TheoremBase

Proof

Claim 1. Fix k∈{1,…,m}k\in\{1,\dots,m\} and j∈{1,…,n}j\in\{1,\dots,n\}, and let eje_j denote the multi-index of length nn whose jjth entry is 11 and whose remaining entries are 00. By Smooth Map on an Open Subset of Euclidean Space, the partial derivative of GkG_k of order α\alpha exists on WW for every multi-index α\alpha of length nn; apply this with α=ej\alpha=e_j. Since ej≠0e_j\ne 0, and since jj is the only index ii with ei≤eje_i\le e_j, the recursive clause defining partial derivatives of order α\alpha, which Smooth Map on an Open Subset of Euclidean Space invokes, requires at α=ej\alpha=e_j exactly that the ordinary partial derivative of GkG_k with respect to the jjth variable exist at every point of WW: that is, for every a∈Wa\in W there is a real number LL such that for every real ε>0\varepsilon>0 there is a real δ>0\delta>0 for which every real hh with 0<∣h∣<δ0<|h|<\delta satisfies

∣Gk(a1,…,aj−1,aj+h,aj+1,…,an)−Gk(a)h−L∣<ε.\left|\frac{G_k(a_1,\dots,a_{j-1},a_j+h,a_{j+1},\dots,a_n)-G_k(a)}{h}-L\right|<\varepsilon .

This is the condition of Partial Derivative on a Euclidean Open Set except for the additional requirement there that the perturbed point lie in WW. That requirement is met after shrinking δ\delta: since WW is open and a∈Wa\in W, there is a real ρ>0\rho>0 such that every y∈Rny\in\mathbb{R}^n with ∑i=1n(yi−ai)2<ρ2\sum_{i=1}^n(y_i-a_i)^2<\rho^2 lies in WW, and for 0<∣h∣<ρ0<|h|<\rho the point y=(a1,…,aj−1,aj+h,aj+1,…,an)y=(a_1,\dots,a_{j-1},a_j+h,a_{j+1},\dots,a_n) satisfies ∑i=1n(yi−ai)2=h2\sum_{i=1}^n(y_i-a_i)^2=h^2, with h2<ρ2h^2<\rho^2 because multiplying the strict inequality ∣h∣<ρ|h|<\rho first by the positive number ∣h∣|h| and then by the positive number ρ\rho gives h2<∣h∣ρ<ρ2h^2<|h|\rho<\rho^2 (claims 10 and 2 of the order arithmetic lemma, and h2=∣h∣2h^2=|h|^2 by claims 1 and 4 of the absolute value lemma). Replacing δ\delta by the smaller of δ\delta and ρ\rho, which is positive by claim 9 of the order arithmetic lemma, therefore gives the condition of Partial Derivative on a Euclidean Open Set. Hence the partial derivative of GkG_k with respect to the jjth variable exists at every point of WW in that sense.

As kk and jj were arbitrary, the hypothesis of Jacobian Matrix of a Map Between Euclidean Spaces is satisfied at every a∈Wa\in W, so DG(a)DG(a) is defined there. This proves claim 1.

Claim 2. Throughout write x=(x1,…,xn)x=(x_1,\dots,x_n), and let DD be the ambient set of Ω\Omega in the sense of Continuous n-Form, Support, and Zero Extension on a Euclidean or Half-Space Domain.

Step 1: a common radius. Since WW is open and x∈Wx\in W, there is a real r1>0r_1>0 such that every y∈Rny\in\mathbb{R}^n with ∑i=1n(yi−xi)2<r12\sum_{i=1}^n(y_i-x_i)^2<r_1^2 lies in WW; similarly there is a real r2>0r_2>0 doing the same for W~\widetilde{W}. Next, if D=RnD=\mathbb{R}^n then Ω\Omega is open in Rn\mathbb{R}^n, and we let U=ΩU=\Omega; if D=HnD=H^n then, by Continuous n-Form, Support, and Zero Extension on a Euclidean or Half-Space Domain, Ω\Omega is open in the closed upper half-space HnH^n, so by that definition there is an open subset U⊆RnU\subseteq\mathbb{R}^n with Ω=Hn∩U\Omega=H^n\cap U. In both cases x∈Ux\in U and UU is open, so there is a real r3>0r_3>0 such that every y∈Rny\in\mathbb{R}^n with ∑i=1n(yi−xi)2<r32\sum_{i=1}^n(y_i-x_i)^2<r_3^2 lies in UU. Let rr be the least of r1,r2,r3r_1,r_2,r_3; then r>0r>0 by claim 9 of the order arithmetic lemma, applied twice.

Step 2: nearby points of the positive jjth coordinate direction lie in W∩W~∩ΩW\cap\widetilde{W}\cap\Omega. Fix j∈{1,…,n}j\in\{1,\dots,n\} and let tt be a real number with 0<t<r0<t<r. Put

y=(x1,…,xj−1,xj+t,xj+1,…,xn).y=(x_1,\dots,x_{j-1},x_j+t,x_{j+1},\dots,x_n).

Then ∑i=1n(yi−xi)2=t2\sum_{i=1}^n(y_i-x_i)^2=t^2. Let uu be any one of r1,r2,r3r_1,r_2,r_3. Since r≤ur\le u and t<rt<r, mixed transitivity gives t<ut<u, and multiplying the strict inequality t<ut<u first by the positive number tt and then by the positive number uu gives t2<tu<u2t^2<tu<u^2, whence t2<u2t^2<u^2; here we used claims 10 and 2 of the order arithmetic lemma. Taking u=r1u=r_1, u=r2u=r_2 and u=r3u=r_3 in turn gives y∈Wy\in W, y∈W~y\in\widetilde{W} and y∈Uy\in U.

If D=RnD=\mathbb{R}^n then U=ΩU=\Omega and so y∈Ωy\in\Omega. If D=HnD=H^n, then x∈Ω⊆Hnx\in\Omega\subseteq H^n gives xn≥0x_n\ge 0, and the nnth coordinate of yy equals xnx_n when j<nj<n and equals xn+tx_n+t when j=nj=n; in the second case xn+t>0x_n+t>0 by claim 3 of the order arithmetic lemma applied to 0≤xn0\le x_n and 0<t0<t. In either case the nnth coordinate of yy is nonnegative, so y∈Hny\in H^n and therefore y∈Hn∩U=Ωy\in H^n\cap U=\Omega.

Thus y∈W∩W~∩Ωy\in W\cap\widetilde{W}\cap\Omega, and the hypothesis gives G(y)=G~(y)G(y)=\widetilde{G}(y). The point xx itself lies in W∩W~∩ΩW\cap\widetilde{W}\cap\Omega, so also G(x)=G~(x)G(x)=\widetilde{G}(x).

Step 3: the partial derivatives at xx agree. Fix k∈{1,…,m}k\in\{1,\dots,m\} and j∈{1,…,n}j\in\{1,\dots,n\}. By claim 1 the partial derivatives

L=∂Gk∂xj(x),L~=∂G~k∂xj(x)L=\frac{\partial G_k}{\partial x_j}(x),\qquad \widetilde{L}=\frac{\partial \widetilde{G}_k}{\partial x_j}(x)

exist in the sense of Partial Derivative on a Euclidean Open Set. Suppose, for contradiction, that L≠L~L\ne\widetilde{L}, and set ε=∣L−L~∣/2\varepsilon=|L-\widetilde{L}|/2, which is positive by claim 1 of the absolute value lemma together with claim 8 of the order arithmetic lemma. Choose δ>0\delta>0 for LL and δ~>0\widetilde{\delta}>0 for L~\widetilde{L} as in Partial Derivative on a Euclidean Open Set for this ε\varepsilon, let σ\sigma be the least of δ\delta, δ~\widetilde{\delta} and rr, which is positive, and choose a real tt with 0<t<σ0<t<\sigma, for instance t=σ/2t=\sigma/2 (claim 8 of the order arithmetic lemma).

With yy as in Step 2 for this tt, set

q=Gk(y)−Gk(x)t.q=\frac{G_k(y)-G_k(x)}{t}.

By Step 2, G(y)=G~(y)G(y)=\widetilde{G}(y) and G(x)=G~(x)G(x)=\widetilde{G}(x), so their kkth coordinates agree and therefore qq also equals (G~k(y)−G~k(x))/t(\widetilde{G}_k(y)-\widetilde{G}_k(x))/t. Since 0<t<δ0<t<\delta and 0<t<δ~0<t<\widetilde{\delta}, the defining condition of Partial Derivative on a Euclidean Open Set gives ∣q−L∣<ε|q-L|<\varepsilon and ∣q−L~∣<ε|q-\widetilde{L}|<\varepsilon. By symmetry of the absolute value and the triangle inequality (claims 2 and 5 of the absolute value lemma),

∣L−L~∣≤∣L−q∣+∣q−L~∣<ε+ε=∣L−L~∣,|L-\widetilde{L}|\le |L-q|+|q-\widetilde{L}|<\varepsilon+\varepsilon=|L-\widetilde{L}|,

which is impossible. Hence L=L~L=\widetilde{L}.

Conclusion. By Step 3 every entry of DG(x)DG(x) equals the corresponding entry of DG~(x)D\widetilde{G}(x), since by Jacobian Matrix of a Map Between Euclidean Spaces the entry in row kk and column jj of each is the partial derivative of the respective kkth coordinate function with respect to the jjth variable at xx. Two real matrices of the same shape with equal entries are equal, so DG(x)=DG~(x)DG(x)=D\widetilde{G}(x). This proves claim 2.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…