Let (X, ρ) be a non-empty complete metric space and K : X → X be a contrac- tion mapping. That is, there exists a constant C < 1 such that for all x, y ∈ X, ρ(K(x), K(y)) ≤ Cρ(K(x), K(y)). Then it is a classical result that K has a unique fixed-point x. Uniqueness is obvious. To show existence, let x0 be any point in X and consider the sequence {xk } defined inductively by xk = K(xk−1 ). The following corollaries are immediate consequences. Corollary 1. Let D(p, r) be a closed ball in X. If ρ(K(p), p) ≤ (1 − C)r, then K has a fixed-point in D(p, r). Corollary 2. Let B(p, r) be an open ball in X. If ρ(K(p), p) < (1 − C)r, then K has a fixed-point in B(p, r). Theorem 1. Let S be a non-empty topological space. Suppose K : S × X → X is a contraction mapping for which there exists C < 1 such that for all s ∈ S, x, y ∈ X, ρ(K(s, x), K(s, y)) ≤ Cρ(x, y). Further assume that for each x ∈ X, s 7→ K(s, x) is continuous. Let ps be the fixed-point at s. Then the mapping s 7→ ps is continuous. Proof. For all x, t ∈ S, ρ(ps , pt ) = ρ(K(s, ps ), K(t, pt )) ≤ ρ(K(s, ps ), K(t, ps )) + Cρ(ps , pt ) . Rearrange to finish the proof. Corollary 3. Let B(p, r) be an open ball in X and S and K : S × B(p, r) → X be as in Theorem 1. Suppose that for any s ∈ S, ρ(K(s, p), p) < (1 − C)r. Then for each s, there is a unique point ps ∈ B(p, r) such that K(s, ps ) = ps , and the mapping s 7→ ps is continuous.
2. The implicit function theorem
Throughout this section, we denoted Banach spaces by V , W and X. Complete- ness is not needed for every theorem, but we assume it anyway. For a function F : V → W , we denote its derivative at α ∈ V by dFα , provided that it exists. If QN V = i=1 Vi , the i-th partial derivative of F at α is denoted by dFαi . In this case, the sum of the norms on Vi is taken as the norm on V . Suppose that F : V → W and G : V × W → X are differentiable. Define H by H(x) = G(x, F (x)). Then H is differentiable, and dHx = dG1(x,F (x)) + dG2(x,F (x)) ◦ dFx . If F and G are such that dG2F (x) is invertible and G(x, F (x)) = 0 for all x, then we can solve for dFx : −1 dFx = − dG2(x,F (x)) ◦ dG1(x,F (x)) . The implicit function theorem addresses the converse case. 1 2 THE IMPLICIT FUNCTION THEOREM
Theorem 2. Let A × B be an open subset in V × W . Suppose that G : A × B → X
is differentiable at (α, β), G(α, β) = 0 and dG2(α,β) is invertible. Also, suppose that F : A → B is a continuous function such that F (α) = β and G(x, F (x)) = 0 for all x ∈ A. Then F is differentiable at α, and −1 dFα = − dG2(α,β) ◦ dG1(α,β) .
Proof. Set η(ξ) = F (α + ξ) − F (α). Since G(α + ξ, β + η(ξ)) − G(α, β) = 0, we have
dG(α,β) (ξ, η(ξ)) + u(ξ, η(ξ)) = dG1(α,β) (ξ) + dG2(α,β) (η(ξ)) + u(ξ, η(ξ)) = 0 , where u is such that ku(ζ)k/kζk → 0 as ζ → 0. Put −1 v = η + dG2(α,β) ◦ dG1(α,β) , −1 and dG2(α,β) = T . Then
v(ξ) = −T (u(ξ, η(ξ))) .
By continuity of F , η(ξ) → 0 as ξ → 0. Hence for any > 0, we can choose δ > 0 such that kξk < δ implies ku(ξ, η(ξ))k < (kξk + kη(ξ)k) . Then kv(ξ)k < C1 (kξk + C2 (kξk + kv(ξ)k)) for some non-negative constants C1 , C2 independent of . Rearrange to get kv(ξ)k C1 (1 + C2 ) < < C1 (1 + C2 ) kξk 1 − C1 C2 for small enough. Therefore, kv(ξ)k/kξk → 0 as ξ → 0, and F is differentiable at α with derivative as desired. Completeness is needed in the following theorem so the contraction mapping fixed-point theorem applies. Theorem 3 (The implicit function theorem). Let G and (α, β) be as in Theorem 2. In addition, let G be continuously differentiable in A×B. Then there is an open ball M about α and a uniquely determined continuously differentiable function F : M → B such that F (α) = β and G(x, F (x)) = 0 for all x ∈ M . −1 Proof. Set T = dG2(α,β) and define K : A × B → W by
K(x, y) = y − T (G(x, y)) .
2 Then K is continuously differentiable, and dK(α,β) = 0. Choose a product of open 2 balls M × N about (α, β) such that for any (x, y) ∈ M × N , kdK(x,y) k ≤ 1/2 and 2 dG(x,y) is invertible. Since K is continuous and K(α, β) = β, we can also improve M so that kK(x, β) − βk < r/2, where r is the radius of N . Then by the mean value theorem, for all x ∈ M and y1 , y2 ∈ N , 1 kK(x, y1 ) − K(x, y2 )k ≤ ky1 − y2 k , 2 THE IMPLICIT FUNCTION THEOREM 3
so K is a contraction mapping in y. Since kK(x, β) − βk < (1 − 1/2)r for each
x ∈ M , K(x, ·) has a unique fixed point px in N , and the mapping F : x 7→ px is continuous. Since K(x, y) = y if and only if G(x, y) = 0, F is the desired function on M . More specifically, −1 dFx = − dG2(x,F (x)) ◦ dG1(x,F (x)) implies that F is continuously differentiable. Suppose that H : M → B is another function as desired. Then by continuity of F and H, the set E = {x ∈ M : F (x) 6= H(x)} is open. Also, H −1 (N ) is an open neighbourhood of α on which F and H must agree, because the fixed point of K is unique. Hence α is an interior point of M − E. The same argument applies to all x ∈ M − E, so M − E is open. But M is connected and M − E is non-empty. Hence E = ∅, and F is the uniquely determined. The inverse function theorem is a special case of the implicit function theorem. Theorem 4 (The inverse function theorem). Let B be an open subset of W and H : B → V be a continuously differentiable function such that dHβ is invertible for some β ∈ B. Then (i) there exist an open ball M about α = H(β) and an open neighbourhood U of β such that H is a homeomorphism from U to M ; (ii) the inverse F of H on M is continuously differentiable, and −1 dFx = dHF (x) . Proof. Define G : V × B → V by G(x, y) = x − H(y). Then G is continuously differentiable, and dG2(α,β) = −dHβ is invertible. Let K : V × B → W and F : M → B be as in the implicit function theorem and put U = F (M ). Then F has the desired derivatives, and H(F (x)) = x for all x ∈ M . Clearly the restriction of H on U is onto M . H is also injective, because the fixed point of K in U for each x ∈ M is unique. Therefore, H : U → M is a continuous bijection with a continuous inverse F .