Honors Linear Algebra Main Problem Set Overview

Honors Linear Algebra Main Problem Set
Note: Problems are usually referenced by section and number, e.g., C6. If a problem is referenced
by a number alone, that means it is in the current section. Thus a reference to Problem 5 within
Section DA means Problem DA5.
Section A. Matrix Operations, More Theoretical
The next few problems (17) ask you to prove various basic properties of matrix operations. A
satisfactory proof must rely on denitions or earlier theorems, but at this point you have no theorems
about matrices, so you must rely on denitions. All the properties say that two matrices are equal,
so you need to apply the denition that A = B if corresponding entries are equal as numbers (e.g.,
that a
ij
= b
ij
for all i, j). If a problem is about, say, two matrix sums being equal, then you will
also need to invoke the denition of A+B, which is also an entrywise denition. Thus good proofs
of these results at this stage will almost surely be entrywise arguments.
1. Prove that A+ B = B + A (more precisely, if A and B are the same shape, say mn, so that
both sums are dened, then both sums are the same). This fact is pretty obvious, so it is a
good one for concentrating on appropriate proof style. Do what you think constitutes a good
writeup and we will discuss it in class.
2. Prove that scalar multiplication distributes over addition of matrices: k(A+ B) = kA+kB.
3. Prove that matrix multiplication distributes over matrix addition: C(A+ B) = CA +CB.
4. Prove: k(AB) = (kA)B = A(kB). We say that scalars pass through matrix multiplication
even though matrix multiplication itself does not commute.
5. Prove: a(bM) = (ab)M. This is sort of an associative law. Its not the associative law, because
that involves just one sort of multiplication and this involves two (why?).
6. Prove 1M = M. Thats 1, not I, so this theorem is about scalar multiplication. The theorem
is rather obvious, but it does show up again later.
7. Prove the Associative Law: A(BC) = (AB)C. (More precisely: For any matrices A, B, C, if
A(BC) exists, then so does (AB)C, and they are equal.)
8. If B and B
both commute with A, for each of the matrices below prove or disprove that it
must commute with A:
kB, B +B
, BB
, B
2
.
Note: Since the hypothesis (that B, B
commute) algebraicizes into a statement at the matrix

level, namely BB
= B
B, rather than at the entry level, it is reasonable to look for a proof

entirely at the matrix level. Most earlier proofs have been at the entry level.
9. Does the high school algebra identity a
2
b
2
= (a + b)(a b) generalize to nn matrices?
10. Show that
A
1
+B
1
= A
1
[A+B]B
1
.
(Just multiply out the righthand side.) This looks pretty weird but it does generalize a familiar
high school algebra fact. What fact?
S. Maurer, Math 28P, Fall 2005
page 2 Section A, Matrix Operations, More Theoretical
11. The transpose A
T
of a matrix A is obtained by interchanging rows and columns. Thus, if
A = [a
ij
] is mn, then A
T
is nm and its ji entry is a
ij
. Another name for the ji entry of A
T
is a
T
ji
, so by denition we may write A
T
= [a
T
ji
] where a
T
ji
= a
ij
. (Why call the typical entry of
A
T
the ji entry when usually we use ij? Because i has already been associated with the rows
of A and j with the columns of A. Since A is mn, then i ranges from 1 to m and j from 1
to n. By naming the typical entry of A
T
as a
T
ji
we are saying that A
T
is nm.)
Find a formula for (AB)
T
in terms of A
T
and B
T
. Hint: It helps to think of A as made up
of rows and B as made up of columns, for then
AB looks like
_
_
_
_
_
_
.
.
.
_
_
Now, what does B
T
A
T
look like?
12. Prove: Regardless of the shape of matrix A, both products AA
T
and A
T
A exist. In fact, they
are both the same shape in some way; what way?
13. a) Come up with a 22 nonzero matrix A, with all entries nonzero, and a 21 vectors b,
also with nonzero entries, such that Ab = 0.
b) For the same A, come up with a nonzero 22 matrix B, with all entries nonzero, such
that AB = 0. Show the multiplication as a check. Note what we have shown: the claim
ab = 0 = [a = 0 or b = 0],
true and so useful for real number multiplication, is false for matrix multiplication.
c) Leave your b from above alone, and modify your A slightly to obtain a C ,= I such that
Cb = b.
14. Show that the cancellation law
AC = BC = A = B
is false for matrices, even nonzero matrices. Hint: use Prob. 13.
15. Let
P =
_
1 1
1 2
_
, Q =
_
1 0
2 1
_
, R =
_
1 0
1 1
_
, S =
_
3 1
2 1
_
, A =
_
0 1
1 0
_
.
By direct calculation, verify that
PQ = RS,
but
PAQ ,= RAS.
This shows that mid multiplication doesnt work (though it does work for ordinary numbers;
why?).
S. Maurer Main 28P Problem Set, Fall 2005
page 3 Section AA, Matrix Operations, Numerical Practice
16. The trace of a matrix is the sum of its main diagonal entries. Let A, B be any two square
matrices of the same size. Prove that AB and BA have the same trace.
17. Suppose the columns of A are c
1
, c
2
, . . . , c
n
and the rows of B are r
1
, r
2
, . . . , r
n
. Show that
AB =

n
i=1
c
i
r
i
.
Section AA. Matrix Operations, Numerical Practice
1. Let the rows of B be r
1
, r
2
, r
3
. You want to premultiply B to obtain
_
_
r
1
+r
2
r
2
2r
3
r
1
+r
3
2r
2
_
_
What is the premultiplier matrix?
2. What does premultiplying by
_
_
0 0 1
1 0 0
0 1 0
_
_ do to a matrix?
3. Determine a premultiplier matrix that converts
_
_
1 1 1
0 0 0
0 1 1
_
_ to
_
_
1 0 0
0 1 1
0 0 0
_
_
by elementary row operations.
4. Determine the premultiplier matrix that does the downsweep from the rst pivot in
_
_
1 2 3
2 3 4
3 4 4
_
_
5. Suppose mystery matrix A has the property that
A
_
_
1
2
3
_
_
=
_
_
1
0
_
_ and A
_
_
4
2
1
_
_
=
_
_
0
1
_
_.
Name a right inverse of A.
6. Suppose mystery matrix A
has the property that

(1, 2, 3) A
= (1, 0) and (4, 2, 1) A
= (0, 1).
Name a left inverse of A
.
page 4 Section AA, Matrix Operations, Numerical Practice
7. Suppose mystery matrix B satises
B
_
_
1
0
_
_ =
_
_
2
3
_
_ and B
_
_
0
1
_
_ =
_
_
5
2
_
_.
a) Quickly nd X such that
BX =
_
5 4
2 6
_
.
b) Quickly nd B.
8. Suppose
C
_
1 3
2 4
_
=
_
1 0
0 1
_
.
Quickly nd x and y so that
Cx =
_
_
1
0
_
_ and Cy =
_
_
2
1
_
_.
9. If
D
_
_
4
5
_
_ =
_
_
1
1
_
_ and D
_
_
2
7
_
_ =
_
_
1
2
_
_,
Find x that satises
Dx =
_
_
1
0
_
_.
10. State and solve a row analog of Problem AA9.
11. Assume that the following chart shows the number of grams of nutrients per ounce of food
indicated. (Actually, the numbers are made up.)
meat potato cabbage
protein 20 5 1
fat 30 3 1
carbohydrates 15 20 5
If you eat a meal consisting of 9 ounces of meat, 20 ounces of potatoes, and 5 ounces of cabbage,
how many grams of each nutrient do you get? Be sure to make clear what this problem is doing
in a linear algebra course.
12. Continuing with the data from Problem 11, suppose the Army desires to use these same
delectable foods to feed new recruits a dinner providing 305 grams of protein, 365 grams of
fat, and 575 grams of carbohydrates. How much of each food should be prepared for each
recruit? Be sure to make clear what this problem is doing in a linear algebra course?
page 5 Section AB, Special Matrices
Section AB. Special Matrices
1. A matrix is symmetric if A
T
= A. Explain why every symmetric matrix is square.
2. Prove: if A is symmetric, so is A
2
. (Hint: see Problem A11.) Can you generalize?
3. Prove or disprove: the product of symmetric matrices is symmetric.
4. A matrix is antisymmetric (also called skew symmetric) if A = (A
T
). Explain why an
antisymmetric matrix must be square. What can you say about the main diagonal entries of
an antisymmetric matrix?
5. Prove or disprove: If A is skew symmetric, then its inverse (if any) is skew symmetric.
6. If A is skew symmetric, what nice property (if any) does A
2
have?
7. Let A be a 2n2n matrix. Then we may write A as
_
P Q
R S
_
, where each of P, Q, R, S is
an nn matrix. In this case, A is called a block matrix (or a partitioned matrix) and
P, Q, R, S are the blocks. Similarly, let A
=
_
P
_
. Prove
AA
=
_
PP
+QR
PQ
+QS
RP
+SR
RQ
+SS
_
; (1)
that is, for the purpose of doing multiplication, you can simply pretend that the blocks are real
numbers and use the regular rule for matrix multiplication, obtaining the product expressed in
block form.
a) Prove (1). Or rather, prove it is correct for the top left block. In other words, prove
that the nn top left corner of AA
is PP
+QR
. (The remaining parts of the proof are

similar.)
b) There is nothing special about having all the blocks be the same size. Come up with as
general a correct statement about multiplication of block matrices as you can. No proof
requested.
8. A matrix is diagonal if the only nonzero entries (if any) are on the main diagonal.
a) Prove that all nn diagonal matrices commute.
b) Determine, with proof, the set of all matrices that commute with all nn diagonal ma-
trices. (As usual, n is arbitrary but xed, so you are looking at square matrices of some
xed size.) For instance, all the nn diagonal matrices are themselves in this set, but
maybe there are other matrices in the set, even of other shapes.
9. A square matrix is lower triangular if all the nonzero entries (if any) are on or below the
main diagonal. Let L be a simple lower triangular matrix if L is lower triangular and all
diagonal entries are 1s.
a) Prove that the product of lower triangular matrices is lower triangular (pre- or postmul-
tiplication approaches can help.)
b) Prove that the product of simple lower triangular matrices is simple lower triangular.
page 6 Section AC, The Origins of Matrix Multiplication (Parallel Worlds)
Section AC. The Origins of Matrix Multiplication (Parallel Worlds)
1. The M-product of two ordered pairs, (c, d) and (a, b) in that order, is dened by
(c, d) (a, b) = (ac, bc+d).
(We use as the special symbol for this sort of product.)
a) Prove or disprove: the M-product commutes.
b) Prove or disprove: the M-product is associative.
2. In linear algebra, a function in the form y = f(x) = ax + b is called an ane function. (A
function y = (x) = ax is called linear.)
Now, if z = g(y) = cy + d is another ane function, we can compose g and f to get z as a
function of x.
a) Show that the composition of two ane functions is ane.
b) Let us represent the ane function y = ax+b by the symbol A = (a, b). If we wish products
of symbols to correspond to composition of ane functions (when the symbols are written
in the same order as the functions are written in the composition), then how must the
product of these symbols be dened? Show the work which leads to your conclusion).
c) Now go back and do AC1b another way.
3. Instead of associating the ane function y = ax + b with the symbol (a, b), for some strange
reason let its symbol be the matrix
M =
_
a b
0 1
_
.
a) Show that with this choice, the correct product of symbols is just the matrix multiplication
we have already dened!
b) Give yet another solution to AC1b.
c) It may seem strange to associate an ane function, which maps a one-dimensional line to
itself, with a matrix, which maps 2-dimensional space to itself. But the correspondence
makes sense if you just look at a certain one-dimensional slice (that is, line) in R
2
. Let S
be the slice (x, 1)
T
[ x R. thus S consists of all the points on a line, represented as
column vectors x. Conrm that the mapping x Mx maps S to S, and in just the ane
way we want.
4. The matrix world and the function world are parallel worlds. Most years we discuss this in
class. Matrices in one world correspond to multivariable linear functions (systems of equations
describing linear substitutions) in the other, and matrix multiplication corresponds to func-
tion composition. Since all function composition is associative, we may conclude that matrix
multiplication is associative
Show that + in the matrix world corresponds to + in the parallel world of functions. You
may work with linear functions from R
2
to R
2
and 22 matrices (just as on the handout
showing how function composition and matrix multiplication are parallel), but if you can, show
it for the general case (as at the end of that handout)
page 7 Section AC, The Origins of Matrix Multiplication (Parallel Worlds)
5. For functions, scalar multiplication is dened as follows. If k is a scalar and f is a function,
then kf is another function, dened by
(kf)(x) = k(f(x))
for all domain values x. E.g., if f is the sine function, then (3f)(x) = 3 sinx. If f(x) is a vector,
(kf)(x) multiplies each coordinate of f(x) by k.
Show that scalar multiplication in the world of matrices corresponds to scalar multiplication
in the parallel function world.
Remark: The conclusion from the previous two problems is: any fact about linear functions
that uses only plus, composition, and scalar multiplication is also true in the matrix world
(using plus, matrix multiplication, and scalar multiplication), and vice versa.
6. Does everything correspond perfectly between the matrix and the function worlds? Is everything
true in one world true in the other? No, because the function world is much bigger. There is
a perfect correspondence between the matrix world and the linear function world (which is the
part of the function world you actually get to from the matrix world), but there are properties
that hold for linear functions that dont hold for all functions. This problem points out one
such property. It shows that a distributive property that holds for matrices does not hold for
all functions.
Consider real functions, that is, functions with domain and range the real numbers. The
claims in the problem work for other functions too, but lets stick to this familiar setting. This
allows us to avoid a number of issues that would only divert us from the main point.
The sum of two functions f and g is dened to be the function f + g that satises, for all
real x,
(f+g)(x) = f(x) +g(x).
For instance, if f(x) = sin x and g(x) = x
2
, then (f+g)(x) = sinx + x
2
. Also, fg means the
composition of f and g. Thus
(fg)(x) = f
_
g(x)
_
.
a) Prove: for any functions f, g, h, (f+g)h = fh+gh. That is, composition right distributes
over addition.
b) Show that there is no corresponding left distributive law. Find functions f, g, h such that
f(g+h) ,= fg + fh.
Remember, two functions are equal if they agree no matter what input they are given. That
is, f = g if for all x, f(x) = g(x). Consequently, they are unequal if the outputs are dierent
for even one x input.
page 8 Section B, Gaussian Elimination, Numerical
Section B. Gaussian Elimination, Numerical
(Most problems in this section are adapted from Schaums Linear Algebra, Chapter 1. The Schaum
numbers are in parentheses.)
1. (1.15) Solve both in algebraic form and in matrix form, using Gaussian elimination exactly in
the latter case. You may get some fractions, but nothing too hard to handle.
x 2y + z = 7
2x y + 4z = 17
3x 2y + 2z = 14.
2. (1.11) Solve in algebraic form and then with Gaussian elimination. The special form of the
equations actually makes the problem easier. Why?
2x 3y + 5z 2t = 9
5y z + 3t = 1
7z t = 3
2t = 8.
3. (1.36) Solve using Gaussian elimination:
x + 2y z = 3
x + 3y + z = 5
3x + 8y + 4z = 17.
4. (1.14) Find the general solution to
x 2y 3z + 5s 2t = 4
2z 6s + 3t = 2
5t = 10.
Use Gaussian elimination, followed by the technique of moving the free (nonpivot) variables to
the right-hand side.
5. (1.21) Use Gaussian elimination to row reduce the following matrix to an echelon form, and
then to a reduced echelon form.
_
_
1 2 3 0
2 4 2 2
3 6 4 3
_
_
page 9 Section BA, GE: Echelon Form
6. (1.26) Reduce by Gaussian elimination to row-reduced echelon form:
_
_
2 2 1 6 4
4 4 1 10 13
6 6 0 20 19
_
_
7. (1.35) Solve by Gaussian elimination:
x + 2y 3z 2s + 4t = 1
2x + 5y 8z s + 6t = 4
x + 4y 7z + 5s + 2t = 8.
8. Find the unique polynomial of degree at most 3 which goes through the points (1,1), (2,3),
(3,6), (4,10). Use Gaussian elimination (but if you recognize these points as involving special
numbers, maybe you can also guess the answer). The order in which you list your unknowns
makes a big dierence in how tedious the Gaussian elimination is by hand. If you have a
calculator that solves matrix equations, it doesnt matter.
9. Suppose Gaussian elimination reduces A to
R =
_
_
1 2 0 3
0 0 1 1
0 0 0 0
_
_.
a) Name all solutions to Ax = 0.
b) But you were never told the entries of A! What theorem justies your answer to part a)?
c) Suppose further that Gaussian elimination required no row switches to reach R. Find a
vector b such that Ax = b has no solutions. Caution: If all you show about your b is
that Rx = b has no solutions, youre not done.
Section BA. GE: Echelon Form
1. Give an example of a 33 lower triangular echelon matrix. Give an example of a 33 lower
triangular row-reduced echelon matrix.
2. a) In a 67 rref (row-reduced echelon matrix) with ve pivotal 1s, at most how many entries
are nonzero?
b) At most how many entries are nonzero in an mn rref with p pivots?
3. There are innitely many 22 row-reduced echelon matrices. For instance,
_
1
0 0
_
is such a
matrix for every value of . However, when an entry can take on any value, let us agree that
dierent values do not give us a dierent reduced echelon form. All that matters is: where 1s
page 10 Section BA, GE: Echelon Form
are, where forced 0s are (the 0s in positions where the algorithm deliberately creates a 0), and
where arbitrary values are. Thus, there are exactly four 22 reduced echelon forms:
_
1 0
0 1
_
,
_
1
0 0
_
,
_
0 1
0 0
_
,
_
0 0
0 0
_
.
a) How many 33 reduced echelon forms are there? One of them is
_
_
1 0
0 1
0 0 0
_
_,
where the two s represent arbitrary values, not necessarily the same.
b) How many 34 reduced echelon forms are there?
There are smarter ways to answer these questions than brute force listing.
4. The solutions of an unaugmented matrix A are the solutions to the augmented system [A[ 0].
For instance, the solutions to
M =
_
1 1
2 2
_
are the solutions to the system
x + y = 0
2x + 2y = 0.
(1)
which, by GE or by inspection, are all ordered pairs (y, y). Equations which are all equal to
0, like (1), are called homogeneous. (Note the second e; homogenous means something
else.)
The solutions of matrices A and B below clearly have the same solutions:
A =
_
_
1 4 7
2 5 8
3 6 9
_
_, B =
_
_
2 5 8
3 6 9
1 4 7
_
_.
a) Why is it clear that A and B have the same solutions?
b) Nonetheless, the steps and matrices in Gaussian elimination will be very dierent. Carry
out both Gaussian eliminations. What is the same? Is this surprising?
5. The systems
x + y +z +w = 0
x + 2y z +w = 0
and
w z + 2y + x = 0
w + z + y + x = 0
clearly have the same solution.
a) Why is it clear that A and B have the same solutions?
page 11 Section C, Matrix Inverses
b) Nonetheless, the steps and matrices in Gaussian elimination will be very dierent. Carry
out both Gaussian eliminations. Is anything surprising?
6. Consider two mn rref matrices M, M
and suppose M has no pivot in the rst (leftmost)

column, and M
does have a pivot in that column. Prove: M and M
do not have the same

solutions. Hint: nd a specic n-tuple that is a solution to M but not to M
. Explain how you

know you are right despite knowing so little about either matrix.
This problem is a special case of a general result that some of you may have already conjec-
tured, and which will be the subject of a handout.
Section C. Matrix Inverses
1. Every real number except 0 has a multiplicative inverse. Does every square matrix which is not
all 0s have an inverse? Does every square matrix which has no 0 entries have an inverse?
2. Theorem: (A
1
)
1
= A. State precisely the if-then statement for which this is shorthand.
Then prove it.
3. Let A, B be nn and invertible. Show that AB is also invertible by showing that B
1
A
1
is
its inverse. (The natural converse If AB is invertible then so are both A and B is false,
but there are some partial inverses that are true; see CC 68.)
4. Alas, the result in Problem3 is often stated as (AB)
1
= B
1
A
1
. Why is this a bad statement?
How might it be misinterpreted?
5. Prove or disprove: If A is symmetric, then its inverse (if any) is symmetric. Remark: Since both
symmetry and inverses are dened at the matrix level, there should be a proof of this claim at
the matrix level. Heres a hint: Show that, if B meets the denition of being an inverse of A,
then so does B
T
.
6. Prove: If A is invertible and AB = 0, then B = 0.
7. Prove: if AB = 0 and B is a not an all-zeros matrix, then A is not invertible. Hint: Use proof
by contradiction, that is, suppose the hypotheses are true but the conclusion is false, and reach
a contradiction.
8. Let A be any 22 matrix, that is, A =
_
a b
c d
_
. Let = ad bc. Let B =
_
d b
c a
_
.
Prove: A is invertible i ,= 0, and when A
1
exists, it equals
1
B. (Only if is the hard

part; one approach is to show that ad bc = 0 is necessary for A to be reducible to I. Another
approach is to use Prob. 7.)
Note: There is a general formula for A
1
, but only in the case n = 2 (this problem) is
the formula easy to use, or quick to state. For n = 2 you just switch the values on the main
diagonal, negate the values on the other diagonal, and scale by 1/.
page 12 Section CA, Partial Inverses, Solutions, and Cancellation
9. Suppose A and B are invertible matrices, not necessarily the same size. Show that the block
matrix C =
_
A
B
_
is invertible. (Blank space means 0s.) Hint: Using A
1
and B
1
,
guess a block formula for C
1
and check that you are right. Another approach is to use your
knowledge of what Gaussian elimination must do to A and B to determine what it does to C.
10. Suppose A and B are mm and nn invertible matrices. Show that the block matrix
_
A C
B
_
is invertible, where C is any mn matrix.
Section CA. Partial Inverses, Solutions, and Cancellation
1. a) Show that if A has a right inverse R, then for every conformable b, the system Ax = b
has at least one solution because Rb is a solution. Method: plug it in and check. We say
that A satises the existence condition (for solutions x for all conformable b).
b) By considering the rref of A =
_
_
1 2
3 4
5 6
_
_
, argue that for this A, Ax = b does not have a
solution for every b. Can this A have even one right inverse?
2. a) Show that if A has a left inverse L, then for every conformable b, the system Ax = b has
at most one solution because Lb is the only candidate. Method: Solve Ax = b for x by
left multiplying. We say that A meets the uniqueness condition (for solutions x for all
conformable b).
b) By considering the rref of A =
_
1 3 5
2 4 6
_
, argue that for this A, Ax = b has more than
one solution for some bs. Can this A have even one left inverse?
3. Let
L =
_
1 1 1
5/2 5/2 1
_
, A =
_
_
1 2
3 4
5 6
_
_, b =
_
_
1
1
2
_
_.
a) Check that LA = I.
b) So, to solve Ax = b, just premultiply by L:
LAx = Lb
Ix = x = Lb.
Therefore, Lb should be the solution. Well, compute Lb and plug it in to Ax = b.
Does it work? If it doesnt, what went wrong? If it does, what do you know about the
existence of other solutions? If it doesnt, what do you know how many solutions Ax = b
has? (Please note there werent any powers or other nonlinear things in this problem of
the sort normally associated with extraneous roots. We solved this problem by linear
operations.)
c) In fact, matrix A of this problem has innitely many left inverses. Find them all. Heres
one way, but maybe you can nd others that are simpler. To solve XA = I, you can
just as well solve A
T
X
T
= I
T
= I; Then to solve for the unknown matrix X
T
you can
solve column by column. For instance, letting x be the rst column, you need to solve
A
T
x = (1, 0)
T
. Now you have a standard system of linear equations and can solve by
Gaussian elimination.
d) Let L
be any of the left inverses found in c) other than the L given in part a). Is L
b = Lb?
Is L
b a solution to Ax = b?
Note that this problem, along with Problem 1, show that the A of this problem has innitely
many left inverses and no right inverses.
4. In Problem CA3, you showed that the existence of a left inverse L is not enough to be sure
that Lb is a solution to Ax = b; we get uniqueness but not existence. Now you will show that
the existence of a right inverse R is not enough to be sure that Rb is the only solution; we get
existence but not uniqueness.
Let
A =
_
1 0 1
0 1 0
_
, R =
_
_
1 0
0 1
0 0
_
_, b =
_
_
p
q
_
_.
a) Check that AR = I and that x = Rb = (p, q, 0)
T
is a solution to Ax = b.
b) Show that x = b
= (0, q, p)
T
is another solution. Find innitely many others.
c) There are innitely many right inverses to A. Find them all.
5. Prove: If A is nonsquare and has a right inverse, then it has innitely many right inverses, and
no left inverse.
6. (Existence condition) Improve on Problem 1 by proving: Ax = b has at least one solution x
for each b if and only if A has a right inverse. (In this and similar problems, it is assumed that
A is a xed but arbitrary mn matrix, and each b, or every b, means each b for which
the equation Ax = b makes sense, that is, all m1 column vectors.)
7. (Uniqueness condition) Improve on Problem 2 by proving: Ax = b has at most one solution x
for each b if and only if A has a left inverse.
8. It is a theorem (hopefully we proved it already in class) that
Given matrix A, do GE on [A [ I ], obtaining [A
R
[ C ]. If A
R
= I, then A
1
= C;
else A
1
does not exist.
a) Come up with a theorem/algorithm about computing left inverses that parallels this the-
orem. Come up with means discover, then state, then prove!. Parallel means the
theorem is of the form
Given matrix A, do GE on [A [ B], obtaining [A
R
[ C ]. If A
R
. . . , then such-
and-such is a left inverse; else a left inverse does not exist.
You have to gure out the right choice of B and what goes in the dot-dot-dots and what
is such-and-such!
b) Apply your algorithm/theorem to
A =
_
_
1 2
3 4
5 6
_
_ and A
=
_
_
1 2
2 4
3 6
_
_.
9. a) Parallel to Problem 8a, come up with a theorem about computing right inverses. The
procedure for computing the right inverse from C is a bit complicated; you may wish to
assume at rst that the pivot columns of A are the leftmost columns.)
b) Apply your algorithm/theorem to
A =
_
1 3 5
2 4 6
_
and A
=
_
1 2 3
2 4 6
_
.
10. In Problems 89 all the results were stated in terms of what A reduces to by GE. In fact, all
that is necessary is that the reduction is obtained by elementary row operations. For instance,
the invertibility theorem may be generalized to say:
A is invertible i by some sequence of elementary row operations A may be reduced
to I, in which case the inverse is C, where C is what is obtained on the right if the
same sequence of row operations is applied to [A [ I ].
Explain why GE may be replaced by elementary row operations in all claims in these two
problems.
11. Denition: The left cancellation property holds for matrix C if, for any matrices A and B
that can postmultiply C,
CA = CB = A = B.
(In high school algebra this property is called a law because it is always true. For matrices
it is at best a property because it is not always true.) Prove: The left cancellation property
holds for C if and only if C has a left inverse. This claim means that
i) If C has a left inverse, then the property holds for all A, B conformable with C for
multiplication; and
ii) If C does not have a left inverse, then there exist two matrices A ,= B for which CA = CB.
(However, we do not claim that for all distinct A, B, necessarily CA = CB.)
page 15 Section CB, Extraneous Roots
Hint for ii): for each non left-invertible C it is sucient to nd an X ,= 0 for which CX = 0,
for then let A be anything and let B = A+X. For instance, if you let A = 0, then we nd that
C0 = CX = 0 but 0 ,= X. Here we follow the semi-mystical principle that problems about
equations are easier if we can make them into equations equalling 0.
12. State, interpret, and prove a theorem, similar to that in the previous problem, about the right
cancellation property.
13. In light of the previous two problems, for only what sort of matrices C can you rightfully go
about canceling it on the left and canceling it on the right?
14. For what matrices do the right and left cancellation properties hold for addition? E.g., for
which C does A+C = B + C = A = B (for all conformable A and B)?
15. Let A be mn.
a) Prove:
1) If Ax = b has at least one solution for every b, then m n (i.e., then A has at least
as many columns as rows).
b) State the contrapositive of 1). (Recall that an if-then statement and its contrapositive are
both true or both false. The contrapositive form of 1) is perhaps the more useful form.)
c) State the converse of 1). Come up with a single example to show that the converse is false.
16. Let A be mn.
a) Prove:
2) If Ax = b has at most one solution for every b, then n m (i.e., A has at least as
many rows as columns).
b) State the contrapositive of 2).
c) State the converse of 2). Come up with a single example to show that the converse is false.
Section CB. Extraneous Roots
1. Consider the following attempt to solve x = 2 +
x.
x = 2 +
x (i)
x 2 =
x (ii)
x
2
4x + 4 = x (iii)
x
2
5x + 4 = 0 (iv)
(x 4)(x 1) = 0 (v)
x = 1, 4 (vi)
a) Is each step legitimate? That is, does each line follow correctly from the previous line?
b) Do lines (i) and (vi) have the same solutions?
c) If your answer to b) is No, which line or lines dont reverse? That is, which lines dont
imply the previous line?
page 16 Section CC, Gaussian Elimination and Inverses
2. Problems with radicals in them are not the only problems in elementary algebra where extra-
neous roots show up. Consider
2x
x 1
+
1
x 1
=
3
x 1
.
Solve this equation in the usual way. Which, if any, of your answer(s) satisfy this equation?
What step(s) in your solution fails to reverse?
3. A certain matrix A has the property that its transpose is its inverse: AA
T
= A
T
A = I. For
this A, and some b, we want to nd all solutions x to Ax = b. Which, if any, of the following
algebraic calculations amount to an argument that there is exactly one solution, x = A
T
b.
a)
Ax = b
A
T
(Ax) = A
T
b
(A
T
A)x = A
T
b
Ix = A
T
b
x = A
T
b.
b)
x = A
T
b
Ax = AA
T
b
Ax = b.
c)
x = A
T
b
x = A
T
Ax
x = x.
d)
Ax = b
A(A
T
b) = b
b = b.
e)
Ax
?
= b
A(A
T
b)
?
= b
b
= b.
4. In fact, we will learn later that, for any square matrix A such that AA
T
= A
T
A = I, and
for any b, there is a unique solution x to Ax = b. (We will prove this by an argument that
doesnt involve GE or multiplying A by its inverse.) Knowing this, which of the arguments in
Problem 3 are valid proofs that that this unique solution is A
T
b?
Section CC. Gaussian Elimination and Inverses
In the rst section on (two-sided) inverses, Section C, the results could be proved at the matrix level
using the denition of inverse. The theoretical results in this section are deeper and you will need
your knowledge of Gaussian elimination to prove them. What you will mostly need is knowledge of
what various rref congurations imply about solutions.
1. Use Gaussian elimination to determine if the following matrices have inverses, and to nd the
inverse when it exists.
a)
_
1 3
2 4
_
b)
_
_
1 2 3
2 3 4
3 4 5
_
_ c)
_
_
1 2 3
2 3 4
3 4 4
_
_
2. Recall that a simple lower triangular matrix is lower triangular with all diagonal entries 1.
Prove that a simple lower triangular matrix is always invertible and its inverse is simple lower
triangular. (Hint: use GE.)
page 17 Section D, Canonical Forms
3. Prove that an uppertriangular nn matrix with no 0s on the main diagonal is invertible, and
the inverse is uppertriangular. (The same theorem is true for lowertriangular, as one could
prove directly, or by applying transposes to the matrices in the uppertriangular proof.)
4. Prove that if matrix M is square and has a left inverse, then it is invertible. Hint: What does
the existence of a left inverse tell you about the number of solutions to Ax = b for every b?
Combine this with knowledge of rrefs from GE.
5. Show: If a matrix A has a right inverse and is square, then A is invertible.
6. Show that the following is false: if AB is invertible, then so are A and B. Hint: Think simple
and small. Find A, B which are obviously not invertible but AB is.
7. Show that if (AB)
1
exists, then at least A has a right inverse and B has a left inverse.
8. Prove: if AB is invertible and A is square, then both A and B are invertible.
9. Lets loosen up the denition of inverse matrix. For this problem, let us say that an mn
matrix A is invertible if there is a matrix B such that BA and AB are both identity matrices
(but not necessarily the same size)
a) (easy) Show that B is necessarily nm.
b) (harder) Show that A must be square anyway (thus, no additional generality is gained
over the usual denition that a matrix A is invertible if it is square and there exists a B
such that AB = BA = I).
Section D. Canonical Forms
1. For representing points on the plane, are cartesian coordinates (x, y) a canonical form? Are
polar coordinates (r, ) a canonical form. Explain.
2. Is y = mx + b a canonical form for straight lines in the plane? If not, do you know another
form which is? (By the meaning of canonical, this form is canonical if every equation of this
form represents a straight line in the plane, and every straight line in the plane can be written
in this form in one and only one way.)
3. Name a canonical form for circles in the plane. Explain briey why it is canonical.
4. You know one canonical form for quadratics: y = ax
2
+ bx + c with a ,= 0. However, there is
another important form for quadratics:
y = a(x d)
2
+ e, a ,= 0.
Is this a canonical form? Whether it is or not, what advantage does this form have?
5. Someone asks you to describe (as an intersection of planes) the set of points (x, y, z) such that
z = 2y and y = 3x. Your friend says this set in the intersection of the three planes
3x + y z = 0
3x 3y + z = 0
6x 2z = 0.
page 18 Section E, Complexity of Matrix Algorithms; Middle View
a) Is your friend right? (You might be able to answer this ad hoc, but do it using canonical
forms.)
b) Your friend changes his mind before he hears your answer. He now claims that the third
plane is
6x z = 0.
Is your friend right now?
6. Is the solution set to
x + 2y + 3z + 4w = 0
2x + 3y + 4z + 5w = 0
the set of (x, y, z, w) satisfying
_
_
x
y
z
w
_
_
= x
_
_
1
2
1
0
_
_
+w
_
_
0
1
2
1
_
_
?
7. A high school algebra teacher gives a problem to solve, with radicals in it. One student gets
the answer 1
1
2
2 . Another student gets

1
2 +
2
. A third student gets
2 +
2
6 + 4
2
. Could
they all be right? Why is this problem in this section?
Section E. Complexity of Matrix Algorithms; Middle View
Note: Restrict attention in this section to nn matrices. and n1 column vectors. In this
section, assume that matrix multiplication takes n
3
steps, that Gaussian elimination takes n
3
/3
steps on any system Ax = b, and that matrix inversion takes n
3
steps. (This is correct if we
do matrix multiplication directly from the denition, if we use Gaussian elimination to compute
inverses, if we count only real-number multiplications and divisions as steps, and if we count only
the highest order terms. While these counts are an appropriate way to measure eciency for
large matrix computations on a computer, it is not appropriate for matrix computations on your
calculator.)
1. Let A and B be nn matrices, and let c be an n1 column.
a) How much work does it take to compute ABc as (AB)c? As A(Bc)? Which is less work?
b) Now suppose A and B are xed, but c varies. In fact, you have to compute ABc for M
dierent vectors c, where M n (that is, M is very much bigger than n). This is the
situation in computer graphics applications.
How much work does it take to compute the answers in the form (AB)c; in the form
A(Bc)? Which is less work?
c) Now suppose B and c are xed but A varies, over M dierent matrices A. Compare the
amount of work by the two methods.
page 19 Section E, Complexity of Matrix Algorithms; Middle View
d) Now suppose A and c are xed but there are M dierent Bs. Compare the work by the
two methods.
2. Which side of the identity
(AB)
1
= B
1
A
1
is faster to compute?
3. Suppose you are given invertible matrix A and asked to solve Ax = b. Which way is faster,
Gaussian elimination, or nding and then premultiplying by A
1
?
4. Again, you are given an invertible matrix A and asked to solve Ax = b. Which way is faster
on your calculator, Gaussian elimination, or nding and then premultiplying by A
1
? What if
you dont know in advance whether A is invertible?
5. Suppose you are given invertible matrices A, B and asked to solve (AB)x = c. Which way is
faster:
i) Find AB and then do Gaussian elimination to solve (AB)x = c;
ii) First use Gaussian elimination to solve Ay = c for y, and then use Gaussian elimination
to solve Bx = y. (Convince yourself that the x you get here really does solve the original
problem.)
6. Your boss gives you a huge square matrix A and tells you to solve Ax = b. Just as you are
about to start, your fairy godmother arrives and hands you A
1
on a silver platter. To solve
the problem most eciently, what should you do?
7. Your boss gives you two square matrices A and B and asks you to verify that B = A
1
. Which
is smarter, checking directly whether B meets the denition, or solving for A
1
from scratch
using Gaussian elimination?
8. This problem is intended to show that complexity is an issue for ordinary arithmetic, not just
for matrix arithmetic; and that what should count as a time-consuming step depends on the
tools you are using.
Consider the identity a(b + c) = ab + ac for the particular values a = 3.729, b = 2.638 and
c = 5.893.
a) Compute both sides of the identity by hand. Well, at least imagine carry out the work.
Which side takes longer to compute? By a lot? What part of the work takes the most
time?
b) Compute both sides of the identity with a hand calculator. Use a memory button if that
makes things faster. Which side takes longer to compute? By a lot? What part of the
work takes the most time?
page 20 Section EA, Complexity of Matrix Algorithms; Detailed View
Section EA. Complexity of Matrix Algorithms; Detailed View
1. In the following parts, you are asked to consider variants of Gaussian elimination in which
the operations are done in dierent orders. Each variant involves the same input: an nn
coecient matrix A and a n1 constant column b. Thus each variant is carried out on a
n(n+1) augmented matrix [A[b]. Assume that Gaussian elimination reduces A to I (the
usual case when A is square).
Each variant involves a Downsweep, an Upsweep, and a Scale, but they may have to be
somewhat dierent from those in regular Gaussian elimination. In every case, Scale(i) means
whatever needs to be done at that point so that the pivot entry in row i gets value 1 (and so
that the solution set is unchanged). Downsweep means whatever needs to be done so that the
entries below the pivot in column i are turned into 0s. Upsweep is whatever needs to be done
so that the entries above the pivot in column i are turned into 0s.
For each version, determine how many multiplication/division operations are required, for
the A matrix, for the b column, and then jointly. In general you need only consider the highest
order term (h.o.t.), but when two methods have the same h.o.t., you may wish to compute
the number of steps exactly. For this you may use the formulas

n
k=1
k = n(n+1)/2 and
n
k=1
k
2
= n(2n+1)(n+1)/6.
Algorithm Gauss-A
for i = 1 to n
Scale(i)
Downsweep(i)
endfor
for i = n downto 1
Upsweep(i)
endfor
Algorithm Gauss-B
for i = 1 to n
Downsweep(i)
endfor
for i = 1 to n
Scale(i)
endfor
for i = n downto 1
Upsweep(i)
endfor
Algorithm Gauss-C
for i = 1 to n
Downsweep(i)
endfor
for i = n downto 1
Upsweep(i)
endfor
for i = 1 to n
Scale(i)
endfor
AlgorithmGauss-D (Gauss-Jordan)
for i = 1 to n
Scale(i)
Upsweep(i)
Downsweep(i)
endfor
Notice, for instance, that in variant A the scaling is done early, and no multiplier needs to
page 21 Section EB, Asymptotics
be computed in Downsweep before subtracting. Also notice that version D has its own name,
Gauss-Jordan reduction.
2. (Complexity of matrix inversion) To nd A
1
, you do Gaussian elimination to [A[I]. Since I
consists of n b-columns, it follow from the analysis in class that the inversion algorithm takes
order (4/3)n
3
steps (n
3
/3 for the A matrix, and n
2
more for each of the n columns of I).
However, the columns of I are not any old b-columns. They have a lot of 0s, so maybe some
steps can be skipped.
Prove that, by skipping vacuous steps, the inversion algorithm in fact takes asymptotically
n
3
steps.
3. Many books discuss matrix inversion using Gauss-Jordan reduction instead of Gaussian elim-
ination. (The Gauss-Jordan reduction method is Algorithm D of Problem EA1.) Although
Gauss-Jordan reduction is slower for solving Ax = b, maybe it pulls ahead in the case of
matrix inversion. Devise a special version of Gauss-Jordan reduction for inverting matrices.
As with Gaussian elimination used for matrix inversion, there may be savings from replacing
computations with assignments. Analyze the eciency of your algorithm and compare it to the
eciency of Gaussian elimination for inversion.
4. Let A be an nn tridiagonal matrix that can be reduced without nding any 0s where pivots
should be (e.g., the matrices D
n
in Problem G9).
a) Determine the number of steps to reduce A.
b) Determine the number of steps to reduce b in Ax = b.
c) Show that A
1
will in general have no zero entries. That is, show that reducing [A [ I] is
not guaranteed to leave any 0s, and demonstrate for a specic 33 matrix that A
1
has
no zero entries.
d) We know that, for general matrices A, GE takes n
3
/3 steps to solve Ax = b, whereas
nding A
1
and then A
1
b takes n
3
. However, we also know that, if for some reason
you already have A
1
, then computing A
1
b takes only n
2
more steps.
Show that for tridiagonal matrices A, even if someone has already given you A
1
, you
shouldnt use it to solve Ax = b. It is still faster to do GE.
Note: Tridiagonal matrices, or more generally, band matrices, are important in the
numerical solution of dierential equations. See Section G.
Section EB. Asymptotics
1. Denition. To say that g(x) = f(x)+ l.o.t. (lower order terms) means that g(x)f(x), i.e., g
is asymptotic to f, meaning
lim
x
g(x)/f(x) = 1, or equilvalently, lim
x
(g(x) f(x))/f(x) = 0.
(We often use n as the variable instead of x, especially when the input values will always be
positive integers.)
a) Show that if g(x) = x
n
+ Ax
m
, where A is any constant, even a very big one, then Ax
m
is a l.o.t. (to x
n
), so long as m < n. Note: n and m can be any real numbers.
b) If g(n) = n
2
+ nlog n, determine if nlog n is a lower order term.
page 22 Section EC, Matrix Algorithms for Hand Held Calculators
2. Prove that is transitive, that is, if fg and gh, then fh.
3. Show that

n
k=1
k
m
and

n1
k=1
k
m
are the same asymptotically. The point is, a change by 1 in
the bounds of a power sum (or in fact, a change by any xed nite amount) makes no dierence
asymptotically.
4. Let p(k) be a polynomial whose highest power term is ak
m
. Show that
n
k=1
p(k)

n
k=1
ak
m
.
5. Prove: if f
1
and f
2
are positive functions (i.e. f(x) > 0 for all x), and g
1
f
1
, g
2
f
2
, then
(g
1
+g
2
) (f
1
+f
2
).
6. Show that, in general, it is not true that
(f
1
g
1
and f
2
g
2
) = (f
1
+f
2
g
1
+g
2
).
Also show that, even for positive functions, it is generally not true that
(f
1
g
1
and f
2
g
2
) = (f
1
f
2
g
1
g
2
).
7. We say that g is little oh of f, written g = o(f), if
lim
x
g(x)
f(x)
= 0.
Another way to say this, using a previous denition, is that (f+g) f.
Prove: if g = o(f) and both

ff and gg, then (

f g) (f g). (Compare this with the
negative result in Problem 6.)
8. Prove: if g = o(f),

ff and gg, then g = o(

f).
9. It is temping to try to prove Problem 4 by proving the more general result: if g = o(f) then
n
k=1
g(k) = o
_
n
k=1
f(k)
_
. Alas, this claim can be false. Give an example.
10. Look back over calculations you have done to simplify expressions while doing asymptotic
analysis on them. Are there any steps you have taken that are not justied by the results in
the problems in this section? If so, try to prove that the steps you took are correct.
Section EC. Matrix Algorithms for Hand Held Calculators
1. (How to solve Ax = b on your hand calculator) The only sort of linear equations that your
hand calculator can solve directly are square invertible systems. That is, if A is invertible
so that Ax = b has a unique solution for every b, then you can nd it on your calculator by
computing A
1
b. You can program your calculator to solve Ax = b in the general case (that is,
you can write programs to do Gaussian eliminationas we have done in class and thus determine
if there are any solutions and nd all of them if there are), but there are no built-in commands
you can use on any calculator I know. My TI-86 can compute the reduced echelon form, but
it doesnt know what to do with it. It also doesnt know about augmentation lines; that is, it
doesnt know how to carry out GE only on the A part of [A [ b] or [A [ B], which can lead to
a dierent rref than if [A B] is to be considered as a single matrix.
page 23 Section F, LU methods
Figure out how to trick your calculator into solving the general system. That is, using just
the built-in commands, gure out how to determine if Ax = b has any solutions, and if it does,
to produce output that expresses them all. You can use built-in commands to tell you how to
replace Ax = b with some other system A
= B
, with A
invertible, that tells you what you

need.
I know one way to do this, but you may think of other and better ways.
Section F. LU methods
1. Let A =
_
_
1 2 3
2 3 2
3 2 1
_
_
.
a) Find the LU factorization of A. Check that A = LU.
b) Find the LDU factorization.
2. Let
A =
_
_
1 2 3
2 6 7
1 4 3
_
_
, b =
_
_
1
0
0
_
_
, c =
_
_
2
0
1
_
_
.
a) Suppose you know that you must solve Ax = b and Ay = c. Solve them simultaneously
(by Gaussian elimination, not the LU method).
b) Solve Ax = b again, by the LU method. Suppose you only learn that you must solve
Ay = c after you solve Ax = b. Show how you can now solve Ay = c with a limited
amount of extra work.
3. When LU-Gauss is applied to a certain matrix A, the resulting matrix that stores all the
information is
_
_
2 0 1
1 1 1
2 3 3
_
_.
a) What is the solution to
Ax =
_
_
4
1
5
_
_
?
b) What is A?
4. Repeat Problem E6, but this time it so happens that yesterday you solved Ax = c by the LU
method and youve saved the nal matrix. Now what should you do?
5. For the A and U of Problem 3 (U of LU, not LDU), we know that the rst part of GE changes
A into U. Thus there is an L
1
such that U = L
1
A. In fact, L
1
is lowertriangular.
a) Find L
1
. Be careful; use some method youre sure is right. Does the answer surprise you?
b) How is L
1
related to the L of A = LU?
6. Two alternative approaches (to the LU method) have been suggested for avoiding wasted eort
when you have to solve Ax = b at dierent times with dierent bs but the same A.
i) Find the premultiplier matrix such that PA = A
R
= I (that is, nd A
1
) and then apply P
to b;
ii) Initially solve Ax = (b
1
, b
2
, b
3
)
T
, that is, use literals instead of numbers, and then solve
Ax = b by substituting the specic constants in b for b
1
, b
2
, b
3
.
a) Carry out both approaches by hand for the example on the handwritten handout, that is,
for A =
_
_
2 1 1
4 1 0
2 2 1
_
_ and b =
_
_
1
2
7
_
_
.
b) I claim that solving Ax = b with literals for b is eectively the same as row reducing
[A [ I]. In fact, I claim that method ii) is eectively the same as method i); for instance,
applying P to b amounts to substituting the numerical b into the solution for the literal
b. Explain.
c) So what is the asymptotic complexity of methods i) and ii) when solving Ax = b, where
A is nn? How does this compare to the complexity of the LU method? (As always in
this part of the course, assume A
GE
I without row switches or 0 pivots.)
7. The strong LU theorem says:
If A is nn and A reduces to U with a pivot in every column and no row switches,
then there is a unique factorization A = LU. That is, there is at least one such
factorization (the one in the basic theorem), and if A = L
is any factorization
where L
is simple lowertriangular, and U is uppertriangular with nonzeros on the

main diagonal, then L
= L and U
= U.
In some sense, then, the LU factorization is like the prime factorization of numbers, which is
also unique.
Prove the strong theorem using previous results as follows:
a) L and U are invertible (a previous result, but you should be able to reconstruct the proof
in your head), so premultiply LU = L
by L
1
and postmultiply by (U
)
1
.
b) L
1
is a special sort of matrix what sort? (Again you should be able to reconstruct the
previous proof.) What sort of matrix is (U
)
1
?
c) Again by previous problems, what dierent sorts of special matrices are
L
1
L
and U(U
)
1
?
d) But U(U
)
1
= L
1
L
. How can this be?

e) Assuming you got, say, L
1
L
= I, what can you do to prove L
= L?
8. Prove the Strong LU Theorem of Problem 7 again by giving an entrywise proof that L
= L
and U
= U. If you start looking at the n

2
real number equations that comprise A = LU, and
look at them in the right order, you will nd that it is not hard to show that the entries in A
force the values in L and U. Actually, there are several right orders.
9. Use the strong LU Theorem to prove the strong LDU theorem:
If A is nn and A reduces to U with a pivot in every column and no row switches,
then there is a unique factorization A = LDU, where L is simple lowertriangular, U
is simple uppertriangular, and D is diagonal.
10. The basic LU Theorem says:
If A is nn and A reduces by the downsweep part of Gaussian elimination to U,
with a pivot in every column and no row switches, then there is a factorization
A = LU, where L is simple lowertriangular (1s on the main diagonal) and U is
uppertriangular (and has nonzeros on the main diagonal).
Prove the converse:
If A = LU where L is simple lowertriangular and U is uppertriangular with nonzeros
on the main diagonal, then A reduces by the downsweep part of GE to U, with a
pivot in every column and no row switches. (A, L, U are all understood to be square.)
Hint: Its easy to see what the downsweep of GE does to L. Apply the same sequence of steps
to LU.
11. Prove: if A is symmetric and has an LDU factorization, then necessarily U = L
T
.
12. Prove a converse to F11:
If A = LDL
T
, then A is symmetric.
Note: this is quite easy using transpose properties.
13. Find in the standard way the LDU factorization of the symmetric M =
_
_
1 2 3
2 2 0
3 0 2
_
_ and
verify by sight that L = U
T
.
page 26 Section G, Applications
Section G. Applications
1. Suppose the economy has two goods, food and steel. Suppose the production matrix is
M =
_
.6 .4
.5 .6
_
,
where food indexes the 1st row and column, and steel indexes the 2nd row and column.
a) Is this economy going to work, that is, for every nonnegative vector y, is it possible for
the economy to create this y as its net production by actually producing (and partially
using in the process) another nonnegative amount x? Are there any nonnegative goods
vectors y that the economy can make as its net production?. In other words, are there some
production plans the economy can carry out, even if it cant carry out every production
plan? (You can answer both questions by using GE to nd if (I M)
1
exists and, if so,
looking at the signs of its entries.)
b) (I M)
1
=

n=0
M
n
if the sum converges. (This is shown in class most years, but also
follows by considering a matrix geometric series.) Compute enough powers of M above
(use a computer or calculator) to get a sense whether M
n
0, which is a necessary
condition for convergence of the sum.
2. Repeat the previous problem for the only slightly dierent matrix
M
=
_
.6 .3
.5 .6
_
.
3. Repeat the previous problem for the matrix
M
=
_
.5 .3
.4 .6
_
.
4. Sometimes the series

n=0
M
n
converges quickly because M
n
= 0 for all n starting at some
value. (Such a matrix M is said to be nilpotent, and if p is the lowest integer for which
M
p
= 0, we say M is nilpotent of degree p.)
a) Find the degree p of nilpotency for
M =
_
_
0 1 1
0 0 1
0 0 0
_
_.
For the rest of this problem, x p at this value.
b) Verify that (I M)
1
=

p1
n=0
M
n
. Do this two ways:
i) Compute (I M)
1
by GE and compute

p1
n=0
M
n
directly.
ii) Verify by matrix algebra rules that (I M)(
p1
n=0
M
n
) = I. This shows that
(
p1
n=0
M
n
) is a right inverse of I M, and since I M is square, we know from a
previous result that any right inverse is the inverse.
5. In class we may have approximated the dierential equation (and initial condition)
f
(x) = x, f(0) = 0,
with the dierence equation
f(x
i
+h) f(x
i
)
h
= x
i
, f(0) = 0, h = 1/4, x
i
= i/4 for i = 0, 1, 2, 3.
The idea is that the solution to the dierence equation would surely approximate the solution
to the dierential equation. Furthermore, the dierence equation is easy to solve on any nite
interval (we used the interval [0,1]). First, this method is Eulers method in disguise. But
more important from the 16H point of view, this system of equations reduces to a single matrix
equation Mx =
1
4
b, where
M =
_
_
1 0 0 0
1 1 0 0
0 1 1 0
0 0 1 1
_
_
, x =
_
_
f(1/4)
f(1/2)
f(3/4)
f(1)
_
_
, b =
_
_
0
1/4
1/2
3/4
_
_
.
Note the scalar
1
4
in front of b. In class we actually had
1
1/4
Mx = b, but the calculation is
easier if you rst multiply both sides by 1/4.
Solve this system and compare the answer with the actual solution of the dierential equation.
6. In class we showed (I hope) that it is reasonable to approximate the 2nd order dierential
equation (and boundary conditions)
f
(x) = x, f(0) = f(1) = 0, (1)

with the dierence equation
f(x
i
+h) 2f(x
i
) + f(x
i
h)
h
2
= x
i
, f(0) = f(1) = 0, h = 1/4, x
i
= i/4 for i = 1, 2, 3.
This system of equations reduces to a single matrix equation Mx = (
1
4
)
2
b, where
M =
_
_
2 1 0
1 2 1
0 1 2
_
_, x =
_
_
f(1/4)
f(1/2)
f(3/4)
_
_
, b =
_
_
1/4
1/2
3/4
_
_
.
Note the scalar (
1
4
)
2
in front of b. In class we actually had
1
(1/4)
2
Mx = b, but the calculation
is easier if you rst clear the complex fraction.
a) Show or verify that the exact solution to (1) is f(x) =
1
6
(x
3
x).
b) Solve the dierence equation using the matrix system Mx = (
1
4
)
2
b and compare the
answer with the actual solution of the dierential equation.
The next two problems are adapted from Strang, Linear Algebra and Its Applications, 2e.
7. Consider the 2nd order dierential equation
f
(x) + f(x) = x, f(0) = f(1) = 0.

As in class, solve approximately on the interval [0,1] with a dierence equation, using step size
h = 1/4. You may use a calculator or software to solve the resulting matrix equation.
Bonus: If you know how to nd the exact solution, nd it and compare with the approxima-
tion.
8. Consider the 2nd order dierential equation
f
(x) = 4
2
sin2x, f(0) = f(1) = 0.
a) Check that a solution (and thus the solution, since by general theory there is only one for
a system like this) is f(x) = sin2x.
b) Use the method from class, with step size h = 1/4, to approximate the solution. Compare
with the actual solution.
c) Repeat b) with step size h = 1/6.
9. Let D
n
be the nn tridiagonal matrix with 2s on the main diagonal and 1s on the diagonals
just above and below the main diagonal. (D is for dierence; this is just the matrix we have
been using for discrete approximations to the 2nd derivative, except the signs are reversed to
make the main diagonal positive.)
a) Find the LDU factorization of D
4
.
b) Make a conjecture about the LDU factorization of D
n
c) Prove your conjecture.
10. In our rst example of a numerical solution to a dierential equations in class, we approximated
the solution to
f
(x) = x, f(0) = 0. (2)

We approximated by replacing f
(x) with
f(x + h) f(x)
h
.
However, it turns out that a much closer approximation to f
(x) is the symmetric secant slope

f(x+h) f(xh)
2h
.
For instance, with this latter approximation, instead of saying
f(1/4) f(0)
1/4
f
(0) (x = 0, h = 1/4),
we say
f(1/4) f(0)
1/4
f
(1/8) (x = 1/8, h = 1/8).

Redo the class solution to (2) using this better approximation throughout. You get the same
matrix M as in class, but a dierent b, and hence dierent values for f. Compare this new
approximate solution with the exact solution. That is, compare the approximate values you get
for f(x) at x = 1/4, 1/2, 3/4, and 1 with the exact values.
11. Heres another approach to coming up with a nite dierence to approximate f
(a) at any a.
For simplicity, we will x a = 0, but the same argument works at any point.
Heres the idea. The approximation of f
(0) by (1/h)[f(h) f(0)] is an approximation

by a linear combination of f(h) and f(0). Specically, we are using the linear combination
Af(h) + Bf(0) where A = 1/h and B = 1/h. Lets try to generalize and look for a linear
combination approximation to f
(0) as well. For f
we probably need one more point, so, lets

hope that some linear combination of f(0), f(h) and f(h) is a good approximation of f
(0).
In other words
Af(h) +Bf(0) + Cf(h) f
(0). (3)
Now, we have 3 unknowns, so we can insist that the approximation be exact for three functions
f. Since, as you learned in calculus, the polynomials are a good basis for approximating
most functions, lets insist that display (3) be an equality for the three basic polynomials
f(x) = 1, x, x
2
.
Write down the 3 equations obtained this way and solve for A, B, C. Compare your answer
to the dierence formula we have used so far for second derivatives.
Note: Finding A, B, C by GE is harder than usual, because of the presence of the literal h.
You may nd it easier to solve by ad hoc methods. However, GE is not actually that hard in
this case.
12. So far our second order dierential equations have had two boundary conditions, namely f(0) =
f(1) = 0. Often instead they have two initial conditions, for instance, f(0) = f
(0) = 0.
(Second order dierential equations always need two special conditions of some sort to uniquely
specify an answer.)
So the question is, what is the right matrix setup to approximate the solution to a 2nd order
equation with 2 initial conditions?
As before, lets develop a method with a specic example. Again we use the equation
f
(x) = x, where we want to solve on [0,1] using the step size h = 1/4, but now the special
conditions are f(0) = f
(0) = 1. Find and explain A, x, b so that the approximation is found

by solving Ax = b.
Note: I can see at least two dierent but good answers to this question, depending on how
you decide to approximate f
(0).
13. Consider the following electric circuit:
.......................................................................................................................................................................... .......................................................................................................................................................................... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
............................................................................................................................................................................................................
A
B
........................................... . . . . . . . . . . .. . . . . .
. . . . . . . . . . . . . . . . .
........................................... . . . . . . . . . . . . . . . . .
. . . . . . . . . . .. . . . . .
......................................................... . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
......................................................... . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
....................................................... . . . .. . .. . .. . .. . ..
.................
C
D
1
2
3
4
5
.........................................................................................................................................................................................................................................................................................................................................................................................................................................................................
battery
The battery connecting D directly to A, makes the voltage at A 1 unit higher than at D. The
edges are wires; The number on each edge is the resistance on that edge, and in this example
can also be used to name the edge. For instance, let i
3
represent the current on the vertical
edge with resistance 3 units. Let the positive direction for current be as shown by the arrows.
The curved wire connecting the battery is assumed to have no resistance.
a) Write down the linear equations for the current. Remember, at each (interior) node net
current must be 0 (Kirchkos Current Law) and along each loop, the net voltage change
must be 0 (Kirchkos Voltage Law). Finally, Ohms law says that V = ir, where V is the
voltage change along an edge (in the direction of the arrow), i is the current on that edge
(in the direction of the arrow), and r is the resistance along that edge. (The battery is
assumed to provide no resistance, but it does increase the voltage by 1 unit from D to A.)
b) Use a calculator or software to nd the currents. (You dont want to do this one by hand.)
14. In the diagram below, something causes the voltage to be higher by amount V at point A than
at C, causing current to ow from A to B to C. (Perhaps there is a battery connecting C back
to A by a wire not shown, or perhaps there are a whole lot of other wires not shown allowing
for current to ow.) The resistences on the two wires are R
1
and R
2
.
.............................................................................................................................................................................................................................................................................................................................................................................................................

A B C
R
1
R
2
i
1
i
2
Show that, if wires AB and BC are to be replaced by a single wire AC so that the same amount
of current ows from A to C as currently, that new wire must have resistance R
1
+R
2
. This is
the law of series circuits. Hint: Solve for i
1
, i
2
with Kirchhos laws you get one current
equation and one voltage equation and then nd R so that V = IR for the given voltage and
the total current ow I out of A.
15. In the diagram below, something causes the voltage to be higher by amount V at point A than
at B, causing current to ow separately over each wire from A to C. (Perhaps there is a battery
connecting C back to A by a wire not shown, or perhaps there are a whole lot of other wires
not shown allowing for current to ow.) The resistences on the two wires are R
1
and R
2
.
........................................................................................................................................................................................................................................................................................................................ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

A B
R
1
R
2
i
1
i
2
Show that, if both wires AB are to be replaced by a single wire AB so that the same total
current ows from A to B as currently, the resistance R in that new wire must satisfy
1
R
=
1
R
1
+
1
R
2
.
This is the law of parallel circuits. Hint: Solve for i
1
, i
2
with Kirchhos laws you get no
current equation and two voltage equations and then nd R so that V = IR for the given
voltage and the total current ow I out of A.
The laws is this and the previous problem allow one to compute currents in many complicated
circuits by hand by repeatedly simplifying the circuit.
page 32 Section H, Vector Space Axioms
Section H. Vector Space Axioms
1. Prove that the following are vector spaces.
a) R
2
, meaning the set of ordered pairs (x, y) of real numbers, where + is dened by
(a, b) + (c, d) = (a+c, b+d)
and scalar multiplication is dened by
a(x, y) = (ax, ay).
The proof can be very tedious if you write everything out in detail, but the key is that
coordinatewise all the vector space properties are known real number properties. Show
this carefully in one or two cases, and thereafter you can just say it. On the other hand,
when an axiom introduces a new quantity, like 0 or u, you have to at least state what
that quantity is as an ordered pair and explain briey why you are right.
b) The set of all real functions, that is, functions from real number inputs to real number
outputs. You need to understand what it means to add two functions, or multiply a
function by a scalar, but you have been doing these operations for a long time, as in
f(x) = 3 sinx + x
2
. Again, show one or two parts in detail.
c) The set of all innite sequences of real numbers, where addition and scalar multiplication
are dened coordinatewise.
2. Prove: in any vector space V , 0 is unique. That is, if 0 and 0
can both be substituted truthfully

for in the statement
for each u V , u + = u,
then 0 = 0
. Note: the uniqueness of 0 is often included in the axioms.

3. Prove that in any vector space, 0 is unique in even a stronger sense: if for even one u, u+x = u,
then x = 0. We summarize by saying: in a vector space, any local zero is the global zero.
(Denition: For any particular u, we say x is a local zero for u if u +x = u. A global zero is
just a 0 as previously dened.)
4. Show from the axioms of a vector space that, for any u, its inverse is unique.
5. Prove: if u ,= v then u +w ,= v +w. That is, equals added to unequals are unequal, just as in
real number algebra. Hint: We know how to prove statements containing equalities, so can be
replace the claim to be proved with an equivalent claim involving only equalities?
6. For any vector space V , prove
a) For any a R, a0= 0.
b) For any v V , 0v= 0.
c) For any a R and v V , (a)v = a(v) = (av).
d) For any v V , (1)v = v.
7. Comment on the following proof of Problem 6a.
Without loss of generality, we may use ordered pairs instead of n-tuples. Then for
any a we have a0 = a(0, 0) = (a0, a0)
rnf
= (0, 0) = 0.
8. Prove: if u, v are vectors in any vector space, then
(u + v) = u +v.
Use only our axioms or theorems proved from those axioms in earlier problems in this problem
collection.
9. The operation subtraction is dened in a vector space by
u v = u + (v).
Use this denition to prove
a) a(u v) = au av,
b) (1)(u v) = v u,
c) (a b)v = av bv.
The symbol is used in mathematics for two rather dierent purposes: additive inverses and
subtraction. This problem shows that no great harm comes from using the same symbol for
those two purposes.
10. Prove: If u ,= v then u v ,= 0.
11. Scalar division: if d ,= 0 is a scalar, dene
u
d
to be (1/d)u. Prove from the axioms of a vector
space, and this denition, that
u + v
d
=
u
d
+
v
d
.
12. The (vector addition) Cancellation Law states: For any vectors a, b, c,
if a + c = b +c then a = b.
a) Prove this Law for arbitrary vector spaces; that is, prove it from the axioms.
b) Suppose vectors are dened to be sets (of numbers, say) and + is dened to be set union.
For instance,
1, 3 + 1, 2, 5 = 1, 2, 3, 5.
Show by example that, with these denitions, the Cancellation Law is false. In light of
part a), this shows that when vectors and + are dened this way, they cant form a vector
space, no matter how scalar multiplication is dened. Therefore, at least one of the rst 5
axioms on our sheet must be false for this +. Which ones?
13. Let ^ be the nonvector-space of H12b: the vectors are sets of numbers and + is dened to be
set union.
a) Prove that ^ has a unique 0 in the sense of Problem H2. (First, nd a set that does
what 0 is supposed to do, and then see if the proof you used for H2 still works.)
a) Prove or disprove that the 0 of ^ is unique in the sense of Problem H3.
14. In class (most years) I claimed that the set of positive real numbers is a vector space with all
real numbers as scalars, if + and scalar multiplication are dened properly. Namely, if we let [x]
be the positive number x when viewed as a vector in this space (and we let a without brackets
represent the real number a when it is viewed as a scalar), then we dene
[x] + [y] = [xy], a[x] = [x
a
].
We proved in class that some of the axioms of a vector space hold for this system.
Go through all the axioms and verify that each one not proved in class is correct for the
denitions I set up.
15. Here is another proposal for a bizarre vector space, call is 1. The vectors are all non-zero real
numbers, and + and scalar are dened as in Problem H14. Well, not quite. Scalar has to
be dened a little dierently, because r
c
is not dened for negative r and most nonintegers c.
So we dene
a[x] =
_
sign(x) [x[
a
,
where
sign(x) =
_
1 if x > 0
1 if x < 0.
Prove or disprove: 1 is a vector space.
16. Prove that the set of real numbers is a vector space with the following denitions of + and
scalar . (When using these denitions, we will represent the number x as x); in other words,
when you see x) +y) you will know that the special meaning of + is intended, but when you
see x +y you will know that ordinary real-number + is intended.)
x) +y)
def
= x+y+1),
kx)
def
= kx +k1).
17. Change the example in Problem 16 as follows. The set, the notation, and + remain the same,
but scalar multiplication becomes kx) = kx), i.e., the normal denition. Prove or disprove:
With this revision we still have a vector space.
18. Prove: In any real vector space, if x ,= 0 and k ,= j, then kx ,= jx. Explain why it follows that
in any real vector space, no matter how bizarre the denition of scalar multiplication, there are
innitely many vectors (unless the space consists of 0 alone).
page 35 Section HA, Complex Numbers
19. For real numbers, we have the theorem
ab = 0 (a = 0 or b = 0).
We have already shown that this result is false for matrices where? Prove or disprove the
vector space analog:
kv = 0 (k = 0 or v = 0).
20. Basically, a eld is any set of number-like things which obey all the usual laws of high school
algebra (commutativity, distributivity, etc.), and such that when you add, subtract, multiply or
divide any pair of them (except you cant divide by 0), you get another thing in the same set.
In other words, a eld has to be closed under these 4 operations. For instance, the positive
integers are not a eld; they are closed under addition and multiplication, but not subtraction
and division. Sure, 4 7 is a number, but it is not in the set of positive integers.
The best known elds are: the rationals, the reals, and the complex numbers.
Now determine if the following are elds:
a) The complex rationals, that is, a + bi [ a, b rational numbers.
b) The complex integers (also known as the Gaussian integers same guy). This is the
set p + qi [ p, q integers.
c) S = a + b
2 [ a, b rational numbers. Note: somewhere in your solution, you will have

to use the fact that

2 is not a rational number. Where?
d) T = p +q
2 [ p, q integers.
e) (Challenge) U = a + b
2 +c
3 [ a, b, c rational numbers.
f ) (Challenge) V = a +b
2 + c
3 + d
6 [ a, b, c, d rational numbers.
Section HA. Complex Numbers
1. The complex numbers are dened to be the set C (some books write C) of all objects
z = a + bi, where a, b R and i is a special symbol dened to satisfy i
2
= 1. By denition,
two complex numbers z = a + bi and w = c + di are equal i a = c and b = d. Addition and
multiplication are dened the only way they can be if the usual rules of real-number algebra
are to hold in C. That is, one denes these operations according to the usual rules and then
has to prove (you do it below) that all the usual rules do indeed hold for C. Indeed, you will
show that C is a eld, as dened in Problem H20.
Useful Notation. If z = a + bi, with a, b R as usual, then the real and imaginary parts of
z are dened by
Re(z) = a, Im(z) = b.
a) Give the denition of addition, i.e., what does (a + bi) + (c + di) equal? It better be
something that meets the denition to be in C.
b) Find the zero of C and show that every z has an additive inverse, called z naturally
enough.
c) Give the denition of multiplication. Show that C has a multiplicative identity.
page 36 Section HA, Complex Numbers
d) Show that every z ,= 0 has a multiplicative inverse. Caution: It doesnt suce to say that
a + bi has the inverse
1
a + bi
. To show that C is closed under multiplicative inverses you
must show that there is an inverse of the form c + di.
e) Show that addition in C is commutative.
f ) Show that multiplication in C is commutative.
g) Show that multiplication in C is associative. This is sort of ugly to show unless you gure
out some particularly creative approach. (I know of some creative approaches, but none
that I am fully satised with.)
h) Dene division in C and prove that
z +w
u
=
z
u
+
w
u
.
The facts you are asked to show above do not cover all the things that must be shown to prove
that C is a eld, but they give the gist.
2. The complex conjugate of a complex number z = a + bi, denoted z, is dened by
a + bi = a bi.
Prove that
z = z, z +w = z +w, zw = z w, [z[ = [z[,
where [z[, the absolute value (or modulus) of z = a + bi is dened by [z[ =
a
2
+b
2
.
3. Let z = 2+i, w = 1+3i. Compute zw. Compute the magnitudes of z, w, zw directly and check
that DeMoivres Thm holds for these magnitudes. Estimate the arguments of z, w, zw from a
sketch and check that DeMoivres Thm seems to hold for these arguments.
4. Prove the magnitude part of DeMoivres as follows. Let z = a + bi, w = c + di. Compute [z[
and [w[ from the denition. Now compute zw in terms of a, b, c, d and use your computation
to compute [zw[ from the denition. Now do high school algebra to show that [z[[w[ = [zw[.
5. Prove both parts of DeMoivres Thm as follows. Let (r, ) be the magnitude and argument of
a + bi. Let (r
) be the magnitude and argument of c +di.

a) Explain why
a +bi = r(cos +i sin) and c +di = r
(cos
+ i sin
) (2)
and conversely, if you are given a complex number in the form r(cos +i sin), you know
right away that the magnitude is r and the argument is .
b) Multiply (a + bi)(c + di) in the trigonometric forms (2), and use some trig identities to
prove the theorem.
page 37 Section HB, Complex Vector Spaces, Part I
Section HB. Complex Vector Spaces, Part I
A complex vector space satises the same axioms as a real vector space except that the set of
scalars is C. (In general, a vector space is anything that satises those axioms, using as the set of
scalars some eld.) All the denitions we have used for real vector spaces (e.g., dependent, basis,
linear transformation) go through unchanged. Therefore, all the results we have proved go through
unchanged. The problems in this section illustrates this lack of change. Part II of this section will
come later, after we have more vector space concepts, like dimension. Then you can consider the
interesting interplay of complex and real vector spaces.
1. Let V be a complex vector space. Prove: if x V , x ,= 0, and z, w are scalars with z ,= w,
then zx ,= wx.
2. Solve Ax = b, where
A =
_
1 1
i 1
_
, b =
_
_
i
1
_
_.
(Gaussian elimination is yukier by hand over the complex numbers than over the reals. Can
you calculator do it?)
Section I. Subspaces
1. a) Show that W = (x, y, z) [ x+y+z = 0 is a subspace of R
3
.
b) Show that U = (x, y, z) [ x+y+z = 1 is not a subspace of R
3
.
2. a) Let /
nn
be the set of all nn matrices (with real number entries). Show that /
nn
is a
vector space.
b) Determine which of the following subsets of /
nn
is a subspace:
i) The symmetric matrices
ii) Those matrices that commute with an arbitrary but xed matrix T
iii) Those matrices for which M = M
2
.
3. For any matrix A, the null space of A, denoted ^
A
or ^(A), is dened by
^
A
= x [ Ax = 0.
That is, if A is mn, then ^
A
is the set of n-tuple column vectors x that satisfy Ax = 0.
Prove that ^
A
is a vector space.
4. Which of these sets of functions are vector spaces? Give reasons.
a) All polynomials (call the set T)
b) All polynomials of degree n (call the set T
=n
)
c) All polynomials of degree at most n (call the set T
n
)
d) All polynomials p(x) that satisfy p(1) = 0.
e) All polynomials p(x) that satisfy p(1) = 2.
f ) All polynomials p(x) that satisfy p(2) = 2p(1).
page 38 Section I, Subspaces
5. For each of the following, is it a vector space? Give reasons briey for your answers.
a) The uppertriangular nn matrices, for some xed positive integer n.
b) The simple uppertriangular nn matrices, for some xed positive integer n.
c) All uppertriangular matrices of all square sizes.
6. Function f is a real function if its domain and codomain are R. Show that the (set of) real
functions f satisfying
3f
(x) +x
2
f
(x) f(x) = 0 for all x

are closed under addition and scalar multiplication. This is the key step to showing that these
functions form a vector space.
7. Prove or disprove: the set of odd real functions is a vector space. Recall a real function is odd
if, for all x, f(x) = f(x).
8. Its a pain, when proving subsets are subspaces, to have to prove two closure properties. For-
tunately, we have
Theorem. A nonempty subset U of a vector space V is a subspace for all
u, v U, and all scalars c, cu +v is in U.
In other words, one combined closure property is enough. Prove this theorem.
9. Prove: if U, U
are subspaces of V , then so is U U
. Is the intersection of an arbitrary number

of subspaces of V , even innitely many, also a subspace?
10. Prove by example that the union of two subspaces need not be a subspace.
11. Let U, V be subspaces of vector space W. We dene their (vector subspace) sum by
U + V = u +v [ u U, v V .
a) Let W = R
3
, let U be the x-axis, and let V be the y-axis. What is U +V ?
b) Let W = R
4
, where we will refer to the coordinates as x, y, z, w. Let U be the xy-plane in
W, and let V be the yz-plane. What is U +V . Explain briey.
c) Why insist in the denition of sum that U, V be subspaces of the same space?
d) Prove in general: U + V is a vector space.
12. Let U, V be arbitrary vector spaces. Dene their cartesian product UV (also called their
exterior sum U V ) as follows. The set of vectors is the set of ordered pairs (u, v) [
u U, v V , where as usual two ordered pairs are equal i they are equal coordinatewise.
Addition and scalar multiplication are also dened coordinatewise. Prove that UV is a vector
space.
page 39 Section J, Linear Combinations and Spans
13. The functions in the following set are solutions to f
(x) = f(x):
S = f [ f(x) = Asinx + Bcos x, A, B arbitrary real numbers.
But in physics class one typically argues that the following are solutions:
T = g [ g(x) = Asin(x+c), A, c arbitrary real numbers.
The physical argument goes: The dierential equation f
= f is a special case of Hookes Law,

with k = 1. All solutions to Hookes Law are sinusoidal. Since here k is xed, the frequency of
the solution is xed, but not the amplitude A nor the phase shift c.
Determine the relationship between S and T. For instance, is one contained in the other?
Section J. Linear Combinations and Spans
1. Let A be a mn matrix. The row space of A, denoted
A
or (A), is the span of the rows
of A. The column space of A, denoted (
A
or ((A), is the span of the columns of A.
a) It is immediate that
A
and (
A
are vector spaces. Why?
b) For what k is
A
R
k
? For what k is (
A
R
k
?
2. Come up with a simple geometric description of Span
_
(1, 1, 0), (1, 2, 0)
_
. Convince me that
you are correct.
3. Express
_
_
1
2
5
_
_
as a linear combination of
_
_
1
3
2
_
_
,
_
_
2
4
1
_
_
,
_
_
1
5
7
_
_
, if it can be done.
4. Is
_
_
1
2
_
_ in the span of
_
_
1
3
_
_,
_
_
2
4
_
_,
_
_
1
5
_
_?
5. Show that the polynomial q(t) = t
2
+ 4t 3 is a linear combination of
p
1
(t) = t
2
2t + 5
p
2
(t) = 2t
2
3t
p
3
(t) = t + 3.
6. Let F be the set of functions that satisfy
f
(x) 3f
(x) + 2f(x) = 0 (1)

(As always, we mean that (1) holds for all real x.)
a) Check that f(x) = e
x
and f(x) = e
2x
happen to be in F.
b) Use a), and the fact that F is a vector space, to name a very large collection of functions
in F.
page 40 Section JA, Dependence and Independence
c) From this large collection, nd an f that satises not only (1) but also the boundary
conditions
f(0) = 1 and f(1) = 5.
Note that a) and b) reduce answering c) to solving linear equations, even if they are messy
ones.
7. In class we have/will prove that two matrices have the same row space i they have the same
rref (except for zero rows). Prove or disprove: Two matrices have the same column space i
they have the same rref.
8. If the point of the denition of Span(S) is to augment S just enough to get a vector space, why
dont we dene Span S to be
cv [ c R, v S u + v [ u, v S ?
If thats no good, why not
cu + dv [ c, d R, u, v S ?
9. Determine if Span
_
(1, 1, 2), (2, 1, 1)
_
= Span
_
(4, 1, 5), (0, 1, 1)
_
.
10. Devise and justify an algorithm for determining if the span of one set of vectors is contained in
the span of another. You may assume both sets are nite sets of vectors in R
n
.
11. Prove or disprove: If S, S
are nite subsets of vector space V , then

Span(S S
) = Span(S) Span(S
).

Span(S S
) = Span(S) Span(S
).

Span(S S
) = Span(S) + Span(S
),
where the sum of vector spaces is dened in I11.
Section JA. Dependence and Independence
1. Determine if the vectors (1, 2, 1), (2, 1, 1) and (7, 4, 1) are independent or dependent.
2. Determine whether the matrices
_
1 1
1 1
_
,
_
1 0
0 1
_
,
_
1 1
0 0
_
are independent or dependent.
3. Determine (without a calculator!) if the columns of the following matrix are independent.
_
3.7
2
10
6
4/3 log 10 sin10
_
page 41 Section JA, Dependence and Independence
4. T
2
, the set of all polynomials of degree at most 2, is a vector space. Show that the polynomials
x
2
, x, 1 are an independent spanning set for T
2
. (Here 1 means the constantly one function,
p(x) = 1 for all x. This is a degree 0 polynomial, since it may be written as p(x) = x
0
. Also,
S is a spanning set for vector space V means V = Span(S).)
5. Show that sin x and cos x are independent.
6. Show that sinx, sin(x +
1
3
), and sin(x +
1
4
) are dependent. (More generally but Im not
asking you to prove this any three or more functions of the form sin(x + c
1
), sin(x + c
2
), . . .
are dependent.)
7. Prove that the functions e
x
, e
2x
and e
3x
are linearly independent.
8. Show that 1, x, x
2
, e
x
, e
x
is independent. Hint: the plugin method will be a mess, requiring
5 values. But with another method, pursued with a vengeance, you can dispatch this problem
quickly if you also apply facts that we have probably already proved in class.
9. Prove that the functions e
x
, xe
x
, and x
2
e
x
are independent. (This is easy given what we know
about independence of polynomials.)
10. Prove that the functions sinx, sin 2x, and sin 3x are independent.
11. Prove that the functions sinx, sin 2x, and sin x are independent. Generalize.
12. Denition. A pair of vectors are parallel is one of them is a scalar multiple of the other.
a) Prove: if a set of two vectors is dependent, then they are parallel.
b) Show that a set of 3 vectors can be dependent without any two of them being parallel.
c) One might instead propose the following slightly dierent wording for the denition of
parallel: vectors u and v are parallel if v is a scalar multiple of u. Is this proposal
equivalent to the denition above, that is, are exactly the same pairs parallel under both
denitions?
13. Prove or disprove: Let u,v,w be vectors in some space V . If the sets u, v, u, w, and v, w
are each independent, then the single set u, v, w is independent.
14. Show: If w depends on u, v, and x depends on u, v, w, then x depends on u, v.
15. Suppose u, v, w are dependent. Come up with an example where u does not depend on
v, w. Come up with an example where, in addition, v does not depend on u, w.
16. Explain why every superset of a dependent set is dependent. (More precisely, if S is a dependent
set in vector space V , and S T V , then T is dependent. That is, the elements we add to
the dependent set S must be in the same vector space, or else it doesnt make sense to ask if
the new set is dependent. Also, you may assume S and T are nite.)
17. Explain why every subset of an independent set is independent.
18. Prove: if S = v
1
, v
2
, . . . , v
k
is an independent set in V , and v
k+1
is in V Span(S), then
T = v
1
, v
2
, ..., v
k
, v
k+1
is independent. Hint: Proof by contradiction. If T were dependent,
the coecient of v
k+1
in the sum equalling 0 would either be the 0 number or a nonzero number.
page 42 Section JB, Basis and Dimension: n-tuple Spaces
19. Suppose V = Span(S) and S is dependent. Show there is some v S that can be thrown out
without aecting what is spanned. That is, for some v S we have Span(Sv) = V .
20. a) Come up with a denition of independent of (as in u is independent of v
1
, v
2
, . . . , v
k
)
that relates to independent the same way depends on relates to dependent. (Cau-
tion: independent of is not standard usage in math, just as depends on is not. Unlike
in ordinary English, the only standard mathematical usage is independent and depen-
dent followed by set, not followed by a preposition.)
b) Show that the following claim is false, and then modify it so that its conclusion is true: If
v
k+1
is independent of v
1
, v
2
, . . . , v
k
, then v
1
, v
2
, . . . , v
k
, v
k+1
is independent.
21. Prove: If matrices A and B each separately have independent columns, then so does the matrix
_
A C
B
_
, no matter what C is. (Of course, C must have as many rows as A and as many
columns as B.)
22. Use the result of Problem 21 to give an alternative proof for Problem C10.
23. Suppose that both matrices A and B = [B
1
[B
2
] have m rows, that A has independent columns
and that B has independent columns. Prove that
_
A B
1
0
A 0 B
2
_
has independent columns.
Section JB. Basis and Dimension: n-tuple Spaces
1. In R
3
, nd the coordinates of (1,2,3) relative to the standard basis e
1
, e
2
, e
3
. (This is easy.)
2. In R
3
, nd the coordinates of (1,2,3) relative to the basis b
1
= (1, 0, 0), b
2
= (1, 1, 0) and
b
3
= (1, 1, 1).
3. Prove or disprove: (1, 1, 0), (1, 0, 1), (0, 1, 1) is a basis of R
3
.
4. Let U be the set of three vectors
_
_
1
2
_
_,
_
_
2
3
_
_,
_
_
4
5
_
_.
Explain why every vector in R
2
has coordinates relative to U but these coordinates are not
unique. Specically, nd all sets of coordinates of
_
_
1
3
_
_ relative to U.
5. Find the coordinates of the standard unit vectors of R
3
relative to the basis
B = v
1
= (1, 1, 0), v
2
= (1, 1, 1), v
3
= (1, 0, 1).
page 43 Section JC, Matrix Spaces
6. Let the columns of B be a basis B of R
n
. Suppose the coordinates of x relative to B are
c
1
, c
2
, . . . , c
n
. Let c = (c
1
, c
2
, . . . , c
n
)
T
.
a) Explain why x = Bc.
b) More likely, you know x (that is, you know its representation using the standard basis)
and you want to nd c. In light of a), how do you do it?
7. Find a basis for the span of (1, 2, 3), (2, 3, 4), (3, 4, 5).
8. For any nite subset U = u
1
, u
2
, . . . , u
n
of R
m
, prove that some subset of U is a basis of
Span(U). Hint: Let A be the matrix with u
1
, u
2
, . . . , u
n
as its columns; use GE.
9. Determine if the following vectors span R
4
.
_
_
1
1
0
0
_
_
,
_
_
0
1
1
0
_
_
,
_
_
0
0
1
1
_
_
,
_
_
1
0
0
1
_
_
.
10. The set T
3
of 33 diagonal matrices is a vector space (you need not show this). Find a simple
basis for this space and show that you are right.
11. Prove: any n independent vectors in R
n
must be a basis. Hint: Write the n vectors as the
columns of a matrix. What does independence of the columns tell you about GE for this
matrix?
12. Let u = (1, 0, 1, 1) and v = (0, 1, 1, 1). It is easy to see that u, v are independent. Let A be
the 42 matrix with u, v as the columns.
a) Find a basis of R
4
that contains u, v by row-reducing [A[ I ].
b) By the method of a) prove: Any independent set u
1
, u
2
, . . . , u
n
in R
m
can be extended
to a basis of R
m
(that is, u
1
, u
2
, . . . , u
m
is a subset of a basis).
Section JC. Matrix Spaces
1. Find a basis for the column space, the row space, and the null space of
A =
_
_
1 2 1 2 3
1 2 2 3 4
1 2 3 4 5
_
_.
2. Find a dierent basis for the column space of A in Problem JC1, and a dierent basis of its
row space, by reducing A
T
. Note: I still want bases for spaces associated with A; thus I said
the column space of A, not the column space of A
T
.
3. Find a basis for the column space, the row space, and the null space of the following matrix A.
_
_
1 2 3 4
2 3 4 5
3 4 5 6
4 5 6 7
_
_
4. Prove: For every matrix A, there is another matrix B so that the null space of A is the column
space of B. Hint: this is easy; think how you would show the claim for the specic matrix
A =
_
_
1 0 2 1
0 1 1 1
1 0 2 1
_
_
5. An inconsistent system Ax = b has 13 equations and 17 unknowns. What is the largest possible
rank of A? Explain briey.
6. This problem and the next are practice for understanding the proof of the main theorem about
how to nd bases for matrix spaces. The calculations themselves could have been done in
Section B.
Suppose the augmented matrix for Ax = b row-reduces to
_
_
0 1 2 0 3 5 A
0 0 0 1 4 6 B
0 0 0 0 0 0 0
_
_.
a) Write down the general solution for x = (x
1
, x
2
, . . . , x
6
)
T
.
b) Show that no matter what A, B are, there is among the many solutions for x a unique one
in which only the variables for the pivot columns are nonzero.
c) Why does b) imply that the pivot columns of A are a basis for the span of the columns of
A?
d) Write down the general solution to Ax = 0. Verify that the vectors that get multiplied by
free variables in your solution are independent by writing them as the columns of a new
matrix B and identifying rows to cross out so that the remaining matrix is I. Verify that
row i is crossed out x
i
is a pivot variable.
7. Suppose now that
rref(A) =
_
_
1 2 0 3 0
0 0 1 4 0
0 0 0 0 1
_
_.
Write down the general solution to Ax = 0. Verify that the constant vectors in your solution
are independent by writing them as the columns of a new matrix B and identifying rows to
cross out so that the remaining matrix is I. Verify that row i is crossed out x
i
is a pivot
variable.
8. Let V be the row space of the matrix
M =
_
1 2 3
2 3 4
_
.
The two rows happen to be independent and thus form a basis B of V .
a) Let N = rref(M). As we have proved, the nonzero rows of N are also a basis of V . Call
this basis B
. Find B
.
b) Find the coordinates of the vectors in B
relative to B.
c) Find the coordinates of the vectors in B relative to B
.
9. Determine whether U = (1, 0, 1), (1, 1, 1) is a basis for the span of
V = (1, 2, 3), (2, 3, 4), (3, 4, 5) .
10. The set of solutions to
x + 2y + 3z + 4u + 5v = 0
2x + 3y + 4z + 5u + 6v = 0
3x + 4y + 5z + 6u + 7v = 0
is a at thing in R
5
, in fact, a vector space. Why is it a vector space? What is its dimension?
11. Prove: rref(C) has a 0 row rref(C
T
) has a nonpivot column. Hint: I dont think this
can be done by analyzing what Gaussian elimination does on C and C
T
, because it will be
so dierent. But you can use your knowledge of the dimensions of some of the vector spaces
associated with C and C
T
to prove this easily.
12. Prove
Theorem. Two matrices have the same row space i they have the same rref (up
to 0-rows)
by rst proving
Lemma. Let R, R
be two matrices in row reduced echelon form. Then R and R
have the same row space if and only if they are identical except that they may have
a dierent number of 0 rows.
13. The left nullspace of a matrix A is y [ yA = 0. This is the 4th fundamental subspace
associated with A. Bases for the other three can be read o from rref(A), but not a basis for
the fourth. However, there is a slight extension of the rref computation (one we have already
done for other purposes) that does exhibit a basis for the left null space.
a) Identify the extended computation and prove that it provides a basis. You must show that
the rows you describe are in the left null space, span it, and are independent.
b) The usual notation for the left null space of A is ^
A
T. Why? This also suggests another
correct way to nd a basis (but one which requires a whole new computation, rather than
an extension of the usual GE on A). What way?
page 46 Section JD, Basis and Dimension: General Spaces
Section JD. Basis and Dimension: General Spaces
1. Prove: If

n
i=1
c
i
v
i
=

n
i=1
d
i
v
i
, where the sums on the two sides are not the same, then
v
1
, v
2
, . . . , v
n
is dependent. That is, if v
1
, v
2
, . . . , v
n
failes the uniqueness part of the
unique representation property, then v
1
, v
2
, . . . , v
n
is dependent. Hint: I want you to practice
an important algebraic technique (that has been discussed in class, or will be after you do this
problem).
2. Suppose V = Span(S) where S = v
1
, v
2
, . . . , v
n
R
m
. Suppose further that there is some
u V for which u is a linear combination of S in two dierent ways. Prove: every w V is a
linear combination of S in innitely many ways.
3. Consider the vector space of period 2 harmonic functions:
V = f(x) = Asin(x+c) [ A, c real numbers
(so x is the variable within each function and dierent values of A and c give dierent functions
in V ).
We know from problem I13 that another representation of V is
V = f(x) = Bsinx + Dcos x [ B, D real numbers .
The latter representation is good because it is in terms of a basis and thus solving for B, D is
easy because the equations are linear.
a) Find A and c in the rst representation for the function f in V that satises both f(0) = 1
and f
(/4) = 1.
b) Find B and D giving the same f in the second representation.
Do you agree that the second representation is easier to work with?
4. Determine if the three polynomials
x
2
+ x + 1, x
2
+ 2x 1, x + 2
are a basis of T
2
, the polynomials of degree at most 2.
5. Show that (x2)
2
, (x2), and (x2)
0
(the last of these is just the constantly 1 function) are
a basis of the space of polynomials of degree 2 or less.
6. The set of all polynomials p(x) of degree at most 2 for which p(3) = 0 form a vector space.
Find a basis and show that you are right.
7. The set of all polynomials p(x) of degree at most 2 for which p(2) = 2p(1) form a vector space.
Find a basis and show that you are right.
8. The falling factorial functions are the polynomials
x
(0)
= 1 [the constantly 1 function]
x
(1)
= x
x
(2)
= x(x1)
and in general
x
(k)
= x(x1)(x2) (xk+1).
Show that x
(0)
, x
(1)
, x
(2)
, x
(3)
are a basis of T
3
, the polynomials of degree at most 3. (Once you
do this, you will probably be correctly convinced that x
(0)
, . . . , x
(k)
are a basis of T
k
. The only
question will be: do they have an advantage over the standard basis 1, x, . . . , x
k
?)
9. Suppose f
1
, f
2
, f
3
is a basis of a certain function space F. Suppose g
1
, g
2
, g
3
are three other
functions in F whose coordinates relative to the fs are
g
1
: 1, 1, 0 (i.e., g
1
= f
1
+ f
2
)
g
2
: 1, 1, 1
g
3
: 1, 0, 1.
a) Show that g
1
, g
2
, g
3
is also a basis of F.
b) Since g
1
, g
2
, g
3
is a basis, then each of f
1
, f
2
, f
3
must have unique coordinates relative
to the gs. Find these coordinates.
Note that we havent the slightest idea what these functions are or what space they are in, yet
the question can be answered.
10. Find a basis for the bizarre vector space V of Problem H14.
11. Prove: any n independent vectors in an n-dimensional vector space V must be a basis. Hint:
As usual, represent the vectors by their n-tuples relative to some basis, known to exist because
we are told that the space is n-dimensional. Write these n n-tuples as the columns of a matrix
A. Since the original vectors are independent, what happens when you row-reduce A and what
can you therefore conclude?
12. (parallel worlds) Fact (which you arent asked to prove): V = p(x) [ p T
2
and p(2) = 0 is
a vector space, and a basis is B = x2, x
2
4x+4.
a) Relative to B, nd the coordinates of q(x) = x
2
x 2 and r(x) = x
2
+ x 6. (Dont
worry, q(x) and r(x) are in V .) Write these coordaintes as vectors. These coordinate
vectors are in the parallel world to V .
b) s(x) = q(x) + r(x) must be in V also, so nd the coordinate vector of s(x) relative to B
as well, by the same method you used for a).
c) In the parallel universe of coordinates, it is much easier to add the images of q(x) and
r(x). Do so, and verify that the result is the same one you got in b). This should be
so, because vector operations, not just the vectors themselves, should be mimicked in the
parallel world.
13. Let B = v
1
, v
2
, . . . , v
n
be a basis for V . Suppose c
1
, c
2
, . . . , c
n
are the coordinates of v
relative to B, that is v =

n
i=1
c
i
v
i
. Suppose d
1
, d
2
, . . . , d
n
are the coordinates of u relative to
B. Explain why c
1
+d
1
, . . . , c
n
+d
n
are the coordinates of v +u.
14. Let V be any nite dimensional vector space that consists of more than a 0 vector. Using the
results of Problems JA18 and JD11 show that the following algorithm must terminate, and
when it does, S is a basis for V . (All this algorithm does is keep picking vectors from the part
of V not yet spanned, going on forever unless at some point V is spanned.)
Input V [V a subspace of R
n
, V ,= 0]
Output S [should be a basis of V ]
Algorithm Grow-Basis
Pick any nonzero v V
S v [rst vector goes into S]
repeat until Span(S) = V
Pick any v V Span(S)
S S v
endrepeat
15. Modify the algorithm of Problem 14 slightly to prove the following: If U is a nontrivial proper
subspace of n-dimensional vector space V , then dim(U) < n. (Proper means not equal to.)
16. Consider the following algorithm.
Input S, V [S a nite spanning set of vector space V ]
Algorithm Shrink-to-Basis
repeat until S is independent
Pick v S such that v depends on Sv
S Sv
endrepeat
Output S
a) Show that the output of this algorithm is a basis for V . (See Problem JA19.)
b) Explain why you have proved: Every nite spanning set of a vector space contains a basis.
17. (Alternative proofs that dimension is well dened)
a) Show that for any matrix A with independent rows, rref(A) has no 0 rows.
b) Let B
2
and B
2
be two bases of the same space, and let B
1
(respt. B
2
) be the matrix
obtained by writing the vectors of B
1
(respt. B
2
) as rows. Using the Theorem in JC12,
prove that B
2
and B
2
have the same number of vectors (and thus dimension is well dened).
c) Now we prove dimension is well dened using Gaussian elimination but without the The-
orem from JC12. Let C
1
=
_
B
1
B
2
_
(that is, the rows of B
2
are stacked under B
1
), and
let C
2
=
_
B
2
B
1
_
. Prove by properties of row reduction that rref(B
1
) = rref(C
1
) except
for 0 rows. By exactly the same reasoning (so dont do it), rref(B
2
) = rref(C
2
) except for
0 rows. Finally, show by other properties of row reduction that rref(C
1
) = rref(C
2
). By
transitivity, rref(B
1
) = rref(B
2
).
d) Why does part c) prove that the dimension of a vector space is well dened?
e) Well, actually, the arguments above prove that dimension is well dened only for vector
spaces in some R
n
, for otherwise how can the vectors in the bases B
2
and B
2
be written as
rows of a matrix? Nonetheless, show how to make these proofs work for any vector space
which has a basis with a nite number of vectors in it.
page 49 Section JE, Complex Vector Spaces, Part II
Section JE. Complex Vector Spaces, Part II
Denition. C
n
= (z
1
, z
2
, . . . , z
n
) [ z
i
C. C
n
is a complex vector space: just as the proof that
R
n
is a vector space followed coordinatewise from real number facts, the proof that C
n
is a complex
vector space follows coordinatewise from the same facts, since they are also complex number facts.
1. Are
_
_
1
i
_
_,
_
_
i
1
_
_ independent in C
2
? What about
_
_
1
i
_
_,
_
_
i
1
_
_?
2. Whats the dimension of the complex plane as a complex vector space? Name a basis.
3. dim(C
n
) = n; the proof that dim(R
n
) = n goes through unchanged.
But C
n
can also be regarded as a real vector space, by using the same + and scalar multi-
plication, but restricting the scalars to R. Prove that, as a real vector space, C
n
has dimension
2n.
4. Suppose that W is a complex vector space with dimension n. Then W is also a real vector space.
Prove: if z
1
, z
2
, . . . , z
n
is a basis for W as a complex vector space, then z
1
, iz
1
, z
2
, iz
2
, . . . , z
n
, iz
n
is a basis of W as a real vector space.
Section JF. Minimality, Maximality, and Spaces without Dimensions
1. A minimal spanning set in a vector space V is any set S V such that
i) S spans V , that is, V = Span(S), and
ii) If any element is removed from S, the remaining set does not span V .
a) Which of the following sets in R
2
are minimal spanning sets?
S
1
= (1, 2), (2, 3), (3, 4) ,
S
2
= (1, 2), (2, 3) ,
S
3
= (1, 2), (2, 4)
S
4
= (1, 2).
b) Prove: S is a minimal spanning set of V S is a basis. Hint: See Problem JA19or
JD16.
c) Why does it follow that every minimal spanning set in a given vector space V is the same
size?
2. A maximal independent set in a vector space V is any set S V such that
i) S is independent, and
ii) If any other element in V is added to S, the enlarged set is dependent.
a) Which of the following sets in R
2
are maximal independent sets?
S
1
= (1, 2), (2, 3), (3, 4) ,
S
2
= (1, 2), (2, 3) ,
S
3
= (1, 2), (2, 4) ,
S
4
= (1, 2).
page 50 Section JF, Minimality, Maximality, and Spaces without Dimensions
b) Prove: S is a maximal independent set of V S is a basis. Hint: See Problem JA18.
c) Why does it follow that every maximal independent set in a given vector space V is the
same size?
3. A set S is minimal with respect to some property T if
i) S satises the property, and
ii) If any element is removed from S, the remaining set does not satisfy the property.
In the rest of this problem, let T be the property the elements have no common factor other
than 1.
a) Which of the following sets of integers are minimal with respect to this property?
S
1
= 6, 10, 15,
S
2
= 3, 5,
S
3
= 3, 5, 7,
S
4
= 15, 21.
Note: A set consisting of a single integer n > 1 does not have this property, because this
set has the common factor n.
b) Do all minimal sets for this T have the same size? (Compare with Problem JF1c.)
4. A set S (within a collection of allowed sets o) is minimum with respect to some property T if
ii) No other set in o that has the property has a smaller number of elements.
Please compare with the denition in Problem JF3; note that the endings al and um
makes a big dierence. This is a distinction that does not arise in ordinary English.
a) Let the property be the elements have no common factor other than 1 and let the
allowed sets be nite nonempty sets of positive integers. Which if any, of the following
sets of integers are minimum with respect to this property?
S
1
= 6, 10, 15,
S
2
= 3, 5,
S
3
= 3, 5, 7,
S
4
= 15, 21.
b) If none of the sets in a) were minimum, name a minimum set.
5. Let o be the set of all nite nonempty sets of vectors in a vector space V . Let T be the property
of spanning V . In this context, is there a distinction between minimal and minimum?
6. Let T be the set of 9 positions on a tic-tac-toe board, and for S T let T be the property that
S intersects with every possible winning 3-in-a-row. (Think of this as a game where you get to
put down all your Xs rst and you try to block all winning congurations for O with as few
Xs as possible.)
a) Find a set S which has property T but is not minimal.
page 51 Section JF, Minimality, Maximality, and Spaces without Dimensions
b) Find another set S which is minimum with respect to T.
c) Find another set S which is minimal but not minimum with respect to T.
7. A set S within some larger set T is maximal (in T) with respect to some property T if
ii) If any other element in T is added to S, the enlarged set no longer satises the
property.
Again, let T be the set of 9 positions on a tic-tac-toe board, but now let T be the property
that S contains no 3-in-a-row. (Think of this as a solitaire game in which you just put down
Xs and try to keep yourself from winning as long as you can.)
a) Find a set S T which has property T but is not maximal.
b) Find another set S T which is maximal with respect to T.
8. A set S within some larger set T is maximum with respect to some property T if
ii) No other subset of T that has the property has a larger number of elements.
Compare with the denition in Problem JF7; note that the endings al and um makes a
big dierence. This is a distinction that does not arise in ordinary English.
Let T be the set of 9 positions on a tic-tac-toe board, and let T be the property that S
contains no 3-in-a-row.
a) Find a maximum set and prove that you are right that it is maximum.
b) Find a maximal set which is not maximum.
c) What should we mean by a minimum maximal set for a property T? Dene this concept,
and for the T of this problem, nd, with justication, a minimum maximal set.
9. Let o be the set of all nite nonempty sets of vectors in a vector space V . Let T be the property
of being independent. In this context, is there a distinction between maximal and maximum?
10. Call a property an independence-like property if, for every set S that satises the property,
every nonempty subset of S also satises the property.
a) Is independence in vector spaces an independence-like property?
Call a property a dependence-like property if, for every set S that satises the property,
every superset of S satises the property. (Actually, you have to set a universe in which you
take supersets. For instance, if S is a set of vectors, your universe will be the vector space they
sit in. You dont want to start adding dogs, airplanes, etc., to the set S.)
b) Within any vector space, is dependence a dependence-like property?
c) Within any vector space, is spanning a dependence-like property?
d) Weve just named one independence-like property and two dependence-like properties in a
vector space. But everything in vector spaces is supposed to come in pairs. Name another
independence-like property.
e) Is being a basis an independence-like property? Is it dependence-like?
page 52 Section K, Talk Right!
11. For this and the next problem, let our space T be the 9 positions on a tic-tac-toe board, and
dene a set S in this space to be independent if it contains no winning line. Independent
is a reasonable name because such sets avoid the redundancy of having 3 in a row. We also
dene a set to be spanning if it intersects every winning line. Spanning is a reasonable
name because such sets reach every winning line.
a) Are these names reasonable in another sense? Is independence an independence-like prop-
erty? Is spanning a dependence-like property?
b) In a vector space, a basis is an independent spanning set. So, dene S T to be a basis
of T if S is independent and spanning. Does every basis of T have the same size? That
is, can we dene dimension within T?
12. In a vector space, we have proved that the following conditions are equivalent, that is, the
collection of sets that satisfy any one of these conditions is the same as the collection that
satises any other of them:
i) S is independent and spanning,
ii) S is minimal spanning,
iii S is minimum spanning,
iv) S is maximal independent,
v) S is maximum independent.
Furthermore, the sets that meet any one of these conditions all have the same size the
dimension of the space.
With the denitions of independent and spanning for tic-tac-toe in Problem JF11, show
that no 2 of the 5 conditions above are equivalent for the tic-tac-toe space.
Section K. Talk Right!
In this section, the instruction correct the wording means gure out what the wording is supposed
to mean and then say it right.
1. Correct the wording in the following statements.
a) v V .
b) v
1
, v
2
, . . . , v
n
V .
b) Consider the vector space u
1
, u
2
, . . . , u
n
.
2. Suppose v
1
, v
2
, . . . , v
k
are in vector space V . State what is means for v
1
, . . . , v
k
to be a basis
of V
a) using the words span and independent;
b) using instead the phrase linear combination.
a) A basis of a matrix is the columns with pivots.
b) (1,0) and (0,1) are the span of R
2
.
c) Addition is closed in R
2
.
d) R
2
spans (1,0) and (0,1).
page 53 Section K, Talk Right!
e) R
2
is the spanning set of (1, 0), (0, 1).
f)
_
1 2 1
3 6 2
_
=
_
1 2 1
0 0 1
_
, so the matrix has no solutions.
4. Correct the wording in the following statement: u + v is a vector space because it is closed.
(This one needs a lot of work!)
a) A matrix A is invertible if it has unique solutions.
b) Every basis of a vector space V has the same dimension.
c) To create a matrix A whose row space is V , let the rows of A be the span of V .
d) A nontrivial linear combination of vectors is independent if it does not equal 0.
e) The number of vectors in ^
A
is the number of free variables in A.
f) An independent set v
1
, v
2
, . . . , v
n
is the dimension of V if it is the span.
page 54 Section L, Dot Products and Norms
Section L. Dot Products and Norms
1. For u = (1, 2, 3), v = (1, 0, 1) and w = (0, 2, 1), compute
a) (u + v) w
b) u w + v w
c) 2 (v w)
d) v (2w)
2. Prove: u (kv) = (ku) v = k(u v).
3. Let u = (5, 4, 1), v = (3, 4, 1) and w = (1, 2, 3). Which pairs of these vectors, if any, are
perpendicular?
4. Find [v[ if v = (1, 2, 3).
5. Normalize (2, 1, 3, 2).
6. For u = (1, 1, 1), v = (0, 1, 2) and k = 3 verify that
a) [kv[ = [k[ [v[
b) [u + v[ [u[ +[v[.
7. The distance between two points p and q is dened to be [p q[. Find the distance between
(1,7) and (6, 5).
8. Find k such that (1, 2, k) and (2, 4, 3) are 3 units apart. Before you solve, can you determine
by geometric reasoning (i.e., a picture in your head) how many solutions for k will exist?
9. Find cos , where is the angle between (1, 1, 1) and (1,2,2).
10. Find the projection of u = (1, 1, 1) on v = (1, 2, 2).
11. The Triangle Inequality says: For all u, v R
n
, [u + v[ [u[ + [v[. The Cauchy-Schwarz
Inequality (CSI) says: For all u, v R
n
, [u v[ [u[ [v[.
Prove that the Triangle Inequality implies CSI.
12. Whether or not we have proved it yet, assume that the angle between vectors u and v in R
n
is well dened. (That is, assume that 1 1, where =
u v
[u[[v[
, so that cos
1
makes
sense.) Prove that the Law of Cosines is true in R
n
.
13. You probably know (and Euclid knew) that the converse of the Pythagorean Theorem is true.
State this converse in traditional language and in vector language. Then prove it in the latter
form.
14. Prove in R
n
using dot products: A parallelogram is a rhombus i the diagonals are perpendic-
ular.
15. Prove the following geometry theorem by dot product methods. Let AB be the diameter of a
circle, and let C be any other point on the circumference. Then ABC is a right triangle, with
right angle at C.
page 55 Section L, Dot Products and Norms
16. Prove the following variation of the triangle inequality:
[u v[ [u[ + [v[.
You can prove this variation from the CSI (parallel to the proof of the regular triangle inequality)
or by a one-step substitution into the regular triangle inequality.
17. Prove: for any vectors u, v R
n
that
[u v[ [u[ [v[.
Hint: this follows easily from the triangle inequality, but not if you apply the triangle inequality
to u and v.
to recapitulate.
18. Let p be the projection of u on v.
a) Show that [p[ = [u v[/[v[.
b) Without using a), show that [p[ [u[. Hint: Projections were dened so that u p has
what property?
c) Prove CSI again.
19. Using the gure below, prove the Law of Cosines by traditional methods, as follows. In triangle
ABC, let D be the foot of the perpendicular from B. Use a and to express the lengths of
BD and CD. Now use this knowledge to express the length of DA. Now use the Pythagorean
Theorem to express c
2
. Simplify.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ............................................................................................................................................................................................................................................................................................................................................................. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...............................................................................................................................................................................
A
B
C
D
a c
b
20. Here you will do another traditional proof of the Law of Cosines, this time by coordinate
geometry.
We may place ABC with C at the origin and CA along the x-axis as follows:
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ............................................................................................................................................................................................................................................................................................................................................................. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A
B
C

a c
b (0,0) (b, 0)
(x, y)
a) Express (x, y) in terms of a and .
b) Use the distance formula to express c
2
Then simplify.
page 56 Section LA, Lines and Planes
Section LA. Lines and Planes
Note. Several problems below ask for an equation of a plane. Give a cartesian equation, that is,
something of the form ax + by + cz = d.
1. Two airplanes y along straight lines. At time t plane 1 is at (75, 50, 25) + t(5, 10, 1) and
plane 2 is at (60, 80, 34) +t(10, 5, 1).
a) Do the planes collide?
b) Do their ight paths intersect?
2. Find an equation of the plane in R
3
which passes through p = (3, 2, 1) and has normal
n = (2, 5, 6).
3. Find an equation of the plane in R
3
which contains p = (1, 5, 2) and is parallel to the plane
3x 7y + 4z = 5.
4. Find an equation of the plane with normal direction i + 2j + 3k and passing through (2,5,2).
5. Let P and Q be, respectively, the planes x + 2y 4z = 5 and 2x y + 3z = 7. Find the
angle between the planes. You may express the angle as an inverse trig quantity. Hint: draw
a picture to connect this angle to the normal vectors.
6. Find the angle between the two planes x +y + z = 3 and 3x 2y +z = 4.
7. Consider the sphere S (in R
3
) centered at the origin with radius 3. One point on this sphere is
p = (1, 2, 2).
a) Name outward and inward facing normals to S at p.
b) Name the tangent plane to S at p in cartesian form.
8. A line segment AB, where A is a point on plane P, is said to be perpendicular to P if it is
perpendicular to every line on the plane through A. A famous hard theorem in Euclid (hard
to prove by the methods available to Euclid) is: If AB is perpendicular to two lines in P, then
it is perpendicular to P.
Prove this theorem easily by dot methods.
9. Show: If x and x
are two points that satisfy

a
1
x
2
+a
2
x
2
+ +a
n
x
n
= b,
then the vector y from one to the other satises a y = 0, where a = (a
1
, . . . , a
n
).
10. Convert the line x + 2y = 3 in R
2
into parametric form. There are many right answers.
11. Consider the line given parametrically in R
2
by x = (1, 0) +t(1, 2). Find a cartesian equation
of this line.
12. Find the distance from point (1,5) to the line in Problem LA11.
page 57 Section LA, Lines and Planes
13. Given a plane n x = d in R
3
, up is dened to be in the direction of n and down is in the
other direction. (Of course, this choice is relative, since the same plane could be represented
by (n) x = d, and then up and down are reversed). More precisely, the point p is said to
be above the plane n x = d (relative to n) if the projection on n of any vector from the plane
to the point is a positive multiple of n, and below the plane if this projection is a negative
multiple of n.
a) Determine if (1, 2, 1) is above or below the plane x +y + 2z = 0.
b) Determine if (1, 2, 1) is above or below the plane x +y + 2z = 2.
c) In general, show that p is above n x = d i n p > d.
14. Let x
0
be any point on the plane n x = d and let p be a point in space. Let f be the closest
point on the plane to p (f for foot of the perpendicular). Explain why the projection P
n
(px
0
)
is the vector from f to p, so that f = p P
n
(p x
0
).
15. Let P be the plane x +y + 2z = 3 in R
3
.
a) Find the distance from the origin to P
b) Find the closest point on P to the origin.
c) Find the distance from (1,2,3) to P.
d) Find the closest point on P to (1,2,3).
16. Let L be the line x 2y = 3 in R
2
. Find the distance from (2,3) to L and the closest point to
(2,3) on L.
17. Find a normal equation (i.e., one of the form ax + by + cz = d) for
a) the plane (1, 2, 1) + s(2, 1, 1) +t(1, 1, 2)
b) the plane containing the points (1, 1, 1), (4, 1, 3) and (1, 5, 1).
c) Is either plane a subspace of R
3
?
18. For this question you will consider the following lines in R
3
:
L
1
: x = (1, 0, 0) +t(1, 1, 1),
L
2
: x = (0, 1, 0) +t(1, 1, 1),
L
3
: x = (0, 1, 1) +t(1, 1, 1),
L
4
: x = (2, 0, 0) +t(1, 2, 3),
L
5
: x = (1, 3, 2) +t(1, 2, 3).
For each of the following pairs of these lines, determine if they i) are the same, ii) intersect
without being the same, iii) are parallel (without being the same), or iv) are skew.
a) L
1
and L
2
b) L
1
and L
3
c) L
1
and L
4
d) L
1
and L
5
Even if you can answer some parts in your head, gure out a systematic matrix method; youll
need it for Problem LB15.
page 58 Section LB, R
n
: Lines, Hyperplanes and Flats
19. Consider the skew lines (1, 0, 0) +s(1, 1, 0) and (0, 1, 0) +t(0, 1, 1).
a) Show directly that they dont intersect (even though part b will also show this).
b) Find the distance between them.
c) Find the points on the two lines that are this distance apart (they are unique).
20. Find the angle between the line (1, 2, 3) +t(1, 0, 1) and
a) The plane x + 2y +z = 2
b) The plane through (2, 3, 4) with direction vectors (1, 0, 0) and (2, 1, 0).
21. Suppose I ask you for the angle between a plane and a line, but the line is parallel to the plane,
or on the plane.
a) Will the method you used for Problem LA20 alert you to this?
b) Do LA20a again, but make the plane x + 2y z = 2.
22. Let
1
and
2
be two skew lines in R
3
. Suppose p
1
, p
2
are points on
1
,
2
respectively such
that the vector p
1
p
2
is perpendicular to both lines. Fill in the following proof that p
1
p
2
is
the shortest line segment between the two lines. (You probably assumed this fact in answering
Problem 19 above.)
Let the parametric equations of the lines be x = p
1
+sd
1
and x = p
2
+td
2
(we can always
pick the previously noted points p
1
, p
2
as the base points of the lines). Call the vector p
1
p
2
perpendicular to both lines n for short. Let q
1
, q
2
be any two points on the line.
a) Explain why q
1
q
2
= n + r, where r = s
0
d
1
+t
0
d
2
for some scalars s
0
, t
0
.
b) Explain why n r.
c) Use the Pythagorean theorem to prove that [n[ [q
1
q
2
[. This inequality says that n is
a shortest segment between the skew lines.
d) In fact, if
1
,
2
are skew (not parallel), and q
1
, q
2
,= p
1
, p
2
, then [n[ < [q
1
q
2
[;
that is, n is the unique shortest segment between the lines. Prove this by explaining why,
under these circumstances, s
0
d
1
+t
0
d
2
,= 0.
Section LB. R
n
1. The matrix equation Ax = 0 represents the intersection of 4 planes in R
3
and this intersection
is a line. What is the rank of matrix A and what is its nullity?
2. Find the angle in R
4
between
a) The vectors (1, 1, 1, 1) and (1, 1, 1, 1).
b) The hyperplanes x y + 2z w = 3 and x 3z + 2w = 4
3. a) Find the angle between the line t(1, 1, 1) and the positive axes in xyz-space.
b) More generally, in R
n
nd the angle between the line t(1, 1, . . . , 1) and each of the positive
axes. Does the angle get bigger or smaller as n increases? What is the limit as n ?
4. Find the angle in R
4
between the line s(1, 2, 3, 4) and the hyperplane x 2y + 2z w = 5.
5. Generalize the denitions of above and below in Problem LA13 to hyperplanes in R
n
.
Then determine whether (1, 2, 3, 4, 5) is above or below the hyperplane xy +2z 2u+3v = 4
in R
5
.
n
6. Consider the hypersphere S in R
4
centered at the origin with radius

7. One point on this
sphere is p = (1, 1, 2, 1).
a) Name outward and inward facing normals to S at p.
b) Name the tangent hyperplane to S at p in cartesian form.
7. Let B
4
be the ball of radius 2 in R
4
, centered at (1,2,3,4) Find the tangent hyperplane to the
ball at (2,3,4,5).
8. Find the distance in R
4
from the point (1,2,3,4) to the line p = t(1, 1, 1, 1).
9. Find all vectors in R
4
perpendicular to (1, 1, 0, 0), (1, 0, 2, 0) and (0, 0, 1, 1). How many such
vectors are there? How many dimensions worth of such vectors are there? How many such unit
vectors are there?
10. Suppose p, q, r, s are mutually orthogonal nonzero vectors in R
4
. Suppose further that L is a
line on the plane through 0 generated by p, q; and that L
is a line on the plane through 0

generated by r, s. Can the angle between L and L
be determined? If so, what is it? Explain.

11. Weve really been using two dierent denitions of hyperplane in R
n
i. A translation of the span of n 1 independent vectors,
ii. The set of points x that satisfy n x = d for some xed nonzero vector n ,= 0 and some
scalar d.
Prove that the two denitions are equivalent, that is,
a) Any set x [ n x = d can be expressed as the translation of the span of n1 independent
vectors.
b) For every set S of vectors which is the translation of the span of n1 independent vectors,
there exists a nonzero n and a scalar d so that S = x [ n x = d.
In fact, while n in b) is not unique, all the dierent n dier only by scalar multiples.
12. Suppose P is an invertible matrix. Let p
j
be the jth column of P. Let S
j
be the span of all
the columns of P except p
j
and let H
j
be the at which is the translate of S
j
by p
j
. That is,
H
j
= p
j
+v [ v S
j
.
a) Why is H
j
a hyperplane?
b) Let p
j
be the jth row of P
1
. Explain why p
j
x = d is a correct normal equation for
H
j
. What is the value of d?
13. In R
3
, it is not possible for planes to intersect in just one point. Not so in higher dimensions.
In R
3
, so-called perpendicular planes actually have lines, one on each plane, that make every
possible angle. Not so for real perpendicular planes.
In R
5
, consider the xy-plane P
xy
= (x, y, 0, 0, 0) [ x, y R and the zw-plane P
zw
=
(0, 0, z, w, 0) [ z, w R.
a) Show that P
xy
and P
zw
intersect in exactly one point. (This is easy, as is everything in
this deliberately simple example.)
b) Show that every vector on P
xy
is perpendicular to every vector on P
zw
. (Remember, every
vector on either plane can be represented as a vector from the common point.)
n
c) Explain how to generalize this example to every R
n
for n 4. (That is, nd two planes
in R
n
which have exactly one point in common and such that all vectors on one are
perpendicular to all vectors on the other.
14. The situation in Problem 13 was not particularly unusual; it was just easy to see it there. Let
P = Span (1, 1, 1, 0), (0, 1, 1, 1) . Let P
= P
, the set of all vectors perpendicular to all

vectors in P.
a) P is a plane, that is, a at with dimension 2. Show that P
is also a plane.
b) Show that P and P
have exactly one point in common.

c) Show that every vector on P is perpendicular to every vector on P
.
d) P and P
have yet another property: every vector v in the starting space (in this case R
4
)
is the sum of a vector in P and a vector in P
, in fact, in a unique way. Show this. Do

this by showing that every v R
4
is a unique linear combination of the 4 specic basis
vectors of P, P
you got earlier. (use Gaussian elimination!).

Two subspaces of R
n
with the properties in bd) are called orthogonal complements. It is
a theorem (not hard to prove with what you know) that for any subspace U of R
n
, we have
that U and U
are orthogonal complements. The subject of orthogonal subspaces is taken up

in more detail in Section NA.
15. For this question you will consider the following planes (2-dimensional) in R
4
:
P
1
: x = (1, 0, 0, 0) +s(1, 0, 1, 2) +t(0, 1, 2, 3)),
P
2
: x = (2, 3, 5, 7) +s(1, 2, 3, 4) + t(1, 1, 1, 1),
P
3
: x = (0, 1, 2, 3) +s(1, 2, 3, 4) + t(1, 1, 1, 1),
P
4
: x = (0, 1, 1, 0) + s(1, 0, 0, 0) +t(1, 1, 0, 0),
P
5
: x = (2, 0, 0, 0) +s(1, 1, 1, 0) + t(1, 1, 1, 1),
P
6
: x = (2, 3, 5, 7) +s(1, 1, 1, 0) + t(1, 1, 1, 1).
For each of the following pairs of these planes, determine if they i) are the same, ii) intersect
in a single point (planes in R
3
cant do that!), iii) intersect in a line, iv) are parallel (without
being the same), v) are skew, or vi) are semi-skew. (For planes, semi-skew means dont
intersect and have only one dimension of direction vectors in common. The same concept is
called semiparallel in Problem LB18.
a) P
1
and P
2
b) P
1
and P
3
c) P
1
and P
4
d) P
1
and P
5
e) P
1
and P
6
16. Below are a point, a line through it, and a plane, all in R
4
:
p = (1, 4, 0, 5),
L = p + k(3, 1, 2, 6),
T = (1, 4, 2, 1) +k(1, 1, 0, 0) + j(1, 0, 2, 1).
n
a) Find the distance of p from T.
b) Find the closest point to p on T. Show that you are right.
c) Show that L and T intersect. Where?
d) Find the angle between L and T
17. A line is parallel to a plane if the line is parallel to a line on the plane. A line and a plane
are skew if they neither intersect nor are parallel. (In R
3
, lines can be skew but there is not
enough room for a line and a plane to be skew.) Show that
T :(1, 0, 3, 0) +k(0, 0, 1, 0) + j(0, 0, 0, 1) and
L :(1, 0, 0, 0) +k(1, 1, 0, 0)
are skew. Find the distance between them.
18. In R
3
, if two planes dont intersect, then they are parallel, i.e., have all their direction vectors
in common. In R
4
, two planes can fail to intersect and have only one lines worth of direction
vectors in common, i.e., they are semiparallel. In R
5
, planes can be skew: not intersect and
have no direction vectors in common. (Lines can be skew in R
3
, but planes need more room.)
a) Consider the planes
P
1
: s(1, 0, 0, 0) + t(0, 1, 1, 0) and P
2
: (0, 0, 0, 1) + u(0, 1, 0, 0) +v(1, 0, 1, 0).
Show that P
1
P
2
= and that P
1
, P
2
have just one dimension of directions in common.
Which directions?
b) Show that any two nonintersecting planes in R
4
must have at least one dimension of
directions in common.
c) In R
5
consider the planes
P
1
: s(1, 0, 0, 0, 0) +t(0, 1, 1, 0, 0) and P
2
: (0, 0, 0, 1, 1) +u(0, 1, 0, 0, 0) + v(0, 0, 0, 1, 0).
Show that they are skew.
d) What is the distance between the planes in part a). Why?
e) What is the distance between the planes in part c). Why?
19. If a line intersects a plane in R
3
, there are lots of pairs of direction vectors through the point of
intersection, one on the line, one on the plane, between which we could measure the angle. Show
that the way we have dened the angle between the line and the plane gives us the smallest of
all these angles.
20. Consider the following planes in R
4
, Q
1
: s(1, 0, 0, 0)+t(0, 1, 0, 0) and Q
2
: u(1, 0, 1, 0)+v(1, 0, 0, 1).
a) These planes have just one point in common. Show this. (Of course, this situation cant
happen to planes in R
3
.)
b) (Hard) Dene the angle between Q
1
and Q
2
to be the smallest angle between direction
vectors, one on each plane, through the point of intersection. Find the angle.
page 62 Section M, Inner Products
21. Let A be mn, let b be m1, and let S be the solution set to Ax = b. Geometrically, S is the
intersection of m hyperplanes in R
n
. (If a
i
is the ith row of A, then the solutions to a
1
x = b
i
form the ith of these hyperplanes.) Show that intersecting S with one more hyperplane results
in a set S
of dimension at most 1 less, unless S
is empty. In particular, it follows that the

dimension of S itself is at least n m, unless S is empty. Note: since S is a translation of ^
A
(if it is nonempty), its dimension is dened to be dim(^
A
).
Section M. Inner Products
An inner product on a vector space V (with real scalars) is any function V V R with the
properties below. The function is traditionally written u, v), that is, the value of the function for
the ordered pair of vectors u, v is a real number denoted as u, v).
For all u, v V ,
1) u, v) = v, u)
2) ku, v) = u, kv) = ku, v) i.e., scalars pass through
3) u, v +v
) = u, v) +u, v
)
4) u, u) 0 and (u, u) = 0 = u = 0).
As usual, two vectors are said to be perpendicular if u, v) = 0, and the length [u[ of u is dened
to be
_
u, u). Especially in nonintuitive situations, another word for perpendicular is orthogonal,
another word for length is norm, and another symbol for the length of u is [[u[[.
1. Let F be the vector space of continuous real-valued functions dened on the interval [0, 2].
a) Verify that
f, g) =
_
2
0
f(x)g(x) dx
is an inner product on F.
b) Find [f[ where f(x) = sinx.
c) Find a function g(x) that is perpendicular to f(x) = sinx with this inner product. (Guess
and check. But dont make a wild guess. The integral of f(x)g(x) has to be 0, so what
does that say about the signed area of the individual curves?)
2. If we let V be the vector space of continuous real functions, and dene
f, g) =
1
_
2
0
f(x)g(x) dx,
then the set sinx, cos x is orthonormal. (Actually, some years we have taken the bounds to
be and , but for no good reason [0, 2] is more common. Do you see that for these two
functions, the inner product will be the same with either set of bounds?)
Assuming these assertions, determine quickly [[ sinx + cos x[[ (where [[ is used to indicate
inner-product length for functions since [f(x)[ already has another meaning).
page 63 Section M, Inner Products
3. Consider the innite set of functions
sinnx, cos mx [ n, m nonnegative integers
Verify that, with the inner product in Problem 2 this big set is orthornomal. You can try
Mathematica, or the CAS on a TI-89 if you have one. You can look up the method for integrating
this products, or gure out the integrals by hand (a bit of a challenge). If Mathematica balks
at n and m being unspecied, limit n, m to 1, 2 and check 10 integrals to determine that the
set of 4 functions you get is orthonormal.
4. Let V = R
2
, let A =
_
2 1
1 1
_
.
a) Verify that the matrix product u, v) = u
T
Av is an inner product.
b) Find a vector perpendicular (under this inner product) to (1, 0).
c) What is the length (under this inner product) of (1, 0)? What is the length of the perpen-
dicular vector you found in b)?
5. Check that the following things really do hold in any inner product space. We may have used
the particular formula u v =

u
i
v
i
in our previous proofs, but now we know that not every
inner product is of this form. Therefore, we now need to check that these claims follow using
only the axioms and denitions of inner product spaces.
a) The Pythaogrean theorem and its converse: u, v) = 0 [u[
2
+ [v[
2
= [u+v[
2
.
b) The Cauchy Schwarz Inequality: [u, v)[ [u[[v[.
c) The angle between two vectors can always be dened by arc cosine, because
1
u, v)
[u[[v[
1.
d) The standard properties of norms:
1) [ku[ = [k[[u[
2) [u[ 0 and ([u[ = 0 u = 0)
3) [u +v[ [u[ + [v[.
page 64 Section N, Orthogonality: Vectors and Matrices
Section N. Orthogonality: Vectors and Matrices
Note. The rst several problems here ask you to prove the main theoretical results about orthogonal
vectors. If you did that in class, you can skip to the concrete example problems starting at Problem
5.
1. Show that 0 is orthogonal to any other vector in R
n
. (Thus, if we hope that an orthogonal set
of vectors is a basis, we better make sure to exclude 0 in dening what can be in the set.)
2. Prove: Any set of nonzero orthogonal vectors is independent.
3. If a vector v is a linear combination

c
i
u
i
of a nonzero orthogonal set of vectors u
1
, u
2
, . . . ,
u
n
, there is a special, simple formula for the coecients. Namely, c
i
= v u
i
/[u
i
[
2
.
a) Prove this
b) Give an even simpler formula for c
i
when u
1
, u
2
, . . . , u
n
is orthonormal.
4. Suppose B = v
1
, v
2
, . . . , v
n
is a basis of R
n
and we want to nd the coordinates of some u
relative to B. The standard way is to row reduce [B [ u]; this takes n
3
/3 steps.
a) If B is orthogonal, there is another way to nd the coordinates. How many steps does it
take?
b) If B is orthonormal, the special method is even simpler. How many steps does it take
now?
c) Yet another general method, instead of reducing [B [ u], is to compute B
1
u. While
this product is ecient to compute, in general we still have to nd B
1
and this usually
requires GE, an even longer GE than for [B [ u]. Well, how well does this method work
when B is orthonormal?
5. Let S be a set of three vectors, u
1
= (1, 1, 1), u
2
= (1, 2, 3), u
3
= (5, 4, 1).
a) Show that S is orthogonal
b) Quickly nd the coordinates of (1, 5, 7) relative to S.
6. Consider the vectors (1, 2, 2)/3, (2, 1, 2)/3 and (2, 2, 1)/3.
a) Show that these vectors are orthonormal.
b) Express (1,2,3) as a linear combination of these vectors. Find the coecients by an easy
method.
7. Let the columns of A be u
1
= (1, 1, 1), u
2
= (1, 2, 3), u
3
= (5, 4, 1).
a) Show that the columns are orthogonal (i.e., perpendicular).
b) Is A an orthogonal matrix?
c) Are the rows of A orthogonal?
8. Modify the columns of matrix A of Problem N7 slightly so that A is orthogonal. Are the rows
now orthogonal too?
9. Many orthogonal matrices are symmetric, in which case the rows are the same set of orthonormal
vectors as the columns. This problem explores the extent to which the row set can be a dierent
orthogonal set than the column set.
page 65 Section NA, Orthogonal Spaces
a) Prove that for a 22 orthogonal matrix Q, the set of rows can be dierent from the set
of columns, but only by the placement of minus signs. That is, if we strip out all minus
signs from Q, then the rows and columns are the same.
b) Once we go to nn orthogonal matrices with n > 2, the situation frees up.
i) Check that
Q
=
1
15
_
_
2 10 11
14 5 2
5 10 10
_
_
is orthogonal. How would you describe the relationship between the row vectors and
the column vectors?
ii) Check that
Q
=
1
75
_
_
10 55 50
41 38 50
62 34 25
_
_
is orthogonal. Notice no row shares more than one numerical value with any column.
By the way, I hope you are wondering how one nds matrices like Q
. You cant just put down

any old integers and hope to have the sums of squares for all rows and columns come out the
same, not to mention the perpendicularity conditions! (Of course, the entries in orthogonal
matrices do not have to be rational, but rational values make for intriguing examples.)
10. Let Q be an orthogonal matrix. Prove that, for any vectors x and y:
a) Qx Qy = x y, and
b) The angle between x and y = the angle between Qx and Qy.
Note: This problem proves that lengths and angles are invariant under change in basis, so long
as you change to an orthonormal basis.
Section NA. Orthogonal Spaces
1. Let S = (1, 2, 3, 1, 2), (2, 4, 7, 2, 1). Explain how you can guess the dimension of S
. Now
nd S
and check that you are correct.

2. For P and P
= P
of Problem LB14, nd a basis of P
= (P
(called P-perp-perp and

written P
). Verify that the generating set of P given in LB14 is contained in P
. Why
does it follow by a dimension argument that P = P
?
3. Explain why for any set S R
n
, we have S S
. (Just check that every u S satises the

denition to be in S
.
4. For any S R
n
we proved that S
is a vector space. When S is a nite set, this new theorem

is actually a special case of one of our matrix space theorems. Which one, and how?
page 66 Section NB, Implicit and Parametric Form
5. Prove: for any set S in any vector space with a dot product, S
= S
. (In fact, the proof

of this doesnt use anything specic to vector space theory or to perpendicularity. The same
proof works for any symmetric relationship, that is, any property pairs of objects might have,
call it , such that whenever u v then v u.)
6. Correct the following sentence: Vectors u, v are orthogonal complements if u
T
v = 0.
7. Let S = (x
1
, x
2
, x
3
, x
4
) [ x
1
+ x
2
+ x
3
+ x
4
= 0 and let T = Span(3, 2, 1, 0). Find a basis
for S T
.
8. Find the intersection of
Span
_
(1, 1, 0, 0), (0, 1, 1, 0), (1, 0, 1, 1)
_
and Span
_
(0, 0, 1, 1), (2, 1, 0, 1), (1, 1, 1, 1)
_
.
This can be solved without orthogonality, but it is harder.
9. Consider the hyerplanes
H
1
= Span
_
(1, 1, 0, 0), (0, 1, 1, 0) (0, 0, 1, 1)
_
,
H
2
= Span
_
(1, 1, 1, 0), (0, 1, 1, 1) (1, 0, 1, 1)
_
.
a) Find H
1
H
2
using implicit representations.
b) Find H
1
H
2
using parametric representations throughout.
10. We have proved, by two separate and lengthy proofs, that rref is canonical for null space and row
space. For null space means that any two matrices with the same null space (homogeneous
solutions) have the same rref up to 0 rows; this was proved in a handout. The proof for row
space was in Problem JC12.
Now show that only one long proof was necessary, by proving quickly (or realizing that we
have already proved) that two matrices have the same row space i they have the same null
space. (The trickier part is the if-part.)
Section NB. Implicit and Parametric Form
1. Let A =
_
1 2 3 4
2 3 4 5
_
.
a) Find
A
in parametric form.
b) Find ^
A
in parametric form.
c) Find
A
in implicit form.
d) Find ^
A
in implicit form.
2. Let V = Span(1, 0, 1, 0), (1, 1, 2, 2). Find an implicit representation of V and explain why
this representation expresses V as an intersection of hyperplanes.
3. Let W be the row space of
_
1 1 2 2
1 0 1 0
_
. Find
a) an implicit representation of W
b) a matrix B so that W is the null space of B
page 67 Section NB, Implicit and Parametric Form
c) cartesian equations for W
Note: this question doesnt really have three parts; why not?
4. Let D =
_
_
1 2
2 3
2 1
1 1
_
_
a) Find (
D
in parametric form.
b) Find (
D
in implicit form.
5. Prove that every subspace of R
n
that can be represented parametrically as the row space of a
matrix A can also be represented implicitly as the null space of a matrix B.
6. Let A be as in Problem NB1 and set C =
_
_
1 1 0 0
1 0 1 0
1 0 0 2
_
_.
a) Find a implicit form for
A

C
.
b) Find an parametric form for
A

C
.
7. Let A be as in Problem NB1 and set B =
_
_
1 1 0 0
1 0 1 0
1 0 0 1
_
_. With the sum of vector spaces

dened as in I11, nd
a) a parametric form for
A
+
B
.
b) an implicit form for
A
+
B
.
8. Let U, V be subspaces of R
n
.
a) Find and prove a formula for (U +V )
in terms of U
and V
.
b) Find and prove a formula for (U V )
in terms of U
and V
.
c) Find and prove a formula for (U V )
in terms of U
and V
. Note: U V is not a
vector space, but you dont have to be a space to have a perp. You just have to be a set
of vectors.
page 68 Section O, Complex Inner Product Spaces
Section O. Complex Inner Product Spaces
C
n
as an inner product space. We could dene z w =

z
i
w
i
, just as in R
n
, and all the properties
like distributivity and commutativity would go through ne. But there would be one glaring prob-
lem: z z would not always be a nonnegative real number, and thus could not be the square of a
norm (represent length). The connection between dot products and, for instance, the Pythagorean
Theorem, would completely break down.
The x begins by noting that for z C, [z[
2
= zz. Therefore, we dene the inner product in C
n
as
z, w) =
n
i=1
z
i
w
i
.
We also dene the norm [z[ of z C
n
by
[z[
2
=
n
i=1
z
i
z
i
=
n
i=1
[z
i
[
2
.
Finally, we continue to say that z, w are orthogonal i z, w) = 0.
1. Determine which of the axioms of an inner product from Section M still hold. For those that
dont hold, what modication does hold?
Note: In some books z, w) is dened as

z
i
w
i
, that is, the conjugation is on the second
argument. In such books the results of this problem are slightly dierent.
2. Determine the statement of, and prove, the Pythagorean Theorem and its converse in C
n
. Be
careful! The right theorem may not be what you think.
3. As usual, the projection of z on w is that vector kw (with k C) such that w and z kw
are orthogonal. Find the formula for Proj
w
z.
4. Use Problem 3 to prove CSI in C
n
. But be careful. Not every way to state CSI in R
n
is correct
in C
n
. Why not?
5. Can you come up with a denition of angle between vectors for complex inner product spaces?
The denition in real inner product spaces doesnt immediately work why not? This question
is admittedly open-ended, but thats how it really is when mathematicians are trying to extend
denitions and theories to more general contexts.
Section OA. Unitary Matrices
To be Written!
Section OB. Dual and Double Dual Spaces, Real and Complex
To be Written!
page 69 Section P, Projections and Gauss Normal Equations
Section P. Projections and Gauss Normal Equations
1. We have come up against the problem of projecting a vector on a at; so far we only know how
to project on a line. To discover the right theorem, do some experimenting. One would hope
to project on a at by reducing to the previous case: projecting on some lines in the at. For
instance, maybe you can project on a basis of the at (if it is a subspace) and sum the results.
Specically, let u = (1, 2, 3) and let the at be the xy-plane. In this special case, you already
know what the projection vector should be. You should always start with an example where
you already know the answer.
a) What should the projection vector be? Draw a picture to show why.
b) One basis of the xy-plane is (1, 0, 0), (0, 1, 0). Project u separately on these basis vectors
(using our formula for projection on vectors). How do your answers relate to the answer
in a)?
c) Another basis of the xy-plane is (1, 1, 0), (1, 1, 0). Project u on these basis vectors.
Do the results relate to the answer to a) in the same way?
d) A third basis of the xy-plane is (1, 0, 0), (1, 1, 0). Project u on these basis vectors. Do
the results relate to the answer to a) in the same way?
e) Any conjectures?
2. Find the projection of (1, 1, 1, 1) on the span of (1, 1, 1, 0) and (1, 1, 0, 1),
a) Using the fact that (1, 1, 1, 0) (1, 1, 0, 1);
b) Using the general formula (Gauss normal equations).
3. We desire to nd the projection of (3, 2, 1) on the plane 5x 4y z = 0.
a) Verify that (1, 1, 1) and (1, 2, 3) are perpendicular vectors on that plane. Use this infor-
mation to nd the desired projection using the theorem about decomposing a projection
into projections onto orthogonal vectors.
b) Find the projection using the general formula A(A
T
A)
1
A
T
p.
c) Of course, somebody had to do some work to nd those perpendicular vectors. Think of
at least one other way, not using those vectors, to nd the projection. Try to think of
methods that are not much work for you.
4. Let P = Span(1, 1, 0), (1, 2, 2). Use Gauss normal equations to nd
a) the matrix M such that Mx is the projection of x on P for any x in R
3
.
b) the projection of (1,1,1) on P
Despite the simple numbers, this is a bit tedious without a calculator or computer.
5. Let v = (1, 3, 5, 7) and let W = Span(1, 1, 1, 1), (1, 2, 3, 2). Find the projection of v onto W
a) using Gauss normal equations
b) by rst using GS to get an orthogonal basis for W.
Note: Method b) is more work if you only want to project one vector, but the better method
if you expect to project on W many times, and when roundo error is taken into account.
page 70 Section P, Projections and Gauss Normal Equations
6. We now prove the theorem you may have conjectured in Problem 1: The projection on a
subspace is the sum of the projections on an orthogonal basis of that subspace. Let v be
any vector in R
n
. Let u
1
, u
2
, . . . , u
k
be mutually perpendicular (orthogonal) in R
n
and let
U = Spanu
1
, u
2
, . . . , u
k
. Let v
i
be the projection of v on u
i
, that is,
v
i
=
v u
i
[u
i
[
2
u
i
.
Finally, dene
p =
k
i=1
v
i
.
Prove by direct check that v p is perpendicular to U. (We show in Problem 7 that there is
only one vector p U such that v p is perpendicular to U, so p must be the same vector we
have called the projection on U elsewhere.)
7. a) Prove: If p is the projection of b onto the subspace U, then [bp[ is the shortest distance
between b and U. That is, if q is any other point on U, then [b q[ > [b p[. In short,
just as in high school Euclidean geometry, the foot of the perpendicular is the closest point.
Hint: Prove [bq[
2
> [bp[
2
using the Pythagorean Theorem (which we have proved
in R
n
). Also, recall that the denition of p is that it is a vector in U such that bp U.
b) Use part a) to prove that projections are unique. That is, prove: given any b and U, there
cannot be two distinct p, p
U such that both b p U and b p
U.
8. Prove: if A is an nk matrix with independent columns, then A
T
A is invertible.
Hint: It is sucient to prove that (A
T
A)x = 0 = x = 0 (Why? This harks back
to various theorems about inverses using GE.) So, assume (A
T
A)x = 0, then uglify to get
x
T
A
T
Ax = x
T
0 = 0. What can you now conclude about Ax ? So what?
9. Let V be a subspace of R
n
. Suppose Ann projects orthogonally from R
n
to V using one basis of
V and Bob projects orthogonally using another basis. Recall that a formula for the projection
of x is W(W
T
W)
1
W
T
x, where W is the matrix with your basis as the columns.
a) Will Ann and Bob necessarily get the same projection function? The issue is: while they
are both using the same formula W(W
T
W)
1
W
T
x, they have dierent matrices W.
b) Prove your claim in part a) by algebra.
10. Let A be the matrix in the general formula for the projection of a vector b on a subspace:
p = A(A
T
A)
1
A
T
b. Suppose A is invertible. What does the formula simplify to? Explain
why the answer is reasonable even without the formula.
11. If the columns of A are independent and span R
n
, then the projection on the columns space of
A should just be the identity function. Prove this algebraically, that is, prove by algebra that
A(A
T
A)
1
A
T
= I
Hint: A is now invertible. Why? So what?
page 71 Section PA, Statistics and Least Squares
12. Prove: if the columns of A are orthogonal, then A(A
T
A)
1
A
T
b reduces to the formula we came
up with earlier for projection onto the span of an orthogonal set (namely, sum the projections
onto the individual vectors in the set).
13. Previously we have discussed two ways to convert from our standard basis e
1
, e
2
, . . . , e
n
to
another basis, a general method and a special method.
General method: If we want to nd the coordinates of our u relative to some basis B, let
matrix M have as its columns our representation of the vectors in B. Then u
B
, the vector
of coordinates of u relative to B, is the solution x to Mx = u.
Special, alternative method: if B happens to be orthonormal, we can use a completely dierent
and easier method: just dot u against each basis vector in turn to gets the coordinates.
Fact: in light of one of the theorems about orthogonal matrices, the Special Method is not a
completely dierent method at all. It is just a special case of the general method. Explain.
Section PA. Statistics and Least Squares
1. Some years I start the discussion of least squares best t by asking what you should answer if
someone tells you they really want a solution to the system
2x = 6
3x = 5.
a) Well, if someone really wants a solution, what is it?
b) Just for kicks, nd x by calculus. That is, nd x that minimizes f(x) = (62x)
2
+(53x)
2
.
No multivariate calculus is needed because there is no constant term for which the constant
has to be determined.
2. Use the normal equations to nd the best tting line to the three points (1,1), (2,3), (3,4).
When you are done, you might check your answer against a plot, or against what your graphics
calculator tells you is the best line (most calculators have this as a built-in function).
3. Repeat Problem 2, but weight the rst point (1,1) by a factor of 5. That is, 5(y
1
(mx
1
+b))
should go into the least squares calculation (or rather, the matrix equivalent) instead of y
1

(mx
1
+b). Show the matrix equation A
_
m
b
_
= y you want to solve, but only solve it if you
have a calculator or computer to help.
4. Find the best tting line through the three data points (1, 1), (2,0), (3,4).
a) rst getting the line in the form y = mx +b
b) then getting it in the form y y = m(x x)+d. (bars in the sense of means). That is, get a
new matrix A for the revised data points (x
1
x, y
1
y), (x
2
x, y
2
y), . . . Youll discover
that the columns are orthogonal, the projection calculations are easier, and the formula
for m and d look dierent. However, m by both methods has got to be the same why?
page 72 Section PB, The Gram-Schmidt Algorithm
5. You have a bunch of data on three variables x, y, z and you want to t a linear model z =
ax + by + c to the data. Find the least squares best t if the data points are
(1, 1, 1), (1, 2, 3), (2, 1, 4), (2, 2, 2).
That is, nd the best constants a, b, c. Actually, just set up the equation Aa = z that you want
to solve. Solve it for a = (a, b, c) only if you have a calculator or computer.
6. You want the best tting quadratic y = ax
2
+ bx + c to the (x, y) data points
(0, 0), (1, 1), (2, 3), (5, 10).
That is, nd the best constants a, b, c. Actually, just set up the equation Aa = z that you want
to solve. Solve it for a = (a, b, c) only if you have a calculator or computer.
7. In a traditional development of the least squares best t formulas, one starts with a set of data
points and predicts that there should be a linear function y = mx+b that ts them pretty well.
For each data x-value x
i
, the actual y-value is called y
i
and the predicted value that comes out
of the function is called y
i
. That is, y
i
= mx
i
+b. The quantity y
i
y
i
is called the error. One
picks m, b that make

i
(y
i
y
i
)
2
minimum. The right m and b are often found by calculus.
(Why squares? So that large errors with opposite signs dont cancel out and make you think
you have a small error. Why not

i
[y
i
y
i
[ as another way to avoid cancelation? Because
absolute values are much harder to do algebra with than squares.)
Explain why our projection approach will get the same answer. That is, explain why nding
m, b as coecient of the projection vector minimizes

i
(y
i
y
i
)
2
.
Section PB. The Gram-Schmidt Algorithm
1. Let u
1
= (1, 0, 0), u
2
= (1, 1, 0), u
3
= (1, 1, 1).
a) Use Gram-Schmidt on u
1
, u
2
, u
3
to nd an orthogonal basis of R
3
. (You can probably
guess what the basis will be; why?)
b) Use Gram-Schmidt on u
3
, u
2
, u
1
, in that order, to nd an orthogonal basis of R
3
.
2. Use Gram-Schmidt to nd an orthonormal basis for the Span of
v
1
= (1, 1, 1, 1), v
2
= (1, 1, 2, 4), v
3
= (1, 2, 4, 3).
(If you are doing this by hand, it will be easier to use a GS variant that does not normalize
until the end.)
Eciency of the Gram-Schmidt algorithm. Below are two versions of the Gram-Schmidt algorithm.
The rst is good for hand computation. (Of course, almost nobody does hand computations anymore,
except when rst learning an algorithm.) The second includes a number of improvements that reduce
the number of computations and work well for machine computation. Instructions for how to count
the amount of work follow the algorithms.
Algorithm GS-1
Input v
1
, . . . , v
k
Output orthonormal vectors w
i
with same span
for i = 1 to k
w
i
v
i
for j = 1 to i 1
w
i
w
i
_
v
i
w
j
w
j
w
j
_
w
j
[thus nal w
i
= v
i
i1
j=1
_
v
i
w
j
w
j
w
j
_
w
j
]
next j
next i
for i = 1 to k
w
i
w
i
/[w
i
[ [this version works only if all [w
i
[ ,= 0]
next i
End algorithm
Algorithm GS-2
Input v
1
, . . . , v
k
Output orthonormal vectors v
i
(the printed vectors)
for i = 1 to k
for j = 1 to i 1
v
i
v
i
(v
i
v
j
) v
j
next j
if v
i
,= 0 then
v
i
v
i
/[v
i
[
print v
i
endif
next i
End algorithm
Note: In Algorithm 1, the point of copying v
i
to w
i
and changing only the ws was so that, in the
rst commented line, one could keep subtracting o projections of the original v
i
. In Algorithm 2,
for each i, we keep subtracting o projections of dierent values of v
i
.
In what follows, assume that all vectors are in R
n
. Assume that the only activities that take
time are square roots, multiplications, and divisions of real numbers. Assume each multiplication or
division of reals takes 1 unit of work, and that each square root takes s units. (Specically, we are
assuming that additions and subtractions take 0 time.) Thus, to nd the length of a vector takes
n +s units of work.
For each question it is sucient to nd the highest power term. In most of the problems below
there are two variables, n and k. Highest power term means any and all terms in which the sum
of the exponents of n and k is maximum. Thus, the highest power term of 2n
2
k + 3nk
2
+nk +Sn
is 2n
2
k + 3nk
2
.
3. How many steps does GS-1 require? You need only answer this when the algorithm works.
4. Explain why GS-1 will work if all the vs are independent.
5. How much work does GS-2 require? You may assume that the input vectors are independent
so that the normalization line is never skipped.
6. Run both algorithms on the set
v
1
= (2, 0, 0, 0), v
2
= (1, 1, 1, 0), v
3
= (1, 2, 3, 4).
7. Describe the dierences between GS-1 and GS-2. Give a brief, informal argument that, given
the same input, they each produce the same output. (A careful argument involves a double
induction and a fair amount of algebra.)
8. Let A be the matrix with the initial vs as its columns. Let Q have the nal orthonormal
vectors as its columns. Then Gram-Schmidt amounts to a factorization A = QR, where R is a
square uppertriangular matrix.
a) Explain why.
b) Write an extension of GS-2 (in your favorite computer language or pseudocode) which
produces Q and R.
The problems above only scratch the surface for eciency issues for Gram-Schmidt. There are
also additional eciency issues for the broader problem of doing projections; recall that our
main use for Gram-Schmidt has been to provide bases which make it easy to project on ats.
See me for additional information and problems.
page 75 Section Q, Linear Transformations: Basic Theory and Examples
Section Q. Linear Transformations: Basic Theory and Examples
Notation: L(U, V ) denotes the collection of all linear transformations from vector space U (the
domain) to vector space V (the codomain or image space). For use below, recall that the axioms a
function T must satisfy in order to be a linear transformation are:
T(u +v) = T(u) + T(v) (1)
T(ku) = kT(u), (2)
1. Suppose T L(U, V ). Using only the axioms for vector spaces and the denition of linear
transformation, prove that
a) T(0) = 0 b) T(u) = T(u)
2. Prove: for any LT, T(u v) = T(u) T(v).
3. Suppose T is an LT from R
3
to R
2
. Suppose further that
T(e
1
) =
_
_
1
2
_
_, T(e
2
) =
_
_
3
4
_
_, T(e
3
) =
_
_
2
1
_
_.
What is T(v), when v = (1, 2, 3)
T
?
4. From the axioms for a function to be a linear transformation it follows that, if T is a linear
transformation, then
T
_
c
i
u
i
_
=

c
i
T(u
i
).
This is pretty obvious, and you are welcome to use it henceforth, but a careful proof actually
takes some work, since the axioms only address the sum of two vectors. A full proof is by
induction (take Math 9). Here just show that the axioms imply
T(u + v +w) = T(u) +T(v) +T(w).
Hint: Write u +v +w as u + (v+w).
5. Consider the function f : R
n
R
m
dened by f(x) = Ax + b, where A is some xed mn
matrix and b R
m
. Prove that [f is a linear transformation] b = 0.
6. For any vector space V , the mapping V V dened by f(v) = kv is called a dilation by a
factor of k. For instance, if V = R
2
and k = 3, then f stretches every vector starting at the
origin in the plane by a factor of 3.
Prove: a dilation is a linear transformation.
7. Let /
m,n
be the vector space of mn matrices. Let A be some xed np matrix. For
M /
m,n
, dene f by f(M) = MA. Show that f is a linear transformation.
8. Let /
n,n
be the vector space of nn matrices. Let A be some xed invertible nn matrix. For
M /
n,n
, dene g by g(M) = A
1
MA. Prove or disprove that g is a linear transformation.
9. Show that T(x) = x
2
is not a linear transformation of R
1
to R
1
.
page 76 Section Q, Linear Transformations: Basic Theory and Examples
10. Let R
+
be the vector space of Problem H14. Show that T([x]) = [x
2
] is a linear transformation
from R
+
to R
+
!
11. For sequences a
n
R
, dene the dierence operator : R
by
a
n
= b
n
, where b
n
= a
n+1
a
n
.
That is, each term in the output sequence a
n
is the dierence of consecutive terms of the
input sequence. For instance, if
a
n
= 1, 4, 9, 16, 25, 36, . . . ,
then
a
n
= 3, 5, 7, 9, 11. . . . .
Prove: is a linear transformation.
Note: is the discrete analog of the derivative operator D in calculus. Just as D is an linear
transformation, so is . Just as there is a product rule for D, so is there one for . And so on.
12. Let T be the space of real functions. Dene L : T T as follows: For every function f, we
get the function L(f) by the rule
_
L(f)
_
(x) = x
2
f(x).
For instance, if f is the cubing function (y = x
3
), then L(f) is the fth-power function (y = x
5
).
If f(x) = e
x
, then L(f) is the function y = x
2
e
x
.
Prove that L is a linear transformation.
13. Let T
n
be the vector space of n-times dierentiable real functions. For f T
n
, dene T
a
(f)
to be the nth-degree Taylor polynomial for f centered around x = a. Prove that T
a
is a linear
transformation.
14. For any function f, let p = T(f) be the unique polynomial of degree 2 such that
p(0) = f(0), p(1) = f(1), p(2) = f(2).
For instance, if f(x) = sin(x/2), then p(x) = 1 (x1)
2
= x
2
+ 2x, and if g(x) = x
3
, then
(T(g))(x) = 3x
2
2x. T(f) is called an interpolating polynomial of f through x = 0, 1, 2.
Interpolate means to agree at certain values and ll in between.
Prove that T is a linear transformation.
15. Show: a function L from vector space U to a vector space W is a linear transformation i
for all u, v U and all k R, L(ku+v) = kL(u) + L(v). ()
That is, just one condition is needed instead of the original two.
16. Prove: if T : U V and S : V W are linear transformations, then so is their composition
ST : U W. Recall that the composition of functions S and T in that order is the function
dened by (ST)(u) = S(T(u)).
17. If S, T are linear transformations, why isnt ST a quadratic? After all, in high school aglebra,
the product of two linear factors is a quadratic.
page 77 Section QA, Matrix Representations of Linear Transformations
Section QA. Matrix Representations of Linear Transformations
1. For a linear transformation T from R
n
to R
m
, the standard matrix A is dened columnwise
as follows: for j = 1, 2, . . . , n, column j of A is the vector T(e
j
). With this denition, prove
x R
n
, T(x) = Ax.
This fact is the basic connection between the linear transformation world and the matrix world.
2. What is the (standard) matrix representation of the linear transformation in Problem Q3?
3. If T : R
n
R
m
is dened by T(x) = Ax, what is the matrix representation of T? (easy!).
4. Suppose A is an invertible matrix. Dene a function T by the rule that, for every vector b,
T(b) is the unique vector x such that Ax = b. It is a fact (dont stop to prove it) that this T
is a linear transformation.
If A is
_
_
1 1 1
0 1 1
0 0 1
_
_ nd the matrix of T.
5. What is the matrix representation of reection of R
3
in the xy-plane?
6. Find the matrix of the projection of
a) R
3
onto the xy-plane sitting in R
3
(easy!).
b) R
3
onto the subspace of solutions (x, y, z) to x + 2y z = 0.
7. Rotation around the origin is a linear transformation. Find the matrix representation of the
counterclockwise rotation of R
2
around the origin by angle .
8. Let T(f) be the antiderivative of f that equals 0 when x = 0. For instance, T(x
2
) = x
3
/3. Let
T
n
be the vector space of polynomials of degree at most n. Find the matrix of T when it is
restricted to functions from T
2
to T
3
. Use the standard bases for T
2
and T
3
.
9. A linear functional is a linear transformation whose image space is R; in otherwords, a linear
transformation T: U R
1
. For now, assume that U = R
n
for some n. Prove: For any linear
functional T, there exists a constant vector c R
n
such that T(x) = c x.
10. Let F = e
x
, xe
x
, x
2
e
x
and let T = Span(F). It happens that the functions in F are indepen-
dent, and so they are a basis of T. Also note that dierentiation maps T into T.
a) Find the matrix of dierentiation relative to the basis F.
b) Use your answer to a) to nd all functions f T which satisfy f
(x) = x
2
e
x
. (So this
problem shows that integration by parts can be done by linear algebra!)
c) Find the matrix of double dierentiation (f
) relative to F.
d) Using c), nd all f T which satisfy f
(x) = x
2
e
x
.
e) Solve f
(x) = x
2
e
x
by calculus.
page 78 Section QB, Linear Transformations and Geometry
11. Consider the following basis of R
3
:
u
1
= (1, 2, 3), u
2
= (2, 3, 1), u
3
= (1, 2, 1).
Consider the linear transformation T: R
3
R
4
dened by
T(u
1
) = (1, 1, 2, 1), T(u
2
) = (1, 0, 1, 1), T(u
3
) = (2, 1, 0, 1).
Find the matrix of T in the standard bases. (There are more and less ecient ways to do this.)
Section QB. Linear Transformations and Geometry
1. By a line in vector space U we mean the set of points p + td [ t R U, where p, d are
arbitrary but xed. Let T be any linear transformation from U to V . Prove: the image under
T of a line is a line or a point.
2. Prove: Linear transformations map ats to ats. The image of a k-at (that is, a translation
of a k-dimensional subspace) is a j-at, where j k.
3. Let R
be the counterclockwise rotation of R

2
around the origin by angle . Clearly, the
composition R
= R
+
. State the matrix identity version of this fact. Now multiply out
the matrix product and compare the two sides entry by entry. What famous theorems have
you proved?
4. Find the matrix representation of the reection of R
2
in the 30
line through the origin (that

is, the line that makes a 30
counterclockwise angle from the positive x-axis).

5. Let R be the reection of R
2
in the line y = 3x.
a) Find the matrix of R.
b) What is the reection of the line s(1, 2)?
c) What is the reection of the line y = 4 x?
6. Let R be the reection of R
3
in the plane x + 2y + 2z = 0. Let P be the projection onto this
plane.
a) Find the matrix M
R
of R. Compute M
2
R
. Why is the result not surprising?
b) What is the reection of the line (1, 1, 1) +t(1, 0, 0)?
c) What is the reection of the plane (1, 2, 3) x = 0?
d) What is the matrix M
P
of P? Verify that M
2
P
= M
P
. Why is this not surprising?
7. Let T be the reection of R
3
in the line 0 + s(1, 2, 2).
a) What is the matrix of T?
b) What is the image of (1,2,3)?
8. Find the matrix of the reection of R
4
in the hyperplane x +y z +w = 0.
9. Find the matrices of the following rotations in R
3
. When I say a rotation of around an axis,
I mean a counterclockwise rotation as you look at the origin from the point I name on the
axis. For instance, 90
around (0,0,1) means that the axis of rotation sticks up from the origin
towards (0,0,1), and you rotate around that axis 90
counterclockwise when looking down at

the xy-plane from above, that is, from the point (0,0,1).
page 79 Section QB, Linear Transformations and Geometry
a) 180
around the axis (1,1,0)

b) 90

c) 180

d) 90

10. A linear transformation T from R
n
to R
n
is said to be orthogonal if it preserves angles and
lengths, e.g., [T(x)[ = [x[ for all x R
n
. Note: This denition doesnt override the previous
denitions of orthogonal (for vectors, then for matrices) because it is for a dierent object
transformations.
Prove: If T is orthogonal, then its standard matrix [T] is orthogonal.
In other words, back in N10 we showed that if a matrix is orthogonal, the linear transforma-
tion it represents preserves angles and lengths. (We didnt state it this way because we hadnt
introduced linear transformations yet; but convince yourself that this is what we showed.) Now
you are proving the converse.
11. Strengthen Problem 10: Prove that if T L(R
n
, R
n
) preserves lengths, its matrix is orthogonal.
12. Strengthen Problem 10 even further: prove that if a function f from R
n
to R
n
(no presumption
that f is an LT) preserves lengths, and f(0) = 0, then f is an LT and is thus orthogonal.
Caution: This is not really a proof in linear algebra, since you dont know that f is an LT.
Youll have to make creative use of facts you know from geometry.
13. Is it possible to have an orthogonal linear transformation from R
n
to R
m
where m ,= n?
Consider two cases.
14. You wish to scale by a factor of k
x
along the x
-axis, and by a factor of k

y
along the y
-axis,
where the (x
, y
) axis system is gotten by rotating the usual xy-axes by angle counterclockwise

around the origin. This scaling is a linear transformation T. Find the matrix of T.
Procedure: Dont try to nd T(e
1
) and T(e
2
) directly. Rather, reason as follows: First rotate
R
2
by degrees. Then scale by k
x
along the ordinary x-axis and by k
y
along the ordinary
y-axis. Then rotate the result by +.
Claim: this approach is equivalent to a change-of-basis approach. First change from xy-
coordinates to x
coordinates, then express the scaling relative to the new coordinates, than
translate back to xy-coordinates. If we have studied change-of-basis theorems for transforma-
tions this year, can you justify this claim?
15. Let a x = b and a
x = b
be hyperplanes H
1
and H
2
in R
n
.
a) Find the equation of the hyperplane H
3
such that H
2
is the reection of H
1
in H
3
.
b) Find the equation of the hyperplane H
4
such that H
4
is the reection of H
1
in H
2
.
c) Find the normal to H
1
H
2
in H
2
. Youll have to describe rst what this question means.
page 80 Section QC, Linear Transformations: More Theory and Examples
Section QC. Linear Transformations: More Theory and Examples
1. Suppose T: R
n
R
m
and the matrix of T is A.
a) Describe the kernel and cokernel of T in terms of A-concepts (e.g., the row-space of A
is an A-concept).
b) What condition on rref(A) is equivalent to T being an onto function? Explain.
c) What condition on rref(A) is equivalent to T being 1-to-1? Explain.
d) What condition on rref(A) is equivalent to T being 1-to-1 and onto?
2. A function is 1to1 if x ,= y = f(x) ,= f(y) (equivalently, f(x) = f(y) only if x = y). Prove:
a linear transformation T is 1to1 if and only if its kernel is trivial, i.e., i T(x) = 0 = x = 0.
This fact ts in with the general principle: to determine if something is happening in a linear
problem, it suces to nd out if it is happening at the origin. In this case the something is
images bunching together.
3. Let T be an LT between R
n
and R
m
. Prove
dim(Kernel) + dim(Cokernel) = dim(domain space),
where dim = dimension. (Actually, you have proved this before, in matrix language, so you
merely have to recognize the translation.)
4. This problem concerns arbitrary functions; they need not be linear and the sets they are dened
on need not be vector spaces.
The identity function on a set S is the function J : S S dened by f(s) = s s S.
We use J instead of I or i because the latter two symbols already have other meanings.
A function f : S T is left invertible if there exists a function g : T S such that
gf = J, that is, if s S, g(f(s)) = s.
Also, f is right invertible if there exists a function h : T S such that fh = J, that is, if
t T, f(h(t)) = t.
A function is invertible if it is both left invertible and right invertible.
Let L : R R
2
be the injection L(x) = (x, 0). Let P : R
2
R be the projection P(x, y) = x.
a) Find two dierent left inverses for L. (In fact, there are innitely many.)
b) Explain why L does not have any right inverses.
c) Find two dierent right inverses for P. (In fact, there are innitely many.)
d) Explain why P does not have any left inverses.
5. Let f : S T; f need not be linear and the sets need not be vector spaces. Prove:
a) f is right invertible f is onto.
b) f is left invertible i f is one-to-one
c) If f is invertible, then the right and left inverses are the same and unique.
6. Suppose that T L(U, V ) is an invertible function. Prove that its inverse T
1
is also a linear
transformation.
7. Consider a linear transformation T : U V . Prove:
page 81 Section QC, Linear Transformations: More Theory and Examples
a) If T is 1-to-1, and u
1
, u
2
, . . . , u
k
is independent in U, then T(u
1
), T(u
2
), . . . , T(u
k
) is
independent in V . Hint: Algebraicize the given and the conclusion, and see if you can get
from the former to the latter by going back and forth with T.
b) Show that 1-to-1 is essential in a); that is, give a simple not-1-to-1 linear transformation
T and an independent set u
1
, u
2
, . . . , u
k
for which T(u
1
), T(u
2
), . . . , T(u
k
) is dependent.
c) If T is onto, and u
1
, u
2
, . . . , u
k
spans U, then T(u
1
), T(u
2
), . . . , T(u
k
) spans V .
d) Show that onto is essential in c); give a simple not-onto linear transformation T and a
spanning set u
1
, u
2
, . . . , u
k
for which T(u
1
), T(u
2
), . . . , T(u
k
) is not spanning in V .
This problem says that 1-to-1 transformations (and only those) preserve independence, whereas
onto transformations (and only those) preserve spanning. So, duality pops up in the world of
linear transformations too.
8. If linear transformation T : U V is one-to-one and onto, it is called a vector space isomor-
phism. (There are isomorphisms between other sorts of structures too, but not in this course.)
Isomorphisms preserve all essential vector space structure. In fact, if U is isomorphic to V (i.e,
there is an isomorphism between them) then eectively they are the same except that the names
have been changed. In this problem we merely ask you to get the avor of what isomorphism
implies.
Prove: if T : U V is an isomorphism, then
a) u
1
, u
2
, . . . , u
k
is an independent set in U i T(u
1
), T(u
2
), . . . , T(u
k
) is an independent set
in V .
b) S U is a subspace of U i T(S) is a subspace of V . (Here T(S) means T(u) [ u S.)
9. Let U be an arbitrary vector space with basis B = u
1
, u
2
, . . . , u
n
. By the coordinate map
T
B
: U R
n
, we mean the function that assigns to each u U its coordinates relative to B.
Example 1. Let U = T
2
, the space of polynomials of degree at most 2, and let B = 1, x, x
2
.
Then T
B
(x
2
3x + 2) = (2, 3, 1).
Example 2. U same as before but now use basis B
= x
2
, (x+1)
2
, (x+2)
2
. Then
T
B
(2x + 1) = (1, 1, 0).
a) Continuing Example 2, nd T
B
(x
2
+x + 1) and T
B
(x).
b) Prove that for any n-dimensional vector space U and any basis B = u
1
, u
2
, . . . , u
n
of U,
the coordinate map T
B
is a linear transformation. Suggestion: Review Problem JD13.
Note: We have probably used the fact that the coordinate map is linear several times in
proofs in class without pointing it out. Can you think of any?
c) Prove that any coordinate map is an isomorphism (see Problem 8).
10. Let U be any n-dimensional vector space with basis B = u
1
, u
2
, . . . , u
n
. The decoordina-
tization map D
B
: R
n
U maps c = (c
1
, c
2
, . . . , c
n
) to

n
i=1
c
i
u
i
.
a) Let U = T
2
and let B = x
2
, (x+1)
2
, (x+2)
2
. Find D
B
(1, 2, 3).
b) Prove that for any n-dimensional vector space U and any basis B = u
1
, u
2
, . . . , u
n
of
U, the decoordinatization map D
B
11. Reconsider the denition of the decoordinatization map of Problem QC10. If B is not a basis,
but just a nite set in U,
a) is D
B
still dened? If dened, is it still a linear transformation? Explain.
page 82 Section QD, Linear Transformations and Approximation
b) Under what circumstances is D
B
one-to-one?
c) Under what circumstances is D
B
onto?
12. Let T be the set of real functions. Let a be any number. The evaluation map E
a
takes any
function f T and returns f(a), its value at a. That is, E
a
is a function from functions to
numbers and E
a
(f) = f(a). This may seem strange. We usually think of functions acting on
numbers, but now we are thinking of numbers acting on functions.
Prove that E
a
13. Let S be a set and V a vector space. Let T be the set of all functions from S to V . Addition
and scalar multiplication of functions in T are dened the way they always are for functions:
(f+g)(x) = f(x) +g(x),
(cf)(x) = c f(x).
a) Prove: T is a vector space.
b) Why do we insist that the codomain V be a vector space?
14. Let U, V be vector spaces. Show that the set L(U, V ) of all linear transformations between
them is itself a vector space. You may use Problem 13.
15. a) Show that for linear transformations, composition left-distributes over addition, that is,
S(T
1
+T
2
) = ST
1
+ ST
2
.
b) Show by example that this distributiviity is false for functions in general. It is simplest to
look for examples where all the domains and codomains are R.
c) Show that right-distributivity,
(S
1
+S
2
)T = S
1
T + S
2
T,
does hold for arbitrary functions.
Section QD. Linear Transformations and Approximation
Denitions. Interpolation is nding or estimating the value of a function between known values.
Extrapolation is nding or estimating the value of a function beyond known values.
1. Determine with as simple algebra as possible the constants c
1
, c
0
, c
1
so that, for all polynomials
p(x) T
2
,
_
1
1
p(x) dx = c
1
p(1) + c
0
p(0) + c
1
p(1). (1)
2. a) Prove that Eq. (1), with the particular c values obtained in Problem 1, is a correct state-
ment for p(x) = x
3
.
b) Explain why (1) is therefore correct for all p(x) T
3
.
page 83 Section QD, Linear Transformations and Approximation
3. Nothing in the linear approximation theory we have developed says that we have to use only
three points to sample our function, or that we need to limit ourselves to quadratics. Dene
F(y
0
, y
1
, y
2
, y
3
) to be the value of
_
3
0
p(x) dx for the unique cubic p(x) that goes through the
points (0, y
0
), (1, y
1
), (2, y
2
) and (3, y
3
). Since every function f(x) can be approximated by a
cubic, F ought to be a good approximation to
_
3
0
f(x) dx for any function.
a) F is a linear functional, so nd with as little work as possible the constants c
0
, c
1
, c
2
, c
3
so
that
F(y
0
, y
1
, y
2
, y
3
)
_
that is,
_
3
0
p(x) dx
_
=
3
i=0
c
i
y
i
. (2)
You have discovered Cotes 4-point quadrature rule.
b) Is Cotes 4-point rule also exact for 4th powers?
4. Without doing extra computation (that is, just think about it), you ought to be able to state
Cotes 4-point quadrature rule for any 4 equally spaced points a
0
, a
1
, a
2
, a
3
(not just 0, 1, 2, 3).
Do so. It is usually stated in the form
_
b
a
f(x) dx
b a
m
4
i=0
d
i
f(a
i
), (2)
where a = a
0
, a
1
, a
2
, a
3
= b are the 4 equally spaced points, f is any function, and m is an
integer chosen as small as possible so that d
0
, d
1
, d
2
, d
3
are integers.
5. Just as the full Simpsons Rule uses 2n intervals and uses the sequence 1, 4, 2, 4, 2, . . . , 2, 4, 1,
so does the full 4-point rule use 3n intervals. What is the sequence of coecients? State the
full 4-point rule.
6. Linear algebra is useful for approximating derivatives as well as integrals. In fact, your TI
calculators use the linear theory outlined below for their numerical derivative capability.
Let F(y
0
, y
1
) = p
(0), where p(x) is the unique linear function that passes through the
points (0, y
0
) and (1, y
1
). Since every function can be approximated by a linear function (but
not always very well), F ought to approximate derivatives at 0 for any function (but not very
well however, this example is just to get us started). That is, if f is any function that passes
through (0, y
0
) and (1, y
1
), then F(y
0
, y
1
) ought to approximate f
(0).
a) Prove that F is a linear functional and nd a formula for it.
b) Use F to approximate f
(0) for f(x) = 2x + 3; for g(x) = x

2
.
7. We should get better derivative approximators by using more points. So let G(y
1
, y
0
, y
1
) =
p
(0) where p(x) is the unique p(x) T

2
that passes through the points (1, y
1
), (0, y
0
), and
(1, y
1
).
a) Show that G is a linear functional and nd a formula for it.
b) Use G to estimate f
(0) for f(x) = 2x

2
x; for h(x) = sinx. Compare your answers with
the exact values you know from calculus.
8. Integral Extrapolation. Nothing in the theory we developed requires the integral to be over the
interval in which the three sample points lie.
page 84 Section QE, Change of Basis of LTs
a) Find constants c
1
, c
0
, c
1
so that, for all p(x) T
2
,
_
3
2
p(x) dx = c
1
p(1) +c
0
p(0) + c
1
p(1). (3)
b) Simpsons Rule makes a good approximation theorem as well as an exact theorem for
some polynomials. Does the extrapolation theorem in part a) make a good approximation
theorem?
c) Is Eq. (3) exact for cubics?
9. Let F(y
1
, y
0
, y
1
) denote the value at x = 1/2 of the quadratic interpolation of the three points
(1, y
1
), (0, y
0
), and (1, y
1
). That is, F is an interpolating function that, given the values
of a function at 1, 0, 1, approximates its value at x = 1/2. (The interpolation will be exact
for functions in T
2
and thus is likely to be a good approximation for other functions.) Find a
formula for F(y
1
, y
0
, y
1
) two ways:
a) First, nd it by nding the quadratic function and then evaluating that function at x = 1/2.
b) Now nd it by linear algebra. By smart choices, make the algebra as simple as you can.
10. Dene f(x
1
, y
1
, x
2
, y
2
) to be the linear function that passes through the points (x
1
, y
1
) and
(x
2
, y
2
). So f is a function from R
4
to T
1
. Is it a linear transformation?
11. Consider the space H of functions Asinx + Bcos x for arbitrary constants A, B. This is the
space of all simple harmonic functions of period 2, since one can show that these are the same
functions that can be written in the form C sin(x +) for amplitude C and phase shift .
a) Interpolation in H. Suppose we are given the points (0, y
0
) and (/2, y
/2
). Suppose we
want to estimate the y-value at x = /6 under the assumption that the function through
the given points is in H. Explain why there have to be constants c
0
and c
/2
so that the
interpolation is c
0
y
0
+ c
/2
y
/2
. Find the constants.
b) Integration in H. Find a simple formula, good for all functions h(x) H, that expresses
_
0
h(x) dx in terms of h(0) and h(/2).
Section QE. Change of Basis of LTs
Notation. Suppose T : U V is a linear transformation, B is a basis of U, and B
is a basis of V .
For u U,
_
u
_
B
is the vector of coordinates of u relative to basis B. Also, [T]
B,B
is the matrix of T
relative to B and B
, that is, [T]

B,B
is the matrix M such that for every u U, M

_
u
_
B
=
_
T(u)
_
B
.
It is a theorem that, for each triple T, B, B
a unique such matrix exists, and there is a formula for

it: the jth column is
_
T(b
j
)
_
B
, where b
j
is the jth vector in B.
1. Let M =
_
1 2
3 4
_
. Let o = e
1
, e
2
be the standard basis of R
2
and let B be the basis of R
2
consisting of b
1
= (1, 1)
T
and b
2
= (1, 2)
T
. Dene T : R
2
R
2
by T(x) = Mx.
a) Find T(b
1
).
b) Find [T]
B,S
and use this to conrm your answer to a).
page 85 Section QE, Change of Basis of LTs
c) Use Gaussian elimination to nd x such that T(x) = b
2
.
d) Find [T]
S,B
, and use this to answer part c) again.
e) Find [T]
B,B
, and use this to nd the B coordinates of T(b
2
).
2. Continuing with o and B as dened in Problem 1,
a) Suppose L is an LT such that [L]
S,B
=
_
4 3
2 1
_
. Find N such that L(x) = Nx. That is,
nd the standard matrix of L.
b) Suppose T is an LT such that [T]
B,S
=
_
4 3
2 1
_
. Find M such that T(x) = Mx.
c) Suppose [S]
B,B
=
_
4 3
2 1
_
. Find A such that S(x) = Ax.
3. Suppose T : V V and for some u V and some scalar c, T(u) = cu. Suppose u is the kth
vector in basis B of V . Explain why the kth column of [T]
B,B
is ce
k
.
4. Do Problem QA11 again, using the general change of basis methods of this section.
5. Let H be a hyperplane through the origin in R
n
. Let P be the projection of R
n
onto H, and
let R be the reection through H. Let B be any basis of R
n
for which the rst n1 vectors in
B are on H, and the last is a normal to H.
a) What is [P]
B,B
? Why? (This is easy!)
b) What is [R]
B,B
? Why? (also easy)
6. Let M =
_
1 2
3 4
_
. Let J : R
2
R
2
be the identity transformation, J(x) = x. Is there a basis
B of R
2
for which M = [J]
B,B
?
7. Let T: U V be an invertible linear transformation, and let B, B
be bases of U, V respectively.
Prove: [T
1
]
B
,B
=
_
[T]
B,B
_
1
That is, any matrix of an invertible linear transformation is
invertible, and it represents the inverse transformation (with the bases reversed).
8. The general theory from a handout about how the matrix of an linear transformation changes
when the bases change can be applied to obtain the simpler theory about how the coordinates
of a vector change when the basis changes. Let B be a basis of R
n
, and let B be the matrix
whose columns are the vectors in B written relative to the standard basis. Use the LT theory
to explain why, for any vector u R
n
(written relative to the standard basis),
u = B
_
u
_
B
and so
_
u
_
B
= B
1
u.
Hint: u is the image of
_
u
_
B
under the identity LT but using two dierent bases.
page 86 Section R, Ane Functions
9. A linear operator is a linear transformation whose domain and codomain are the same space.
Let S, T be operators on R
n
. Let M, N be the matrices of S, T relative to some basis B. Let
M
, N
be the matrices of S, T relative to some other basis B
. Suppose M, N commute. Prove

that M
, N
also commute. Prove it two ways:

a) In the matrix world. That is, express M
and N
in terms of M, N and show that

the two expressions are equal.
b) Using the function world.
10. Square matrix B is said to be similar to square matrix A if there exists an invertible matrix
P such that B = P
1
AP.
a) Prove: if B is similar to A, then A is similar to B. (This ought to be true, given the way
the word similar is used in ordinary English; if it werent true, mathematicians could be
criticized for causing confusion. Because it is true, we often simply say A, B are similar.)
b) Prove: if B is similar to A, and C is similar to B, then C is similar to A.
c) Prove: if A
2
= I and A, B are similar, then B
2
= I.
d) Prove: if A
2
= A and A, B are similar, then B
2
= B.
11. Suppose some Aliens use a dierent basis B for R
n
than we do. Naturally, they will take dot
products relative to their basis. In other words, if their coordinates for our x are (c
1
, . . . , c
n
)
and their coordinates for our y are (d
1
, . . . , d
n
), they will say that the dot product is

c
i
d
i
.
Lets write x, y) for their dot product.
a) We would like a formula for their dot product using x,y, (that is, using our coordinates
for these vectors) and, probably, some matrix multiplication. Let P be the matrix whose
columns are B, written our way. Find a formula for x, y) using x,y, and P.
b) Using your formula, give a nice description of when their dot product result is the same
as ours. That is, for what sort of Alien basis B for R
n
will it be true, no matter what x, y
we pick, that x y = x, y)?
Section R. Ane Functions
1. Function F : U V is ane if F is a linear transformation followed by a translation. That
is, F is ane if U, V are vector spaces and there is some linear transformation T : U V and
some b V such that, x U, F(x) = T(x) + b.
a) Prove: the set of all ane transformations from U to V is a vector space.
b) Let G : V W be ane. Prove that the composition GF is ane. (As usual, GF(x)
means G(F(x)).)
2. So far it seems that linear transformations are a special case of ane transformations, but ane
transformations can be made to be a special case of linear transformations by adding another
dimension! If x R
m
, then by (x, a) we mean the vector in R
m+1
whose rst m coordinates
are x and whose last coordinate is the number a.
page 87 Section R, Ane Functions
Prove: for every ane transformation F : R
n
R
m
, there is a linear transformation
L
F
: R
n+1
R
m+1
such that
L
F
(x, 1) = (F(x), 1).
That is, the action of F is just a slice at height 1 of the action of L
F
. Hint: dene L
F
by
dening its action on a basis of R
n+1
. The last basis vector e
n+1
can take care of the translation
part of F.
Note: the result of this problem is used every day in computer graphics. Ane transfor-
mations are needed left and right, and this result allows their use to be subsumed under the
theory of linear transformations.
3. Th dilation D
k,p
with scalar k and center p is the function (from any vector space V to itself)
that stretches everything by a factor of k around p. For instance, if k = 2, D
2,p
expands the
universe by a factor of 2 with p as the center. (If 0 k < 1, shrinking would be a better
word than stretching, and if k < 0 neither word is quite right.)
a) Algebraically, the denition is
D
k,p
(x) = k(xp) +p, (1)
which can be further rewritten symmetrically as kx + (1k)p. Why does (1) capture the
geometric idea?
b) There is exactly one dilation on V with k = 1; that is, you get the same transformation
no matter what p is. What transformation is that?
c) What does D
0,p
do? That is, describe in your own brief words what this function does.
d) Prove: the composition of dilations is a dilation. This is geometrically obvious to me when
both dilations are centered at the same point, but it is not obvious to me (though still
true) when they are centered at dierent points.
i) If D
k
,p
D
k,p
= D
j,q
, then you should be able to name j and q in terms of k, k
, p, p
.
Can you give a geometric way to nd q?
ii) Actually, there is one case in which the composition of dilations is not a dilation.
When? Why?
e) Explain why every dilation is an ane function.
page 88 Section S, Determinants, Elementary
Section S. Determinants, Elementary
Many of the questions in this section are adapted from Schaums Linear Algebra, 2nd ed., Chapter 7
problems.
1. Compute the following two determinants by the method explained on the rst page of the
handout.
a)
1 2
3 4
. Check using the formula
a b
c d
= ad bc.
b)
1 2 1
2 3 2
3 2 1
.
2. Use any appropriate method to nd the determinant of
a)
_
6 4
2 3
_
b)
_
t 5 7
1 t + 2
_
c)
_
_
1 2 3
2 4 1
1 6 2
_
_
3. Find the values of k for which det(A) = 0 if A =
_
k k
4 2k
_
.
4. Find
0 1 0 1
1 0 1 0
0 1 0 2
2 0 1 0
.
5. Find the values of k for which the system below has a unique solution:
kx + y + z = 1
x +ky + z = 1
x + y +kz = 1.
6. Consider the parallelogram in R
2
with sides (4,1) and (2,3).
a) Draw this parallelogram with the southwest corner at (0,0).
b) Compute its area with a determinant.
c) Compute its area without determinants. This takes some ingenuity since no side of the
parallelogram is lined up with the axes. You either have to nd the length of base and
height, or you have to do some trig, or you have to add and subtract areas of some nicer
shapes. The point of this problem, of course, is to instill proper respect for the determinant
volume theorem.
7. Find the volume in R
3
of the parallelepiped with direction vectors (1, 2, 4), (2, 1, 3) and
(5, 7, 8).
page 89 Section SA, Determinants, Intermediate
8. Here is a mnemonic for computing 33 determinants. First, copy the 1st and 2nd columns
over again and draw arrows as follows:
a a
b b
d d
e e
g g
h h
c
f
i
_
_
_
_
............................................................................................................................................................................................................................................................ . . . . . . . . . . . . . . . . .
.. . .. . . .. . .. . . .. .
............................................................................................................................................................................................................................................................ . . . . . . . . . . . . . . . . .
. . .. . . .. . .. . . .. . .
............................................................................................................................................................................................................................................................ . . . . . . . . . . . . . . . . .
. .. . .. . . .. . .. . . ..
............................................................................................................................................................................................................................................................ .. . .. . . .. . .. . . .. .
. . . . . . . . . . . . . . . . .
............................................................................................................................................................................................................................................................ . . .. . . .. . .. . . .. . .
. . . . . . . . . . . . . . . . .
............................................................................................................................................................................................................................................................ . .. . .. . . .. . .. . . ..
. . . . . . . . . . . . . . . . .
Then multiply the 3 entries under each downarrow and add the products. Then multiply the
entries under each uparrow and subtract the products from the previous sum. The result is the
determinant.
Prove that this method is correct. Use one or more of the properties on the determinants
handout.
9. Show without much computation that
1 1 1
a b c
b + c c + a a +b
= 0.
Section SA. Determinants, Intermediate
Most of the early questions in this section are adapted from Strang, Linear Algebra and Its Appli-
cations, 2nd ed. All matrices are square, unless stated otherwise.
1. If A is nn, how are det(2A), det(A) and det(A
2
) related to det(A)? The answer may depend
on n.
2. Let Z be the set of nn matrices whose determinants are 0. Prove or disprove that Z is
a) closed under multiplication,
b) closed under multiplication by scalars,
c) closed under addition.
3. Prove from the axioms of determinants that if A =
_
a b
c d
_
, and if furthermore a = 0, then
[A[ = ad bc. (Most years the case a ,= 0 is done in class.) You will need to consider the
subcases, c ,= 0 and c = 0. And youre not done breaking into cases yet! This illustrates that
GE is not really the right tool for proving formulas about dets. In fact, it is not the right tool
even for computing dets when some or all entries of the matrix are symbolic (e.g., xs, not just
numbers).
4. Prove: for every square uppertriangular matrix, the det is the product of the diagonal entries
(whether they are nonzero or not so you cant just say this result follows because the det is
the product of the pivots). Prove this. (Naturally, the same theorem holds for lowertriangular
matrices.)
5. Show that, if A is any skew-symmetric 33 matrix, then det(A) = 0. Generalize.
6. Find the determinant of the nn matrix with 1s on the minor diagonal (the one from lower
left to upper right) and 0s elsewhere. The answer may depend on n.
7. Find det(Q) for all square orthogonal matrices. There are two answers; show that they both
can happen.
8. This problem is meant to help you understand my sketch of a proof that [AB[ = [A[[B[ by
making you carry out an example of the algorithm that proves the theorem (at least in the case
that A is invertible).
Let A =
_
1 2
3 3
_
, B =
_
0 1
2 3
_
. Then AB =
_
4 7
6 12
_
. Find the determinant of AB two
ways:
a) Row reduce AB to I by GE, but put bars instead of brackets around each matrix so that
it is a determinant, and put equal signs and scalar multiples between each determinant.
At the last stage, replace [I[ by 1 and you have the determinant evaluated numerically.
For instance, the rst step is
4 7
6 12
4 7
0 3/2
because subtracting one row from another doesnt change the det.
b) Row reduce AB the following way. First do the steps you would do if you were reducing
A. For instance, begin by subtracting the rst row three times from the 2nd, not 4 times.
To keep straight what to do, you might put A o to the side, and reduce it by the same
steps, e.g., as if you were reducing [AB [ A]. However, use det notation on AB and equal
signs just as in part a). For instance, the rst step in the augmented row reductions would
be
_
4 7 1 2
6 12 3 3
_

_
4 7 1 2
6 9 0 3
_
so what you would write with determinants is
4 7
6 12
4 7
6 9
.
When you have nished this part of the reduction, AB will have been reduced to B and
the scalar in front will be [A[. Now continue by the normal GE to reduce B to I and thus
nish computing [AB[. Do you see that your answer is also [A[[B[?
9. (Challenge) Let A
n
be the nn matrix with 2s on the main diagonal, 1s on the diagonals
just above and below the main diagonal, and 0s elsewhere. Find det(A
n
). Prove that you are
right.
10. Prove: if A, B, C are nn matrices, then det
_
A B
0 C
_
= det(A) det(C).
11. Prove: if A is square and Ax = 0 has a nontrivial solution, then det(A) = 0.
12. How are det(A) and det(B) related if B = M
1
AM?
13. Show that, if C = M
T
DM, where all matrices are square and M is invertible, then det(C) and
det(D) have the same sign both positive, or both 0, or both negative.
14. Give a counterexample to det(A+B) = det(A) + det(B).
15. Prove: if the entries of A are integers, and det(A) = 1, then the entries of A
1
are integers.
16. Prove: if the entries of A are integers, and the entries of A
1
are also integers, then det(A) = 1.
17. Suppose each row of A sums to 0. What is det(A)? Why?
18. For a square matrix A, dene
A
=
_
(1)
i+j
[A
ji
[
In other words, the ij entry is the ji-cofactor of A as dened on the handout.

a) Prove that A
A = [A[ I. (This looks daunting, but it follows fairly quickly from results on
the handout.)
b) If A is invertible, derive from a) a formula for A
1
.
c) For n = 2, show that this formula is one that we already know.
Warning: Never use the formula from b) to actually compute a numerical inverse for n > 2; it
is horribly inecient. But it has its theoretical uses. For instance, it is just the tool needed to
prove that the mapping A A
1
is continuous; see Problem SA20.
Note: A
is called the classical adjoint. It is unwise to call it simply the adjoint, because
that term is also used for a dierent and much more valuable companion matrix to A (which
we probably wont mention explicitly in 16HS). Another name used for A
, which gets around

this problem, it to call it the adjugate.
19. Explain why
det
_
_
x y 1
2 8 1
4 7 1
_
_
= 0
is an equation for the line through (x, y) = (2, 8) and (x, y) = (4, 7).
20. Consider the matrices
M =
_
_
1 2 3
1 2 4
2 1 0
_
_, M
+
=
_
_
1 2 3
1 2.01 4
2 1 0
_
_, M
=
_
_
1 2 3
1 1.99 4
2 1 0
_
_.
a) Take Gaussian elimination just far enough on each matrix to convince yourself that
i) They all are invertible, but
ii) Gaussian elimination works very dierently on M than on M
+
and M
, in that
reducing M involves a row switch and reducing the others doesnt.
page 92 Section SB, Determinants and Volume in R
n
Thus the behavior of Gaussian elimination is not a continuous function of the values of the
matrix, since a small change in the values can produce an abrupt change in the behavior.
b) Nonetheless, illustrate that the mapping A A
1
is a continuous function, by using
MATLAB to show that M
1
, (M
+
)
1
and (M
)
1
are very close. (There is an easy
proof that A A
1
is continuous using determinant theorems; Gaussian elimination is
the wrong tool.)
Section SB. Determinants and Volume in R
n
1. Let A be nn and let P be the parallelepiped in R
n
associated with the columns of A. That is,
the dierent edges of P, viewed as vectors, are the columns of A. Show that the n-dimensional
volume of P is
_
det(A
T
A). (You may ask: why nd the volume this hard way, requiring a
matrix multiplication as well as a determinant, when we can get more information, the signed
volume, just by taking det(A) directly? Well, hang on.)
2. Let P be the parallelepiped in R
n
whose edge directions, viewed as vectors, are v
1
, v
2
, . . . , v
n
.
Let M = [v
i
v
j
], that is, M is a matrix whose ij entry is v
i
v
j
. Show that
_
det(M) is the
hypervolume of P.
3. (Cross products put in a generalizable context) The cross product in R
3
of vectors u =
(u
1
, u
2
, u
3
) and v = (v
1
, v
2
, v
3
) is most memorably dened by
u v = det
_
_
i j k
u
1
u
2
u
3
v
1
v
2
v
3
_
_, (1)
where i, j, k are the alternative notation for e
1
, e
2
, e
3
.
a) Use (1) to nd (1, 2, 3) (3, 2, 1).
b) Notice: if we now dot uv with any vector (a, b, c), the result is what we would get if we
substituted a, b, c for i, j, k in (1) before we evaluate. Check this out by evaluating
[(1, 2, 3) (3, 2, 1)] (1, 1, 2)
rst by dotting with the result of part a) and second by substituting for i, j, k in (1).
c) We will now use the observation in b) to give a quick proof of the three most notable
properties of the cross product.
i) Show that (u v) u, v by using (1) to show that
(u v) u = (u v) v = 0.
ii) Let w be any third vector in R
3
. Use (1) to explain why (uv) w is the volume of
the parallelepiped with edges u, v, w.
By the way, u v w is usually written without parentheses, because it couldnt possibly
mean u (v w). Why not?
iii) Show that [u v[ is the area of the parallelogram in R
3
with sides u, v. Hint: Let
w be a unit vector in R
3
perpendicular to both u and v. With this w, show that iii)
is just a special case of ii). You dont even have to compute w; just use the fact that
page 93 Section T, Eigentheory, Elementary
w has unit length and is perpendicular to the plane of the parallelogram. How will
the numerical value of the volume of the (u, v, w)-parallelepiped compare with the
numerical value of the area of the parallelogram?
By the way, this is the rst time we have come up with a formula for the k-volume
of an k-dimensional gure sitting in general position in R
n
. We have just done the
case k = 2, n = 3. See below for a more general result.
4. Let A be the nk matrix whose columns are the edge directions of a k-dimensional paral-
lelepiped P in R
n
.
a) Why cant we say that the signed volume of P is det(A)?
b) Show that the k-volume of P is
_
det(A
T
A). Hint: Augment A to an nn matrix

A by
adding as additional columns on the right an orthonormal basis of (
A
. Let

P be the
n-dimensional parallelepiped whose sides are the columns of

A. What is the relationship
between the n-volume of

P and the k-volume of P? Now use Problem 1 to nd the volume
of

P.
5. Consider the parallelogram P in R
3
with vectors (1,2,3) and (3,2,1) as its sides.
a) Compute the area.
b) Project P onto, in turn, the xy-plane, the xz-plane, and the yz-plane. You get parallelo-
grams P
xy
, P
xz
, P
yz
. Find the area of each of these.
c) Find a relationship between the square of the area of P and the squares of the areas of
P
xy
, P
xz
, P
yz
. Does this seem like any theorem you know and love?
6. Problem 3 began by claiming it would present cross-products in a generalizable way. Whats
the generalization? (Actually, there are several.)
7. A linear operator is a linear transformation whose domain and codomain are the same space.
For any n-dimensional vector space V and any linear operator T : V V , dene the deter-
minant of the operator, denoted [T[, to be [M[ where M = [T]
B,B
for any basis B of V you
choose. (See Section QE for the notation just used. If V = R
n
, note that [T[ is just the signed
hypervolumn of the parallelepiped to which T maps the unit hypercube.)
Prove: [T[ is well-dened. That is, show that if you choose a dierent B the next day, you
still get the same number for [T[.
Section T. Eigentheory, Elementary
A number of the problems in Sections TTD are adapted from Strang, Linear Algebra and Its
Applications, 2nd ed. and Schaums Linear Algebra, 2nd ed., Chapter 8.
1. Verify that
_
_
2
1
_
_ is an eigenvector for
_
1 4
2 3
_
. What is its eigenvalue?
2. Let A be any nn matrix. Is the n1 zero vector an eigenvector for A?
page 94 Section T, Eigentheory, Elementary
3. What are all the eigenvalues and eigenvectors of the nn identity matrix? Of the nn all-zeros
matrix?
4. If u is an eigenvector of A with eigenvalue 3, and v is an eigenvector of A with eigenvalue 2,
what is
a) A(2u + 3v) [express the answer in terms of u and v]
b) A
2
(u)
Moral: its easy to gure out what A and its powers do to linear combinations of eigenvectors.
5. Let M be the matrix of the linear operator from R
2
to R
2
that reects points in the 45
line through the origin. (We know this mapping is a linear transformation, therefore it has a
standard matrix.) From geometric reasoning alone (no algebra), determine all eigenvalues and
eigenvectors of M.
6. Find all eigenvalues and eigenvectors of the diagonal matrix
D =
_
_
1
e
_.
Even if we have already learned how to determine eigenvalues and eigenvectors by a standard
algebraic technique, for now answer the question directly from the denitions. Show that the
vectors you select are eigenvectors, and show that no other vectors can be eigenvectors.
7. Suppose A has eigenvalues 5 and 1. Suppose further that B = A 7I. What are the
eigenvalues of B? How do you know you have all of them?
8. For any matrix A and any number , show that the eigenvectors of A with eigenvalue form
a vector space. This is called the eigenspace for A and .
9. If A is nn, then det(AI) is an nth degree polynomial in .
a) Verify this in the case n = 3 for the particular 33 matrix A whose entries are all 1.
b) The polynomial det(A I) is called the characteristic polynomial of A. Find the
roots of the characteristic polynomial of the particular A from part a), that is, nd the
eigenvalues of that A.
c) Find a basis for each eigenspace from b)
10. Explain why is an eigenvalue of A ^
AI
,= 0.
11. The following statement, often heard, is false. Why?
is an eigenvalue of A if there is a vector x such that Ax = x.
12. Linear operators (linear functions T from vector space U to U) can have eigenvalues and
eigenvectors too: k is an eigenvalue of T if there is a nonzero vector u such that T(u) = ku, in
which case all such u are eigenvectors.
Let U be the space of dierentiable real functions (that is, the vectors are functions like
f(x) = x
2
). The dierentiation operator, D(f) = f
, is a linear operator on U.
page 95 Section TA, Eigentheory, Modeling
a) Find a nonzero f in U such that f is an eigenvector of D with eigenvalue 1.
b) Find a nonzero f in U such that f is an eigenvector of D with eigenvalue 2.
Eigenvectors in spaces of functions are often called eigenfunctions.
Section TA. Eigentheory, Modeling
1. Consider the following population transition model:
u
n+1
= Au
n
, where A =
_
.8 .1
.2 .9
_
and u
n
=
_
_
c
n
r
n
_
_, with c
n
the population in California at
time n and r
n
the population in the rest of the US at time n. This is the same problem that is
usually the rst example we do in class for eigentheory, except no initial conditions have been
given.
a) Prove that it does not matter what the initial population distribution u
0
is: so long as
the transition matrix A is unchanged, in the limit California will have half the population
of the rest of the country. This is an example of a general phenomenon: Markov Chains
forget their initial state.
b) More specically, show that in the limit California has 1/3 of the original population in
the United States.
2. Your hand calculator has built-in commands to nd eigenvalues and associated eigenvectors.
Learn these commands and use them to once again nd the eigenvalues and eigenvectors for
the California population problem above. Does you calculator nd the same basic eigenvectors
we found by hand?
3. Let A =
_
1 4
2 3
_
.
a) Solve the recurrence relation u
n+1
= Au
n
with initial condition u
0
=
_
_
4
2
_
_.
b) Solve the dierential equation
dv
dt
= Av with initial condition v(0) =
_
_
0
6
_
_.
4. Consider the linked recurrence
a
(t) = 3 a(t) + 2 b(t)

b
(t) = a(t) 2 b(t)

a(0) = 1, b(0) = 4.
a) If a(t) and b(t) are the amounts of military spending at time t by countries A and B, explain
why this recurrence system could model an arms race. Its the signs of the coecients
that you should justify; dont worry about trying to justify the specic magnitudes.)
b) Rewrite the linked dierential equation as a single matrix equation.
c) Solve the equation. Does this arms race have a happy ending or an unhappy ending?
d) Solve the discrete analog of this model:
a
n+1
= 3a
n
+ 2b
n
b
n+1
= a
n
2b
n
a
0
= 1, b
0
= 4.
In light of the solutions you get, is this discrete version a plausible model for an arms race?
5. Modify the arms race of Problem 4 to
u
=
_
3 4
3 2
_
u, u(t) =
_
_
8
1
_
_.
a) How does the story associated with the modication compare to the story for the original.
Is each country more or less sensitive to armament by the other country than it was in
Problem 4?
b) Solve the recurrence. Does this arms race have a happy ending or an unhappy ending?
6. Reverse all the signs in the matrix of Problem 4:
u
=
_
3 2
1 2
_
u.
a) Interpret the revised problem in terms of the population of two competing species.
b) Find the general solution to the recurrence. (You cant nd a particular solution because
no initial conditions were given yet.)
c) Suppose further that u(0) =
_
_
6
3
_
_. Find the particular solution. What is the longterm

outcome of the competition?
d) Do c) again, but with u(0) =
_
_
3
6
_
_.
e) Do c) again but with u(0) =
_
_
4
4
_
_.
7. Here perhaps is a competing-species model with more realistic numbers:
p
=
_
1.2 .4
.1 1.2
_
p
a) Find the general solution.
b) Show that there is only one ratio of the initial populations for which both populations
survive.
c) To what extent can you generalize the result of b) to any model p
n+1
= Ap
n
or p
= Ap
for which the entries in A have the sign pattern
_
+
+
_
?
8. Rabbits and wolves compete. Let r(t) and w(t) be the populations of rabbits and wolves in
some environment at time t. Consider the system of dierential equations
dr/dt = 4r 2w
dw/dt = r + w.
a) Why is this system plausible? What does it say that is correct about how rabbits and
wolves interact?
b) If initially r = 300 and w = 200, what are r(t) and w(t)?
c) What happens as t ?
d) Is this system stable, neutrally stable, or unstable?
9. Change the initial conditions of Problem 8, but keep the dierential equations the same:
a) Let
_
_
r(0)
w(0)
_
_ =
_
_
300
100
_
_. What is the solution and what is the long-term behavior?

b) Same as a) but
_
_
r(0)
w(0)
_
_ =
_
_
100
200
_
_.
c) In this model, the initial ratio r(0)/w(0) determines whether both populations survive.
Explain.
10. Consider the system
v
= 4v 5w v(0) = 8
w
= 2v 3w w(0) = 5.
a) Solve it.
b) Can you give an example of how this could be a reasonable population model for two
species; that is, how at least the signs of the coecients could be right for two species that
are dependent in a certain way?
11. Use the explicit formula for f
n
, the nth bonacci number, to determine lim
n
f
n+1
/f
n
.
12. Suppose Fibonacci had started his sequence with F
0
= 1, F
1
= 3 while keeping the recurrence
F
n+1
= F
n
+F
n1
. Find a formula for the new Fibonacci numbers. What is lim
n
F
n+1
F
n
now?
page 98 Section TB, Eigentheory, Intermediate
Section TB. Eigentheory, Intermediate
1. Suppose A is nn, u
0
R
n
and Au
0
= u
0
. Verify that the vector-valued function u(t) =
e
t
u
0
satises u
(t) = Au(t). This explains how nding eigenvalues and eigenvectors for A (a
problem that involves no continuous variables) can lead to a solution of a system of dierential
equations.
2. a) Prove: If
1
,
2
are distinct eigenvalues of A, and u
1
, u
2
are associated nonzero eigenvec-
tors, then u
1
, u
2
are independent.
b) Prove: If
1
,
2
,
3
are distinct eigenvalues of A, and u
1
, u
2
, u
3
are associated nonzero
eigenvectors, then u
1
, u
2
, u
3
are independent. After doing this, you should see how the
induction goes that proves the n-eigenvalue result.
3. Let
1
,
2
, . . . ,
n
be eigenvalues of A and let v
1
, v
2
, . . . , v
n
be associated eigenvectors. Let P be
the matrix with v
1
, v
2
, . . . , v
n
as the columns. Let be the diagonal matrix with
1
,
2
, . . . ,
n
down the diagonal.
a) Prove: AP = P.
b) If A is nn and v
1
, v
2
, . . . , v
n
are independent, then P
1
exists (why?). Use this fact
to write a formula for from a) and thus conclude that there is a basis of R
n
in which
the mapping x Ax just stretches (or shrinks, or reverses) by various amounts along the
dierent axes.
c) Assuming P
1
exists, write a formula for A
k
u
0
in terms of P and . Although this formula
is longer on paper than just writing A
k
u
0
, it is much shorter to compute and easier to use
in determining the behavior of A
k
u
0
as k . Why?
4. Let A be square.
a) Show that 0 is an eigenvalue of A if and only if (You nd an appropriate
condition; there are several good ones.)
b) Show: If A is invertible and is an eigenvalue with eigenvector v, then
1
is an eigenvalue
of A
1
, also with eigenvector v.
5. Use a computer or calculator to nd the eigenvalues and eigenvectors for
C =
_
_
1 2 3 4 5
2 3 4 5 6
1 0 2 0 3
1 1 2 3 5
1 1 1 0 2
_
_
.
6. Find the characteristic polynomial for
a) A =
_
2 3
5 1
_
b) B =
_
_
1 6 2
3 2 0
0 3 4
_
_. c) C =
_
_
1 2 3 4
0 2 1 3
0 0 3 5
0 0 0 4
_
_
.
page 99 Section TC, Eigentheory and Diagonalization
7. It is not likely you will get multiple roots to a polynomial if you pick a polynomial at random.
This problem (which is solved by calculus) begins to get at a careful demonstration of this
claim.
Consider all quadratics p(x) = x
2
+bx+c, where b and c are real numbers chosen at random
and independenly from the interval [1, 1].
a) Find the probability that both roots are real (not complex) Hint: in the 22 square of
points (b, c) on the plane, nd the area of the points for which x
2
+ bx + c = 0 has real
roots. Use your knowledge of the quadratic equation.
b) Find the probability that the two roots are distinct.
Section TC. Eigentheory and Diagonalization
1. Consider the matrix A =
_
3 0
0 1/2
_
. The matrix induces a mapping of the xy-plane to itself
by x Ax.
a) What does this mapping do to the x-axis? To the y-axis?
b) Here is the unit square with diagonals drawn in:
....................................................... .................
. . .. . .. . .. . .. . .. .
....................................................... .. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
............................................................................................................. . . .. . .. . .. . .. . .. .
.................
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .................
.................
.............................................................................................................. ............................................................................................................. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. ...................................................................................................................................................................................................................................................................... ..........................................................................................................................................................
Show the image of this square and its diagonals under the mapping. You should be able
to do this without actually computing any products Ax.
The point of this problem is that it is easy to understand the action of diagonal matrices. The
great achievement of eigentheory is to show that almost all square matrices behave the same
way as diagonal matrices, only with dierent bases.
2. Let A =
_
1 4
2 3
_
. Find
a) all eigenvalues of A and corresponding eigenvectors,
b) an invertible matrix P such that D = P
1
AP is diagonal.
3. For each of the following matrices, Find all eigenvalues, and if they exist, a basis of eigenvectors
for R
2
and a diagonalizing matrix P.
a)
_
5 6
3 2
_
b)
_
5 1
1 3
_
page 100 Section TC, Eigentheory and Diagonalization
4. Let A =
_
2 2
1 3
_
. Find
a) all eigenvalues and corresponding eigenvectors,
b) an invertible P such that D = P
1
AP is diagonal,
c) A
6
(computed in a way that is reasonable to do by hand),
d) a square root of A, that is, a matrix B such that B
2
= A.
5. Determine if matrix C in Problem TB6c is diagonalizable.
6. Find the matrix A whose eigenvalues are 1 and 4 and whose eigenvectors are
_
_
3
1
_
_ and
_
_
2
1
_
_
respectively.
7. For a timed paper-and-pencil test on eigentheory, it is only fair that the eigenvalues and eigen-
vectors should be simple but unguessable. Whats a good way for a professor to create matrices
A with this property? (Trial and error is not a good way try it). Use your method to create
a simple 22 matrix with small-integer but unguessable eigenvalues and eigenvectors.
8. Let A be square.
a) Show: A and A
T
have the same eigenvalues.
b) Suppose A is diagonalizable with eigenvalues
1
,
2
, . . . ,
n
and P is a square matrix whose
jth column is an eigenvector of A for eigenvalue
j
. By a), A
T
has the same eigenvalues
as A. Name associated eigenvectors for A
T
using something related to P.
9. Let A =
_
3 2
2 3
_
. Find an orthogonal matrix P for which P
T
AP is diagonal.
10. Find a symmetric 22 A for which the eigenvalues are 1 and 9 and u = (1, 3)
T
is an eigenvector
for = 1.
11. Suppose A is a symmetric matrix with eigenvalues
1
,
2
, . . . ,
n
and associated orthonormal
basis v
1
, v
2
, . . . , v
n
. Then it turns out that
Ax =
n
i=1
i
Proj
v
i
x.
Prove this two ways:
a) Argue conceptually (no algebra; just from what have a diagonalization means about a
mapping and from what orthogonal projection means).
b) Algebraically, starting with A = PP
T
, where as usual P is the matrix whose columns
are the v
i
.
12. Is A =
_
1 1
2 1
_
diagonalizable over the reals? Over the complex numbers? (In each case,
either compute a diagonalization or explain why none exists.)
page 101 Section TD, Matrix Exponentials
13. If A is diagonalizable, show that A
T
is diagonalizable with the same eigenvalues. If P is a
diagonalizing matrix for A, what is a diagonalizing matrix for A
T
?
14. Suppose A, B are both diagonalizable nn matrices with the same set of n independent eigen-
vectors. Show that A, B commute.
15. Suppose A, B commute and A is diagonalizable with distinct eigenvalues. Show that B is
diagonalizable with the same set of eigenvectors.
Section TD. Matrix Exponentials
1. a) If A is diagonalizable, show that e
tA
e
sA
= e
(t+s)A
.
b) However, give a 22 example that, in general, e
B
e
C
,= e
B+C
. Sigh, the law of exponents
breaks down for matrix exponents.
2. Explain why, if A and B commute, then e
A
e
B
= e
A+B
. Hint: Expand out all three exponentials
with their power series denitions, and match terms on the two sides up to, say, 3rd powers to
see what is going on. Why is commutativity necessary?
3. Convert y
= 0 into a matrix system; call your matrix A. Find the general solution by nding
e
At
u
0
. (You wont be able to compute e
At
by diagonalizing, but in this case you can compute
it directly. Of course, you know the solutions to y
= 0 in advance from 1st-year calculus, so

you can check your answer.)
Section TE. Reducing Systems to First Order
1. Consider the 3rd order recurrence relation
a
n+1
= a
n
+a
n1
+ 2a
n2
, a
0
= a
1
= a
2
= 1.
Rewrite this as a 1st order matrix recurrence relation
u
n+1
= Au
n
, u
0
= something.
2. Consider the 2-variable 2nd-order recurrence relation
a
n+1
= a
n
2b
n1
b
n+1
= 3b
n
+ 2a
n
a
n1
.
Rewrite this as a 1st-order matrix recurrence relation.
3. Consider the 3rd order dierential equation
f
(t) f
(t) f
(t) 2f(t) = 0, f(0) = f
(0) = f
(0) = 1.
Rewrite this as a 1st order matrix dierential equation
f
(t) = Af (t), f (0) = something.

page 102 Section TF, Computing Eigenvalues and Eigenvectors
4. Transform the equation y
+ y = 0 into a rst-order matrix dierential equation, and use

that matrix equation to nd the particular solution when the initial conditions are y(0) = 2,
y
(0) = 0.
Note: If you have taken any calculus-based physics, y
+ y = 0 is a dierential equation
you have seen before and know the solutions to. Thus you should make sure that the answer
you get by linear algebra methods really is the same as the solution you already know. Also,
compare the linear algebra method to the method you used before.
Section TF. Computing Eigenvalues and Eigenvectors
This section is about how eigenvalues are actually computed, given that symbolic determinants
(those with letters like as well as numbers) are too inecient to compute for even modest size
ns. We use toy examples (n = 2, 3, where determinants are not bad) made up to illustrate the
methods Still, you will want to carry out the calculators with a graphics calculator or with Matlab
or Mathematica.
1. Let A =
_
1 1
1 4
_
. Let u =
_
_
1
0
_
_.
a) Compute Au. Normalize the result by scaling so that the rst entry is 1. That is, if
v = Au = (2, 7)
T
, replace v by (1, 3.5)
T
. (You could also normalize the ocial way, by
scaling to make [v[ = 1, but the way I propose may be a little simpler. Call the result
u (that is, update u). Now iterate many times. Does u approach a limit? If so, call the
limit u
. Compute Au
. What do you nd?

b) Compute the eigenvalues and eigenvectors of A. Explain what happened in a) using the
eigentheory formula for A
n
u (for any initial u).
Find it on oce computer!
Section TG. Jordan Canonical Form
1. Recall that E
(a Jordan block in Jordan canonical form) is a square matrix with each entry on
the main diagonal being , each entry in the superdiagonal being 1, and all other entries 0.
Find the pattern in the entries of (E
)
n
. No proof required, but ideally you will describe the
entries so well that if I ask you for the (2,5) entry of the 8th power of the 77 Jordan block
E
4
, you can tell me its numerical value by hand computation.
2. Find the Jordan canonical form A
of the matrix A =
_
4 4
1 0
_
. That is, nd a matrix A
consisting of Jordan blocks, and an invertible P, such that A = PA
P
1
. I have not given you
an algorithm for doing this, but for such a small matrix it is not hard to do it with what you
already know. Hint: Find all the eigenvalues of A and then nd all possible 22 Jordan forms
using these eigenvalues. Then nd which one works for A.
page 103 Section TH, Eigentheory with Complex Numbers
Section TH. Eigentheory with Complex Numbers
1. Let A =
_
0 1
1 0
_
.
a) Find the eigenvalues (real or complex), associated eigenvectors, and a factorization A =
PP
1
where is diagonal.
b) Express (4, 2) as a linear combination of the vectors in the basis of eigenvectors from a).
c) Same as part b) for (1, 3). Notice anything about the coecients in b) and c)?
2. Let A be an nn real matrix (i.e., real entries). Suppose is a complex eigenvalue, that is, the
imaginary part is nonzero. Then, since complex roots of real polynomials (like the characteristic
polynomial) come in conjugate pairs,

is also an eigenvalue of A. In this problem, we prove,
roughly speaking, that everything about the eigenspaces also comes in conjugate pairs.
For this purpose, consider v C
n
, that is, v = (c
1
, c
2
, . . . , c
n
) where each c
i
is a complex
number. Then dene the conjugate vector v to be (c
1
, c
2
, . . . , c
n
)
a) Prove: v is in the eigenspace E
of i v is in the eigenspace E
of

.
b) Prove: If v
1
, v
2
, . . . , v
n
is a basis of E
, then v
1
, v
2
, . . . , v
n
is a basis of E
.
c) Suppose A is diagonalizable (allowing complex eigenvalues and eigenvectors, that is, if we
work in the vector space C
n
). Let v
1
, v
2
, . . . , v
n
be a basis of C
n
consisting of eigenvectors,
where, whenever we have chosen some v as a basis eigenvector for , we have also chosen
v as a basis eigenvector for

. (By part b we can do this.) Suppose u R
n
and we wish
to write u as a linear combination of these eigenvectors. Since u is also in C
n
, we know
u =

n
i=1
c
i
v
i
in a unique way. Prove: In this linear combination, if c is the coecient
of v, then c is the coecient of v. Hint: if u =

n
i=1
c
i
v
i
, then u =

n
i=1
c
i
v
i
is a
representation of the real vector u in a seemingly dierent linear combination.
d) Show that, in consequence of part c, the contribution of v and v to u in

n
i=1
c
i
v
i
is twice
the real part of cv. By the real part of a vector (a
1
+b
1
i, . . . , a
n
+b
n
i) we mean (a
1
, . . . , a
n
).
3. a) Show that every linear operator in R
3
has at least one real eigenvector. Hint: What do
you know about roots of cubic polynomials?
b) Let T be a length preserving linear transformation of R
3
, that is [T(x)[ = [x[ for all
x R
3
. Show that the only possible real eigenvalues of T are 1.
c) Show that every rotation of R
3
has an axis. That is, show that any distance preserving
linear operator on R
3
that doesnt reect any vectors must have a xed axis around which
points o the axis may move.
In fact, every distance preserving function in R
n
that holds the origin xed is necessarily a
linear function, and these linear transformations are compositions of rotations and reections.
page 104 Section U, Quadratic Forms
Section U. Quadratic Forms
When we say, for instance, that a matrix A is positive denite, we mean that the associated quadratic
form x
T
Ax is positive denite. A quadratic form Q(x) = x
T
Ax is positive denite if Q(x) > 0 for
all x ,= 0. See the handout for more details.
The following quadratic form problems are adapted from Schaum, Linear Algebra 2/ed. Original
problem numbers are shown in parentheses.
1. (4.48) Consider the quadratic form in traditional notation
Q(x, y, z) = 3x
2
+ 4xy y
2
+ 8xz 6yz + z
2
.
Find a symmetric matrix A so that if x = (x, y, z), then Q(x) = x
T
Ax.
2. (4.47) Let A =
_
5 2
8 8
_
and consider the quadratic form Q(x) = x
T
Ax.
a) Find a symmetric matrix B so that Q(x) = x
T
Bx.
b) Write Q(x) in traditional algebraic form.
3. (4.53+) Determine if Q(x, y) = 2x
2
12xy + 5y
2
is positive denite by
a) traditional completing the square,
b) row reduction to reduce the associated symmetric matrix A to U
T
DU,
c) Finding the eigenvalues.
4. (4.50) Let
Q(x, y) = 3x
2
+ 2xy y
2
. (1)
a) Rewrite Q in matrix form.
b) Now consider the substitution x = s 3t, y = 2s +t.
i) Carry out this substitution in (1) and rearrange by high school algebra to get a new
quadratic form Q
(s, t). Rewrite this quadratic in matrix form Q
(s) = s
T
Bs.
ii) Do the substitution entirely in matrix notation, starting with the matrix form of Q
from part a). This should help you understand better why, when we get diagonaliza-
tions, they always correspond to factorizations P
T
DP.
5. (4.46) Determine if A =
_
_
1 3 2
3 7 5
2 5 8
_
_
is positive denite by doing GE to factor it into
the form P
T
DP.
6. Let Q(x, y) = 3x
2
6xy +11y
2
. Apply eigentheory to the associated symmetric matrix to nd
a factorization P
T
P where P is orthogonal and the diagonal matrix tells you if Q is positive
denite.
The following quadratic form problems are adapted from Strang, Linear Algebra and Its Applications,
2nd ed.
page 105 Section U, Quadratic Forms
7. Show that the quadratic form Q(x, y) = 2x
2
+4xy +y
2
has a saddle point at the origin, despite
the fact that its coecients are positive. Show this by completing the square, and thus getting
the dierence of two squares.
8. Use one or another of the characterizations of positive-deniteness for a symmetric matrix to
prove the following.
a) If A is positive, then so is A
2
.
b) If A and B are positive denite, then so is A+ B.
9. Prove: if A is a positive denite matrix (thus, among other things, symmetric), then it has a
positive denite square root. That is, there is a positive denite matrix R such that R
2
= A.
Hint: Finding a square root of a matrix is easy if the matrix is diagonal with positive diagonal
entries.
10. Prove: if A is positive denite and C is invertible, then C
T
AC is positive denite.
11. A symmetric matrix A is negative denite if, for all x ,= 0, x
T
Ax < 0. Give several if and
only if conditions for being negative denite. Hint: A is positive denite.
12. Let A =
_
_
1 3 0
3 8 7
0 7 6
_
_. Find nonzero vectors u, v such that u

T
Au > 0 and v
T
Av < 0. That
is, A is neither positive denite, nor negative denite, nor positive indenite, nor negative
indenite. Hint: Use any one of several standard algorithms will convince you that u, v exist,
but you may have to do a little more work to nd specic vectors.

Honors Linear Algebra Main Problem Set Overview

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Honors Linear Algebra Main Problem Set Overview

Uploaded by

Copyright:

Available Formats

Honors Linear Algebra Main Problem Set

commute) algebraicizes into a statement at the matrix

B, rather than at the entry level, it is reasonable to look for a proof

has the property that

= (1, 0) and (4, 2, 1) A

. (The remaining parts of the proof are

and suppose M has no pivot in the rst (leftmost)

does have a pivot in that column. Prove: M and M

do not have the same

. Explain how you

B. (Only if is the hard

2 . Another student gets

invertible, that tells you what you

is simple lowertriangular, and U is uppertriangular with nonzeros on the

. How can this be?

= I, what can you do to prove L

= U. If you start looking at the n

(x) = x, f(0) = f(1) = 0, (1)

(x) + f(x) = x, f(0) = f(1) = 0.

(x) = x, f(0) = 0. (2)

(x) is the symmetric secant slope

(1/8) (x = 1/8, h = 1/8).

(0) by (1/h)[f(h) f(0)] is an approximation

(0) as well. For f

we probably need one more point, so, lets

(0) = 1. Find and explain A, x, b so that the approximation is found

can both be substituted truthfully

. Note: the uniqueness of 0 is often included in the axioms.

2 [ a, b rational numbers. Note: somewhere in your solution, you will have

) be the magnitude and argument of c +di.

(x) f(x) = 0 for all x

are subspaces of V , then so is U U

. Is the intersection of an arbitrary number

= f is a special case of Hookes Law,

(x) + 2f(x) = 0 (1)

are nite subsets of vector space V , then

are nite subsets of vector space V , then

are nite subsets of vector space V , then

be two matrices in row reduced echelon form. Then R and R

are two points that satisfy

is a line on the plane through 0

be determined? If so, what is it? Explain.

, the set of all vectors perpendicular to all

have exactly one point in common.

, in fact, in a unique way. Show this. Do

you got earlier. (use Gaussian elimination!).

are orthogonal complements. The subject of orthogonal subspaces is taken up

of dimension at most 1 less, unless S

is empty. In particular, it follows that the

. You cant just put down

and check that you are correct.

of Problem LB14, nd a basis of P

(called P-perp-perp and

). Verify that the generating set of P given in LB14 is contained in P

. (Just check that every u S satises the

is a vector space. When S is a nite set, this new theorem

. (In fact, the proof

_. With the sum of vector spaces

U such that both b p U and b p

, dene the dierence operator : R

be the counterclockwise rotation of R

line through the origin (that

counterclockwise angle from the positive x-axis).