LECTURE: IMPLICIT FUNCTION THEOREM

1. Inverse Function Theorem

Recall: Inverse Function Theorem, which says that if y = f (x) and
f ′ (x0 ) ̸= 0 for some x0 then can locally solve for x in terms of y near
x0 , that is x = g(y) for some smooth g and moreover
−1
g ′ (y) = (f ′ (g(y)))
Inverse Function Theorem:
Let F : Rn → Rn be C 1 and suppose DF (x0 ) is invertible at x0

Then F is invertible in a neighborhood of x0 . More precisely:

(1) There exist neighborhoods U of x0 and V of y0 = F (x0 )

such that the restriction F |U : U → V is a bijection

(2) The inverse function G : V → U is also C 1 and for y ∈ V

DG(y) = [DF (G(y))]−1

2. Implicit Function Theorem

On the other side of the coin is the Implicit Function Theorem.

Goal: Suppose you have an equation of the form F (x, y) = 0, can you
solve for one variable in terms of the other one(s)?

1
2 LECTURE: IMPLICIT FUNCTION THEOREM

Example 1: Let F (x, y) = x2 + y 2 − 1 = 0 ⇒ x2 + y 2 = 1

√
Then you can solve for y in terms of x because y = ± 1 − x2

This expression fails precisely when y = 0 that is when Fy = 0

Notice here Fy is the derivative of F with respect to the variable you

want to solve for
dy
Moreover, we can calculate dx in terms of partial derivatives:

′
x2 + y 2 − 1 = (0)′
 
dy
2x + 2y =0
dx
dy 2x
=−
dx 2y
dy Fx
=−
dx Fy

Here the x and y get switched in the right-hand-side. This is defined

precisely when Fy ̸= 0

Example 2: To prep for the notation of the Implicit Function Thm:

Let n = 3 (number of x variables) and m = 2 (number of y variables)

Define F : R3+2 → R2 by F = (F1 , F2 ) where

F1 (x1 , x2 , x3 , y1 , y2 ) =x1 y2 − 4x2 + 3 + 2ey1

F2 (x1 , x2 , x3 , y1 , y2 ) =2x1 − x3 + y2 cos(y1 ) − 6y1
 LECTURE: IMPLICIT FUNCTION THEOREM 3

Notice F (x0 , y0 ) = 0 where x0 = (3, 2, 7) and y0 = (0, 1)

Question: Can we solve for y in terms of x, for x near x0 = (3, 2, 7)?

The Implicit Function Theorem says yes provided that “Fy ̸= 0” (the
derivative with respect to the variable you want to solve for is nonzero)
   
y2 −4 0 2ey1 x1  
[DF (x, y)] = | = Fx | Fy
2 0 −1 −y2 sin(y1 ) − 6 cos(y1 )

   
1 −4 0 2 3  
[DF (x0 , y0 )] = | = Fx (x0 , y0 ) | Fy (x0 , y0 )
2 0 −1 −6 1

Here all you need to check here is that Fy (x0 , y0 ) is invertible, but

2 3
|Fy (x0 , y0 )| = = 2 + 18 = 20 ̸= 0 YES
−6 1

Then the implicit function theorem then that there is a function

y = G(x) from a neighborhood W of x0 = (3, 2, 7) to Rm such that
F (x, G(x)) = 0 that is the equation is satisfied

Moreover, we can calculate DG(3, 2, 7) via

 −1    
2 3 1 −4 0 1 5 4 −3
DG(3, 2, 7) = − (Fy )−1 Fx = − =−
−6 1 2 0 −1 20 −10 12 2

∂y1 4
Looking at the (1, 2) entry for example, this tells us ∂x2 = − 20 = − 15
dy
Compare this once again with the dx = − FFxy condition from Ex 1
4 LECTURE: IMPLICIT FUNCTION THEOREM

Implicit Function Theorem:

Suppose F : Rn+m → Rm is C 1 and F (x0 , y0 ) = 0 for some (x0 , y0 )

If det Fy (x0 , y0 ) ̸= 0, then there is a neighborhood U of (x0 , y0 )

and a neighborhood W of x0 and a function G : W → Rm differ-
entiable at x0 such that

{(x, y) ∈ U | F (x, y) = 0} = {(x, G(x)) | x ∈ W }

Moreover DG(x0 ) = − (Fy (x0 , y0 ))−1 Fx (x0 , y0 )

In other words, if the derivative with respect to the variable you want
to solve for is invertible, then the equation F (x, y) = 0 is locally the
graph of a function y = G(x).

Application: This theorem is extremely useful in PDEs. Lots of

PDEs, especially first-order ones, are given by implicit equations of
the form F (x, u, ∇u) = 0. The Implicit Function Theorem can then
be used to solve for u in terms of x, provided some “nondegeneracy”
condition holds, which is usually equivalent to the assumption above.

Proof:

Surprisingly, the Implicit function theorem and Inverse function theo-

rem are equivalent, since they both solve for one variable in terms of
another one. So our goal is to apply the Inverse Function Theorem to
a cleverly designed function

STEP 1: Given our F define f : Rn+m → Rn+m by

 LECTURE: IMPLICIT FUNCTION THEOREM 5

f (x, y) = (x, F (x, y))

Goal: Apply the Inverse function theorem to f at a = (x0 , y0 )

First show that Df (a) is invertible. However

 
In×n 0n×m
[Df (x, y)] =
Fx Fy
 
In×n 0n×m
Hence [Df (a)] =
Fx (x0 , y0 ) Fy (x0 , y0 )
It then follows from cofactor expansion along the first n rows that

det[Df (a)] = det[Fy (x0 , y0 )] ̸= 0

Where the last step follows precisely because Fy (x0 , y0 ) is invertible

STEP 2: Hence Df (a) is invertible and therefore by the Inverse Func-

tion Theorem there is an open set U containing (x0 , y0 ) and an open
set V containing f (x0 , y0 ) such that f : U → V is invertible.

Moreover, f −1 : V → U is differentiable at f (x0 , y0 ) and

Df −1 (f (x0 , y0 )) = (Df (x0 , y0 ))−1

Note: f (x0 , y0 ) = (x0 , F (x0 , y0 )) = (x0 , 0)

Write f −1 in terms of components as f −1 =: (h, g)

STEP 3: Define W and G as follows, think of it like an x−axis of V

W =: {x ∈ Rn | (x, 0) ∈ V }
6 LECTURE: IMPLICIT FUNCTION THEOREM

G(x) =: g(x, 0) for x ∈ W

Notice W is nonempty since x0 ∈ W and W is open since W is just a

projection of V on Rn

Since f −1 is differentiable at (x0 , 0) and g is a component of f −1 it

follows that g is differentiable at (x0 , 0) ∈ V and so G is differentiable
at x0

STEP 4: Let’s show

{(x, y) ∈ U | F (x, y) = 0} = {(x, G(x)) | x ∈ W }

A =B

A ⊆ B : If (x, y) ∈ A then (x, y) ∈ U and F (x, y) = 0 from which it

follows that f (x, y) = (x, F (x, y)) = (x, 0)
| {z }
0
Since (x, 0) ∈ V (range of f ), by definition x ∈ W and from f (x, y) =
(x, 0) we get (x, y) = f −1 (x, 0) = (h(x, 0), g(x, 0))

Comparing components, this implies y = g(x, 0) = G(x) which implies

that (x, y) = (x, G(x)) and since we’ve shown x ∈ W , we get that
(x, y) ∈ B

Hence A ⊆ B and similarly we have B ⊆ A

STEP 5: The only thing left to show is the formula for the derivatives

Note: G(x0 ) = y0 because f (x0 , y0 ) = (x0 , 0) implies f −1 (x0 , 0) =

(x0 , y0 ) and comparing the second component we get g(x0 , 0) = y0 so
G(x0 ) = y0
 LECTURE: IMPLICIT FUNCTION THEOREM 7

Since F (x, G(x)) = 0 for all x ∈ W , F is differentiable at (x0 , G(x0 )) =

(x0 , y0 ), and G is differentiable at x0 , by the Chain Rule, we have

D (F (x, G(x))) =0
Fx (x0 , y0 ) + Fy (x0 , y0 )DG(x0 ) =0
Fy (x0 , y0 )DG(x0 ) = − Fx (x0 , y0 )
DG(x0 ) = − (Fy (x0 , y0 ))−1 Fx (x0 , y0 ) □

Note: It is also true that if F is C k then G is C k as well.

As mentioned above, the Implicit Function Theorem implies the In-

verse Function Theorem, so both of them are equivalent.

3. Uniform Contraction Mapping Principle

The following is a generalization of the Banach fixed point theorem,
which was used in the proof of the Inverse Function Theorem.

Main Idea: Suppose we have family of contraction maps F (x; µ) in-

dexed by a parameter µ. For each µ, F (·; µ) has a unique fixed point
G(µ). The Uniform Contraction Mapping Principle says that the map
G is as smooth as the original map F

We will first need one technical result involving Lipschitz functions

Fact:
Let X and Y be Banach spaces, U ⊆ X open and F : U → Y
differentiable. If F is Lipschitz with constant L, then

∥DF (x)∥ ≤ L for all x ∈ U

8 LECTURE: IMPLICIT FUNCTION THEOREM

Proof: Let ϵ > 0 and x ∈ U . Since F is differentiable at x, we can

find δ > 0 such that, whenever ∥h∥X < δ

∥F (x + h) − F (x) − DF (x)h∥Y
< ϵ.
∥h∥X

Let u ∈ U be a unit vector, and let h = 2δ u so that ∥h∥X < δ.

Then since u = h/∥h∥X , we have

∥DF (x)h∥Y
∥DF (x)u∥ =
∥h∥X
∥DF (x)h − (F (x + h) − F (x))∥Y + ∥F (x + h) − F (x)∥Y
≤
∥h∥X
∥F (x + h) − F (x) − DF (x)h∥Y ∥F (x + h) − F (x)∥Y
= +
∥h∥X ∥h∥X
LIP L∥h∥X
≤ ϵ+
∥h∥X
= ϵ + L,

Since ϵ is arbitrary, we have ∥DF (x)∥ ≤ L for all x ∈ U □

Definition:
F = F (u, µ) is a uniform contraction if there is L < 1 such
that for all µ and all u, v, we have

|F (u, µ) − F (v, µ)| ≤ L|u − v|

 LECTURE: IMPLICIT FUNCTION THEOREM 9

Definition:
F is uniformly Lipschitz in µ if there exists a constant M > 0
such that for all u and µ1 , µ2 , we have

|F (u, µ1 ) − F (u, µ2 )| ≤ M |µ1 − µ2 |

Uniform Contraction Mapping Principle:

Let X and Y be Banach spaces and D ⊆ X be closed and B ⊆ Y
(parameter space) be open

Suppose F = F (u, µ) : D × B → D is a uniform contraction

Let G = G(µ) : B → D be the map of fixed points of F (u, µ)

(1) If F is uniformly Lipschitz in µ then G is Lipschitz

(2) If F ∈ C k (D × B, X) for k ≥ 0, then G ∈ C k (B, X)

Proof:

STEP 1: Define G(µ) as in the statement of the theorem. By the

Banach fixed point theorem, G : D → B is the unique function such
that F (x; µ) = x if and only if x = G(µ)

|G(µ1 ) − G(µ2 )| = |F (G(µ1 ); µ1 ) − F (G(µ2 ); µ2 )|

≤ |F (G(µ1 ); µ1 ) − F (G(µ1 ); µ2 )| + |F (G(µ1 ); µ2 ) − F (G(µ2 ); µ2 )|
≤ |F (G(µ1 ); µ1 ) − F (G(µ1 ); µ2 )| + L|G(µ1 ) − G(µ2 )|

(1 − L)|G(µ1 ) − G(µ2 )| ≤ |F (G(µ1 ); µ1 ) − F (G(µ1 ); µ2 )|

10 LECTURE: IMPLICIT FUNCTION THEOREM

Finally, divide by (1 − L) to get

1
|G(µ1 ) − G(µ2 )| ≤ |F (G(µ1 ); µ1 ) − F (G(µ1 ); µ2 )|
1−L
Since F is continuous in both variables, then the RHS above → 0 as
µ2 → µ1 , thus G is continuous.

STEP 2: Proof of (1)

If F is uniformly Lipschitz in µ, then the above becomes

M
|G(µ1 ) − G(µ2 )| ≤ |µ1 − µ2 |
1−L
STEP 3: Proof of (2)

First consider the case k = 1, i.e. F is C 1 . We will first find a candi-

date for DG(µ) and then show that this is the derivative of G

Recall that G(µ) = F (G(µ), µ). If G were differentiable, then

DG(µ) = DF (G(µ), µ) = FX (G(µ), µ)DG(µ) + FB (G(µ), µ)
This means that DG(µ) would be a fixed point of the mapping
Φ : L(B, X) × B → L(B, X) defined by
Φ(A; µ) = FX (G(µ), µ)A + FB (G(µ), µ)
The map Φ is a uniform contraction, since for A1 , A2 ∈ L(B, X)
|Φ(A1 ; µ) − Φ(A2 ; µ)|
=|FX (G(µ), µ)A1 + FB (G(µ), µ) − (FX (G(µ), µ)A2 + FB (G(µ), µ))|
=|FX (G(µ), µ)(A1 − A2 )|
≤∥FX (G(µ), µ)∥ |A1 − A2 |
≤L|A1 − A2 |
 LECTURE: IMPLICIT FUNCTION THEOREM 11

Where the last line follows from the previous proposition and the fact
that F (·; µ) is Lipschitz with constant L

STEP 4: Since L < 1, by the Banach fixed point theorem, there exists
a function Z : B → L(B, X) where Z(µ) is the unique fixed point of
Φ(·; µ). Since F is C 1 , Φ is continuous, thus by STEP 2 but with Z
instead of G, Z(µ) is continuous.

The function Z(µ) is our candidate for DG(µ). All that remains is to
use the definition of the derivative to show that Z(µ) is actually the
derivative of G. This is technical and will be omitted.

We then repeat this argument for k > 1 for higher order derivatives.
□