15.

Implicit Functions and Their

Derivatives

When y is written as a function of (x1, . . . , xm ),

y = f(x1 , . . . , xm )

we say that y is an explicit function of (x1, . . . , xm ).

Things are different when y and (x1 , . . . , xm ) are combined in a single
function so that
f(x1 , . . . , xm , y) = 0. (15.0.1)

If the x1 , . . . , xm determine y in equation (15.0.1), we say that y is an

implicit function of (x1, . . . , xm )
With luck, we will be able to solve for y in terms of (x1, . . . , xm ). But
that is not always possible. For example, the quintic equation

y5 − 5xy + 4x2 = 0

does not have an explicit solution, although we can say that (x, y) = (1, 1)
is a solution, as is (1/4, 1), suggesting that y(1) = 1 and y(1/4) = 1. It’s
also clear that y(0) = 0. There are hints of a function here, but we can’t
solve for it.
When the equation implicitly defines y in terms of x, but we cannot
write an expression for y(x), we might still be able to determine the
derivatives. The Implicit Function Theorem gives conditions for finding
local functions for y and their derivatives.
2 MATH METHODS

15.1 Is there an Implicit Function?

One issue with equation (15.0.1) is that it is difficult to determine whether
there even is an implicit function.
◮ Example 15.1.1: No Implicit Function for a Circle. Consider the equation
x2 + y2 = 25. We know this has solutions such as (0, 5) and (3, 4). Does
this expression implicitly define a function y(x)? In this case we can solve
for y, obtaining p
y(x) = ± 25 − x2.

There is a problem here. This is not a function!

Just look at the graph. For every x, −5 < x < 5, there are two values
of y, not one. It’s not a function.
y

b (3, 4)

3 x

b
(3, −4)

Figure 15.1.2: The circle is the graph of x2 +y2 = 25, which tries to implicitly
define y as a function of x. As you can see, there are two solutions y(x) for
most values of x. This is illustrated at x = 3.

For every value of x ∈ (−5, +5), there are two values of y(x), not one.
Only at x = ±5 do we have a function. This is illustrated in Figure
15.2.2. ◭
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 3

15.2 Picking an Implicit Function I

One way to work around this is to lower the bar, to give up the search
for a global function and focus on a locally defined implicit function. We
look for a function that solves the equation in a neighborhood of a point
(x0, y0 ). We can use one function near (3, 4) and perhaps a different one
near (3, −4).
◮ Example 15.2.1: Local Implicit Functions on a Circle I. Here y = +(25−
x2 )1/2 is implicitly defined by the equation x2 + y2 = 25 and includes
the starting point (x0 , y0) = (3, 4). It can be defined on open sets as
large as (−5, 5), as in the upper blue arc in the figure below. Similarly, if
(x0, y0 ) = (3, −4), the function y = −(25 − x2 )1/2 works for x ∈ (−5, 5)
and yields the lower light blue arc. The green line segment indicates the
points (x, 0) where both functions are defined, the points with −5 < x <
5.
y
(3, 4)
b

b b b

(3, −4)

Figure 15.2.2: The circle is the graph of x2 +y2 = 25, which tries to implicitly
define y as a function√ of x. We can define such a function on the upper half
= 25 − x2 for −5 < x < 5. This contains the point (3, 4).
of the circle by y √
The function − 25 − x2 does the same thing for the lower half of the circle.
◭
4 MATH METHODS

15.3 Picking an Implicit Function II

◮ Example 15.3.1: Local Implicit Functions on a Circle II. Another problem

occurs at both (5, 0) and (−5, 0). Neither point allows us to define
a function y(x) on an open interval containing x = ±5. The points
x = ±5 cannot be in the interior of the domain of y.
This is connected with the fact that the graph becomes vertical at those
two points.
However, we can turn things around to make it work. Instead of defin-
ing y as function of x, we canp define x as a function of p y at those points.
Indeed the functions x1 (y) = 25 − y2 and x2(y) = − 25 − y2 do the
trick. They can cover both of the red arcs, and more.
y

(−5, 0) (5, 0)
b b b
x

Figure 15.3.2: The circle is the graph of x2 +y2 = 25, which tries to implicitly
define y as a function of x. The points (5, 0) and (−5, 0) pose particular
problems as we are unable to write y as a function of x on a neighborhood
of x = ±5 due to the verticality of the graph of y at x = ±5.
◭
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 5

15.4 The Implicit Function Theorem for R2

The key result for implicit functions is the Implicit Function Theorem.1
Here is a version for R2 . This is a special case of the general Implicit
Function Theorem in section 15.30. Although the two-dimensional ver-
sion is a bit easier to prove, it is difficult enough that we will not give it a
separate proof. The proof of the multidimensional version of the Implicit
Function Theorem (and related Inverse Function Theorem) will suffice.2
Implicit Function Theorem for R2 . Let G(x, y) be a C1 function on a
neighborhood of (x0, y0 ) ∈ R2 . Suppose that G(x0, y0 ) = c. If

∂G
(x0 , y0) 6= 0,
∂y

there exists a C1 function y(x) defined on an interval I containing x0 such

that:


(a) G x, y(x) = c for all x ∈ I,
(b) y(x0) = y0 , and
(c) The function y obeys

∂G (x , y )
∂x 0 0
y′ (x0 ) = − . (15.4.2)
∂G (x , y )
∂y 0 0

1
The basic idea is already present in Newton in 1669. Leibniz’s work includes an
example of implicit differentiation in 1684. The theorem is generally attributed to
Cauchy, who provided a rigorous statement and proof in two dimensions in his first
Turin Memoir (1831).
2
The history of the Implicit Function Theorem and more is covered in Steven G.
Krantz and Harold R. Parks (2002), The Implicit Function Theorem: History, Theory, and
Applications, Birkhäuser, Boston, Basel & Berlin.
6 MATH METHODS

15.5 More on the Implicit Function Theorem for R2

The condition (∂G/∂y)(x0, y0 ) 6= 0 rules out vertical graphs at (x0 , y0),
making it possible to write y as a function of x. As long as that works,
we’re in business. We not only get y as a function of x, but if the original
function was continuously differentiable, so is y(x).
Point (c) follows from the Chain Rule
once we establish that y(x) is C1.
To see this, just differentiate G x, y(x) = c to find
 
∂G ∂G
(x0 , y0) + (x0 , y0) y′ (x0 ) = 0.
∂x ∂y

Rearrange to obtain equation (15.4.2).

The statement of the theorem is similar to a 2-dimensional version of the
Linear Implicit Function Theorem, expect that it applies to differentiable
functions, and only gives a local inverse, not a global inverse (the cost of
generalizing to differentiable functions)
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 7

15.6 Using the Implicit Function Theorem

◮ Example 15.6.1: The Implicit Function Theorem and the Circle. How
does this apply to the circle x2 + y2 = 25 we studied in Example 15.3.1?
Here we set G(x, y) = x2 + y2 and c = 25. Let’s try (x0, y0 ) = (3, −4)
and see what happens.
Here
∂G
(3, −4) = −8 6= 0,
∂y
so we can apply the Implicit Function Theorem to find y(x) solving x2 +
[y(x)]2 = 25 with y(3) = −4 and y′ (3) = −2(3)/2(−4) = 3/4. Compare
to the solution y1 given by y1(x) = −(25 − x2 )1/2. Then

x
y′1(x) = −(1/2)(25 − x2)−1/2(2x) = √ .
25 − x2

Then y′1 (3) = 3/4, exactly as with y. ◭

The Implicit Function Theorem will be useful for writing one set of
economic variables as a function of other variables. For instance, suppose
we solve the consumer’s problem for prices p and income m. Can we
write the demands for x and y as functions of prices and income? Once
we characterize the solution via first order and second order equations,
we will be able to use the Implicit Function Theorem to find whether we
have proper demand functions.
8 MATH METHODS

15.7 One-dimensional Differentiable Manifolds

Regular Points and Curves. A point (x0 , y0) is a regular point of a C1

function G : R2 → R if either

∂G ∂G
(x0, y0 ) 6= 0 or (x0, y0 ) 6= 0.
∂x ∂y

If every point on C = {(x, y) : G(x, y) = c} is a regular point, we say that

C is a regular curve.
In the case of our circle, G(x, y) = x2 + y2 so ∂G/∂x = 2x and
∂G/∂y = 2y. Since these can’t both be zero on C, C is a regular curve.
Regular curves are examples of one-dimensional differentiable mani-
folds.3

3
The term “manifold” is a direct translation of Riemann’s term Mannigfaltigkeit, intro-
duced in his Göttingen lecture of 1854, Über die Hypothesen, welche der Geometrie
zu Grundeliegen, (On the hypotheses that underlie geometry). The lecture founded the
field of differential geometry.
The German mathematician Bernhard Riemann (1826–1866) was one of the all-time
greatest mathematicians. Only 39 when he died, he founded the field of differential
geometry (later used by Einstein in his theory of General Relativity), proposed the Rie-
mann Hypothesis, which is still being investigated and remains unproven after over 150
years. Some of his other accomplishments are developing the first rigorous definition
of the integral, which he used to prove some results concerning Fourier series, and
introducing Riemann surfaces in complex analysis.
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 9

15.8 Consequences of Regularity

In Theorem 15.9.1, we will show that regularity implies that at every
point of (x, y) of C, we can either write y as a function of x or x as
a function of y. We can handle horizontal segments by writing y as a
function of x and vertical segments by writing x as a function of y.
This means that regular curves can twist and turn around. What they
can’t do is cross themselves. One curve that crosses itself is Bernoulli’s
Lemniscate, defined by the function4

G(x, y) = (x2 + y2 ) − 2a2(x2 − y2) = 0.

We compute DG(0, 0) = (0, 0), showing that the curve is not regular
at the origin. In fact, there is no way to describe this curve that makes it
regular at the origin.5

Figure 15.8.1: Bernoulli’s Lemniscate: (x2 + y2 ) − 2a2 (x2 − y2 ) = 0.

4
This lemniscate is that of the Swiss mathematician Jacob (aka James or Jacques)
Bernoulli (1655–1705 NS), one of the mathematicians in the Bernoulli family. He made
a number of important contributions to mathematics. He discovered the constant e,
base of the natural logarithms. He also provided the original formulation of the Law
of Large Numbers. Finally, he and his brother Johann (1667–1748 NS) founded the
Calculus of Variations, a method of optimization where the optimal point is a function
rather than a number.
5
The only possibility is to be a one-dimensional manifold. If U is a neighborhood of
the origin that is homeomorphic to an open interval, removing the origin breaks U into
four components, but removing the corresponding point in the interval breaks it into
two. The number of components of a set must be preserved under homeomorphism,
so this is impossible.
10 MATH METHODS

15.9 Characterizing Regular Points

The following theorem characterizes the regular points of a curve. It
follows immediately from the Implicit Function Theorem for R2 .
Theorem 15.9.1. Let G : R2 → R be a C1 function. If (x0 , y0) is a regular
point on the curve C = {(x, y) : G(x, y) = c}, Then either
1. (∂G/∂y)(x0, y0 ) 6= 0 and there is a C1 function y(x) with G x, y(x) =

c on some neighborhood of (x0 , y0), or

2. (∂G/∂x)(x0, y0 ) 6= 0 and there is a C1 function x(y) with G x(y), y =

c on some neighborhood of (x0 , y0).

Then either

∂G (x , y ) ∂G (x , y )
∂x 0 0 ∂y 0 0
y′ (x0) = − or x′ (y0) = − ,
∂G (x , y ) ∂G (x , y )
∂y 0 0 ∂x 0 0

respectively.
When the curve defined by G is regular, we can strengthen this as
follows.
Corollary 15.9.2. Let G : R2 → R be a C1 function. It the curve C =
{(x, y) : G(x, y) = c} is regular, at every point (x0, y0 ) on the curve C,
we can parameterize the curve by
a C1 curve defined
on an open set
containing (x0 , y0 ): either as x, y(x) or x(y), y .
The manifold C is considered one-dimensional since it can be locally
described by a single parameter. It is differentiable because there are in-
vertible differentiable functions that describe it locally, in a neighborhood
of each point.
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 11

15.10 Tangent Spaces

What if we have a curve inside our manifold? Can we say anything about
its tangent vector?
One way to define tangent vectors for a manifold M is to define them
as the tangent vectors of all the curves in M. To that end, let

M = {(x, y) ∈ R2 : G(x, y) = c}

be a regular
manifold and x(t) = x(t), y(t) be a C1 curve in M. It obeys

′
G x(t), y(t) = c for all t ∈ I. Then we can
consider v = x (t0 ) to be a
tangent vector to M at x(t0) = x(t0), y(t0 ) .


Now G x(t), y(t) = c for all t. We apply the Chain Rule at t0 to find

DG(x0, y0 ) x′ (t0), y′ (t0) = DG(x0, y0) v = 0

 
 

since G is constant on the manifold. The tangent vector v is in

ker DG(x0, y0). In fact, any element of the
kernel can be represented
this way by appropriate choice of x(t), y(t) .6 We call

T(x0 ,y0 ) M = ker DG(x0, y0 )

the tangent space of M at (x0 , y0 ).

6
This will be easier to see once we introduce coordinate charts.
12 MATH METHODS

15.11 Tangent Space via the Gradient

Alternatively, we can relate the tangent space T(x0 ,y0 ) M to the gradient
vector ∇G(x0, y0):
∇G(x0, y0 )·v = 0.

The gradient gives the direction of fastest increase of G. The tangent

space T(x0 ,y0 ) M is the set of all vectors perpendicular to it (definition 2).

G(x, y) ∇G
b

TM

Figure 15.11.1: Here ∇G is the gradient and the tangent line to G(x, y) = 0,
labelled T M, is shown by the heavy line perpendicular to the gradient vector.
The function on the graph is 0 = G(x, y) = y + 3(x3 − x).
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 13

15.12 m-Dimensional Manifolds

We’ve called regular curves 1-dimensional manifolds. But what is a
manifold? We can give a general definition of a manifold by using
homeomorphisms.7
Manifolds. A metric space M is a manifold if for every x ∈ M, there
is a open neighborhood U of x and an integer m ≥ 0 such that U is
homeomorphic to an open subset of Rm .
In other words, a manifold is a space that locally looks like m-
dimensional Euclidean space, Rm .
One advantage to this definition of a manifold is that it minimizes the
baggage we carry about from Rm . This allows us (forces us?) to define
things in manifolds in a way that is independent of coordinates. If you
check other sources, you will sometimes see manifolds defined as subsets
of some ambient space RA . Manifolds do not have to sit inside some
other space, although they often do.
Early on, manifolds were studied in ways that made it hard to distinguish
intrinsic properties from those based on the coordinates currently in use.
When Einstein proposed his theory of general relativity, he expressed ev-
erything in terms of coordinates, making it harder to see which concepts
were dependent on the coordinate system used, and which were not,
something that has been the bane of many students of general relativity.
The type of manifold we defined above is sometimes called a topo-
logical manifold. There are more specialized types of manifold, such
as differentiable manifolds. Topological manifolds are the basic type of
manifold. Other manifolds will be topological manifolds with additional
properties, just as inner product spaces are vector spaces with additional
properties. In the case of manifolds, we will mostly be interested in
differentiable manifolds. These are topological manifolds which have
differentiable structure added.

7
See page 1-1 of Michael Spivak, A Comprehensive Introduction to Differential Ge-
ometry, vol. I, 1970. Also see page 1 of Morris Hirsch, Differential Topology, 1976.
This definition can be extended to topological spaces, but that ideally involves para-
compactness, which takes us too far afield.
14 MATH METHODS

15.13 Open Sets in Rm are Manifolds

It’s easy to find lots of manifolds. Any open set in Rm is an m-dimensional
manifold, including the entire space Rm .
Theorem 15.13.1. Let U be an open set in Rm . Then U is an m-
dimensional manifold.
Proof. Let x ∈ U. Then U is a open neighborhood of x and U is trivially
homeomorphic to itself (use the identity map id). As U is an open subset
of Rm , this means that every point x ∈ U has a open neighborhood (U)
that is homeomorphic to an open subset of Rm (the same U). This shows
that U is an m-dimensional manifold.
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 15

15.14 Circles are Manifolds I

Any circle in R2 is a one-dimensional manifold. We will show this for
the circle defined by x2 + y2 = 25, but the methods used here apply to
any circle. They can be easily generalized to spheres in Rm .
We focus on the circle from Example 15.3.1, defined by C = {(x, y) :
x + y2 = 25}. We will use four functions gi : Vi → C with Vi ⊂ R that
2

describe C as a manifold.
p

g1(x) = x, 25 − x2 for x ∈ V1 = (−5, 5)
p

g2(x) = x, − 25 − x2 for x ∈ V2 = (−5, 5)
p

g3(y) = 25 − y2, y for y ∈ V3 = (−5, 5)
p

g4(y) = − 25 − y2, y for y ∈ V4 = (−5, 5)

The functions g1 and g2 describe the top and bottom halves of the circle,
respectively. The functions for the right and left sides of the circle are g3
and g4 , respectively. I’ve labeled the domains Vi , which happen to be
identical here, but need not be.
Every point except (0, 5), (0, −5), (5, 0), and (−5, 0) is in the range of
exactly two of the gi . Those points are in the range of only one of the
gi , with the points listed in the same order as g1, . . . , g4 .
The inverses of the gi are the projections ϕ1 (x, y) = ϕ2 (x, y) = x and
ϕ3 (x, y) = ϕ4 (x, y) = y. That means that when (x, y) ∈ C, ϕ1 can be
written p
ϕ1 (x, 25 − x2) = x,

with similar definitions for the other ϕu = i. This guarantees that each of
the four open half-circles is homeomorphic to the interval (−5, 5). This
shows that the circle C is a manifold.
The mappings g1 and ϕ1 are illustrated in Figure 15.15.1.
16 MATH METHODS

15.15 Circles are Manifolds II

y

U1
g1 ϕ1

bc bc

V1 x

Figure 15.15.1: The vertical lines illustrate the bijection between U1 (in red)
and V1 = {(x, 0) : |x| < 5} (green) created by projection ϕ1 onto the x-axis
and g1 , mapping back to the circle. Two examples are highlighted. One
showing ϕ1 mapping down to the x-axis, the other showing g1 mapping up
to the circle.
Here V1 is embedded in R2 using the map ψ(x) = (x, 0), with image
{(x, 0) : x ∈ (−5, 5)}. The subspace topology ensures ψ is a homeomorphism.

The range of g1 is the set U1 = {(x, y) ∈ C : y > 0}. Then g1 ◦

ϕ1 : U1 → U1 and is defined by
 p
  p 
g1 ϕ1 x, 25 − x 2 2
= g1 (x) = x, 25 − x .

Here both ϕ1 and its inverse g1 are continuous, one-to-one, and

onto. Therefore g1 and ϕ1 are homeomorphisms between U1 and
V1 = (−5, +5). Define U2, U3, and U4 similarly for the other gi .8

8
Although the functions gi can be defined on the closures of the intervals Vi , we
would no longer have an open set as the range, and the functions cannot be C1 there.
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 17

15.16 Local Coordinate Systems

One difference between Rm and manifolds is that we have a natural
system of coordinates on Rm , but none on any manifold. The mappings
gi allow us to establish local coordinate systems on our manifold C.
The coordinates are dependent on the maps gi . Different maps mean
different coordinates.
Each gi : Vi → Ui is a homeomorphism with inverse ϕi : Ui → Vi =
(−5, 5). We will think of each ϕi as setting up a one-dimensional coor-
dinate system in Vi for each of the Ui . There is only one coordinate in
the system because it is a one-dimensional manifold.
The pair (Ui, ϕi ) is called a coordinate chart, a term that is meant to
remind you of nautical charts marked with latitude and longitude lines.
The collection of all the charts on a manifold is called an atlas. To be an
atlas M, every point of M must be contained in some chart. Since each
Ui is homeomorphic to an open set Vi ⊂ Rm , and M is covered by the
charts, it is a manifold.9

9
The use of local coordinates such as these dates back at least to Gauss in 1827.
18 MATH METHODS

15.17 Coordinate Charts

As you have seen, we can set up coordinate system by using a system of
coordinate charts. This allow us to describe open sets in a manifold M
by means of a coordinate system in some Rm .
Chart. A chart or coordinate system on a manifold M is a pair (U, ϕ)
where U is a open subset of M and ϕ is a continuous homeomorphism
from U to an open subset of some Euclidean space Rm .

R2

ϕ
U ϕ(U)

Figure 15.17.1: This illustrates a chart (U, ϕ) for the manifold M. It is a

homeomorphism from the open set U ⊂ M onto the open subset ϕ(U) ⊂ R2 .
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 19

15.18 Reconciling Local Coordinate Systems

We now have two different coordinate systems that describe all but four
points of the circle. How do they relate?
When both exist for a point, we can change the i coordinates to the j
coordinates. To make this concrete, consider the point (3, 4). This is in
the domain of both ϕ1 (top half) and ϕ3 (right half). If we have the ϕ1
coordinate, which is 3, we map it to (3, 4), then project on the y-axis to
obtain the ϕ3 coordinate 4.
More generally, when Ui ∩Uj is non-empty, the coordinate change takes
coordinate x ∈ ϕi (Ui ∩ Uj) ∈ Vi , maps it to ϕ−1i (x) = gi (x) ∈ Ui ∩ Uj ,
−1
and then applies ϕj to map it to ϕj(ϕi (x)) ∈ ϕj (Ui ∩ Uj) ∈ Vj . In our
example above, Ui ∩ Uj = C ∩ R2++ and both ϕi (Ui ∩ Uj) ⊂ Vi and
ϕj (Ui ∩ Uj) ⊂ Vj are the interval (0, 5).

ϕ−1 ϕj
Vi −−i→ Ui ∩ Uj −→ Vj .
20 MATH METHODS

15.19 Transition Maps

In short, we change coordinates from ϕi to ϕj by applying the transition
map

ϕj ◦ ϕ−1
i : ϕi (Ui ∩ Uj ) ⊂ Vi → ϕj (Ui ∩ Uj ) ⊂ Vj . (15.19.3)

Transition maps are only defined when Ui ∩ Uj is non-empty.

So what does this coordinate change map look like?√On U1 ∩ U3, the
ϕ1 coordinate x √ ∈ (0, 5) is mapped to ϕ−1 2
1 (x) = (x, 25 − x ). Then
ϕ3 maps that to 25 − x2 . This map is not only a homeomorphism on
U1 ∩ U3 = (0, 5), but is actually C∞ there.
We will shortly define differentiable manifolds, and the defining feature
will be that the transition maps are differentiable. We will need some
more definitions first, but when we have them, we will find that the circle
C, together with the differentiable structure given by the charts {gi }4i=1 is
a C∞ manifold!
y
bc

ϕ3 U1
V3
ϕ−1
1
bc

V1 x

Figure 15.19.1: Here V1 is marked in green and V3 in blue and U1 =

1 (V1 ) ∩ ϕ3 (V3 ) is shown in red. We follow the transition map ϕ3 ◦ ϕ1
ϕ−1 −1 −1

from V1 , up to U1 , and over to V3 . This transition map is only defined on

1 (U1 ), which is the right-hand portion of V1 , not including (0, 0).
ϕ−1
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 21

15.20 Atlases of Charts

As mentioned before, we bundle our charts (coordinate systems) into an
atlas, which can be used to give the manifold a differentiable structure
via the transition maps.
Atlas. An atlas of a manifold M is an indexed family of charts {(Uα, ϕα )}α∈A
that covers M (M ⊂ ∪α∈A Uα).
Since the atlas covers M, every point in M is in the domain of at least
one chart.
22 MATH METHODS

15.21 Smooth Atlases and Transition Maps

We now consider the smoothness of an atlas.
Transition Maps and Ck Atlas. Given an atlas {(Uα , ϕα )}α∈A let Wαβ =
Uα ∩ Uβ . An atlas is a Ck atlas if each transition map ϕβ ◦ ϕ−1 k
α is C as a
map from ϕα (Wαβ) to ϕβ (Wαβ), whenever Wαβ is non-empty.
Since ϕα (Wαβ ) ⊂ ϕα (Uα) ⊂ Rm and ϕβ (Wαβ) ⊂ ϕβ (Uβ) ⊂ Rm , every
transition function maps a subset of Rm into Rm . That means it makes
sense to consider whether transition functions are Ck .

Wαβ = Uα ∩ Uβ
M

ϕα ϕβ

ϕβ ◦ ϕ−1
α

ϕα (Uα) ϕβ (V)

ϕα (Wαβ) ϕβ (Wαβ )
ϕα ◦ ϕ−1
β

Figure 15.21.1: Here (ϕα , U) and (ϕβ , V) are charts with non-empty com-
mon domain Wβ = Uα ∩ Uβ . The transition maps ϕα ◦ ϕ−1 β and ϕβ ◦ ϕα
−1

are indicated. Here ϕα ◦ ϕ−1

β : ϕβ (W) → ϕα (W) and ϕβ ◦ ϕα : ϕα (W) →
−1

ϕβ (W). The darker regions indicate the sets ϕα (W) ⊂ ϕα (U) and ϕβ (W) ⊂
ϕβ (V).
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 23

15.22 Differentiable Manifolds

We are finally ready to define a differentiable manifold.10
Differentiable Manifold. A C1 or differentiable manifold is a manifold
where all of the transition maps ϕβ ◦ ϕ−1 1
α are C . More generally, a
manifold is a Ck manifold if all of the transition maps are Ck functions.
When k = 0, we have a topological manifold, where all of the transition
maps are continuous, but need not be differentiable.
We continue to apply these ideas to the circle C = {(x, y) : x2 + y2 =
25}.
◮ Example 15.22.1: Transition Maps on the Circle. The charts and
their inverse mappings were introduced in Section 15.14 and Exam-
ple 15.26.1. Consider the charts (U1, ϕ1 ) and (U3, ϕ3 ). The inter-
section ϕ−1 −1
1 (U1 ) ∩ ϕ3 (U2 ) is the NE quadrant of the circle, {(x, y) ∈
C : x > 0, y √ > 0}. The transition functions for this pair ofp
charts are
ϕ3 ◦ ϕ−1
1 = 25 − x2 , defined for x ∈ (0, 5) and ϕ1 ◦ ϕ−1
3 = 25 − y2,
defined for y ∈ (0, 5). Both are C∞ , as are all of the other transition
functions. This means C is a not just a topological manifold, but a C∞
manifold. ◭

10
This definition, using transition maps, dates to O. Veblen and J.H.C. Whitehead
(1931) “A set of axioms for differential geometry”, Proc. Nat. Acad. Sci. 17, 551–561.
Oswald Veblen (1880–1960) was an American mathematician who specialized in
topology, differential geometry, and projective geometry. His uncle was the sociologist
Thorstein Veblen (Theory of the Leisure Class).
J.H.C. Whitehead (1904–1960) was a British mathematician. During World War II, he
applied operations research to submarine warfare, and later joined the codebreakers
at Bletchley Park. In algebraic topology, he defined CW complexes and developed
simple homotopy theory. The British mathematician and philosopher Alfred North
Whitehead was his uncle, who is best known in mathematics for the three-volume
Principia Mathematica, written with Bertrand Russell.
24 MATH METHODS

15.23 Dimension of a Manifold

We will show that charts containing the same point must map to the
same Rm . This allows us to unambiguously define the dimension of the
manifold M at each point x ∈ M. If the charts containing x all map to
Rm , we say that M has dimension m at x.
By using Proposition 34.9.1, we can show that every chart containing
a point x must map to the same Rm .
Theorem 15.23.1. If x ∈ U for some chart (U, ϕ) with ϕ : U → Rk and
x ∈ V for some chart (V, ψ) with ψ : V → Rm , then k = m.
Proof. Now x ∈ U ∩ V, which is open. Consider the mapping ϕ ◦ ψ−1
which is defined on the open set ψ(U ∩ V).

ψ−1 ϕ
→ ϕ(U ∩ V) ⊂ Rk .
ψ(U ∩ V) −−→ U ∩ V −

This is a homeomorphism between ψ(U ∩ V) ⊂ Rm and the open set

ϕ(U ∩ V) ⊂ Rk .
Proposition 34.9.1 now shows that k = m.
If the dimension of M is m at all points of M, then the manifold M is
m-dimensional. It’s easy to show that connected manifolds must have
the same dimension at every point.
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 25

15.24 Charts and Coordinates in Vector Spaces

A simple example of charts occurs in finite-dimensional vector spaces,
where we use them to set up coordinate systems.
◮ Example 15.24.1: Charts for Vector Spaces. Suppose we have an m-
dimensional vector space V with the usual topology. Let B1 and B2 be
bases for V and B1 and B2 the corresponding basis matrices. We can
now define two charts on V. The chart ϕi gives us the coordinates in Rm .
From Section 31.23 we know that ϕi (x) = B−1 i x = t, the coordinates in
m
R . Both of these mappings, ϕ1 and ϕ2 , are one-to-one onto mappings
from V to Rm . Moreover, as linear mappings they are continuous. As
both ϕi are onto, Either one by itself would form an atlas. Together, they
are also an atlas.
Let’s examine the transition maps. The two charts are (V, ϕ1 ) and
(V, ϕ2 ). The intersection of their domains is all of V and ϕj ◦ ϕ−1
i (t) =
−1 m
Bj Bi t for any t ∈ R . The transition map is exactly the coordinate
change formula we used in both Equations 31.23.1 and 15.19.3. ◭
26 MATH METHODS

15.25 Graphs are Manifolds 11/03/22

In a sense, a manifold is a generalization of the graph of a function
from Rm to R, just as a curve generalizes a function from R to R.
◮ Example 15.25.1: Graph of a Function. Let f : Rm → R. Let M be the
graph of f,
M = x, f(x) : x ∈ Rm .

The set M is an m dimensional manifold. Define F(x) = x, f(x) . Here
F maps Rm onto the graph of f. We only need one chart for this manifold.
Let π(x, y) = x, which projects the graph onto its first m coordinates, x .
The chart is (M, π) and π−1 = F. The atlas is also (M, π). There are no
transition maps to worry about. ◭
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 27

15.26 Charts and Atlases for Regular Curves

We used charts when treating circles as manifolds.
◮ Example 15.26.1: Charts for a Circle. The function ϕ1 in Section
15.14 is a chart. Let U2 = {(x, y) : y < 0} and define ϕ2 on U2 ∩ C
by ϕ2 (x, y) = x. This yields a chart with inverse g2. Similarly, you can
define open sets U3 = {(x, y) : x > 0} and U4 = {(x, y) : x < 0}.
The projections ϕi : Ui ∩ C defined for i = 3, 4 by ϕi (x, y) = y are
charts with inverses g3 and g4. Together, the four charts cover C, so they
form an atlas. Notice that any two would leave one of the points of C
uncovered. ◭
This method can be made more general by using the Implicit Function
Theorem.
◮ Example 15.26.2: Regular Curves have Charts. Corollary 15.9.2 shows
how the same process can be used for any regular curve C defined as
a level set of a C1 function G(x, y). For any (x0 , y0 ) ∈ C, it gives us an
open set U. Depending on the case we are in, we define
ϕ(x,−1y) = x
−1
or ϕ(x,
y) = y. The inverse on ϕ(U) is ϕ (x) = x, y(x) or ϕ (x) =
x(y), y , respectively.
Since the Implicit Function Theorem can be applied at any point of the
curve C, for each point, there is a chart that covers it. The charts form an
atlas. The transition functions are all C1 , so we have a C1 manifold. ◭
28 MATH METHODS

15.27 Tangent Spaces Revisited

As with curves, we can define tangents by considering tangents of curves
in M. Since M is m-dimensional, we potentially have a much richer
collection of tangents to study.
Let z(t) be a C1 curve ′


 with z(0) = z0 and z (0) = v, then g z(t) = c.
By the Chain Rule, Dg(z0) v = 0, again showing any tangent at z0 is in
the null space of Dg(z0) .
◮ Example 15.27.1: Isoquants. Suppose f : Rm → R is a production
function. We normally assume Df ≫ 0, so Df has rank one. By the
Fundamental Theorem of Linear Algebra, m = rank Df + dim ker Df,
so dim ker Df = m − 1. A basis for the tangent space (ker Df) can be
constructed by considering ∆x1e1 + ∆xi ei . Then

∂f ∂f
∆x1 + ∆xi = 0.
∂x1 ∂xi

It follows that
∆xi f1
= − = − MRTS1i .
∆x1 fi
The slopes of the isoquant in various directions is given by the marginal
rate of technical substitution. ◭
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 29

15.28 Differentiable Functions on Manifolds

We can’t define differentiable functions directly on a manifold, as our
manifold may not have the required vector space structure to define
derivatives. However, by using charts to write coordinates in Rm , we
can define differentiable functions from one differentiable manifold to
another.
Let M be a differentiable m-manifold and N be a differentiable n-
manifold. We say a function f : M → N is differentiable at x ∈ M if
there exist coordinate charts (U, ϕ) with x ∈ U ⊂ M and (V, ψ) with
f(x) ∈ V ⊂ N such that
ψ ◦ (f ◦ ϕ−1 )

is differentiable where it makes sense, meaning on the set ϕ U∩f−1(V) .

Consider the mapping

ϕ−1 f ψ
ϕ(U) −−→ U ⊂ M −
→V ⊂ N −
→ ψ(V).

which is defined on
ϕ U ∩ f−1(V) .

By using transition functions and the Chain Rule, it is easy to see that all
charts containing x and f(x) agree on the differentiability of f. We can
similarly define Ck functions for any k.
As a sanity check, suppose M and N are open subsets U ⊂ Rk and
V ⊂ Rm . The only charts are the respective identity maps, so if f : M →
N, we need only check whether id ◦(f ◦ id−1) = f is differentiable as a
function from U to V. I.e., it must be differentiable in the ordinary sense.
30 MATH METHODS

15.29 Differentiable Functions on the Manifold C

We’ll illustrate how this works with a simple example mapping the circle
C into the real line.
◮ Example 15.29.1: A C∞ function from C to R. The atlas for C is given
by the charts (Ui, ϕi )4i=1 defined in Section 15.14. Recall that ϕ−1i = gi .
Consider the function f(x, y) = x2 + y4 with f : C → R. For R, there
is a single chart (R, id) where id(x) = x is the identity map.
We now compute id ◦f ◦ ϕ−1 i = id ◦f ◦ gi for (x, y) ∈ C. Since id is the
identity, this reduces to f ◦ gi . Now for i = 1, 2, we have

id ◦f ◦ ϕ−1 2 4 2 2 2
i (x) = gi1 (x) + gi2 (x) = x + (25 − x ) ,

while for i = 3, 4,

id ◦f ◦ ϕ−1 2 4 2 4
i (y) = gi1 (y) + gi2 (y) = 25 − y + y .

It is easy to see that these functions are not only differentiable, but C∞ ,
showing that f is a C∞ function from C to R. ◭
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 31

15.30 The Multidimensional Implicit Function Theorem

So far, we have only examined one-dimensional manifolds using the
Implicit Function Theorem. The Implicit Function Theorem can be ex-
tended to systems of k implicit functions of m variables. Then we can use
the Implicit Function Theorem to describe many m-dimensional mani-
folds.11
Implicit Function Theorem. Let g : Rm+k → Rk be C1 . For x ∈ Rm and
y ∈ Rk we write g(x, y). Consider the system of k equations

g(x, y) = c.

If g(x∗, y∗) = c and the k × k matrix (Dyg)(x∗, y∗ ) is invertible, then there

is a C1 function ŷ : Rm → Rk defined on some ball Br (x∗) ⊂ Rm with
r > 0 such that

g x, ŷ(x) = c (15.30.4)

for all x ∈ Br (x∗) and

y∗ = ŷ(x∗).

Moreover,
−1
Dŷ(x∗ ) = − Dy g (x∗, y∗ ) Dx g (x∗, y∗ ).



(15.30.5)

As an exercise, it’s worth considering how the Linear Implicit Function

Theorem (section 7.35) relates to the Multidimensional Implicit Function
Theorem. Is the linear theorem a special case of this one?

11
The Italian mathematician Ulisse Dini (1845–1918) generalized the Implicit Function
Theorem to m dimensions. Dini worked primarily in real analysis. Among other things,
he developed a criterion for the pointwise convergence of Fourier series.
32 MATH METHODS

15.31 The Inverse Function Theorem

The Inverse and Implicit Function Theorems are closely related. It’s fairly
easy to prove one from the other. We will prove a multidimensional
Implicit Function Theorem by first proving a multidimensional Inverse
Function Theorem.
The Inverse Function Theorem gives conditions that ensure a function
will have a C1 inverse. Of the two theorems, the proof of the Inverse
Function Theorem is easier to understand, so it is the one we prove. After
that, the Implicit Function Theorem follows easily.12
The Inverse Function Theorem. Let f : Rm → Rm be C1. If there is y∗
with f(y∗) = x∗ and the m × m matrix Dyf(y∗) is invertible, then there
is a C1 function ŷ : Rm → Rm defined on some ball Br (y∗ ) and an open
neighborhood V of x∗ where f is a bijection between Br (y∗) and V. The
inverse map ŷ : V → Br (y∗ ) is also C1 with


f ŷ(x) = x (15.31.6)

for all x ∈ V. Moreover,

−1
Dŷ(x∗) = Dyf(y∗) .

(15.31.7)

12
The earliest version of the Inverse Function Theorem seems to be that of Joseph Louis
Lagrange in 1770. It was later proved by Picard and Goursat via iteration. For us, the
iteration is replaced by the Contraction Mapping Theorem.
The French mathematician Édouard Jean-Baptiste Goursat (1858–1936) is probably
best known today for the Cauchy-Goursat Theorem of complex analysis. He was also
one of the 19th century mathematicians who explored geometry in more than three
dimensions. In his own day he was best known for his Cours d’analyse mathématique.
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 33

15.32 Proof of the Inverse Function Theorem I

Proof (Inverse Function Theorem). Let T = Dyf(y∗ ). By hypothesis, T

is invertible. Taylor’s formula yields f(y) = f(y∗ ) + T (y − y∗ ) + R1. We
can approximate x = f(y) by dropping the remainder R1 and using

x = f(y∗) + T (y − y∗ ).

Multiplying by T −1 and rearranging, we obtain

y = y∗ + T −1 x − f(y∗) .

This equation approximates y, which obeys f(y) = x. If we knew f−1(x)

existed, we would be approximating it.
Define a new function Fx(y) by replacing y∗ with y in the right hand
side:
Fx(y) = y + T −1 x − f(y) .

The proof uses a form of Newton’s method to estimate the inverse. We

normally think of iterating Newton’s method to increase the estimate’s
precision. Instead of iterating, we apply the Contraction Mapping The-
orem. The iteration is then hidden in its proof. We’ll show that the
mapping from y to Fx(y) is a contraction for each x. As such, it has a
fixed point which will provide the inverse function.
The proof also uses some facts about matrix norms that we establish in
sections 35–25.

Proof continues ...

34 MATH METHODS

15.33 Proof of the Inverse Function Theorem II

Proof continues. By continuity of Dy f, we may choose an r > 0 small

enough that for y ∈ Br (y∗),

1
T − Dyf(y) = Dy f(y∗) − Dy f(y) <
2kT −1k

and so that Dy f is invertible on Br (y∗ ). The latter is possible because

Dy f(y) is continuous in y, Dy f(y∗) is invertible, and the set of invertible
matrices is open.
Take any y1 , y2 ∈ Br (y∗). By the Mean Value Theorem, there is ȳ in
the line segment ℓ(y1 , y2 ) with
 
Fx (y1 ) − Fx(y2 ) = Dy Fx (ȳ) (y1 − y2 )

For any y ∈ Br (y∗), including ȳ,

Dy Fx(y) = I − T −1Dy f(y) = T −1 T − Dy f(y)

 

Putting it together,
 
Fx(y1 ) − Fx(y2 ) = Dy Fx (ȳ) (y1 − y2 )
= T −1 T − Dyf(ȳ)
 
y1 − y2
≤ T −1 T − Dy f(ȳ) y1 − y2
1
< ky1 − y2 k.
2

Proof continues ...

 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 35

15.34 Proof of the Inverse Function Theorem III

Remainder of Proof. We have shown that for each x, Fx is a contraction

on Br (y∗). By the Contraction Mapping Theorem, it has a unique fixed
point, which we call yx .
The fixed point yx obeys

yx = Fx(yx )
= yx + T −1 f(yx ) − x .

This implies

f(yx) = x. Define the function ŷ by ŷ(x) = yx . Then
f ŷ(x) = x, so ŷ is the inverse of f.
By Invariance of Domain, the inverse of f is continuous on V =
f Br (y ) , which is open and contains x∗. Finally, since f is C1 and
∗

Dŷ = (Df)−1 exists and is continuous, ŷ = f−1 is continuously differ-

entiable on V. That is, ŷ is C1 on V with derivative given by equation
(15.31.7).
36 MATH METHODS

15.35 Proof of Implicit Function Theorem I

With the Inverse Function Theorem in hand, we are ready to tackle the
Implicit Function Theorem.
Proof (Implicit Function Theorem). We use the Inverse Function Theo-
rem. The key is to define the proper mapping. Let13
x  x 
G = ,
y g(x, y)

which maps from Rm+k to Rm+k . Here x ∈ Rm and g ∈ Rk . Its derivative

is  I
m 0 
DG = ,
D x g Dy g
which is invertible at (x∗, y∗ ) because (Dy g)(x∗, y∗ ) is invertible. In fact,

−1
 I 0
DG = .
−(Dy g)−1Dx g I

Proof continues ...

13
All of the arguments to g, G, and H in this proof are vectors, not covectors. I’ve
sometimes written them horizontally to save space.
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 37

15.36 Proof of Implicit Function Theorem II

Remainder of Proof. Applying the Inverse Function Theorem to G, there

is an r > 0 so that an inverse function H is defined on
Br (x∗ , y∗) which
maps to a neighborhood of G(x∗, y∗ ) = x∗ , g(x∗, y∗) = (x∗ , c) ∈ Rk .
We now write  u   H (u, v) 
1
H = .
v H2 (u, v)
Then
   
u H1 (u, v) 
G H =

v g H1 (u, v), H2 (u, v)
u
= .
v

Then u = H1 (u, v) and we can write

g u, H2 (u, v) = v.

Setting v = c and u = x, we obtain

c = g x, H2 (x, c) ,

and define ŷ(x) = H2 (x, c).

Then
g(x∗, ŷ(x∗ )) = c = g(x∗, y∗ ),

showing that ŷ(x∗) = y∗ since Dy g(x∗, y∗ ) is invertible. Finally, equation

(15.31.7) follows by the Chain Rule.
38 MATH METHODS

15.37 Regular Manifolds

We can extend the concept of regularity to systems of equations and so
to manifolds.
Regular Points. A point (x0, y0 ) ∈ Rm+k is a regular point of a C1 function
g : Rm+k → Rk if rank Dy g(x0, y0 ) = k.
We can use this to define regular manifolds.
Regular m-Manifolds. If every point of M = {(x, y) ∈ Rm+k : g(x, y) = c}
is a regular point, we say that M is a regular m-manifold.
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 39

15.38 Manifolds as Solution Sets

We will show that every regular m-manifold has a C1 atlas and so is a C1
m-manifold.
Theorem 15.38.1. Suppose g : Rm+k → Rk is a C1 function and M =
{z ∈ Rm+k : g(z) = c} is a regular m-manifold. Then for every point
z∗ ∈ M there is a relatively open set U ⊂ M and a function ϕ : U → Rm
such that (U, ϕ) is a chart. These charts form a C1 atlas, making M a C1
manifold.
Proof. Since rank Dg(z∗ ) = k, we can divide the variables into two
groups, x = (zi1 , . . . , zim ) and y = (zim+1 , . . . , zim+k ) so that Dyg(x∗, y∗ )
is invertible. The Implicit Function Theorem yields a C1 function
∗
ŷ : Rm → k m

R defined on ball∗ Br∗(x ) ∈ R with r > 0 such that
g x, ŷ(x) = c. Set U = Br (x , y ) ∩ M and define ϕ(x, y) = x for
x ∈ U.
I claim (U, ϕ) is a chart. Both ϕ and ϕ−1 = x, ŷ(x) are C1 functions

and ϕ is a homeomorphism between U and the open set V = ϕ−1 (U) ⊂

M. Thus (U, ϕ) is a chart. Since M is regular, we can generate a chart
containing any point of M, showing that the charts cover M, forming an
atlas.
All that is left is to show that the transition functions are C1. Now
suppose we have two charts (U, ϕ) and (V, ψ) where U ∩ V is non-empty.
Now consider
ψ ◦ ϕ−1 : ϕ(U ∩ V) → ψ(U ∩ V).

Since both ϕ−1 and ψ are C1 , the Chain Rule tells us that the transition
maps are C1 , showing that we have a C1 atlas.
40 MATH METHODS

15.39 Manifold or Not?

Graphs provide some of the simplest manifolds.
◮ Example 15.39.1: An Atlas of One Chart. Sometimes an atlas only needs
one chart. Recall Example 15.25.1, where f is a real-valued function on
Rm and M is its graph. We defined a single chart, (M, π) where the

projection is defined by π(x, y) = x. It has inverse F(x) = x, f(x) .
Since the single chart covers M, it suffices to define an atlas for M. ◭
Circles require more than one chart. One way to see this is that any
circle is compact, so its image is also compact, and cannot be an open
set in Rm . The charts we’ve used give us an atlas of 4 charts.
◮ Example 15.39.2: An Atlas for every Circle. Take the circle C about the
origin with radius 5, C ∩ U1 ∩ U3 is the upper right quadrant of the circle
and ϕ−13 U1 = {y : 5 > y > 0}. Then

p p
ϕ1 ◦ ϕ−1
3 (y) = ϕ1 (x0 + 25 − y2 , y) = x0 + 25 − y2 .

This is C1 on {y : 5 > y > 0}. Similarly, the other transition maps are C1.
This atlas makes C a C1 manifold. In fact, it makes C a C∞ manifold.
This same technique can be used to define an atlas for any circle.
In fact the technique can also be used on spheres, although it requires
more charts, 2(m + 1) for the unit m-sphere, which is Sm = {x ∈ Rm+1 :
kxk2 = 1}. ◭
 15. IMPLICIT FUNCTIONS AND THEIR DERIVATIVES 41

15.40 Are these Manifolds?

In contrast, the curve with a cusp in Example 14.31.1 (shown below)
is not a differentiable manifold as it is impossible to define a chart in a
neighborhood of the origin that is compatible with a C1 atlas. This is due
to the fact that the curve is not regular at the origin. Since the curve is
homeomorphic to the real line, it is a topological manifold, but it is not
a differentiable manifold.
x2

x1

As we saw with Bernoulli’s lemniscate (section 15.8), curves that cross

or have a T-intersection fail to even be topological manifolds.
◮ Example 15.40.1: Regular Curves are Manifolds. We saw how to con-
struct charts for any regular curve in Example 15.26.2. Since this con-
struction can be done at any point of C, the charts defined this way form
an atlas for C. This shows that our original definition of a manifold is
encompassed in the second definition using charts and an atlas. ◭
A similar procedure works for m-dimensional manifolds. It uses the
Implicit Function Theorem.

November 3, 2022

Copyright c 2022 by John H. Boyd III: Department of Economics, Florida International

University, Miami, FL 33199