Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
W W L CHEN
c
W W L Chen, 1997, 2008.
This chapter is available free to all individuals, on the understanding that it is not to be used for financial gain,
and may be downloaded and/or photocopied, with or without permission from the author.
However, this document may not be kept on any information storage and retrieval system without permission
from the author, unless such system is not accessible to any individuals other than its owners.
Chapter 3
IMPLICIT AND
INVERSE FUNCTION THEOREMS
Example 3.1.1. Consider the unit circle x2 + z 2 = 1. Write F (x, z) = x2 + z 2 − 1. Then the circle can
be described by the equation F (x, z) = 0. Consider now the point (x0 , z0 ) = (1, 0).
x2 + z 2 − 1 = 0
x
(x0 , z0 ) = (1, 0)
Concentrate on a small neighbourhood near the point x0 = 1 and a small neighbourhood near the point
z0 = 0. If we are given some value x near x0 = 1 such that x < 1, can we find a value z near z0 √
= 0 such
that F (x, z) = 0? The answer in this case is not only yes, but we can find two solutions z = ± 1 − x2 .
In view of the lack of uniqueness, the answer for z cannot be described as a function of x. Next, observe
that the xxxxx
partial derivative
∂F ∂F
= 2z, so that (1, 0) = 0.
∂z ∂z
Consider next a point (x0 , z0 ) where z0 > 0.
(x0 , z0 )
x2 + z 2 − 1 = 0
xxxxx
Again concentrate on a small neighbourhood near the point x0 and a small neighbourhood near the
point z0 . If we are given some value x near x0 , can we find a value z near √z0 such that F (x, z) = 0? The
answer in this case is not only √ yes, but we can precisely one solution z = 1 − x2 , given by the positive
square root. The point z = − 1 − x2 , which also satisfied the equation,√ is nowhere near the value z0 .
It follows that near (x0 , z0 ), the function z = g(x), where g(x) = 1 − x2 , gives the correct value of z
which satisfies F (x, z) = 0. Consider finally a point (x0 , z0 ) where z0 < 0.
x2 + z 2 − 1 = 0
(x0 , z0 )
Again concentrate on a small neighbourhood near the point x0 and a small neighbourhood near the
point z0 . If we are given some value x near x0 , can we find a value z near z0 such√ that F (x, z) = 0?
The answer in this case is not only yes,√but we can precisely one solution z = − 1 − x2 , given by the
negative square root. The point z = 1 − x2 , which also satisfied the equation, √ is nowhere near the
value z0 . It follows that near (x0 , z0 ), the function z = h(x), where h(x) = − 1 − x2 , gives the correct
value of z which satisfies F (x, z) = 0. Observe that in both of these last two cases, we have z0 6= 0, and
so
∂F
(x0 , z0 ) 6= 0. (1)
∂z
Indeed, it appears from observing the pictures that this last condition is sufficient to guarantee that in
suitable neighbourhoods of x0 and z0 , the value of z can be given as a function of x.
1
xxxxx and Vector Analysis
Multivariable c
W W L Chen, 1997, 2008
Example 3.1.2. Consider the parabola x + z 2 = 0. Write F (x, z) = x + z 2 . Then the parabola can be
described by the equation F (x, z) = 0. Consider now the point (x0 , z0 ) = (0, 0).
z
2
x+z =0
x
(x0 , z0 ) = (0, 0)
Concentrate on a small neighbourhood near the point x0 = 0 and a small neighbourhood near the point
z0 = 0. The reader should have little difficulty in concluding that z cannot be described as a function
of x, and that the partial derivative
xxxxx ∂F ∂F
= 2z, so that (0, 0) = 0.
∂z ∂z
Example 3.1.3. Consider the curve x−z 3 = 0. Write F (x, z) = x−z 3 . Then the curve can be described
by the equation F (x, z) = 0. Consider now the point (x0 , z0 ) = (0, 0).
z
x − z3 = 0
x
(x0 , z0 ) = (0, 0)
Concentrate on a small neighbourhood near the point x0 = 0 and a small neighbourhood near the point
z0 = 0. The reader should have little difficulty in concluding that z = g(x), where g(x) = x1/3 , gives the
correct value of z which satisfies F (x, z) = 0. Note here, though, that the partial derivative
∂F ∂F
= −3z 2 , so that (0, 0) = 0.
∂z ∂z
Let us analyze the situation more carefully. From Example 3.1.1, it appears that a condition such as
(1) is sufficient to guarantee that in suitable neighbourhoods of x0 and z0 , the value of z can be described
as a function of x. However, from Examples 3.1.2 and 3.1.3, a condition such as
∂F
(x0 , z0 ) = 0 (2)
∂z
does not give us any firm conclusion as to whether in suitable neighbourhoods of x0 and z0 , the value of
z can be described as a function of x. Indeed, we shall concentrate on situations analagous to (1) and
not those analogous to (2). 1
We shall state our result in full, but will then continue to examine only special cases of it.
THEOREM 3A. (IMPLICIT FUNCTION THEOREM) Suppose that for every i = 1, . . . , m, the
function Fi : Rn × Rm → R has continuous partial derivatives. Suppose further that there exists a point
(x0 , z0 ) ∈ Rn × Rm such that
and
∂(F1 , . . . , Fm )
(x0 , z0 ) 6= 0.
∂(z1 , . . . , zm )
(b) If x ∈ U and z ∈ V satisfy Fi (x, z) = 0 for every i = 1, . . . , m, then (3) must hold.
(c) The functions (3) are continuously differentiable, and for every i = 1, . . . , m and j = 1, . . . , n, we
have
,
∂gi ∂(F1 , . . . , Fm ) ∂(F1 , . . . , Fm )
=− . (5)
∂xj ∂(z1 , . . . , zi−1 , xj , zi+1 , . . . , zm ) ∂(z1 , . . . , zm )
We are interested in the special case m = 1. Here, (x, z) ∈ Rn × R denotes x = (x1 , . . . , xn ) ∈ Rn and
z ∈ R.
THEOREM 3A’. Suppose that the function F : Rn × R → R has continuous partial derivatives.
Suppose further that there exists a point (x0 , z0 ) ∈ Rn × R such that
∂F
F (x0 , z0 ) = 0 and (x0 , z0 ) 6= 0.
∂z
F : R2 × R → R : (x, y, z) 7→ x2 + y 2 + z 2 − 1.
Here n = 2 and m = 1. It is easy to see that F (x0 , y0 , z0 ) = 0 on the surface of a sphere centred at the
origin (0, 0, 0) and with radius 1. On the other hand,
∂F
(x0 , y0 , z0 ) = 2z0 6= 0
∂z
whenever z0 6= 0; in other words, this partial derivative does not vanish on the surface of the sphere
except on the “equator” z0 = 0 and x20 + y02 = 1. Clearly the function
p
z = g(x, y) = 1 − x2 − y 2
satisfies the requirements in a sufficiently small neighbourhood of (x0 , y0 , z0 ) if z0 > 0, and the function
p
z = g(x, y) = − 1 − x2 − y 2
satisfies the requirements in a sufficiently small neighbourhood of (x0 , y0 , z0 ) if z0 < 0. On the other
hand, in a neighbourhood of (x0 , y0 , z0 ) on the “equator”, it is not clear whether we should take the
positive or negative root. But then the partial derivative above vanishes here, so the theorem does not
apply in this case. Let us investigate the case z0 > 0 further. We have
,
∂g x ∂F ∂F x x
= −p and − = − = −p .
∂x 1−x −y2 2 ∂x ∂z z 1 − x2 − y 2
We also have
,
∂g y ∂F ∂F y y
= −p and − = − = −p .
∂y 1 − x2 − y 2 ∂y ∂z z 1 − x2 − y 2
F1 : R2 × R2 → R : (x, y, z, w) 7→ x2 + y 2 + z 2 + w2 − 2,
F2 : R2 × R2 → R : (x, y, z, w) 7→ x2 − y 2 + z 2 − w2 .
satisfy F1 (x, y, g1 (x, y), g2 (x, y)) = 0 and F2 (x, y, g1 (x, y), g2 (x, y)) = 0. Furthermore, it is easily checked
(the reader is advised to fill in the details) that
,
∂g1 ∂(F1 , F2 ) ∂(F1 , F2 ) x
=− = −√ ,
∂x ∂(x, w) ∂(z, w) 1 − x2
,
∂g1 ∂(F1 , F2 ) ∂(F1 , F2 )
=− = 0,
∂y ∂(y, w) ∂(z, w)
,
∂g2 ∂(F1 , F2 ) ∂(F1 , F2 )
=− = 0,
∂x ∂(z, x) ∂(z, w)
,
∂g2 ∂(F1 , F2 ) ∂(F1 , F2 ) y
=− = −p .
∂y ∂(z, y) ∂(z, w) 1 − y2
† Sketch of Proof of Theorem 3A’. We shall show here how one may prove the special case n = 2
and m = 1. For convenience, we shall use the notation x = (x, y) and x0 = (x0 , y0 ). Since
∂F
(x0 , z0 ) 6= 0,
∂z
we shall assume that it is positive (otherwise we simply consider −F instead of F ). By the continuity
of the partial derivative ∂F/∂z, there exist a > 0 and b > 0 such that
∂F
(x, z) > b whenever kx − x0 k < a and |z − z0 | < a.
∂z
Chapter 3 : Implicit and Inverse Function Theorems page 6 of 13
Multivariable and Vector Analysis c
W W L Chen, 1997, 2008
In view of continuity, we may also assume that there exists M > 0 such that
∂F ∂F
(x, z)<M and (x, z)<M
∂x ∂y
in the same region. Note next that since F (x0 , z0 ) = 0, it follows that
To study the term F (x, z) − F (x0 , z), we note that the line segment in R3 from (x0 , z) to (x, z) can be
described by the function
for some θ ∈ (0, 1), in view of the Mean value theorem. On the other hand, it follows from the Chain
rule that
x − x0
∂F ∂F ∂F
h0 (θ) = (DF )(L(θ))(DL)(θ) = (L(θ)) (L(θ)) (L(θ)) y − y0
∂x ∂y ∂z
0
∂F ∂F
= (θx + (1 − θ)x0 , z) (x − x0 ) + (θx + (1 − θ)x0 , z) (y − y0 ).
∂x ∂y
To study the term F (x0 , z) − F (x0 , z0 ), we note that the line segment in R3 from (x0 , z0 ) to (x0 , z) can
be described by the function
for some φ ∈ (0, 1), in view of the Mean value theorem. On the other hand, it follows from the Chain
rule that
0
∂F ∂F ∂F
h0 (φ) = (DF )(L(φ))(DL)(φ) = (L(φ)) (L(φ)) (L(φ)) 0
∂x ∂y ∂z
z − z0
∂F
= (x0 , φz + (1 − φ)z0 ) (z − z0 ).
∂z
Hence
∂F ∂F
F (x, z) = (θx + (1 − θ)x0 , z) (x − x0 ) + (θx + (1 − θ)x0 , z) (y − y0 )
∂x ∂y
∂F
+ (x0 , φz + (1 − φ)z0 ) (z − z0 )
∂z
ba0
a0 ∈ (0, a) and δ < min a0 , .
2M
∂F ∂F
(θx + (1 − θ)x , z) (x − x ) + (θx + (1 − θ)x , z) (y − y ) < ba0 ,
0 0 0 0
∂x ∂y
so that
It now follows from the Intermediate value theorem that there exists z ∈ (z0 − a0 , z0 + a0 ) such that
F (x, z) = 0. Furthermore, this value of z is unique, since a function with positive derivative cannot have
more than one zero. In other words, if we take U = D(x0 , δ) and V = (z0 − a0 , z0 + a0 ), then for every
x ∈ U , there exists a unique z ∈ V such that F (x, z) = 0. We can write z = g(x, y) and this completes
the proof of the first two parts. To prove the last part, note that since F (x, z) = 0, we have
∂F ∂F
(θx + (1 − θ)x0 , z) (x − x0 ) + (θx + (1 − θ)x0 , z) (y − y0 )
∂x ∂y
g(x) − g(x0 ) = z − z0 = − .
∂F
(x0 , φz + (1 − φ)z0 )
∂z
In particular,
,
g(x0 + h, y0 ) − g(x0 , y0 ) ∂F ∂F
=− (x0 + θh, y0 , z) (x0 , y0 , φz + (1 − φ)z0 ).
h ∂x ∂z
Similarly
,
∂g ∂F ∂F
(x0 , y0 ) = − (x0 , y0 , z0 ) (x0 , y0 , z0 ).
∂y ∂y ∂z
The argument can be repeated for every (x, y, z) ∈ U × V . This completes the proof.
Remarks. (1) Suppose that it has been established that the functions (3) are continuously differentiable.
We shall show here how we may deduce (5) by the use of the Chain rule. Consider the functions
and
In view of (4), it is clear that the composite function H = F ◦ G : U → Rm is identically zero, so that
(DH)(x) is the zero m × n matrix. On the other hand, we have, by the Chain rule, that
so that
∂F1 ∂F1 ∂g1 ∂F1
...
∂z1 ∂zm
∂xj
∂xj
.. .. .. ..
. . . = − . . (8)
∂Fm ∂Fm ∂gm ∂Fm
...
∂z1 ∂zm ∂xj ∂xj
(2) Note that in the special case when m = 1, the condition (5) reduces to (7).
In the theory of real valued functions of one real variable, we appreciate the importance of inversion. For
example, we know that the exponential function has as its inverse the logarithmic function. In this last
section, we shall study this problem more closely. More precisely, we shall deduce the following result from
the Implicit function theorem. Throughout this section, (x, z) ∈ Rn × Rn denotes x = (x1 , . . . , xn ) ∈ Rn
and z = (z1 , . . . , zn ) ∈ Rn . For functions fi : Rn → R, where i = 1, . . . , n, we write
∂f1 ∂f1
...
∂z1 ∂zn
∂(f1 , . . . , fn )
.. ..
= det . . (9)
∂(z1 , . . . , zn )
∂fn ∂fn
...
∂z1 ∂zn
∂(f1 , . . . , fn )
∂(z1 , . . . , zi−1 , ej , z1+1 , . . . , zn )
for the determinant when the i-th column of the matrix in (9) is replaced by the vector
ej = (0, . . . , 0, 1, 0, . . . , 0),
| {z } | {z }
j−1 n−j
written as a column.
THEOREM 3B. (INVERSE FUNCTION THEOREM) Suppose that for every i = 1, . . . , n, the func-
tion fi : Rn → R has continuous partial derivatives. Suppose further that the system of n equations
f1 (z1 , . . . , zn ) = x1
..
. (10)
fn (z1 , . . . , zn ) = xn
∂(f1 , . . . , fn )
(z0 ) 6= 0.
∂(z1 , . . . , zn )
Note now that the conditions for Fi in Theorem 3A are satisfied. Note also that for i, j = 1, . . . , n, we
have
Definition. The determinant given by (9) is called the Jacobian determinant of f = (f1 , . . . , fn ).
z 2 + w2
=x and ez sin w = y.
z
z 2 + w2
f1 (z, w) = and f2 (z, w) = ez sin w
z
Chapter 3 : Implicit and Inverse Function Theorems page 11 of 13
Multivariable and Vector Analysis c
W W L Chen, 1997, 2008
It follows that we can solve for z and w in terms of x and y near any point (z0 , w0 ) for which
w2 2w0
1 − 20 cos w0 6= sin w0 .
z0 z0
It follows that we can solve for r and θ in terms of x and y near any point (r0 , θ0 ) for which r0 6= 0.
Furthermore, we have
1 ∂f 1 0 ∂f 1
∂θ ∂θ
det det
0 ∂f 2
1 ∂f 2
∂r ∂θ ∂r ∂θ
= = . . . = cos θ and = = . . . = sin θ,
∂x ∂(f1 , f2 ) ∂y ∂(f1 , f2 )
∂(r, θ) ∂(r, θ)
as well as
∂f1 1 ∂f1 0
∂r ∂r
det det
∂f ∂f
2 2
0 1
∂θ ∂r sin θ ∂θ ∂r cos θ
= = ... = − and = = ... = .
∂x ∂(f1 , f2 ) r ∂y ∂(f1 , f2 ) r
∂(r, θ) ∂(r, θ)
These can be checked by using the formulae
p y
r = x2 + y 2 and θ = tan−1 .
x
F : R2 × R → R : (x, y, z) 7→ x3 z 2 − z 3 yx.
a) Explain why there is no neighbourhoods U of (0, 0) and V of 0 such that there exists a function
z = g(x, y) defined for (x, y) ∈ U and z ∈ V and satisfying F (x, y, g(x, y)) = 0.
b) Explain why the equation is soluble for z as a function of (x, y) near the point (1, 1, 1). Compute
the partial derivatives ∂z/∂x and ∂z/∂y at this point by using the partial derivatives of F .
Fi (x1 , x2 , x3 , z1 , z2 , z3 ) = 0, where i = 1, 2, 3,
3. In R2 , rectangular coordinates (x, y) and polar coordinates (r, θ) are related by x = r cos θ and
y = r sin θ. Discuss when we can solve for (r, θ) in terms of (x, y).
a) Show that
∂(x, y, z)
= r2 sin φ.
∂(r, φ, θ)