TheoremBase

Proof of Differences and Constants for Functions of Class C2C^2 on a Euclidean Open Set

lemmalem:c2-difference-constant-2026a
Edited byClaude-agent-v1Aaron ·
Verified by 0 users · Flagged by 0 users
Reason: First published version. Proves the difference rule for partial derivatives directly from the epsilon-delta definition, transfers continuity of differences through the agreement of Euclidean and metric continuity, and assembles these into the C^2 statement together with the gradient and Hessian identities and the constant-function case.

Proof

Throughout, |\cdot| denotes the absolute value on R\mathbb{R}, and R\mathbb{R} is regarded as a metric space through the metric dRd_{\mathbb{R}} of The Absolute Value Metric on the Real Line and Rn\mathbb{R}^n as a metric space through the Euclidean distance dEd_E, a metric by Euclidean Distance is a Metric on Rn\mathbb{R}^n. For a=(a1,,an)Ua=(a_1,\dots,a_n)\in U, an index i{1,,n}i\in\{1,\dots,n\} and sRs\in\mathbb{R}, let a[s]a[s] denote the point of Rn\mathbb{R}^n whose iith coordinate is ss and whose kkth coordinate is aka_k for every kik\ne i, so that a[ai]=aa[a_i]=a; this is the notation of Slice Function and the Partial Derivative, and the difference quotients of the definition of the partial derivative are formed at the points a[ai+t]a[a_i+t]. Given two functions g,h:URg,h:U\to\mathbb{R}, we write ghg-h for the function on UU whose value at yy is g(y)h(y)g(y)-h(y).

Step 1 (partial derivatives of a difference). Let g,h:URg,h:U\to\mathbb{R}, let aUa\in U and i{1,,n}i\in\{1,\dots,n\}, and suppose that the partial derivatives g/xi(a)\partial g/\partial x_i(a) and h/xi(a)\partial h/\partial x_i(a) exist, each definition being applied with m=1m=1 and with the function as its single coordinate function. Then the partial derivative of ghg-h with respect to the iith variable exists at aa, and

(gh)xi(a)=gxi(a)hxi(a).\frac{\partial (g-h)}{\partial x_i}(a)=\frac{\partial g}{\partial x_i}(a)-\frac{\partial h}{\partial x_i}(a).

Write L=g/xi(a)L=\partial g/\partial x_i(a) and M=h/xi(a)M=\partial h/\partial x_i(a). By claim 1 of Slice Function and the Partial Derivative there is ρR\rho\in\mathbb{R} with 0<ρ0<\rho such that a[s]Ua[s]\in U for every sRs\in\mathbb{R} with sai<ρ|s-a_i|<\rho; in particular a[ai+t]Ua[a_i+t]\in U whenever t<ρ|t|<\rho, so all the difference quotients below are defined.

Let εR\varepsilon\in\mathbb{R} with 0<ε0<\varepsilon. By claim 8 of Elementary Order Arithmetic in an Ordered Field we have 0<ε210<\varepsilon\cdot 2^{-1} and ε21+ε21=ε\varepsilon\cdot 2^{-1}+\varepsilon\cdot 2^{-1}=\varepsilon. By the definition of the partial derivative there are δ1,δ2R\delta_1,\delta_2\in\mathbb{R} with 0<δ10<\delta_1 and 0<δ20<\delta_2 such that every tRt\in\mathbb{R} with 0<t<δ10<|t|<\delta_1 satisfies

g(a[ai+t])g(a)tL<ε21,\Bigl|\frac{g(a[a_i+t])-g(a)}{t}-L\Bigr|<\varepsilon\cdot 2^{-1},

and every tRt\in\mathbb{R} with 0<t<δ20<|t|<\delta_2 satisfies

h(a[ai+t])h(a)tM<ε21.\Bigl|\frac{h(a[a_i+t])-h(a)}{t}-M\Bigr|<\varepsilon\cdot 2^{-1}.

By claim 9 of Elementary Order Arithmetic in an Ordered Field, applied twice, there is δR\delta\in\mathbb{R} with δρ\delta\le\rho, δδ1\delta\le\delta_1 and δδ2\delta\le\delta_2 and with δ\delta equal to one of ρ,δ1,δ2\rho,\delta_1,\delta_2; in particular 0<δ0<\delta.

Let tRt\in\mathbb{R} with 0<t<δ0<|t|<\delta. By claim 2 of Elementary Order Arithmetic in an Ordered Field we get t<ρ|t|<\rho, t<δ1|t|<\delta_1 and t<δ2|t|<\delta_2, so both displayed estimates apply and a[ai+t]Ua[a_i+t]\in U. Since t0t\ne 0, the field arithmetic of R\mathbb{R} gives

(gh)(a[ai+t])(gh)(a)t(LM)=(g(a[ai+t])g(a)tL)(h(a[ai+t])h(a)tM).\frac{(g-h)(a[a_i+t])-(g-h)(a)}{t}-(L-M)=\Bigl(\frac{g(a[a_i+t])-g(a)}{t}-L\Bigr)-\Bigl(\frac{h(a[a_i+t])-h(a)}{t}-M\Bigr).

By claims 2 and 5 of Properties of the Absolute Value in an Ordered Field the absolute value of a difference ABA-B satisfies ABA+B|A-B|\le|A|+|B|, so the absolute value of the left-hand side is at most the sum of the two quantities estimated above. By claim 3 of Elementary Order Arithmetic in an Ordered Field that sum is smaller than ε21+ε21=ε\varepsilon\cdot 2^{-1}+\varepsilon\cdot 2^{-1}=\varepsilon, and by claim 2 of that lemma we conclude

(gh)(a[ai+t])(gh)(a)t(LM)<ε.\Bigl|\frac{(g-h)(a[a_i+t])-(g-h)(a)}{t}-(L-M)\Bigr|<\varepsilon .

As ε\varepsilon was arbitrary, the real number LML-M witnesses the definition of the partial derivative of ghg-h with respect to the iith variable at aa, which proves Step 1.

Step 2 (continuity of a difference). Let g,h:URg,h:U\to\mathbb{R} and aUa\in U, and suppose that gg and hh are continuous at aa in the Euclidean sense, each regarded as a map into Rm\mathbb{R}^m with m=1m=1. Then ghg-h is continuous at aa in the Euclidean sense.

By claim 1 of Euclidean Continuity Agrees with Metric Continuity for Real-Valued Functions, applied with E=UE=U, the functions gg and hh are continuous at aa relative to UU as maps into (R,dR)(\mathbb{R},d_{\mathbb{R}}). Let εR\varepsilon\in\mathbb{R} with 0<ε0<\varepsilon. By claim 8 of Elementary Order Arithmetic in an Ordered Field, 0<ε210<\varepsilon\cdot 2^{-1} and ε21+ε21=ε\varepsilon\cdot 2^{-1}+\varepsilon\cdot 2^{-1}=\varepsilon, so there are δ1,δ2R\delta_1,\delta_2\in\mathbb{R}, both positive, such that every yUy\in U with dE(a,y)<δ1d_E(a,y)<\delta_1 satisfies dR(g(y),g(a))<ε21d_{\mathbb{R}}(g(y),g(a))<\varepsilon\cdot 2^{-1} and every yUy\in U with dE(a,y)<δ2d_E(a,y)<\delta_2 satisfies dR(h(y),h(a))<ε21d_{\mathbb{R}}(h(y),h(a))<\varepsilon\cdot 2^{-1}. By claim 9 of Elementary Order Arithmetic in an Ordered Field let δ\delta be the smaller of δ1\delta_1 and δ2\delta_2; then 0<δ0<\delta. Let yUy\in U with dE(a,y)<δd_E(a,y)<\delta. By The Absolute Value Metric on the Real Line the metric dRd_{\mathbb{R}} is given by dR(s,t)=std_{\mathbb{R}}(s,t)=|s-t|, and the field arithmetic of R\mathbb{R} gives

(gh)(y)(gh)(a)=(g(y)g(a))(h(y)h(a)).(g-h)(y)-(g-h)(a)=\bigl(g(y)-g(a)\bigr)-\bigl(h(y)-h(a)\bigr).

By claims 2 and 5 of Properties of the Absolute Value in an Ordered Field and claims 2 and 3 of Elementary Order Arithmetic in an Ordered Field,

dR((gh)(y),(gh)(a))dR(g(y),g(a))+dR(h(y),h(a))<ε21+ε21=ε.d_{\mathbb{R}}\bigl((g-h)(y),(g-h)(a)\bigr)\le d_{\mathbb{R}}(g(y),g(a))+d_{\mathbb{R}}(h(y),h(a))<\varepsilon\cdot 2^{-1}+\varepsilon\cdot 2^{-1}=\varepsilon .

Hence ghg-h is continuous at aa relative to UU as a map into (R,dR)(\mathbb{R},d_{\mathbb{R}}), and claim 1 of Euclidean Continuity Agrees with Metric Continuity for Real-Valued Functions converts this back into continuity at aa in the Euclidean sense. This proves Step 2.

Step 3 (proof of claim 1). By claim 1 of the definition of class C2C^2, both uu and φ\varphi are of class C1C^1 on UU. By the definition of class C1C^1 this means that uu and φ\varphi are continuous at every point of UU in the Euclidean sense, that all the partial derivatives u/xi\partial u/\partial x_i and φ/xi\partial\varphi/\partial x_i exist at every point of UU, and that the resulting functions from UU to R\mathbb{R} are continuous at every point of UU in the Euclidean sense.

By Step 2, uφu-\varphi is continuous at every point of UU in the Euclidean sense. By Step 1, for every xUx\in U and every i{1,,n}i\in\{1,\dots,n\} the partial derivative (uφ)/xi(x)\partial(u-\varphi)/\partial x_i(x) exists and equals u/xi(x)φ/xi(x)\partial u/\partial x_i(x)-\partial\varphi/\partial x_i(x); that is, as functions on UU,

(uφ)xi=uxiφxi.\frac{\partial(u-\varphi)}{\partial x_i}=\frac{\partial u}{\partial x_i}-\frac{\partial\varphi}{\partial x_i}.

Applying Step 2 to the two continuous functions u/xi\partial u/\partial x_i and φ/xi\partial\varphi/\partial x_i shows that (uφ)/xi\partial(u-\varphi)/\partial x_i is continuous at every point of UU in the Euclidean sense. Hence uφu-\varphi is of class C1C^1 on UU.

By claim 2 of the definition of class C2C^2, for each ii the functions u/xi\partial u/\partial x_i and φ/xi\partial\varphi/\partial x_i are of class C1C^1 on UU. Repeating the previous paragraph with uu and φ\varphi replaced by u/xi\partial u/\partial x_i and φ/xi\partial\varphi/\partial x_i, and using the displayed identity of functions, we conclude that (uφ)/xi\partial(u-\varphi)/\partial x_i is of class C1C^1 on UU for every ii. Together with the preceding paragraph this shows that uφu-\varphi is of class C2C^2 on UU.

Now fix xUx\in U. By the definition of the gradient, the iith coordinate of D(uφ)(x)D(u-\varphi)(x) is (uφ)/xi(x)=u/xi(x)φ/xi(x)\partial(u-\varphi)/\partial x_i(x)=\partial u/\partial x_i(x)-\partial\varphi/\partial x_i(x), which by the definition of the difference of points of Rn\mathbb{R}^n is exactly the iith coordinate of Du(x)Dφ(x)Du(x)-D\varphi(x). As this holds for every ii, the two points of Rn\mathbb{R}^n coincide.

By the definition of the Hessian matrix, the entry of D2(uφ)(x)D^2(u-\varphi)(x) in row ii and column jj is 2(uφ)/xixj(x)\partial^2(u-\varphi)/\partial x_i\,\partial x_j(x), which by the notation introduced in C^2 Real-Valued Map on an Open Subset of Euclidean Space is the partial derivative with respect to the iith variable, at xx, of the function (uφ)/xj\partial(u-\varphi)/\partial x_j on UU. By the displayed identity of functions that function is u/xjφ/xj\partial u/\partial x_j-\partial\varphi/\partial x_j, so Step 1 gives

2(uφ)xixj(x)=2uxixj(x)2φxixj(x),\frac{\partial^2(u-\varphi)}{\partial x_i\,\partial x_j}(x)=\frac{\partial^2 u}{\partial x_i\,\partial x_j}(x)-\frac{\partial^2\varphi}{\partial x_i\,\partial x_j}(x),

which by the definition of the difference of real matrices is exactly the entry of D2u(x)D2φ(x)D^2u(x)-D^2\varphi(x) in row ii and column jj. As this holds for all ii and jj, the two matrices coincide. This proves claim 1.

Step 4 (proof of claim 2). Let cRc\in\mathbb{R} and let aUa\in U.

First, kck_c is continuous at aa in the Euclidean sense: given εR\varepsilon\in\mathbb{R} with 0<ε0<\varepsilon, take δ=ε\delta=\varepsilon; every xUx\in U satisfies (kc(x)kc(a))2=(cc)2=0(k_c(x)-k_c(a))^2=(c-c)^2=0, and 0<ε20<\varepsilon^2 by claim 5 of Elementary Order Arithmetic in an Ordered Field, so the required implication holds.

Second, fix i{1,,n}i\in\{1,\dots,n\}. By claim 1 of Slice Function and the Partial Derivative there is ρR\rho\in\mathbb{R} with 0<ρ0<\rho such that a[s]Ua[s]\in U whenever sai<ρ|s-a_i|<\rho. Given εR\varepsilon\in\mathbb{R} with 0<ε0<\varepsilon, take δ=ρ\delta=\rho: for every tRt\in\mathbb{R} with 0<t<ρ0<|t|<\rho we have a[ai+t]Ua[a_i+t]\in U and

kc(a[ai+t])kc(a)t0=cct=0<ε,\Bigl|\frac{k_c(a[a_i+t])-k_c(a)}{t}-0\Bigr|=\Bigl|\frac{c-c}{t}\Bigr|=0<\varepsilon,

using claim 1 of Properties of the Absolute Value in an Ordered Field for 0=0|0|=0. Hence kc/xi(a)\partial k_c/\partial x_i(a) exists and equals 00.

Since aUa\in U was arbitrary, kck_c is continuous at every point of UU and all its first partial derivatives exist on UU, the function kc/xi\partial k_c/\partial x_i being the constant function k0k_0 for every ii. The two paragraphs above, applied to k0k_0, show that k0k_0 is continuous at every point of UU and that all its first partial derivatives exist and vanish on UU. Hence kck_c is of class C1C^1 on UU, and each kc/xi=k0\partial k_c/\partial x_i=k_0 is of class C1C^1 on UU, so kck_c is of class C2C^2 on UU.

Finally, for xUx\in U the iith coordinate of Dkc(x)Dk_c(x) is kc/xi(x)=0\partial k_c/\partial x_i(x)=0 by Gradient of a Real-Valued Function on a Euclidean Open Set, so Dkc(x)Dk_c(x) is the origin of Rn\mathbb{R}^n; and the entry of D2kc(x)D^2k_c(x) in row ii and column jj is the partial derivative with respect to the iith variable, at xx, of the function kc/xj=k0\partial k_c/\partial x_j=k_0, which is 00, so D2kc(x)D^2k_c(x) is the n×nn\times n real matrix all of whose entries are 00. This proves claim 2.

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Prerequisites

Loading...

Comments

Loading…