General implicit and inverse function theorems

Theorem 1. (Implicit function theorem) Let f : RN R M with N > M. We decompose

RN = RN −M × RM (1)

and denote the first N − M coordinates by vector x and the rest M coordinates by y.
Assume

i. f is differentiable and has continuous partial derivatives;

ii. f (x0, y0) = 0.

(x0, y0)  0 where the Jacobian matrix with respect to y is defined as

 
∂f
iii. det ∂y

∂y∂f
 
∂f1 1
  ∂y1
∂f
7
  
 M 
. (2)
∂y

∂f

∂fM M
∂y1 ∂y M

Then there are open sets U ⊆ RN −M , V ⊆ RM satisfying x0 ∈ U , y0 ∈ V and

i. For every x ∈ U the equation f (x, y) = 0 has one unique solution y = Y (x) ∈ V;

ii. Y (x0) = y0 ;

iii. Y is differentiable with continuous partial derivatives;

iv. For x ∈ U,

∂y∂f

 −1 
∂f1 1 ∂f1 ∂f1
   −1   ∂y1 ∂x1 ∂xN −M
∂Y ∂f ∂f
   
 M   
=− =   . (3)
∂x ∂y ∂x

∂f

   
∂fM M ∂fM ∂fM
∂y1 ∂y M ∂x1 ∂xN −M

Proof. The proof follows exactly the same idea as the R2 case. We only emphsize the difference here. Denote
 
A7
∂f
(x0, y0). (4)
∂y

 Set F (x, y) 7 A f (x, y). Then it is easy to

To make the presentation easier we pre-process as follows. −1
∂F
verify that it suffices to work with F and furthermore ∂y (x0, y0) = I the identity matrix. We choose δ1,
δ2 small to satisfy the following:

1. δ2 small enough so that

 
∂fi 1 1 − ∂fi  < 1
 
< , (5)
∂y j 2 M 2  ∂yi  2 M 2

for all k(x, y) − (x0, y0)k < 2 δ2.

1
2. Fix the above δ2. Now we choose δ1 6 δ2 such that for all x ∈ B(x0, δ1), kF (x, y0)k < 2 .

1
Now we first try to show the existence of a unique y solving

F (x, y) = 0 (6)

for all x ∈ B(x0, δ1). Fix x. Denote

g(y) 7 F (x, y). (7)

We still use the iteration

yn = yn−1 − g(yn−1). (8)

Now we have

yn − yn−1 = yn−1 − yn−2 − [g(yn−1) − g(yn−2)]. (9)

Note that the difference now is that we do not have one single ξ such that
 
∂g
g(yn−1) − g(yn−2) = (ξ) (yn−1 − yn−2). (10)
∂y

However we still have the following mean value theorem:

gi(yn−1) − gi(yn−2) = (grad g)(ξ) · (yn−1 − yn−2). (11)

This way we still could prove that {yn } is Cauchy. 

Exercise 1. Complete the proof of the theorem.

Example 2. Consider the system

x1 y2 − 4 x2 + 2 ey1 + 3 = 0 (12)
2 x1 − x3 − 6 y1 + y2 cos y1 = 0 (13)

Calculate the Jacobian of the implicit function Y (x) at x1 = −1, x2 = 1, x3 = −1, y1 = 0, y2 = 1.

Solution. Let
 
x1 y2 − 4 x2 + 2 e y1 + 3
F (x, y) 7 . (14)
2 x1 − x3 − 6 y1 + y2 cos y1
Then we have
       
∂F y2 −4 0 ∂F 2 e y1 x1
= , = . (15)
∂x 2 0 −1 ∂y −6 − y2 sin y1 cos y1

At the specified point we have

       
∂F 1 −4 0 ∂F 2 −1
= , = . (16)
∂x 2 0 −1 ∂y −6 1
We see that
   
∂Y 1 3 −4 −1
= . (17)
∂x 4 10 −24 −2

2
 dz
Example 3. Let z = f (x, y), g(x, y) = 0. Calculate dx
.
Solution. Let
 
f (x, y) − z
F (x, y, z) = . (18)
g(x, y)
Then we have
   
∂f ∂f
−1
   
∂F ∂x ∂F ∂y
= ∂g
, = ∂g
 (19)
∂x ∂(y, z) 0
∂x ∂y

which gives
 −1  ! 
∂f ∂f ∂f
−1 0 1
∂(Y , Z)
= − ∂y   ∂x = − 1 ∂g ∂f  ∂x . (20)
∂x ∂g
0
∂g ∂g − ∂y ∂y
∂g
∂y ∂x ∂y ∂x

Finally we have
dZ 1 ∂(f , g)
= det . (21)
dx ∂g ∂(x, y)
∂y

Theorem 4. (Inverse function theorem) Let f : RN R N

satisfy

i. f is differentiable with continuous partial derivatives;

ii. f (y0) = x0 ;

(y0)  0.
 
∂f
iii. det ∂y

Then there are two open sets U , V such that x0 ∈ U , y0 ∈ V and there is a function g: U  V which is the
inverse of f. Furthermore we have
 
∂g
h  i−1
∂f
(x0, y0) = ∂y
(x0, y0) . (22)
∂x

Proof. Most of the theorem follow directly from implicit function theorem, from which we obtain the
existence of I , J , g such that

f (g(x)) = x (23)

for all x ∈ I, with g(x) unique and belong to J .

Notice that to show g is the inverse, we need to further check the following: There is V ⊆ J open such that
f is one-to-one on V . (Think: Why do we need this?)
We take V = f −1(U ) ∩ J. Since f is continuous, f −1(U ) is open and thus V is open.
Now we check one-to-one. Assume that f (y1) = f (y2). Then we know there are ξ1, , ξN such that

(grad fi)(ξi) · (y1 − y2) = 0. (24)

If we set
 
(grad f1)(ξ 1)T
A7  (25)
 
,
T
(grad fN )(ξ N )

3
we would have

A (y1 − y2) = 0. (26)

But by our choice of U , det A  0. Consequently y1 − y2 = 0 y 1= y2 . 

Exercise 2. Let A, B be sets. f : A  B a function. If there is g: B  A such that

f (g(y)) = y (27)

for all y ∈ B, can we say g is an inverse of f ? What if we further assume f is one-to-one?

 
∂(r, θ)
Exercise 3. (Polar coordinates) Let x = r cos θ, y = r sinθ. Calculate ∂(x, y)
.

Exercise 4. Let x = r cosθ, y = r sin θ.

 0 for all r > 0.

 
∂(x, y)
a) Show that det ∂(r, θ)

b) Does the inverse function exist globally?

Problem 1. Let f: RN  RN be differentiable with continuous partial derivatives. Assume that det J f (x0)  0. Then
there is r > 0 such that for any open set U ⊆ B(x0, r), f(U ) is open.