TheoremBase

Proof

We prove both parts simultaneously by induction on kk.

Base case (k=1k=1). Fix x∈Ux\in U and i∈{1,…,n}i\in\{1,\dots,n\}.

(i) By the definition of the partial derivative, ∂(fg)∂xi(x)\dfrac{\partial(fg)}{\partial x_i}(x) is the limit as h→0h\to 0 of f(x+hei) g(x+hei)−f(x) g(x)h\dfrac{f(x+he_i)\,g(x+he_i)-f(x)\,g(x)}{h}. We factor the numerator as

f(x+hei) g(x+hei)−f(x) g(x)=[f(x+hei)−f(x)]g(x+hei)+f(x)[g(x+hei)−g(x)].f(x+he_i)\,g(x+he_i)-f(x)\,g(x) =\bigl[f(x+he_i)-f(x)\bigr]g(x+he_i)+f(x)\bigl[g(x+he_i)-g(x)\bigr].

Since ff and gg are C1C^1, both partial derivatives ∂f∂xi(x)\dfrac{\partial f}{\partial x_i}(x) and ∂g∂xi(x)\dfrac{\partial g}{\partial x_i}(x) exist and gg is continuous. Dividing by hh and taking h→0h\to 0: the first term tends to ∂f∂xi(x)⋅g(x)\dfrac{\partial f}{\partial x_i}(x)\cdot g(x) (since g(x+hei)→g(x)g(x+he_i)\to g(x) by continuity of gg), and the second term tends to f(x)⋅∂g∂xi(x)f(x)\cdot\dfrac{\partial g}{\partial x_i}(x). Therefore

∂(fg)∂xi(x)=∂f∂xi(x) g(x)+f(x) ∂g∂xi(x).\frac{\partial(fg)}{\partial x_i}(x)=\frac{\partial f}{\partial x_i}(x)\,g(x)+f(x)\,\frac{\partial g}{\partial x_i}(x).

The right-hand side is a sum of products of continuous functions (namely ∂f∂xi\dfrac{\partial f}{\partial x_i}, gg, ff, ∂g∂xi\dfrac{\partial g}{\partial x_i}, all continuous by C1C^1), hence continuous by continuity of sums and products of continuous real-valued functions. As ii was arbitrary, fgfg is C1C^1, proving (i).

(ii) With g(x)≠0g(x)\ne 0 for all x∈Ux\in U, the difference quotient for f/gf/g at xx in direction eie_i equals

f(x+hei)g(x+hei)−f(x)g(x)h=[f(x+hei)−f(x)]g(x)−f(x)[g(x+hei)−g(x)]h⋅g(x+hei)⋅g(x).\frac{\dfrac{f(x+he_i)}{g(x+he_i)}-\dfrac{f(x)}{g(x)}}{h} =\frac{\bigl[f(x+he_i)-f(x)\bigr]g(x)-f(x)\bigl[g(x+he_i)-g(x)\bigr]}{h\cdot g(x+he_i)\cdot g(x)}.

Dividing the numerator by hh and taking h→0h\to 0: the two fractions tend to ∂f∂xi(x)\dfrac{\partial f}{\partial x_i}(x) and ∂g∂xi(x)\dfrac{\partial g}{\partial x_i}(x) respectively, and g(x+hei)→g(x)≠0g(x+he_i)\to g(x)\ne 0 by continuity. Therefore

∂(f/g)∂xi(x)=∂f∂xi(x) g(x)−f(x) ∂g∂xi(x)g(x)2.\frac{\partial(f/g)}{\partial x_i}(x)=\frac{\dfrac{\partial f}{\partial x_i}(x)\,g(x)-f(x)\,\dfrac{\partial g}{\partial x_i}(x)}{g(x)^2}.

The numerator and denominator g2g^2 are continuous (since gg, ∂f∂xi\dfrac{\partial f}{\partial x_i}, ∂g∂xi\dfrac{\partial g}{\partial x_i} are continuous and g2≠0g^2\ne 0), so ∂(f/g)∂xi\dfrac{\partial(f/g)}{\partial x_i} is continuous by continuity of sums and products. As ii was arbitrary, f/gf/g is C1C^1, proving (ii).

Inductive step. Assume both (i) and (ii) hold for k−1k-1 in place of kk; let f,g:U→Rf,g:U\to\mathbb{R} be CkC^k with k≥2k\ge 2.

(i) For each i∈{1,…,n}i\in\{1,\dots,n\}, the base-case computation (which is a pointwise limit argument valid for any C1C^1 functions) gives ∂(fg)∂xi=∂f∂xi⋅g+f⋅∂g∂xi\dfrac{\partial(fg)}{\partial x_i}=\dfrac{\partial f}{\partial x_i}\cdot g+f\cdot\dfrac{\partial g}{\partial x_i}. Since ff and gg are CkC^k, the partial derivatives ∂f∂xi\dfrac{\partial f}{\partial x_i} and ∂g∂xi\dfrac{\partial g}{\partial x_i} are Ck−1C^{k-1}, and ff, gg themselves are Ck−1C^{k-1}. By the inductive hypothesis (i), both products ∂f∂xi⋅g\dfrac{\partial f}{\partial x_i}\cdot g and f⋅∂g∂xif\cdot\dfrac{\partial g}{\partial x_i} are Ck−1C^{k-1}; their sum is also Ck−1C^{k-1}. As ii was arbitrary, fgfg is CkC^k.

(ii) Similarly, the base-case formula ∂(f/g)∂xi=[∂f∂xi g−f ∂g∂xi]/g2\dfrac{\partial(f/g)}{\partial x_i}=\bigl[\dfrac{\partial f}{\partial x_i}\,g-f\,\dfrac{\partial g}{\partial x_i}\bigr]\big/g^2 holds. The numerator is Ck−1C^{k-1} by (i) applied inductively, and g2g^2 is Ck−1C^{k-1} by (i) inductively, with g2>0g^2>0 on UU. By the inductive hypothesis (ii), the quotient ∂(f/g)∂xi\dfrac{\partial(f/g)}{\partial x_i} is Ck−1C^{k-1}. As ii was arbitrary, f/gf/g is CkC^k.

By induction, both (i) and (ii) hold for all natural numbers k≥1k\ge 1.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…