INVERSE FUNCTION THEOREM

PAUL BRESSLER

1. Fixed point theorem for contracions

1.1. Fixed points. Suppose that X is a set and f : X → X. An element x ∈ X is called a fixed
point of f if it satisfies f (x) = x.
1.2. Contractions. Suppose that (X, d) is a metric space and f : X → X is a map (of sets). The
map f is called a contraction if there exists 0 6 λ < 1 such that for all x, y ∈ X the inequality
d(f (x), f (y)) 6 d(x, y) holds.
Note that a contraction is uniformly continuous.
1.3. Fixed point theorem.
Theorem 1.1. Suppose that (X, d) is a complete metric space and f : X → X is a contraction.
Then, there exists a unique fixed point p ∈ X of f .
Proof. The map f is a contraction, hence there exists 0 6 λ < 1 such that for all x, y ∈ X the
inequality d(f (x), f (y)) 6 d(x, y) holds.
Let x0 ∈ X and let C := d(x0 , f (x0 ). For n, m = 1, 2, . . . let xn := f n (x0 ) = f (f n−1 (x0 )) =
f (xn−1 ). Then,
d(x1 , x2 ) 6 λC, d(x2 , x3 ) 6 λ2 C, . . . , d(xn , xn+1 ) 6 λn C, . . . , d(xn+m−1 , xn+m ) 6 λn+m C, . . . .
Therefore, for  > 0,
1
d(xn , xn+m ) 6 d(xn , xn+1 ) + · · · + d(xn+m−1 , xn+m ) 6 (λn + · · · + λn+m )C 6 λm C<
1−λ
(1 − λ)
if λm < . Since 0 6 λ < 1, it follows that lim λm = 0, and the inequality d(xn , xn+m ) < 
C m→∞
is satisfied for sufficiently large m. Thus, the sequence (xn ) is Cauchy.
Since (X, d) is complete, the sequence converges to p := lim xn ∈ X. We will show that p is the
n→∞
unique fixed point of f .
Since p = lim xn and f is continuous, it follows that f (p) = lim f (xn ) = lim xn+1 = p.
n→∞ n→∞ n→∞
Suppose that p, q ∈ X are fixed points of f , i.e. f (p) = p and f (q) = q. If p 6= q, then
0 < d(p, q) = d(f (p), f (q)) 6 λ · d(p, q) < d(p, q)
which is impossible, hence p = q.

2. Inverse function theorem
2.1. Diffeomorphisms. Suppose that A and B are finite dimensional vector spaces, U ⊆ A and
V ⊆ B are open subsets and f : U → V is a map of class C r , where r = 1, 2, . . . , ∞.
If the map f
• admits a set-theoretic inverse (equivalenly, is a bijection) and
• the set-theoretic inverse f −1 is of class C r
we say that f is a C r -diffeomorphism.
f
Lema 2.1. Suppose that A ⊇ U → − V ⊆ B is a diffeomorphism. Then, for any a ∈ U the derivative
(Df )a : A → B is an isomorphism. In particular, dim A = dim B.
1
2 P.BRESSLER

Proof. Apply the chain rule to f ◦ f −1 = Id. 

f
Exercise: Construct an example of a map A ⊇ U → − V ⊆ B such that U is connected, for any a ∈ U
the derivative (Df )a : A → B is an isomorphism, but the map is not injective.
Theorem 2.2 (Inverse function theorem). Suppose that
• A and B are finite dimensional vector spaces, U ⊆ A and V ⊆ B are open subsets, a ∈ U ;
• f : U → V is a map of class C k , where k = 1, 2, . . . , ∞;
• (Df )a : A → B is an isomorphism.
Then, there exists an open neighborhood a ∈ V ⊂ U such that
• the map f |V : V → f (V ) is a C k -diffeomorphism (i.e. f (V ) is open in B, f |V is bijective
and (f |V )−1 : f (V ) → V is of class C k );
• for all x ∈ V , (Df −1 )f (x) = (Df )−1
x : B → A.

Proof. (1) Without loss of generality we may assume that

• A = B,
• a = 0, f (0) = 0,
• (Df )0 = IdA .
(2) Let g : U → A denote the map defined by g(x) = x − f (x). Then,
• g(0) = 0,
• g is of class C k ,
• (Dg)0 = 0.
(3) Let  > 0.
Since the map x 7→ k(Dg)x k is continuous1 and 0 7→ k(Dg)0 k = 0, there exists r > 0 such
that, if x ∈ A and kxk < r, then
• x∈U
• k(Dg)x k < 
• (Df )x = Id − (Dg)x is invertible.
(4) Therefore, if x1 and x2 satisfy kxi k < r, then kx1 − x2 k < 2r and
g(x2 ) = g(x1 ) + (Dg)x1 (x2 − x1 ) + kx2 − x1 k · R(x2 , x1 ),
where R satisfies lim R(x2 , x1 ) = 0. Hence,
x2 →x1

kg(x2 )−g(x1 )k 6 k(Dg)x1 k·kx2 −x1 k+kx2 −x1 k·kR(x2 , x1 )k = kx2 −x1 k·(k(Dg)x1 k+kR(x2 , x1 )k).
1
Choose r > 0 so that k(Dg)x1 k + kR(x2 , x1 )k < . Then, for x2 such that kxi k < r the
2
inequality
1
(1) kg(x2 ) − g(x1 )k 6 kx2 − x1 k
2
holds. It follows from (1) that
1
kx2 − x1 k − kf (x2 ) − f (x1 )k 6 k(x2 − x1 ) − (f (x2 ) − f (x1 ))k = kg(x2 ) − g(x1 )k 6 kx2 − x1 k
2
and, therefore,
1
(2) kf (x2 ) − f (x1 )k > kx2 − x1 k.
2
Note that (2) implies that f is injective near 0.
Taking x1 = 0 and x2 = x in (1) and using g(0) = 0 we see that
1 1
kxk 6 r =⇒ kg(x)k 6 kxk 6 r.
2 2
1because it is equal to the composition of continuous maps x 7→ (Dg)x and k · k
 INVERSE FUNCTION THEOREM 3

1
(5) For y such that kyk 6 r let
2
Fy (x) = y + g(x).
Then, Fy (x) = x (i.e. x is a fixed point of Fy ) if and only if
y + g(x) = x ⇔ y = x − g(x) = x − (x − f (x)) = f (x).
In other words,
Fy (x) = x (i.e. x is a fixed point of Fy ) ⇔ y = f (x).
1
(6) Suppose that kyk 6 r. For x1 and x2 such that kxi k < r it follows from (1) that
2
1
kFy (x1 ) − Fy (x2 )k = kg(x1 ) − g(x2 )k 6 kx2 − x1 k,
2
i.e. Fy is a contraction of {x ∈ A | kxk 6 r} and, therefore, has a unique fixed point in
{x ∈ A | kxk 6 r}. Hence,
1
for any y such that kyk 6 r the equation y = f (x) has a unique solution.
2
Thus, we obtain the injective map
1
f −1 : {y ∈ A | kyk < r} → A
2
−1
by f (y) = the unique solution of y = f (x).
1
(7) Let V = {x ∈ A | kf (x)k < r }. Then,
2
1
f : V → f (V ) = {y ∈ A | kyk < r }
2
is a bijection. Note that f (V ) is an open subset of A.
(8) The inequality (2) may be rewritten as
1
(3) ky2 − y1 k > kf −1 (y2 ) − f −1 (y1 )k
2
showing that f −1 is continuous, hence f |V : V → f (V ) is homeomorphism.
(9) We claim that f −1 is, in fact, differentialble.
1
Proof. Let b ∈ f (V ), i.e. kbk < r and let a = f −1 (b) so that f (a) = b. Then,
2
(4) f (x) − f (a) = (Df )a (x − a) + kx − ak · R(x, a),
where lim R(x, a) = 0. Recall that kak < r implies (by the choince of r) that (Df )a is
x→a
invertible. Let y = f (x). Applying (Df )−1
a to (4) we obtain

(Df )−1
a (y − b) = f
−1
(y) − f −1 (b) + kf −1 (y) − f −1 (b)k · (Df )−1
a R(f
−1
(y), f −1 (b))
or, equivalently,
f −1 (y) = f −1 (b) + (Df )−1
a (y − b) − kf
−1
(y) − f −1 (b)k · (Df )−1
a R(f
−1
(y), f −1 (b)).
The inequality (3) implies that kf −1 (y) − f −1 (b)k 6 2ky − bk. Therefore, for y 6= b
1
kkf −1 (y) − f −1 (b)k · (Df )−1
a · R(f
−1
(y), f −1 (b))k 6 2 · k(Df )−1
a k · kR(f
−1
(y), f −1 (b))k.
ky − bk
Since f −1 is continuous and lim R(x, a) = 0, it follows that lim R(f −1 (y), f −1 (b)) = 0. 
x→a y→b
4 P.BRESSLER

(10) It remain to show that f −1 is of class C k if f is. So far we know that (Df −1 )y = (Df )−1
f −1 (y) .
−1
The map y 7→ (Df )y is equal to the composition
(5) y 7→ f −1 (y) 7→ (Df )f −1 (y) 7→ (Df )−1
f −1 (y) ,

where the first map is C 0 (i.e. continuous), the second map is C k−1 and the third map
(inversion) is C ∞ . Therefore, y 7→ (Df −1 )y is continuous, which is to say f −1 is of class C 1 .
We proceed by induction: if l 6 k − 1 and we assume that f −1 is of class C l , then (5), i.e.
y 7→ (Df −1 )y , is of class C l , which is to say f −1 is of class C l+1 .


3. Implicit function theorem

Implicit function theorem (below) contains the inverse function theorem as a special case and
will be deduced from the latter. In other words, the two statements are equivalent.
We begin by introducing notation. Suppose that A and B are finite dimensional vector spaces,
U ⊆ A × B is an open subset, f : U → B is of class C k .  
A linear map Φ : A × B → B may be written in the “matrix” form as Φ = φ ψ , where
φ : A → B, ψ : B → B so that
 
  a
Φ(a, b) = φ ψ = φ(a) + ψ(b).
b
 
In this notation Df = D1 f D2 f , so D1 f and D2 f are some kind of “partial derivatives”.
Theorem 3.1 (Implicit function theorem). Suppose that (a, b) ∈ U satisfies
• f (a, b) = 0;
• (D2 f )(a,b) is an isomorphism.
Then, there exists an open neighborhood a ∈ V ⊂ prA (U ) and a unique map g : V → B such that
(1) g(a) = b
(2) ∀x ∈ V f (x, g(x)) = 0.
Proof. (1) Let F : U → A × B denote the map defined by F (x, y) = (x, f (x, y)). Then,
• F (a, b)= (a, 0), 
Id 0
• DF =
D1 f D2 f
Since (D2 f )(a,b) is an isomorphism, it follows that (DF )(a,b) is an isomorphism.
(2) The inverse function theorem implies that F is a local diffeomorphism near (a, b). More
precisely, there are open subsets a ∈ S ⊂ prA (U ) and b ∈ T ⊂ prB (U ) and an open subset
(a, 0) ∈ W ⊂ A × B such that
F |S×T : S × T → W
is a diffeomorphism. Applying F −1 to F (x, y) = (x, f (x, y)) we see that F −1 (x, f (x, y)) =
(x, y), hence, F −1 is of the form F −1 (x, z) = (x, h(x, z)) for suitable map h : S × T → B.
(3) Let a ∈ V ⊂ prA (U ) be an open neighborhood which
• is connected2
• satisfies V × {0} ⊂ W .
For x ∈ V and, hence, (x, 0) ∈ W
F −1 (x, 0) = (x, h(x, 0)) =⇒ (x, 0) = F (x, h(x, 0)) = (x, f (x, h(x, 0))),
i.e. f (x, h(x, 0)) = 0.
(4) Let g(x) = h(x, 0). Then,
• f (x, g(x)) = 0
• (a, b) = F −1 (a, 0) = (a, h(a, 0)) =⇒ b = g(a).
2The connectedness assumption will be used to show uniqueness.
 INVERSE FUNCTION THEOREM 5

(5) It remains to prove uniqueness of g. Suppose that g1 : V → B satisfies the same conditions
as g, i.e. g1 (a) = g(a) = b and f (x, g1 (x)) = 0.
Claim: Suppose that a0 ∈ V and g1 (a0 ) = g(a0 ). Then, there exists an open neighborhood
of a0 ∈ V 0 ⊆ V such that g|V 0 = g1 |V 0 .
Proof. Since g1 is continuous and W is open, there exists an open neighborhood of a0 ∈ V 0 ⊆
V such that g1 (V 0 ) ⊂ W .
Since, by assumption, f (x, g1 (x)) = 0 for all x ∈ V 0 , it follows that F (x, g1 (x)) = (x, 0)
and, therefore,
(x, g1 (x)) = F −1 (x, 0) = (x, h(x, 0)) = (x, g(x)).

(6) It follows that the subset {x ∈ V | g(x) = g1 (x)} ⊂ V is open. On the other hand, since g,
g1 are continuous, the subset {x ∈ V | g(x) = g1 (x)} is closed. Since V was chosen to be
connected, it follows that g = g1 on V .


Departamento de Matemáticas, Universidad de Los Andes

Email address: [email protected]