Note first that σ 1 = 1 \sigma_{1}=1 σ 1 = 1 , since σ 1 \sigma_{1} σ 1 is positive with σ 1 2 = 1 \sigma_{1}^{2}=1 σ 1 2 = 1 by its defining clause . Hence the Lipschitz image bound reads, in dimension 1 1 1 ,
λ ∗ ( ψ ( S ) ) ≤ 2 M λ ∗ ( S ) \lambda^{\ast}\bigl(\psi(S)\bigr)\le 2M\,\lambda^{\ast}(S) λ ∗ ( ψ ( S ) ) ≤ 2 M λ ∗ ( S )
for a map ψ \psi ψ Lipschitz with constant M > 0 M>0 M > 0 on a nonempty domain and S S S a subset of that domain.
Claim 1. We may assume A ≠ ∅ A\ne\varnothing A = ∅ , the case A = ∅ A=\varnothing A = ∅ being trivial. Let ε ∈ R \varepsilon\in\mathbb{R} ε ∈ R with 0 < ε < 1 0<\varepsilon<1 0 < ε < 1 .
By outer regularity of the outer measure there is an open set O ′ O' O ′ with A ⊆ O ′ A\subseteq O' A ⊆ O ′ and λ ( O ′ ) ≤ λ ∗ ( A ) + ε \lambda(O')\le\lambda^{\ast}(A)+\varepsilon λ ( O ′ ) ≤ λ ∗ ( A ) + ε ; put O = O ′ ∩ I O=O'\cap I O = O ′ ∩ I , which is open, being a finite intersection of open sets (The Open Subsets of a Metric Space Form a Topology ), and contains A A A . Both O O O and O ′ O' O ′ are Borel, so by agreement on Borel sets and monotonicity , λ ( O ) = λ ∗ ( O ) ≤ λ ∗ ( O ′ ) = λ ( O ′ ) ≤ λ ∗ ( A ) + ε \lambda(O)=\lambda^{\ast}(O)\le\lambda^{\ast}(O')=\lambda(O')\le\lambda^{\ast}(A)+\varepsilon λ ( O ) = λ ∗ ( O ) ≤ λ ∗ ( O ′ ) = λ ( O ′ ) ≤ λ ∗ ( A ) + ε .
Let x ∈ A x\in A x ∈ A . Since ϕ \phi ϕ has the derivative ϕ ′ ( x ) \phi'(x) ϕ ′ ( x ) at x x x , there is a real δ x > 0 \delta_{x}>0 δ x > 0 such that
∣ ϕ ( t ) − ϕ ( x ) t − x − ϕ ′ ( x ) ∣ ≤ ε whenever t ∈ I , 0 < ∣ t − x ∣ < δ x , \Bigl|\frac{\phi(t)-\phi(x)}{t-x}-\phi'(x)\Bigr|\le\varepsilon\qquad\text{whenever }t\in I,\ 0<|t-x|<\delta_{x}, t − x ϕ ( t ) − ϕ ( x ) − ϕ ′ ( x ) ≤ ε whenever t ∈ I , 0 < ∣ t − x ∣ < δ x ,
whence ∣ ϕ ( t ) − ϕ ( x ) ∣ ≤ ( ∣ ϕ ′ ( x ) ∣ + ε ) ∣ t − x ∣ ≤ ( K + ε ) ∣ t − x ∣ |\phi(t)-\phi(x)|\le(|\phi'(x)|+\varepsilon)|t-x|\le(K+\varepsilon)|t-x| ∣ ϕ ( t ) − ϕ ( x ) ∣ ≤ ( ∣ ϕ ′ ( x ) ∣ + ε ) ∣ t − x ∣ ≤ ( K + ε ) ∣ t − x ∣ for those t t t , and trivially also for t = x t=x t = x .
Let F \mathcal{F} F be the set of pairs ( x , h ) (x,h) ( x , h ) with x ∈ A x\in A x ∈ A , 0 < h < δ x 0<h<\delta_{x} 0 < h < δ x and B ˉ ( x , h ) ⊆ O \bar{B}(x,h)\subseteq O B ˉ ( x , h ) ⊆ O . It finely covers A A A : given x ∈ A x\in A x ∈ A and a real η > 0 \eta>0 η > 0 , x x x is an interior point of the open set O O O , so claim 1 of Interior Points in the Metric Topology are Exactly the Centres of Contained Closed Balls supplies a real ρ > 0 \rho>0 ρ > 0 with B ˉ ( x , ρ ) ⊆ O \bar{B}(x,\rho)\subseteq O B ˉ ( x , ρ ) ⊆ O , and any h h h with 0 < h < min { η , δ x , ρ } 0<h<\min\{\eta,\delta_{x},\rho\} 0 < h < min { η , δ x , ρ } gives a pair of F \mathcal{F} F whose ball contains x x x and whose radius is less than η \eta η .
For ( x , h ) ∈ F (x,h)\in\mathcal{F} ( x , h ) ∈ F and t ∈ B ˉ ( x , h ) t\in\bar{B}(x,h) t ∈ B ˉ ( x , h ) we have ∣ t − x ∣ ≤ h < δ x |t-x|\le h<\delta_{x} ∣ t − x ∣ ≤ h < δ x , so ∣ ϕ ( t ) − ϕ ( x ) ∣ ≤ ( K + ε ) h |\phi(t)-\phi(x)|\le(K+\varepsilon)h ∣ ϕ ( t ) − ϕ ( x ) ∣ ≤ ( K + ε ) h . Hence ϕ ( B ˉ ( x , h ) ) \phi(\bar{B}(x,h)) ϕ ( B ˉ ( x , h )) is contained in the closed interval with endpoints ϕ ( x ) ± ( K + ε ) h \phi(x)\pm(K+\varepsilon)h ϕ ( x ) ± ( K + ε ) h , and therefore, by monotonicity, agreement on Borel sets, and claim 4 of Lebesgue Measure on the Real Line: Existence, Lengths of Intervals and Uniqueness ,
λ ∗ ( ϕ ( B ˉ ( x , h ) ) ) ≤ 2 ( K + ε ) h = ( K + ε ) λ ( B ˉ ( x , h ) ) . ( ∗ ) \lambda^{\ast}\bigl(\phi(\bar{B}(x,h))\bigr)\le 2(K+\varepsilon)h=(K+\varepsilon)\,\lambda\bigl(\bar{B}(x,h)\bigr). \tag{$\ast$} λ ∗ ( ϕ ( B ˉ ( x , h )) ) ≤ 2 ( K + ε ) h = ( K + ε ) λ ( B ˉ ( x , h ) ) . ( ∗ )
Since λ ∗ ( A ) < ∞ \lambda^{\ast}(A)<\infty λ ∗ ( A ) < ∞ , the Vitali covering theorem supplies pairs ( x 1 , h 1 ) , … , ( x N , h N ) ∈ F (x_{1},h_{1}),\dots,(x_{N},h_{N})\in\mathcal{F} ( x 1 , h 1 ) , … , ( x N , h N ) ∈ F whose closed balls are pairwise disjoint and satisfy λ ∗ ( A ∖ ⋃ i B ˉ ( x i , h i ) ) ≤ ε \lambda^{\ast}(A\setminus\bigcup_{i}\bar{B}(x_{i},h_{i}))\le\varepsilon λ ∗ ( A ∖ ⋃ i B ˉ ( x i , h i )) ≤ ε . Write A ′ = A ∩ ⋃ i B ˉ ( x i , h i ) A'=A\cap\bigcup_{i}\bar{B}(x_{i},h_{i}) A ′ = A ∩ ⋃ i B ˉ ( x i , h i ) and A ′ ′ = A ∖ ⋃ i B ˉ ( x i , h i ) A''=A\setminus\bigcup_{i}\bar{B}(x_{i},h_{i}) A ′′ = A ∖ ⋃ i B ˉ ( x i , h i ) , so that ϕ ( A ) = ϕ ( A ′ ) ∪ ϕ ( A ′ ′ ) \phi(A)=\phi(A')\cup\phi(A'') ϕ ( A ) = ϕ ( A ′ ) ∪ ϕ ( A ′′ ) .
The balls B ˉ ( x i , h i ) \bar{B}(x_{i},h_{i}) B ˉ ( x i , h i ) are pairwise disjoint members of B ( R ) \mathcal{B}(\mathbb{R}) B ( R ) contained in O O O , so ∑ i λ ( B ˉ ( x i , h i ) ) ≤ λ ( O ) \sum_{i}\lambda(\bar{B}(x_{i},h_{i}))\le\lambda(O) ∑ i λ ( B ˉ ( x i , h i )) ≤ λ ( O ) by claims 1 and 2 of Basic Properties of a Measure . Using countable subadditivity of λ ∗ \lambda^{\ast} λ ∗ , applied to the sequence whose first N N N terms are the sets ϕ ( B ˉ ( x i , h i ) ) \phi(\bar{B}(x_{i},h_{i})) ϕ ( B ˉ ( x i , h i )) and whose remaining terms are empty, and then ( ∗ ) (\ast) ( ∗ ) ,
λ ∗ ( ϕ ( A ′ ) ) ≤ ∑ i = 1 N λ ∗ ( ϕ ( B ˉ ( x i , h i ) ) ) ≤ ( K + ε ) ∑ i = 1 N λ ( B ˉ ( x i , h i ) ) ≤ ( K + ε ) ( λ ∗ ( A ) + ε ) . \lambda^{\ast}\bigl(\phi(A')\bigr)\le\sum_{i=1}^{N}\lambda^{\ast}\bigl(\phi(\bar{B}(x_{i},h_{i}))\bigr)\le(K+\varepsilon)\sum_{i=1}^{N}\lambda\bigl(\bar{B}(x_{i},h_{i})\bigr)\le(K+\varepsilon)\bigl(\lambda^{\ast}(A)+\varepsilon\bigr). λ ∗ ( ϕ ( A ′ ) ) ≤ i = 1 ∑ N λ ∗ ( ϕ ( B ˉ ( x i , h i )) ) ≤ ( K + ε ) i = 1 ∑ N λ ( B ˉ ( x i , h i ) ) ≤ ( K + ε ) ( λ ∗ ( A ) + ε ) .
By the displayed Lipschitz image bound applied to ϕ \phi ϕ and S = A ′ ′ S=A'' S = A ′′ , together with λ ∗ ( A ′ ′ ) ≤ ε \lambda^{\ast}(A'')\le\varepsilon λ ∗ ( A ′′ ) ≤ ε , we get λ ∗ ( ϕ ( A ′ ′ ) ) ≤ 2 L ε \lambda^{\ast}(\phi(A''))\le 2L\varepsilon λ ∗ ( ϕ ( A ′′ )) ≤ 2 L ε . Adding, and using countable subadditivity once more for the two-set union ϕ ( A ) = ϕ ( A ′ ) ∪ ϕ ( A ′ ′ ) \phi(A)=\phi(A')\cup\phi(A'') ϕ ( A ) = ϕ ( A ′ ) ∪ ϕ ( A ′′ ) ,
λ ∗ ( ϕ ( A ) ) ≤ ( K + ε ) ( λ ∗ ( A ) + ε ) + 2 L ε ≤ K λ ∗ ( A ) + ε c , c = λ ∗ ( A ) + K + 1 + 2 L , \lambda^{\ast}\bigl(\phi(A)\bigr)\le(K+\varepsilon)\bigl(\lambda^{\ast}(A)+\varepsilon\bigr)+2L\varepsilon\le K\lambda^{\ast}(A)+\varepsilon\,c,\qquad c=\lambda^{\ast}(A)+K+1+2L, λ ∗ ( ϕ ( A ) ) ≤ ( K + ε ) ( λ ∗ ( A ) + ε ) + 2 L ε ≤ K λ ∗ ( A ) + ε c , c = λ ∗ ( A ) + K + 1 + 2 L ,
using ε < 1 \varepsilon<1 ε < 1 . The real number c c c is positive and does not depend on ε \varepsilon ε ; taking ε = 1 / 2 \varepsilon=1/2 ε = 1/2 in the display shows in particular that λ ∗ ( ϕ ( A ) ) \lambda^{\ast}(\phi(A)) λ ∗ ( ϕ ( A )) is finite. If λ ∗ ( ϕ ( A ) ) > K λ ∗ ( A ) \lambda^{\ast}(\phi(A))>K\lambda^{\ast}(A) λ ∗ ( ϕ ( A )) > K λ ∗ ( A ) we could choose, by The Archimedean Property of the Real Numbers , a real ε \varepsilon ε with 0 < ε < 1 0<\varepsilon<1 0 < ε < 1 and ε c < λ ∗ ( ϕ ( A ) ) − K λ ∗ ( A ) \varepsilon c<\lambda^{\ast}(\phi(A))-K\lambda^{\ast}(A) ε c < λ ∗ ( ϕ ( A )) − K λ ∗ ( A ) , contradicting the display. Hence λ ∗ ( ϕ ( A ) ) ≤ K λ ∗ ( A ) \lambda^{\ast}(\phi(A))\le K\lambda^{\ast}(A) λ ∗ ( ϕ ( A )) ≤ K λ ∗ ( A ) .
Claim 2. Let x , y ∈ I x,y\in I x , y ∈ I . If x = y x=y x = y the assertion is trivial, so assume x < y x<y x < y after exchanging them if necessary; note ∣ y − x ∣ = y − x |y-x|=y-x ∣ y − x ∣ = y − x . Since I I I is an interval containing x x x and y y y , the closed interval [ x , y ] [x,y] [ x , y ] is contained in I I I .
Put A = [ x , y ] ∖ N A=[x,y]\setminus N A = [ x , y ] ∖ N . Then λ ∗ ( A ) ≤ λ ( [ x , y ] ) = y − x < ∞ \lambda^{\ast}(A)\le\lambda([x,y])=y-x<\infty λ ∗ ( A ) ≤ λ ([ x , y ]) = y − x < ∞ by monotonicity and claim 4 of Lebesgue Measure on the Real Line: Existence, Lengths of Intervals and Uniqueness , and by hypothesis ϕ \phi ϕ has at every point of A A A a derivative bounded in absolute value by K K K . Claim 1 therefore gives λ ∗ ( ϕ ( A ) ) ≤ K ( y − x ) \lambda^{\ast}(\phi(A))\le K(y-x) λ ∗ ( ϕ ( A )) ≤ K ( y − x ) . The set [ x , y ] ∩ N [x,y]\cap N [ x , y ] ∩ N is null, being a subset of the null set N N N , by the null-set claim ; hence ϕ ( [ x , y ] ∩ N ) \phi([x,y]\cap N) ϕ ([ x , y ] ∩ N ) is null by the null-image claim . As ϕ ( [ x , y ] ) = ϕ ( A ) ∪ ϕ ( [ x , y ] ∩ N ) \phi([x,y])=\phi(A)\cup\phi([x,y]\cap N) ϕ ([ x , y ]) = ϕ ( A ) ∪ ϕ ([ x , y ] ∩ N ) , countable subadditivity gives
λ ∗ ( ϕ ( [ x , y ] ) ) ≤ K ( y − x ) . \lambda^{\ast}\bigl(\phi([x,y])\bigr)\le K(y-x). λ ∗ ( ϕ ([ x , y ]) ) ≤ K ( y − x ) .
The function ϕ \phi ϕ is continuous on [ x , y ] [x,y] [ x , y ] , being Lipschitz and hence uniformly continuous by A Lipschitz Map is Uniformly Continuous and Continuous . By Intermediate Value Theorem on a Closed Real Interval, and Continuous Images of Intervals , ϕ \phi ϕ attains on [ x , y ] [x,y] [ x , y ] every value between ϕ ( x ) \phi(x) ϕ ( x ) and ϕ ( y ) \phi(y) ϕ ( y ) ; hence ϕ ( [ x , y ] ) \phi([x,y]) ϕ ([ x , y ]) contains the closed interval J J J with endpoints ϕ ( x ) \phi(x) ϕ ( x ) and ϕ ( y ) \phi(y) ϕ ( y ) , whose measure is ∣ ϕ ( y ) − ϕ ( x ) ∣ |\phi(y)-\phi(x)| ∣ ϕ ( y ) − ϕ ( x ) ∣ by claim 4 of Lebesgue Measure on the Real Line: Existence, Lengths of Intervals and Uniqueness . By monotonicity and agreement on Borel sets,
∣ ϕ ( y ) − ϕ ( x ) ∣ = λ ∗ ( J ) ≤ λ ∗ ( ϕ ( [ x , y ] ) ) ≤ K ( y − x ) = K ∣ y − x ∣ . |\phi(y)-\phi(x)|=\lambda^{\ast}(J)\le\lambda^{\ast}\bigl(\phi([x,y])\bigr)\le K(y-x)=K|y-x| . ∣ ϕ ( y ) − ϕ ( x ) ∣ = λ ∗ ( J ) ≤ λ ∗ ( ϕ ([ x , y ]) ) ≤ K ( y − x ) = K ∣ y − x ∣.