· 11,923 chars · 25 deps · depth 12 Reason: First publication of the proof: explicit Lipschitz estimates for the radial map and its inverse, the extreme value theorem for images of compact sets, and a two-case neighbourhood argument for the extended function.
Explicit estimates: the norm is nonexpansive, the radial map is Lipschitz with constant two with an explicit inverse, the extreme value theorem bounds the image of a compact set away from the unit sphere, and the extended function vanishes near the sphere.
Proof
Each result cited below is universally quantified over the data in its own statement, and is applied to the data named where it is used. Throughout, the vector operations of Rm and their identities are those of Euclidean Space Rn is a Real Vector Space, and the claims of Elementary Properties of the Euclidean Norm on Rn are used for the Euclidean norm: claim 2 for dE(x,y)=∥x−y∥, claim 3 for the vanishing of the norm, claim 5 for ∥λx∥=∣λ∣∥x∥ and claim 6 for the triangle inequality.
Inverse. Let x∈Rm, put y=h(x) and s=∥y∥=ax∥x∥. Then
1−s=(1+∥x∥−∥x∥)ax=ax,
so (1−s)−1=1+∥x∥ and g(y)=(1+∥x∥)axx=x. Thus g(h(x))=x for every x∈Rm, so h is injective. Conversely let y∈U, put x=g(y) and t=∥x∥=(1−∥y∥)−1∥y∥, the factor (1−∥y∥)−1 being positive. Then
1+t=(1−∥y∥+∥y∥)(1−∥y∥)−1=(1−∥y∥)−1,
so ax=1−∥y∥ and h(x)=(1−∥y∥)(1−∥y∥)−1y=y. Hence h maps Rm onto U and g is its inverse.
h is Lipschitz with constant 2. Let x,y∈Rm and write a=ax, b=ay. Then h(x)−h(y)=a(x−y)+(a−b)y, and
a−b=((1+∥y∥)−(1+∥x∥))ab=ab(∥y∥−∥x∥).
Since a and b are positive, ∣a−b∣=ab∥y∥−∥x∥. Now 0<a≤1, and b∥y∥=∥h(y)∥≤1 by the first paragraph of the present claim; hence, using claim 1,
g is continuous on U. Let y0∈U, put s0=∥y0∥<1, c=2−1(1−s0)>0 and δ0=c. Let y∈U satisfy dE(y,y0)<δ0. By claim 1, ∥y∥≤s0+∥y−y0∥<s0+c, so 1−∥y∥>1−s0−c=c and therefore β:=(1−∥y∥)−1<c−1. Put β0=(1−s0)−1. As above,
β−β0=((1−s0)−(1−∥y∥))ββ0=ββ0(∥y∥−s0),
so ∣β−β0∣≤c−1β0∥y−y0∥ by claim 1. Since g(y)−g(y0)=β(y−y0)+(β−β0)y0 and ∥y0∥≤1, the triangle inequality and homogeneity give
Given a positive real ε, choose δ to be the least of δ0 and εL−1, which is positive. Every y∈U with dU(y,y0)=dE(y,y0)<δ then satisfies dE(g(y),g(y0))≤L∥y−y0∥<LεL−1=ε. By Continuous Map Between Metric Spaces, g is continuous at y0 relative to U, and as y0∈U was arbitrary, g is continuous on U as a map from (U,dU) into (Rm,dE).
By Minimum of Two Elements of a Totally Ordered Set and Elementary Properties of the Minimum of Two Elements, max{0,ψ(y)}≥0 and χ(y)≤1, while χ(y) is one of the two numbers 1 and max{0,ψ(y)}, both nonnegative; so 0≤χ(y)≤1. If ∥y∥≤ρ then σ−∥y∥≥σ−ρ>0, so ψ(y)≥1 on multiplying by the positive number (σ−ρ)−1, whence max{0,ψ(y)}=ψ(y)≥1 and χ(y)=1. If σ≤∥y∥ then σ−∥y∥≤0, so ψ(y)≤0, whence max{0,ψ(y)}=0 and χ(y)=min{1,0}=0.
Claim 6.Vanishing near the unit sphere. Let y∈Bˉ satisfy σ≤∥y∥. If y∈U then F(y)=χ(y)f(g(y))=0 because χ(y)=0 by claim 5; if y∈/U then F(y)=0 by definition.
Continuity. Let y0∈Bˉ and let ε be a positive real number.
Suppose first that y0∈U, and put δ1=1−∥y0∥>0. If y∈Bˉ satisfies dE(y,y0)<δ1, then ∥y∥≤∥y0∥+∥y−y0∥<1 by claim 1, so y∈U. On U the map y↦χ(y)f(g(y)) is continuous relative to U: the map g is continuous on U as a map into (Rm,dE) by claim 3 and f is continuous on Rm, so f∘g is continuous on U by claim 3 of Semicontinuity and Continuity Under Composition with a Continuous Map; the restriction of χ to U is continuous on U by claim 4 of that lemma; and the pointwise product of two maps continuous on U is continuous on U by claim 5 of Continuity of Sums and Products of Real-Valued Functions on a Metric Space. Hence there is a positive real δ2 such that every y∈U with dE(y,y0)<δ2 satisfies ∣χ(y)f(g(y))−χ(y0)f(g(y0))∣<ε. Let δ be the least of δ1 and δ2, which is positive. Every y∈Bˉ with dBˉ(y,y0)=dE(y,y0)<δ then lies in U, so F(y)=χ(y)f(g(y)) and F(y0)=χ(y0)f(g(y0)), and ∣F(y)−F(y0)∣<ε.
Suppose next that y0∈Bˉ∖U, so that ∥y0∥=1, since ∥y0∥≤1 and not ∥y0∥<1. Then F(y0)=0. Put δ=1−σ>0. If y∈Bˉ satisfies dBˉ(y,y0)=dE(y,y0)<δ, then by claim 1
∥y∥≥∥y0∥−∥y0−y∥>1−(1−σ)=σ,
so F(y)=0 by the vanishing established above, and ∣F(y)−F(y0)∣=0<ε.
In both cases F is continuous at y0 relative to Bˉ in the sense of Continuous Map Between Metric Spaces, read for the metric space (Bˉ,dBˉ); as y0∈Bˉ was arbitrary, F is continuous on Bˉ.