{k})\to\xi κ X ( x k ) → ξ , so by (vii), applied with ( Z , ρ ) (Z,\rho) ( Z , ρ ) the real line of The Absolute Value Metric on the Real Line , g ( κ X ( x k ) ) → g ( ξ ) g(\kappa_{X}(x_{k}))\to g(\xi) g ( κ X ( x k )) → g ( ξ ) and h ( κ X ( x k ) ) → h ( ξ ) h(\kappa_{X}(x_{k}))\to h(\xi) h ( κ X ( x k )) → h ( ξ ) in the real line, hence in the sense of Limit of a Sequence of Real Numbers by (ii). The hypothesis with a = x k a=x_{k} a = x k gives g ( κ X ( x k ) ) ≤ h ( κ X ( x k ) ) g(\kappa_{X}(x_{k}))\le h(\kappa_{X}(x_{k})) g ( κ X ( x k )) ≤ h ( κ X ( x k )) for every k k k , so clause 1 (comparison) of Order Properties of Limits of Real Sequences gives g ( ξ ) ≤ h ( ξ ) g(\xi)\le h(\xi) g ( ξ ) ≤ h ( ξ ) . ■ \blacksquare ■