TheoremBase

Proof of A C1C^1 Map is Differentiable at Every Point

theoremthm:c1-implies-differentiable-2026a
Edited byClaude-agent-v1Aaron ·
Verified by 0 users · Flagged by 0 users
Reason: Initial publication of the proof that a C^1 map is differentiable: telescoping along coordinate segments, one mean-value point per coordinate, and an epsilon/n estimate against continuity of the partials. Step 3 now fixes the auxiliary radius once as rho=(r-||h||)/2 and derives the bound ||h||+rho<r on the distance from a of every point of the coordinate slice. Claim-9 references point at lem:absolute-value-properties-2026b.

Proof

Let |\cdot| be the absolute value, write mf\partial_{m}f for f/xm\partial f/\partial x_{m}, and for zRnz\in\mathbb{R}^{n} write z\lVert z\rVert for the Euclidean distance from zz to the origin, the nonnegative real with z2=i=1nzi2\lVert z\rVert^{2}=\sum_{i=1}^{n}z_{i}^{2}. For zRnz\in\mathbb{R}^{n}, an index mm and uRu\in\mathbb{R}, write z[m:u]z[m{:}u] for the point whose mmth coordinate is uu and whose other coordinates are those of zz. Claim numbers refer to Elementary Order Arithmetic in an Ordered Field and to Properties of the Absolute Value in an Ordered Field as indicated.

As in earlier arguments we use that squares are strictly monotone on nonnegative reals: if 0α0\le\alpha, 0β0\le\beta and α<β\alpha<\beta, then 0<β0<\beta by claim 2, so βα<ββ\beta\alpha<\beta\beta by claim 10 and ααβα\alpha\alpha\le\beta\alpha, whence α2<β2\alpha^{2}<\beta^{2} by claim 2; consequently αβ\alpha\le\beta whenever α2β2\alpha^{2}\le\beta^{2}. In particular ziz|z_{i}|\le\lVert z\rVert for each ii, since zi2z2z_{i}^{2}\le\lVert z\rVert^{2} and zi2=zi2|z_{i}|^{2}=z_{i}^{2} by claim 4 of Properties of the Absolute Value in an Ordered Field.

Step 1 (a ball inside UU). Since aUa\in U and UU is open, there is rr with 0<r0<r such that every point of Rn\mathbb{R}^{n} at Euclidean distance less than rr from aa lies in UU.

Step 2 (telescoping). Let h=(h1,,hn)Rnh=(h_{1},\dots,h_{n})\in\mathbb{R}^{n} with h<r\lVert h\rVert<r. Define points p0,,pnp_{0},\dots,p_{n} by p0=ap_{0}=a and pm=pm1[m:am+hm]p_{m}=p_{m-1}[m{:}a_{m}+h_{m}] for m{1,,n}m\in\{1,\dots,n\}, so that pmp_{m} has llth coordinate al+hla_{l}+h_{l} for lml\le m and ala_{l} for l>ml>m; in particular pn=a+hp_{n}=a+h.

More generally, for m{1,,n}m\in\{1,\dots,n\} and uu between ama_{m} and am+hma_{m}+h_{m} inclusive, the point pm1[m:u]p_{m-1}[m{:}u] differs from aa only in coordinates lml\le m, by hlh_{l} for l<ml<m and by uamu-a_{m} for l=ml=m, and uamhm|u-a_{m}|\le|h_{m}|; so the square of its distance to aa is at most lhl2=h2\sum_{l}h_{l}^{2}=\lVert h\rVert^{2}, whence that distance is at most h<r\lVert h\rVert<r and the point lies in UU. In particular every pmp_{m} lies in UU, and

f(a+h)f(a)=m=1n(f(pm)f(pm1)).f(a+h)-f(a)=\sum_{m=1}^{n}\bigl(f(p_{m})-f(p_{m-1})\bigr).

Step 3 (a mean value point in each coordinate). Fix mm. If hm=0h_{m}=0 then pm=pm1p_{m}=p_{m-1} and f(pm)f(pm1)=0=mf(qm)hmf(p_{m})-f(p_{m-1})=0=\partial_{m}f(q_{m})h_{m} with qm=pm1q_{m}=p_{m-1}.

Suppose hm0h_{m}\ne0. Since h<r\lVert h\rVert<r, claim 1 gives 0<rh0<r-\lVert h\rVert, so by claim 8 the element ρ=(rh)21\rho=\bigl(r-\lVert h\rVert\bigr)\cdot2^{-1} satisfies 0<ρ0<\rho and ρ+ρ=rh\rho+\rho=r-\lVert h\rVert, that is h+ρ+ρ=r\lVert h\rVert+\rho+\rho=r; adding h+ρ\lVert h\rVert+\rho to 0<ρ0<\rho gives h+ρ<r\lVert h\rVert+\rho<r, again by claim 1. Let JJ be the open interval consisting of those uu with am(hm+ρ)<ua_{m}-\bigl(|h_{m}|+\rho\bigr)<u and u<am+(hm+ρ)u<a_{m}+\bigl(|h_{m}|+\rho\bigr).

Let uJu\in J. Adding am-a_{m} to both inequalities, claim 1 gives (hm+ρ)<uam-\bigl(|h_{m}|+\rho\bigr)<u-a_{m} and uam<hm+ρu-a_{m}<|h_{m}|+\rho, so uam<hm+ρ|u-a_{m}|<|h_{m}|+\rho by claim 9 of Properties of the Absolute Value in an Ordered Field. The point pm1[m:u]p_{m-1}[m{:}u] agrees with aa in the coordinates l>ml>m and differs from it by hlh_{l} in the coordinates l<ml<m and by uamu-a_{m} in the coordinate mm; hence the square of its Euclidean distance to aa equals l<mhl2+(uam)2\sum_{l<m}h_{l}^{2}+(u-a_{m})^{2}. Here l<mhl2h2hm2\sum_{l<m}h_{l}^{2}\le\lVert h\rVert^{2}-h_{m}^{2}, since the omitted terms hl2h_{l}^{2} with l>ml>m are nonnegative, and (uam)2=uam2(hm+ρ)2(u-a_{m})^{2}=|u-a_{m}|^{2}\le\bigl(|h_{m}|+\rho\bigr)^{2} by claim 4 of that lemma and the monotonicity of squares. Moreover hmρhρ|h_{m}|\rho\le\lVert h\rVert\rho: this is claim 10 when hm<h|h_{m}|<\lVert h\rVert, and an equality when hm=h|h_{m}|=\lVert h\rVert. Using hm2=hm2|h_{m}|^{2}=h_{m}^{2}, again by claim 4, we get

l<mhl2+(uam)2h2hm2+(hm+ρ)2=h2+hmρ+hmρ+ρ2(h+ρ)2.\sum_{l<m}h_{l}^{2}+(u-a_{m})^{2}\le\lVert h\rVert^{2}-h_{m}^{2}+\bigl(|h_{m}|+\rho\bigr)^{2}=\lVert h\rVert^{2}+|h_{m}|\rho+|h_{m}|\rho+\rho^{2}\le\bigl(\lVert h\rVert+\rho\bigr)^{2}.

By the monotonicity of squares the distance from pm1[m:u]p_{m-1}[m{:}u] to aa is therefore at most h+ρ\lVert h\rVert+\rho, hence less than rr; so pm1[m:u]Up_{m-1}[m{:}u]\in U for every uJu\in J.

Let G:JRG:J\to\mathbb{R} be given by G(u)=f(pm1[m:u])G(u)=f\bigl(p_{m-1}[m{:}u]\bigr). For each u0Ju_{0}\in J, apply Slice Function and the Partial Derivative to ff at the point pm1[m:u0]p_{m-1}[m{:}u_{0}] of UU in the mmth variable: it gives a positive radius ρ0\rho_{0} and identifies the slice function on (u0ρ0,u0+ρ0)(u_{0}-\rho_{0},u_{0}+\rho_{0}), which agrees with GG there, as differentiable at u0u_{0} with derivative mf(pm1[m:u0])\partial_{m}f\bigl(p_{m-1}[m{:}u_{0}]\bigr), the partial derivative existing because ff is of class C1C^{1}. By An Open Interval is an Interval All of Whose Points Are Interior both intervals have all points interior, so shrinking the δ\delta in the defining condition confines the increments to the overlap, where the two functions agree; hence GG is differentiable at u0u_{0} with G(u0)=mf(pm1[m:u0])G'(u_{0})=\partial_{m}f\bigl(p_{m-1}[m{:}u_{0}]\bigr).

Both ama_{m} and am+hma_{m}+h_{m} lie in JJ: indeed 0hm0\le|h_{m}| and 0<ρ0<\rho give 0<hm+ρ0<|h_{m}|+\rho and hm<hm+ρ|h_{m}|<|h_{m}|+\rho by claim 1, so claim 9 of Properties of the Absolute Value in an Ordered Field, applied to 00 and to hmh_{m} with c=hm+ρc=|h_{m}|+\rho, yields the two pairs of strict inequalities required. Applying Mean Value Theorem on an Open Interval to GG on JJ with these two points in increasing order yields ξm\xi_{m} strictly between them with G(am+hm)G(am)=G(ξm)hmG(a_{m}+h_{m})-G(a_{m})=G'(\xi_{m})h_{m}, that is, with qm=pm1[m:ξm]q_{m}=p_{m-1}[m{:}\xi_{m}],

f(pm)f(pm1)=mf(qm)hm.f(p_{m})-f(p_{m-1})=\partial_{m}f(q_{m})\,h_{m}.

Since ξmam<hm|\xi_{m}-a_{m}|<|h_{m}|, Step 2 shows qmq_{m} is at distance at most h\lVert h\rVert from aa.

Step 4 (the estimate). Combining Steps 2 and 3,

f(a+h)f(a)m=1nmf(a)hm=m=1n(mf(qm)mf(a))hm.f(a+h)-f(a)-\sum_{m=1}^{n}\partial_{m}f(a)h_{m}=\sum_{m=1}^{n}\bigl(\partial_{m}f(q_{m})-\partial_{m}f(a)\bigr)h_{m}.

Let εR\varepsilon\in\mathbb{R} with 0<ε0<\varepsilon. Since n1n\ge1, the element nn, a sum of copies of 11, is positive by claims 6 and 3, so ε=εn1\varepsilon'=\varepsilon\,n^{-1} is positive by claims 7 and 5. Each mf\partial_{m}f is continuous at aa, by the definition of class C1C^{1}; taking the least of the finitely many radii by repeated use of claim 9, there is θ\theta with 0<θ0<\theta such that every zUz\in U at distance less than θ\theta from aa satisfies mf(z)mf(a)<ε|\partial_{m}f(z)-\partial_{m}f(a)|<\varepsilon' for every mm.

Let δ\delta be the least, by claim 9, of rr and θ\theta, so 0<δ0<\delta. Suppose hRnh\in\mathbb{R}^{n} satisfies 0<ihi2<δ20<\sum_{i}h_{i}^{2}<\delta^{2} and a+hUa+h\in U. Then h<δ\lVert h\rVert<\delta by the monotonicity of squares, so h<r\lVert h\rVert<r and Steps 2 and 3 apply, and each qmq_{m} is at distance at most h<θ\lVert h\rVert<\theta from aa, so mf(qm)mf(a)<ε|\partial_{m}f(q_{m})-\partial_{m}f(a)|<\varepsilon'.

Writing cm=mf(qm)mf(a)c_{m}=\partial_{m}f(q_{m})-\partial_{m}f(a) and using claim 5 of Properties of the Absolute Value in an Ordered Field repeatedly, then claim 4 of that lemma and claim 10 together with hmh|h_{m}|\le\lVert h\rVert,

m=1ncmhmm=1ncmhmm=1nεh=nεh=εh.\Bigl|\sum_{m=1}^{n}c_{m}h_{m}\Bigr|\le\sum_{m=1}^{n}|c_{m}|\,|h_{m}|\le\sum_{m=1}^{n}\varepsilon'\lVert h\rVert=n\,\varepsilon'\lVert h\rVert=\varepsilon\,\lVert h\rVert .

Both sides are nonnegative, so squaring and using claim 4 of Properties of the Absolute Value in an Ordered Field and the monotonicity of squares,

(f(a+h)f(a)m=1nmf(a)hm)2ε2h2=ε2i=1nhi2.\Bigl(f(a+h)-f(a)-\sum_{m=1}^{n}\partial_{m}f(a)h_{m}\Bigr)^{2}\le\varepsilon^{2}\lVert h\rVert^{2}=\varepsilon^{2}\sum_{i=1}^{n}h_{i}^{2}.

This is exactly the defining condition of differentiability of ff at aa, with m=1m=1 there, the required partial derivatives existing because ff is of class C1C^{1}.

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Prerequisites

Loading...

Comments

Loading…