MATHEMATICS E-23a, Fall 2016

Linear Algebra and Real Analysis I

Module #3, Week 4
Implicit functions, manifolds, tangent spaces, critical points
Author: Paul Bamberg
R scripts by Paul Bamberg
Last modified: December 5, 2016 by Paul Bamberg (added the graph for HW
problem 5)

Reading
• Hubbard, Section 3.1 (Implicit functions and manifolds)
• Hubbard, Section 3.2 (Tangent spaces)
• Hubbard, Section 3.6 (Critical points)
• Hubbard, Section 3.7 through page 354 (constrained critical points)

Recorded Lectures
• November 24, 2015 (watch on November 22)
• December 1, 2015 (watch on November 29)
• December 3, 2015 (watch on December 1)

Proofs to present in section or to a classmate who has done them.

• 12.1 (Hubbard Theorem 3.2.4) Suppose that U ⊂ Rn is an open subset,
F : U → Rn−k is a C 1 mapping, and manifold M can be described as the
set of points that satisfy F(z) = 0. Use the implicit function theorem to
show that if [DF(c)] is onto for c ∈ M , then the tangent space Tc M is the
kernel of [DF(c)]. You may assume that the variables have been numbered
so that when you row-reduce [DF(c)], the first n − k columns are pivotal.
• 12.2(Hubbard, theorems 3.6.3 and 3.7.1) Let U ∈ Rn be an open subset
and let f : U → R be a C 1 (continuously differentiable) function.
First prove, using a familiar theorem from single-variable calculus, that if
x0 ∈ U is an extremum, then [Df (x0 )] = [0].
Then prove that if M ⊂ Rn is a k-dimensional manifold, and c ∈ M ∩ U is
a local extremum of f restricted to M , then Tc M ⊂ ker[Df (c)].
• 12.3(Special case of Hubbard, theorem 3.7.5) Let M be a manifold known
by a real-valued C 1 function F (x) = 0, where F goes from an open subset
U of Rn to R and [DF (x)] is nowhere zero. Let f : U → R be a C 1 function.
Prove that c ∈ M is a critical point of f restricted to M if and only if there
exists a Lagrange multiplier λ such that [Df (c)] = λ[DF (c)].

1
R Scripts

• Script 3.4A-ImplicitFunction.R
Topic 1 - Three variables, one constraint
Topic 2 - Three variables, two constraints

• Script 3.4B-Manifolds2D.R
Topic 1 - A one-dimensional submanifold of R2 – the unit circle
Topic 2 - Interesting examples from the textbook
Topic 3 - Parametrized curves in R2
Topic 4 - A two-dimensional manifold in R2
Topic 5 - A zero-dimensional manifold inR2

• Script 3.4C-Manifolds3D.R
Topic 1 - A manifold as a function graph
Topic 2 - Graphing a parametrized manifold
Topic 3 - Graphing a manifold that is specified as a locus

• Script 3.4D-CriticalPoints
Topic 1 - Behavior near a maximum or minimum
Topic 2 - Behavior near a saddle point

• Script 3.5A-LagrangeMultiplier.R
Topic 1 - Constrained critical points in R2

2
1 Executive Summary
1.1 Implicit functions – review of the linear case.
We have n unknowns, n − k equations, e.g for n = 3, k = 1
2x + 3y − z = 0, 4x − 2y + 3z = 0  
2 3 −1
Create an (n − k) × n matrix: T =
4 −2 3
If the matrix T is not onto, its rows (the equations) are linearly dependent.
Otherwise, when we row reduce, we will find n − k = 2 pivotal columns and
k = 1 nonpivotal columns. We assign values arbitrarily to the “active” variables
that correspond to the nonpivotal columns, and then the values of the “passive”
variables that corresponds to the pivotal column are determined.
Suppose that we reorder the unknowns so that the “active” variables come last.
Then, after we row reduce the matrix, the first n − k columns will be pivotal. So
the first n − k columns will be linearly independent, and they form an invertible
square matrix. The matrix is now of the form  T = [A|B], where A is invertible.
~x
The solution vector is of the form ~v = , where the passive variables ~x come
~y
first, the active variables ~y come second.
A solution to T ~v = ~0 is obtained by choosing ~y arbitrarily and setting
~x = −A−1 B~y. Our system of equations determines ~x “implicitly” in terms of ~y.

1.2 Implicit function theorem – the nonlinear case.

We have a point c ∈ Rn , a neighborhood W of c, and a function F : W → Rn−k
for which F(c) = 0 and [DF(c)] is onto. F imposes constraints.
The variables are ordered so that the n − k pivotal columns in the Jacobian
matrix, which correspond to the passive variables, come first. Let a denote the
passive variables at c; let b denote the active variables at c.
The implicit function g expresses the passive variables in terms of the active
variables, and g(b)
 = a. For y near b, x = g(y) determines passive variables
a
such that F = 0. Tweak y, and g specifies how to tweak x so that the
b
constraints are still satisfied.
Although we usually cannot find a formula for g, we can find its derivative at
b by the same recipe that worked in simple cases.
Evaluate the Jacobian matrix [DF(c)].
Extract the first n − k columns to get an invertible square matrix A.
Let the inverse of this matrix act on the remaining k columns (matrix B) and
change the sign to get the (n − k) × k Jacobian matrix for g.
That is, [Dg(b)] = −A−1 B.

3
1.3 Curves, Surfaces, Graphs, and Manifolds
Manifolds are a generalization of smooth curves and surfaces.
The simplest sort of manifold is a flat one, described by linear equations. An
example is the line of slope 2 that passes through the point x = 0, y = −2: a
one-dimensional submanifold of R2
There are three equivalent ways to describe such a manifold.
• (The definition) As the graph of a function that expresses the passive vari-
ables in terms of the active variables: either y = f (x) = −2 + 2x or
x = g(y) = 12 (y + 2).
 
x
• As a “locus” defined by a constraint equation F = 2x − y − 2 = 0.
y
   
1 1
• By a parametrization function g(t) = +t .
0 2
Definition: A subset M ⊂ Rn is a smooth manifold if locally it is the graph
of a C 1 function (the partial derivatives are continuous). “Locally” means that
for any point x ∈ M we can find a neighborhood U of x such that within M ∩ U ,
there is a C 1 function that expresses n − k passive variables in terms of the other
k active variables. The number k is the dimension of the manifold. In R3 there
are four possibilities:
• k = 3. Any open subset M ⊂ R3 is a smooth 3-dimensional manifold. In
this case k = 3, and the manifold is the graph of a function
f : R3 → {~0}, whose codomain is the trivial vector space {~0} that contains
just a single point. Such a function is necessarily constant, and its derivative
is zero.
 
x
• k = 2. The graph of z = f = x2 + y 2 is a paraboloid.
y
   
x cos 2πz
• k = 1. The graph of the function = ~f (z) = is a helix.
y sin 2πz
• k = 0. In this case the manifold consists of one of more isolated points.
Near any of these points x0 , it is the graph of a function ~f : {~0} → R3
whose domain is a zero-dimensional vector space and whose image is the
point x0 ∈ R3 .
There is no requirement that a manifold be the graph of a single function, or
that the “active” variables be the same at every point on the manifold. The unit
circle, the locus of x2 + y 2 − 1 = 0, is the union of four function graphs, two of
which have x as the active variable, two of which have y. By using a parameter
t that
 is not one of the variables, we can represent it by the parametrization
x cos t
= g(t) =
y sin t

4
1.4 Using the implicit function theorem
Start with an open subset U ⊂ Rn and a C 1 function F : U → Rn−k . Consider
the “locus,” M ∩ U , the set of solutions of the equation F(z) = 0.
If [DF(z)] is onto (surjective) for every z ∈ M ∩ U , then M ∩ U is a smooth
k-dimensional manifold embedded in Rn .
Proof: the implicit function theorem says precisely this. The statement that
[DF(z)] is onto guarantees the differentiability of the implicitly defined function.
If [DF(z)] does not exist or fails to be onto, perhaps even just at a single point,
the locus is not a manifold. We use the notation M ∩ U because F may define
just part of a larger manifold M that cannot be described as the locus as a single
function. To say that M itself is a manifold, we have to find an appropriate U
and F for every point z in the manifold.

1.5 Parametrizing a manifold

For a k-dimensional submanifold of Rn , the parametrization function is γ : U →
M , where U ⊂ Rk is an open set. The variables in Rk are called “parameters.”
The parametrization function γ must be C 1 , one-to-one, and onto M . In other
words, we want γ to give us the entire manifold. Finding a local parametrization
that gives part of the manifold is of no particular interest, because there is, by
definition, a function graph that does that.
An additional requirement: The derivative of the parametrization function
is one-to-one for all parameter values. This requirement guarantees that the
columns of the the Jacobian matrix [Dγ] are linearly independent.

1.6 Tangent space as graph, kernel, or image

Locally, a k-dimensional submanifold M of Rn is the graph of a function
g : Rk → Rn−k . The derivative of g, [Dg(b)], is an (n − k) × k matrix that
converts a vector of increments to the k active variables, ẏ, into a vector of
increments to the n − k passive variables, ẋ. That is, ẋ = [Dg(b)](ẏ)
A point c of M is specified by the active variables b and the accompanying
passive variables a. The tangent space TM (c) is the graph of this derivative. It
is a k-dimensional subspace of Rn .
The k-dimensional manifold M can also be specified as the locus of the equation
F(z) = 0, for F : Rn → Rn−k The tangent space Tc M is the kernel of the linear
transformation [DF(c)].
Finally, the manifold M can also be described as the image of a parametrization
function γ : U ⊂ Rk → Rn ,
In this case any point of M is the image of some point u in the parameter
space, and the tangent space is Tγ(u) M = Img [Dγ(u)]. Whether specified as
graph, kernel, or image, the tangent space Tc M is the same! It contains the
increment vectors that lead from c to nearby points that are “almost on the
manifold.”

5
1.7 Critical points
Suppose that function f : Rn → R is differentiable at point x0 and that the
derivative [Df (x0 )] is not zero. Then there exists a vector ~v for which the direc-
tional derivative is not zero, the function g(t) = f (x0 + t~v − f (x0 ) has a nonzero
derivative at t = 0, and, even if we just consider points that lie on a line through
x0 with direction vector ~v, the function f cannot have a maximum or minimum
at x0 . So in searching for a a maximum or minimum of f at points where it is
differentiable, we need to consider only “critical points” where [Df (x0 ] = 0.
A critical point is not necessarily a maximum or minimum, but for f : Rn → R
there is a useful test that generalizes the second-derivative test of single-variable
calculus. The proof relies on sections 3.3-3.5 of Hubbard, which we are skipping.
Form the “Hessian matrix” of second partial derivatives (Hubbard, p. 348),
evaluated at the critical point x of interest.

Hi,j (x) = Di Dj f (x).

H is a symmetric matrix. If it has a basis of eigenvectors and none of the

eigenvalues are zero, we can classify the critical point.
If H has a basis of eigenvectors, all with positive eigenvalues, the critical point
is a minimum.
If H has a basis of eigenvectors, all with negative eigenvalues, the critical
point is a maximum.
If H has a basis of eigenvectors, some with positive eigenvalues and some with
negative eigenvalues, the critical point is a saddle: it is neither a maximum or a
minimum.

1.8 Constrained critical points

These are of great important in physics, economics, and other areas to which
mathematics is applied.
Consider a point c on manifold M where the function f : Rn → R is differen-
tiable. Perhaps f has a maximum or minimum at c when its value is compared to
the value at nearby points on M , even though there are points not on M where f
is larger or smaller. . In that case we should not consider all increment vectors,
but only those increment vectors ~v that lie in the tangent space to the manifold.
The derivative [Df (c)] does not have to be the zero linear transformation, but
it has to give zero when applied to any increment that lies in the tangent space
Tc M , or
Tc M ⊂ Ker[Df (c)].
When manifold M is specified as the locus where some function F = 0, there
is an ingenious way of finding constrined critical points by using “Lagrange mul-
tipliers,”but not this week!

6
1.9 Constrained critical points - three approaches
We have proved the following:
If M ⊂ Rn is a k-dimensional manifold, and c ∈ M ∩ U is a local extremum of f
restricted to M , then Tc M ⊂ ker[Df (c)].
Corresponding to each of the three ways that we can “know” the manifold
M, there is a technique for finding the critical points of f restricted to M .

• Manifold as a graph
Near the critical point, the passive variables x are a function g(y of the
active variables y. Define the graph-making function
 
x
g̃(y) =
y

Now f (g(y) specifies values of f only at points on the manifold. Just search
for unconstrained critical points of this function by setting [Df ◦ g̃(y)] = 0.
This approach works well if you can represent the entire manifold as a single
function graph.

• Parametrized manifold
Points on the manifold are specified by a parametrization γ(u).
Now f (γ(u)) specifies values of f only at points on the manifold. Just search
for unconstrained critical points of this function by setting [Df ◦ γ(u)] = 0.
This approach works well if you can parametrize the entire manifold.

• Manifold specified by constraints

Points on the manifold all satisfy the constraints F(x) = 0.
In this case we know that
Tc M = Ker[DF(c)], so the rule for a critical point becomes
Ker[DF(c)] ⊂ Ker[Df (c)].
If there is just a single constraint F (x) = 0, both derivative matrices consist
of just a single row, and we can represent the condition for a critical point
as Ker α ⊂ Ker β.
Suppose that ~v ∈ ker α and that β = λα. The quantity λ is called a
Lagrange multiplier. Then by linearity, [Df (c)]~v = β~v = λα~v = 0.
So [Df (c)]~v = 0 for any vector in the tangent space of F = 0, and we have
a constrained critical point.
It is not quite so obvious that the condition β = λα is necessary as well as
sufficient. We will need to do a proof by contradiction (proof 12.3).

7
2 Lecture Notes
1. Implicit functions – review of the linear case.
From linear algebra we are already familiar with the situation where a
system of linear equations has more than one solution. This situation arises
when there are fewer equations than unknowns.
We have n unknowns, n − k equations, e.g for n = 3, k = 1
2x + 3y − 4z = 0
x + 2y − 3z = 0  
2 3 −4
Create an (n − k) × n matrix: T =
1 2 −3
If the matrix T is not onto, its rows (the equations) are linearly dependent.
Otherwise, when we row reduce, we will find n − k = 2 pivotal columns
and k = 1 nonpivotal columns. We assign values arbitrarily to the “active”
variables that correspond to the nonpivotal columns, and then the values
of the “passive” variables that corresponds to the pivotal column are de-
termined.
Suppose that we reorder the unknowns so that the “active” variables come
last. Then, after we row reduce the matrix, the first n − k columns will be
pivotal. So the first n − k columns will be linearly independent, and they
form an invertible square matrix. The matrix is now of the form T = [A|B],
where A is invertible.  
~x
The solution vector is of the form ~v = , where the passive variables ~x
~z
come first, the active variables ~z come second.
Substitute this form of ~v into T ~v = ~0 and show that if we choose ~z arbi-
trarily and set ~x = −A−1 B~z, we have a solution.

Apply this technique to the given matrix and thereby get a solution where
“passive” x and y are expressed in terms of “active” z.

Our system of equations determines ~x “implicitly” in terms of ~z.

8
2. The implicit function theorem
Let W be an open subset of Rn , and let F : W → Rn−k be a C 1 mapping
such that F(c) = 0. Assume that [DF(c)] is onto.
Prove that the n variables can be ordered so that the first n − k columns
of [DF(c)] are linearly independent, and that [DF(c)] = [A|B] where A is
an invertible (n − k) × (n − k) matrix.
 
a
Set c = , where a are the n − k passive variables and b are the k active
b
variables.
Let g be the “implicit function”
from aneighborhood of b to a neighborhood
g(y)
of a such that g(b) = a and F = 0.
y
Prove that [Dg(b)] = −A−1 B.

9
3. Implicit functions – three variables,
  one equation
x
The nonlinear equation F y  = x2 − 4z 2 − 4y 2 − 1 = 0 implicitly deter-
z
mines x as a functionp of y and z. p
If x is positive, x = 1 + 4y + 4z 2 ; if x is negative, x = − 1 + 4y 2 + 4z 2 .
2

Near the
 point x = 3, y = 1, z = 1, which is the appropriate explicit function
y
x=g ?
z
There are three ways, in this simple example, to find the derivative of the
implicitly defined function g. Two of them take advantage of the fact that
there is only one constraint equation and only one active variable.

(a) Solve algebraically to find a formula for the function g that expresses
the passive variable x in terms of active variables y and z.
 
1
Then calculate [Dg ] for the appropriate function.
1

(b) Use the implicit function theorem to calculate the partial derivatives
of g from the partial derivatives of F :
Calculate the Jacobian matrix [DF ] at x = 3, y = 1, z = 1.

Split off a squarematrix

 A on the left, so that [DF ] = [A|B], and
1
confirm that [Dg ] = −A−1 B.
1

 
y
(c) Since there is only one passive variable x, just replace x by g to
z
   2
y y
get h =g − 4z 2 − 4y 2 − 1 = 0.
z z
 
1
Take the partial derivative D1 h , using the chain rule, and solve
  1
1 ∂g
for D1 g . In this context the notation ∂y is standard.
1

10
4. Finding the derivative of an implicitly defined function
The plane x + 2y − 3z + 4 = 0 andthecone x2 + y 2 − z 2 = 0 intersect in a
3
curve that includes the point c = 4. Near that point this curve is the
  5
x
graph of a function = g(z).
y
Use the implicit function theorem to determine g0 (5), then find the approx-
imate coordinates of a point on the curve with z = 5.1.
Check: 2.9+2(4.2) - 3(5.1)+4=0; 2.92 + 4.22 − 5.14 = 0.04.

11
5. Curves, Surfaces, Graphs, and Manifolds

Manifolds are a generalization of smooth curves and surfaces to an arbitrary

number of dimensions.
The simplest sort of manifold is a flat one. Consider an affine subset
X ⊂ Rn . This subset could be a subspace, or it could be a subspace plus a
constant, like a line or plane that does not include the origin.
There are three equivalent ways to describe such a manifold. The first will
generalize to the definition of a manifold, but the second and third are more
common and useful. Try them out on the line of slope 2 that passes through
the point x = 0, y = −2.

(a) As the graph of a function. There are two ways to do this!

 
x
(b) As the locus defined by an equation F = 0.
y

(c) By a parametrization in term of a point and a direction vector, using

a parameter t.

12
Graphs and manifolds in R3 and Rn
   
x   x
Example 1: Consider the graph of f = 2 3 −4
y y
This graph is the plane 2x + 3y − z = 4. It is a subset of the space R3
whose dimension, n = 3, is the sum of the dimension of the domain of f ,
k = 2, and the dimension of the codomain of f , n − k = 1.
Describe this manifold as the locus of a function F : R3 → R.

Invent a parametric description of this manifold.

   
x 3 + 2z
Example 2: Consider the graph of = f (z) =
y 3−z
This graph is a line.
Invent a parametric description of this manifold.

Invent a description of this manifold as a locus.

Again, the manifold is a subset of the space R3 whose dimension, n = 3, is

the sum of the dimension of the domain of f , k = 1, and the dimension of
the codomain of f , n − k = 2.
Make the function f : Rk → Rn−k arbitrary and we have a graph Γ:
the set of pairs (x, y) such that y = f (x). This is a subset of Rn .
Require that the function be C 1 (continuously differentiable) and we almost
have a manifold. But it is asking too much for a single function always to
do the job.

13
6. Definition, and examples in R3
A subset M ⊂ Rn is a smooth manifold if locally it is the graph of a C 1
function (the partial derivatives are continuous). “Locally” means that for
any point x ∈ M we can find a neighborhood U of x such that within
M ∩ U , there is a C 1 function that expresses n − k variables in terms of the
other k. The number k is the dimension of the manifold. Finding such a
function is precisely the job of the implicit function theorem.

• k = 3. Any open subset M ⊂ R3 is a smooth 3-dimensional manifold.

In this case k = 3, and the manifold is the graph of a function
f : R3 → {~0}, whose codomain is the trivial vector space {~0} that
contains just a single point. Such a function is necessarily constant,
and its derivative is zero.
Why does M have to be open?

 
x
• k = 2. Consider the graph of f = x2 + y 2 .
y
What is the name of this surface?

 
cos 2πz
• k = 1. Consider the graph of the function ~f (z) = What is
sin 2πz
the name of this curve?

• k = 0. In this case the manifold consists of one of more isolated points.

Near any of these points x0 , it is the graph of a function ~f : {~0} → R3
whose domain is the trivial vector space and whose image is the point
x0 ∈ R 3 .
This function is differentiable. Its derivative is the zero function.
Whatever  > 0 you choose, ~f (~0 + ~h) − ~f (~0) <  for all ~h, since
the only possible ~h is ~0.

14
7. Manifolds defined by geometry

For the unit circle, we need four functions to implement the preceding
definition.
Show how to break up the circle into four pieces, each of which is a function
graph.

What single equation can characterize the unit circle as a locus?

For the unit sphere, we need six functions in place of the single equation
x2 + y 2 + z 2 = 1.
How would you slice up an apple to illustrate this?

15
8. Other examples of manifolds

• An “indifference” surface. Let x = days of vacation per year, y =

fraction of health insurance paid by the company, z = number of
shares of stock options available
  for purchase annually. The manifold
x
M consists of all points y  that employees regard as neither better

  z
15
nor worse than 0.5 

100
• We are given as F three equations involving five variables, so that
n = 5, k = 2. In this case M ∩ U might be part of a two-dimensional
manifold, a surface in R5 . In economics this is easy to picture: choose
five variables that specify how the Federal budget is divided up among
Cabinet departments, and invent three equations that impose con-
straints specified by the Pentagon, the White House, and the Congres-
sional budget office. Three of the budget variables become “passive”:
they are functions of the other two “active” variables.
• The set of equilibrium states of one mole of helium gas confined to a
cylinder fitted with a movable piston. There are many pairs of vari-
ables that specify the state (e.g. pressure and volume, or temperature
and entropy). This is a 2-dimensional manifold, but it is not an obvi-
ous subset of some Rn .
• The set of positions for a pair of particles joined by a rigid rod of length
l. Specify x1 , y1 ,and x2 , with |x2 − x1 | ≤ l, and you can calculate y2 ,
but you need to know whether y2 − y1 is positive or negative. Make a
sketch to illustrate the problem.

• The configuration of a set of four particles joined into a quadrilateral

by four rigid rods. This instructive example is discussed in detail on
pp. 295-296 of Hubbard, and it provides the basis for some interesting
homework problems. Sketch an example to show how the location of
one joint depends on the others, but only locally.

16
9. Identifying a smooth manifold by the implicit function theorem

This is Hubbard Theorem 3.1.10.

We have a subset M ⊂ Rn defined as a “locus” and would like to show that
M is a smooth k-dimensional manifold.
Suppose there is an open subset U ⊂ Rn and a C 1 function F : U → Rn−k
such that the “locus,” the set of solutions of the equation F(z) = 0, is
M ∩ U.
If [DF(z)] is onto (surjective) for every z ∈ M ∩ U , then M ∩ U is a smooth
k-dimensional manifold embedded in Rn .
Make a sketch to illustrate M ∩ U .

To say that M itself is a manifold, we have to find such a U for every point
z in the manifold, perhaps with different functions at different points.
Suppose that M consists of a circle of radius 1 whose center is the origin
and a circle of radius 1 centered at x = 3, y = 0. What pair of choices for
F and U will define the entire manifold?

Proof: the implicit function theorem says precisely this. The statement
that [DF(z)] is onto guarantees the differentiability of the implicitly defined
function.

17
10. Testing that a locus is a manifold (Example 3.1.11)
 
x
F = x8 + 2x3 + y + y 5 − c = 0. The locus is a smooth manifold, which
y
looks surprisingly normal when plotted for various values of c.
Prove that [DF ] is onto for any value of c.

If you want to represent the manifold as a graph, which should be the active
variable?

Is there a function y = f (x) that describes the manifold? Can you find a
formula for it?

18
11. Testing that a locus is a manifold (Example 3.1.12)
 
x
F = x4 + y 4 + x2 − y 2 = c. Construct [DF ], and find the three points
y
where it fails to be onto because both partial derivatives vanish. For each
point, choose c so that the point is on the locus.

Here is a plot, generated in R, for various positive values of c.

Here is a plot, generated in R, for various negative values of c.

19
12. Parametrizing a manifold

It the spring term we will need to integrate over manifolds in order to

evaluate the line integrals and surface integrals that are crucial in physics.
The only practical way to do this is by parametrizing the manifold as the
image of an open set in Rk . The variables in Rk are called “parameters,”
and the functions on the manifold that give the parameter values are called
“coordinate functions.”
What parameters do geographers use for the surface of the Earth?

Here are the strict requirements for the parametrization of a k-dimensional

manifold M in Rn , as given in Hubbard, Definition 3.1.18.

• The parametrization function is γ : U → M , where U ⊂ Rk is an open

set.
What problem with differentiability would arise if U were not open?

• The parametrization function γ is C 1 , one-to-one, and onto M . In

other words, we want γ to give us the entire manifold. Finding a local
parametrization that gives part of the manifold is of no particular
interest, because there is, by definition, a function graph that does
that.
• The derivative of the parametrization function is one-to-one for all
parameter values.
If k < n, does [Dγ] have more rows or more columns? Could it be
onto in this case?

What does this requirement say about the columns of [Dγ]?

If k = 1 (a smooth curve), how can [Dγ] fail to be one-to-one?

20
13. Examples of parametrizations, with some problems

• A circle, traced out by a particle that goes around in 2π seconds,

parametrized by time. The notation used is standard in physics texts.
 
cos t
~r(t) = , 0 ≤ t < 2π
sin t
What requirement is violated, and why is there no way to fix the
problem?

• Polar coordinates to parametrize the manifold R2 .

   
x r cos θ
= , 0 ≤ θ < 2π, r ≥ 0
y r sin θ
Where is the requirement that [Dγ] is one-to-one not met? What is
the related problem with the coordinate function that assigns θ to each
point in the plane?

• Part of the unit sphere (a surface), parametrized by longitude θ and

latitude φ. The choice of φ is standard in geography but not in physics.
 
  cos θ cos φ
θ π
~r =  sin θ cos φ  , 0 < θ < π, 0 < φ <
φ 2
sin φ
What part of the sphere is the manifold given by this parametrization?

What problems occur if we try to use this parametrization for the entire
sphere?

21
14. Graph of a derivative

The concept of the tangent line to the graph of a function f : R → R is

familiar from single-variable calculus. In general, a tangent line is just an
affine subset, not a vector space. Similarly, the tangent plane to the graph
of a function f : R2 → R is an affine subset but not a vector space.
In either case, if the graph happened to pass through the origin, the tangent
line or plane would be a vector space.
Locally, a k-dimensional submanifold M of Rn is the graph of a function
g : Rk → Rn−k . We can denote the vector of k active variables by b and
the vector of n − k passive variables by a = g(b).
For consistency with the implicit function theorem, we put the active vari-
ables last, so a point of the manifold is
 
g(b)
c=
b
The derivative of the function g for active variables b, [Dg(b)],
is an (n − k) × k matrix. It converts a vector of increments to the k active
variables, ẏ, into a vector of increments to the n − k passive variables, ẋ.
That is, ẋ = [Dg(b)](ẏ)
Previously we used ~h to represent the vector of increments on which a
derivative acts, but now we have adopted Hubbard’s convention of using ẏ
for the active “input” to the derivative and ẋ for the passive “output.”
Prove that the graph of the derivative is a k-dimensional subspace of Rn .
This subspace is called the tangent space.

22
15. Tangent space as a graph (Hubbard Definition 3.2.1)
Given a point on a manifold, represent the manifold locally as a function
graph. Then the tangent space is the graph of the derivative of this function.
It is true, but not immediately obvious from the definition, that the tangent
space comes out the same if we make a different choice of active and passive
variables. Try a special case: for example, the circle of radius 5 at the point
x = 3, y = 4.
What happens if you express x as a function of y?

What happens if you express y as a function of x?

Consider point c on a manifold M . Call the active variables b and the

passive variables a. The manifold is locally the graph of x = g(y).
Distinguish carefully:
ẋ = [Dg(b)]ẏ is the equation whose graph is the tangent space.
x − a = [Dg(b)](y − b)is the equation of the tangent plane.
Example: The graph of x = y 2 + z 2 is a paraboloid, centered on the x axis.
Find the equation of the tangent space at x = 5, y = 2, z = 1.

Find the equation of the tangent plane at x = 5, y = 2, z = 1.

23
16. The equal-dimension lemma
If subspaces V and W both have dimension k and V ⊂ W, then V = W .
Proof: Choose k basis vectors for V . Since V ⊂ W, these vectors are also k
independent vectors in W and therefore also form a basis for W . Thus any
vector in W is a linear combination of these vectors and is also a vector in
V, i.e., W ⊂ V . But V ⊂ W and W ⊂ V means V = W .

17. Tangent space as a kernel

Suppose that a k-dimensional manifold M is specified as the locus of the

equation F(z) = 0 for a function F whose domain is an open subset of Rn
and whose codomain is Rn−k . The implicit function theorem says that such
a function is a satisfactory way to describe a manifold at any point c where
[DF(c)] is onto.
To prove: the tangent space Tc M is the kernel of the linear transformation
[DF(c)].
The proof (your proof 12.1) that this method of finding the tangent space
is equivalent to the definition is on pp. 310-311 in Hubbard.

• Write the matrix [DF(c)] as [A|B], with the columns that correspond
to passive variables coming first. Given that [DF(c)] is onto, the
square matrix A will be invertible.
• The dimension of the image of [DF(c)] is n − k. So the dimension of
Ker [DF(c)], by the rank-nullity theorem, is n − (n − k) = k.
• The derivative of the function g(b) that expresses the passive variables
a locally in terms of the active variables b is given by the implicit
function theorem:
[Dg(b)] = −A−1 B.

 
ẋ
• A vector in the tangent space can be written as ż = , where ẋ and
ẏ
ẏ areincrements
 to passive and active variables respectively. Suppose
ẋ
that is an element of Ker [DF(c)].
ẏ
 
ẋ
Thus [A|B] = Aẋ + B ẏ = 0.
ẏ
So ẋ = −A−1 B ẏ = [Dg(b)]ẏ.
We have established that Ker [DF(c)] is a subspace Tc M . But both
these spaces have dimension k, so they are equal.

24
18. Tangent space as a kernel (Proof 12.1 - Hubbard Theorem 3.2.4)
Suppose that U ⊂ Rn is an open subset, F : U → Rn−k is a C 1 mapping,
and manifold M can be described as the set of points that satisfy F(z) = 0.
Use the implicit function theorem to show that if [DF(c)] is onto for c ∈ M ,
then the tangent space Tc M is the kernel of [DF(c)]. You may assume that
the variables have been numbered so that when you row-reduce [DF(c)],
the first n − k columns are pivotal.

25
19. Tangent space as an image (Hubbard, Proposition 3.2.7)
Let U ⊂ Rk be open, and let γ : U → Rn be a parametrization of manifold
M . Show that

Tγ(u) M = Img[Dγ(u)].

20. Summary - three ways to characterize the tangent space of manifold M

• If M is represented
 as a function graph x = g(y), the tangent space
g(b)
Tc M at c = is the graph of its derivative and ẋ = [Dg(b)]ẏ.
b
• If M is represented as the locus of F(z) = 0, the tangent space at
z = c isthe kernel of the derivative of F, Tc M = Ker [DF(c)], and
ẋ
[DF(c)] = 0.
ẏ
• If M is represented by a parametrization function γ : U → Rn , the
the tangent space at c = γ(u) is the image of the derivative of γ:
Tγ(u) M = Img[Dγ(u)].

26
21. Exploring a manifold
A cometary-exploration robot is fortunate enough to land on an ellipsoidal
comet whose surface is described by the equation

y2 z2
x2 + + = 9.
4 9
Its landing point is x = 2, y = 4, z = 3.

• Prove that the surface of the comet is a smooth manifold.

• The controllers of the robot want it to move to a nearby point on the
surface where y = 4.02, z = 3.06. Use the implicit function theorem to
determine the approximate x coordinate of this point.
(Check: 1.982 + 4.022 /4 + 3.062 /9 = 9.0009.)
• Find a basis for the tangent space at the landing point.
• Find the equation of the tangent plane at the landing point.
(Check: 4(1.98) + 2(4.02) + (2/3)(3.06) = 18.)

27
22. Manifolds – keeping track of dimension
Assume that, at the top level, there are nine categories x1 , x2 , · · · , x9 in the
Federal budget. They must satisfy four constraints:

• One simply fixes the total dollar amount.

• One comes from your political advisors – it makes the budget looks
good to likely voters in swing states.
• One comes form Congress - it guarantees that everyone can have his
or her earmarks.
• One comes from the Justice Department – it guarantees compliiance
with all laws.

These four constraints together define a function F whose derivative is onto

for budgets that satisfy the constraints. The acceptable budgets, for which
F(x) = 0, form a k-dimensional submanifold M of Rn .
Specify the dimension of the domain and codomain for

(a) A function g that specifies that passive variables in terms of the active
variables.
(b) The function F that specifies the constraints.
(c) A parametrization function γ that generates a valid budget from a set
of parameters.

For each alternative, specify the shape of the matrix that represents the
derivative of the relevant function and explain how, given a valid budget c,
it could be used to find a basis for the tangent space Tc M.

28
23. Necessary condition for an extremum (maximum or minimum)
Recall the rule for a function of one variable:
Let U be an open interval and f : U → R a differentiable function. If x0 is
an extremum, then f 0 (x0 ) = 0.
In Rn the necessary condition for a differentiable function to have an ex-
tremum is the same.
If U ⊂ Rn is open and f : U → R is a differentiable function, then at an
extremum x0 of f , [Df (x0 )] = 0.
Define the function g(t) = f (x0 + tei ).
By the chain rule, g 0 (0) = [Df (x0 )]ei = Di f (x0 ).
Since g(t) has an extremum at t = 0, g 0 (0) = Di f (x0 ) = 0. Illustrate this
argument with a diagram for R2 .

Therefore each partial derivative of f is zero. Since f is differentiable, its

derivative is zero at the extremum.

24. Finding critical points and extrema

If U ⊂ Rn is open and f : U → R is a differentiable function,
a critical point of f is any point where its derivative [Df ] vanishes.
We just proved that an extremum of f must be a critical point. The converse
is not true: it is common to have a critical point that is not an extremum.
Show that for f (x) = x3 , x = 0 is a critical point but not an extremum.

 
x
Example: f = xy.
y
Evaluate [Df ] at the origin.

Find points near the origin where f > 0 and where f < 0.

29
25. Finding and classifying critical points
For a function defined on an open subset of Rn , finding and classifying
critical points is just an exercise in differentiation and algebra: evaluate all
the partial derivatives, set them equal to zero, and try to solve the resulting
set of (generally nonlinear) equations.
Here is a test for a critical point of a C 2 function to be a maximum or a
minimum that generalizes what you learned in single-variable calculus. The
proof relies on sections 3.3-3.5 of Hubbard, which we are skipping.
Form the “Hessian matrix” of second partial derivatives (Hubbard, p. 348),
evaluated at the critical point x of interest.

Hi,j (x) = Di Dj f (x).

H is a symmetric matrix. If it has a basis of eigenvectors and none of the

eigenvalues are zero, we can classify the critical point.
If H has a basis of eigenvectors, all with positive eigenvalues, the critical
point is a minimum.
If H has a basis of eigenvectors, all with negative eigenvalues, the critical
point is a maximum.
If H has a basis of eigenvectors, some with positive eigenvalues and some
with negative eigenvalues, the critical point is a saddle: it is neither a
maximum or a minimum.
Why this test works (not quite a proof):
Suppose first that f (~x) = f (~0) + 12 λ1 x21 + 12 λ2 x22 + · · · + 21 λn x2n .

Calculate [Df (~0)] and H.

Show that ~0 is a critical point and that the test is correct in this case.

30
Now suppose that all the eigenvalues are real and distinct.
Why is H symmetric?

Prove that any two eigenvectors ~vi and ~vj are orthogonal.

So we can make an orthonormal basis of eigenvectors. Relative to this basis,

the situation is the same as on the preceding page.
Shortcomings of this argument:

• It assumes that the critical point is at the origin.

• It does not deal with the case where there are fewer than n eigenvalues.
• It applies only to quadratic functions, and we have not shown that
any C 2 function can be approximated by such a function.
• It ignores the case where an eigenvalue is zero.

Quick test in R2 .
Here det H = λ1 λ2 . What can we conclude if

• det H < 0?
• det H > 0?
• det H = 0?

31
26. Critical points
 
x
f = 21 x2 + 13 y 3 − xy
y
Calculate the partial derivatives
  as
 functions
 of x and y, and show that the
0 1
only critical points are and
0 1

Calculate the Hessian matrix H and evaluate it numerically at each critical

point to get matrices H0 and H1 .

 
0
Find the eigenvalues of H0 and classify the critical point at .
0

 
1
Find the eigenvalues of H1 and classify the critical point at .
1

32
27. (Proof 12.2 – Hubbard, theorems 3.6.3 and 3.7.1)
Let U ∈ Rn be an open subset and let f : U → R be a C 1 (continuously
differentiable) function.
First prove, using a familiar theorem from single-variable calculus, that if
x0 ∈ U is an extremum, then [Df (x0 )] = [0].
Then prove that if M ⊂ Rn is a k-dimensional manifold, and c ∈ M ∩ U is
a local extremum of f restricted to M , then Tc M ⊂ ker[Df (c)].

33
28. Constrained critical points - three approaches
We have proved the following:
If M ⊂ Rn is a k-dimensional manifold, and c ∈ M ∩ U is a local extremum
of f restricted to M , then Tc M ⊂ Ker[Df (c)].
Corresponding to each of the three ways that we can “know” the manifold
M, there is a technique for finding the critical points of f restricted to M .

• Manifold as a graph
Near the critical point, the passive variables x are a function g(y) of
the active variables y. Define the graph-making function
 
x
g̃(y) =
y

Now f (g(y)) specifies values of f only at points on the manifold.

Just search for unconstrained critical points of this function by setting
[Df ◦ g̃(y)] = 0. This approach works well if you can represent the
entire manifold as a single function graph.
• Parametrized manifold
Points on the manifold are specified by a parametrization γ(u).
Now f (γ(u)) specifies values of f only at points on the manifold.
Just search for unconstrained critical points of this function by setting
[Df ◦ γ(u)] = 0. This approach works well if you can parametrize the
entire manifold.
• Manifold specified by constraints
Points on the manifold all satisfy the constraints F(x) = 0.
In this case we know that
Tc M = Ker[DF(c)], so the rule for a critical point becomes
Ker[DF(c)] ⊂ Ker[Df (c)].
If there is just a single constraint F (x) = 0, both derivative matrices
consist of just a single row, and we can represent the condition for a
critical point as Ker α ⊂ Ker β.
Suppose that ~v ∈ ker α and that β = λα. The quantity λ is called a
Lagrange multiplier. Then by linearity, [Df (c)]~v = β~v = λα~v = 0.
So [Df (c)]~v = 0 for any vector in the tangent space of F = 0, and we
have a constrained critical point.
It is not quite so obvious that the condition β = λα is necessary as
well as sufficient. We will need to do a proof by contradiction (proof
12.3).

34
29. (Hubbard, theorem 3.7.5 - proof 12.3 is the special case where m = 1)
~
Let M be a manifold known by a real-valued C 1 function F(x) = 0, where
~ n m
F goes from an open subset U of R to R and [DF(x)] is onto.
Let f : U → R be a C 1 function.
Prove that c ∈ M is a critical point of f restricted to M if and only if there
exist m Lagrange multipliers λ1 , · · · , λm such that
[Df (c)] = λ1 [DF1 (c)] + · · · + λm [DFm (c)].

35
30. Two approaches to an elementary maximiztion problem
Farmer Brown wants to build a rectangular pigpen using fencing of total
length 20 meters. One side of the pigpen is his barn, so the width x (side
parallel to the barn) and depth y (other two sides) are constrained to lie
on the manifold x + 2y = 20.  
x
What choice of x maximizes the area f = xy?
y
• Solve the problem by elementary methods, using y as the active vari-
able.
• Solve the problem by using Lagrange multipliers.

36
31. Using a parametrization

What rectangle inscribed in the ellipse x2 +4y 2 = 4 has the greatest perime-
ter 4(x + y)?
Solve the problem by using the parametrization
   
x 2 cos t
= ~γ (t) = ,
y sin t

then get the same solution by using Lagrange multipliers.

37
32. (This problem is equivalent to the derivation of the “Boltzmann factor”
E
e− kT that you may have heard of in a chemistry or physics course, but I
have rewritten it so that there is no mention of entropy or probability.)
You have taken over from the Postal Service the task of sorting mail in
Cambridge. You have 7 million pieces of mail to sort, and you must de-
cide how to divide it among your three “sortation centers.” Center 1, an
abandoned post office, is rent-free. Center 2, rented from UPS, charges
1 kilobuck for every million pieces of mail that you sort there, Center 3,
rented from Harvard, charges 2 kilobucks for every million pieces of mail
that you sort there. You are willing to pay 4 kilobucks of rent.
So your constraints are x1 + x2 + x3 = 7 and x2 + 2x3 = 4.
The
 total
 effort required to do the sorting is
x1
f x2  = x1 log x1 + x2 log x2 + x3 log x3 (natural logarithms).
x3
Using two Lagrange multipliers, find the values of x1 , x2 , x3 that minimize f
while satisfying the two constraints. Hint: if you let t = xx12 you can reduce
the problem to solving a quadratic equation for t.

38
3 Group Problems
If your group has R skills, choose 4a or 4b as your third problem

1. Implicitly defined functions

 
x  2 
x + y2 + z2 − 3
(a) The nonlinear equation F y =  = 0 implicitly
x2 + z 2 − 2
z
determines x and y as a function of z. The first equation describes a
sphere of radius 3, the second describes a cylinder of radius 2 whose
axis is the y-axis. The intersection is a circle in the plane y = 1.
Near the point x = 1, y = 1, z = 1, there is a function that expresses
the two√ passive variables
 x and y in terms of the active variable z.
2−z 2
g(z) = .
1
Calculate g0 (z) and determine the numerical value of g0 (1)
Then get the same answer without using the function g by forming
the Jacobian matrix [DF] evaluating it at x = y = z = 1, and using
the implicit function theorem to determine g0 (z) = −A−1 [B].
(b) Dean Smith is working on a budget in which he will allocate x to the
library, y to pay raises, and z to the Houses. He is constrained.
The Library Committee, happy to see anyone get more funds as long
as the library does even better, insists that x2 − y 2 − z 2 = 1.
The Faculty Council, content to see the Houses do well as long as other
areas benefit equally, recommends that x + y − 2z = 1.
To comply with these constraints, the dean tries x = 3, y = 2, z = 2.
Given theconstraints,
 x and y are determined by an implicitly defined
x
function = g(z).
y
Use the implicit function theorem to calculate g0 (2), and use it to find
approximate values of x and y if z increased to 2.1.

39
2. Critical points
 
x
(a) i. Find the one and only critical point of f = 4x2 + 12 y 2 + x82 y
y
on the square 14 ≤ x ≤ 4, 41 ≤ y ≤ 4.
ii. Use second derivatives (the Hessian matrix) to determine whether
this critical point is a maximum, minimum, or neither.
 
x
(b) The function F = x2 y − 3xy + 12 x2 + y 2 has three critical points,
y
two of which lie on the line x = y. Find each and use the Hessian
matrix to classify it as maximum, minimum, or saddle point.

3. Lagrange Multipliers

(a) Example with a two-dimensional manifold.

At what point on the sphere x2 + y 2 + z 2 = 7 does the function xy 2 z 4
have a maximum?
A useful trick with this sort of function is again to take the logarithm
– a monotone function, so it has the same critical points.
 
x
f y  = log x + 2 log y + 4 log z is the function to be maximized.
z
 
x
F y  = x2 + y 2 + z 2 − 7 is the constraint.

z
Set [Df ] = λ[DF ] and solve.
(b) Reversing the function to be optimized and the constraint
You are building office space and have contracted to supply 10 units of
space, half at the end of year 1 and half at the end of year 2. Because
large building projects are inefficient, the cost of building x units of
space is x2 . However, if you produce more than 5 units in the first
year, you can rent out the excess space during the second year at 8
units of money per unit of space.It might be optimal to do something
like producing x = 6 in the first year, renting out 1 unit for a year,
and producing only y = 4 in the second year.
i. Write down the function of x and y to be minimized and the
constraint, then use a Lagrange multiplier to find the optimal
amount to produce during the first year.
ii. When you use Lagrange multipliers, the cost function and the con-
straint are almost interchangeable. Invent a problem that involves
maximizing a linear function and that has the same solution as
the original problem.

40
4. Manifolds and tangent spaces, investigated with help from R

(a) Manifold
  M is known by the equation  
x 4
F y  = xz − y 2 = 0 near the point c = 2.
z 1
It can also be
 2 described parametrically by
  s
s
γ = st2  near s = 2, t = 1.
t
t4
i. Use the parametrization to find a basis for the tangent space Tc M.
ii. Use the function F to confirm that your basis vectors are indeed
in the tangent space Tc M.
iii. Use the parametrization to do a wireframe plot of the parametrized
manifold near s = 2, t = 1. See script 3.4C, topic 2.
 
z1  
z3
(b) (Hubbard, Example 3.1.14) F z2  =
z3 − z1 z2
z3
Construct [DF]. It has two rows.
Find the point for which [DF] is not onto. Use R to find points on
the manifold near this point, and try to figure out what is going on.
See the end of script 3.4C for an example of how to find points on a
1-dimensional manifold in R3 .

41
4 Homework - due on Wednesday, December 7
Although all of these problems except the last one were designed so that they
could be done with pencil and paper, it makes sense to do a lot of them in R,
and the Week 12 scripts provide good models. For each problem that you choose
to do in R, include a “see my script” reference in the paper version. Put all your
R solutions into a single script, and upload it to the homework dropbox on the
week 12 page.
When you use R, you will probably want to include some graphs that are not
required by the statement of the problem.

1. (Hubbard, exercise 3.12)

Let X ⊂ R3 be the set of midpoints of segments joining a point of the
curve C1 of equation y = x2 , z = 0 to a point of the curve C2 of equation
z = y 2 , x = 0.

(a) Parametrize C1 and C2 .

(b) Parametrize X.
(c) Find an equation for X (i.e. describe X as a locus)
(d) Show that X is a smooth surface.
 
x
2. Manifold M is known by the equation F y  = x2 + y 4 − 2z 2 − 2 = 0 near

  z
3
the point c = 1.

2
 
y
(a) Locally, near c, M is the graph of a function x = g . Determine
z
[Dg(c)] by using the implicit function theorem.
(b) Use [Dg(c)] to find the approximate value of x for a point of M near
c for which y = 1.1, z = 1.8.
(c) Check your answers by finding an explicit formula for g and taking its
derivative.

42
3. Pat and Terry are in charge of properties for the world premiere of the
student-written opera “Goldfinger” at Dunster House. In the climactic
scene the anti-hero takes the large gold brick that he has made by melting
down chalices that he stole from the Vatican Museum and places it in a
safety deposit box in a Swiss bank while singing the aria “Papal gold, now
rest in peace.”
The gold brick is supposed to have length x = 8, height y = 2, and width
z = 4. With these dimensions in mind, Pat and Terry have spent their
entire budget on 112 square inches of gold foil and 64 cubic inches of an
alloy that melts at 70 degrees Celsius. They plan to fabricate the brick by
melting the alloy in a microwave oven and casting it in a sand mold.
Alas, the student mailboxes that they have borrowed to simulate safety-
deposit boxes turn out to be not quite 4 inches wide. Fortunately, the
equation
 
x  
xyz − 64
F y =
  =0
xy + xz + yz − 56
z
specifies x and y implicitly in terms of z.

(a) Use the implicit function

  theorem to find [Dg(4)], where g is the func-
x
tion that specifies in terms of z, and find the approximate dimen-
y
sions of a brick with the same volume and surface area as the original
but with a width of only 3.9 inches.
(b) Show that if the original dimensions had been x = 2, y = 2, z = 16,
then the constraints of volume 64, surface area 136 specify y and z in
terms of x but fail to specify x and y in terms of z.
(c) Show that if the original brick had been a cube with x = y = z = 4,
then, with the constraints of volume 64, surface area 96, we cannot
show the existence of any implicit function. In fact there is no implicit
function, but our theorem does not prove that fact. This happens
because this cube has minimum surface area for the given volume.

43
4. (Physics version)In four-dimensional spacetime, a surface is specified as the
intersection of the hypersphere x2 + y 2 + z 2 = t2 − 2 and the hyperplane
3x + 2y + z − 2t = 2.
(Economics version)A resource is consumed at rate t to manufacture goods
at rates x, y, and z, and production is constrained by the equation
x2 + y 2 + z 2 = t2 − 2.
Furthermore, the expense of extracting the resource is met by selling the
goods, so that 2t = 3x + 2y + z − 2.
In either case, we have a manifold that is the locus of
 
x  2 
y  x + y 2 + z 2 − t2 + 2
F  =
  = 0.
z 3x + 2y + z − 2t − 2
t

(a) Show that this surface is a smooth 2-dimensional manifold.

(b) One point on the manifold is x = 1, y = 2, z = 3, t = 4. Near this
point the manifold is the graph of a function g that expresses x and y
as functions of z and t. Using the implicit function theorem, determine
[Dg] at the point z = 3, t = 4.

5. Consider
 the
 manifold
 specified by the parametrization

x t + et
g(t) = = , −∞ < t < ∞.
y t + e2t
Find where it intersects the line 2x+y = 10. You can get an initial estimate
by using the graph below (generated in R), then use Newton’s method to
improve the estimate.

44
6. Manifold X, a hyperboloid, can be parametrized as
   
x   sec u
y  = γ u = tan u cos v 
v
z tan u sin v
If you use R, you can do a wireframe plot the same way that the sphere
was plotted in script 3.4C, topic 2.

(a) Find the coordinates of the point c on this manifold for which
u = π4 , v = π2 .
π
(b) Find the equation of the tangent plane Tc X as the image of [Dγ π4 ].
2
 
x
(c) Find an equation F y  = 0 that describes the same manifold near
z
c, and find the equation of the tangent plane Tc X as the kernel of
[DF(c)].
 
y
(d) Find an equation x = g that describes the same manifold near
z
c, and
  find the equation of the tangent plane Tc X as the graph of
0
[Dg ].
1

7. Hubbard, Exercise 3.6.2. This is the only problem of this genre on the
homework that can be done with pencil and paper, but you must be pre-
pared to do one like it on the final exam!
The final problem, which requires R, is only for graduate-credit students.
8. Here is another function that has one maximum, one miminum, and two
saddle points, for all of which x and y are less than 3 in magnitude.
 
x
f = x3 − y 3 + 2xy − 5x + 6y.
y
Locate and classify all four critical points using R, in the manner of script
3.4D. A good first step is to plot contour lines with x and y ranging from
-3 to 3. If you do
contour(x,y,z, nlevels = 20)
you will learn enough to start zooming in on all four critical points.
An alternative, more traditional, approach is to take advantage of the fact
that the function f is a polynomial. If you set both partial derivatives equal
to zero, you can eliminate either x or y from the resulting equations, then
find approximate solutions by plotting a graph of the resulting fourth-degree
polynomial in x or y.

45