Math 52: Course Summary

Rich Schwartz
September 2, 2009

General Information: Math 52 is a first course in linear algebra. It is a

transition between the lower level calculus courses and the upper level math
courses. I would say that Math 52 has four general aims:
• To clarify and unify all the stuff you learned in high school about linear
equations and matrices.
• To introduce you to higher dimensional Euclidean spaces and the easiest
kinds of mappings between these spaces – the linear maps.
• To introduce you to the abstract formalism of vector spaces and linear
transformations. The upper level math classes are filled with abstract
structures like this, and the ones in M52 are a good warm-up.
• To introduce you to the idea of proving theorems in mathematics.
Proofs abound in the upper level courses, and M52 gives an easy con-
text in which to get started.
M52 is traditionally taken after finishing the calculus series, but there is some
argument for taking M52 in advance of several variable calculus – e.g. M18,
M20, M35. Several variable calculus uses linear algebra implicitly, and the
constructions are more awkward without the linear algebra formalisms.
Math 54 is a somewhat more advanced version of Math 52. In math 54,
less attention is paid to concrete calculations with matrices and such, and
more attention is paid to the theoretical side of the subject.

High School Algebra Revisited: Suppose you want to solve the linear
system
3x + 4y = 2
7x − 3y = 5

1
It doesn’t hurt to multiply both sides of the first equation by 3 and both
sides of the second equation by 4. This gives

9x + 12y = 6
28x − 12y = 20

It doesn’t hurt to add the equations together. This gives

37x = 26 =⇒ x = 26/37.

A similar trick will get y. Continuing with the example above, the only thing
really important is the 3 × 2 array of numbers

3 4 | 2
.
7 −3 | 5

You can simplify this array by adding multiples of the first column to the
second column or vice versa. After some effort, you get down to an array
that looks like 26
1 0 | 37
0 1 | −137
This final array tells us that x = 26/37 and y = −1/37.
The same idea works for systems of equations with more variables. You
might have an equation like

a11 x1 + a12 x2 + a13 x4 + a14 x4 = b1

a21 x1 + a22 x2 + a23 x4 + a24 x4 = b2
,
a31 x1 + a32 x2 + a33 x4 + a34 x4 = b3
a41 x1 + a42 x2 + a43 x4 + a44 x4 = b4

where a11 etc. and b1 etc are numbers and the variables x1 , x2 , x3 , x4 are the
unknowns. The important thing in solving this system is the 5 × 4 array

a11 a12 a13 a14 | b1

a21 a22 a23 a24 | b2
a31 a32 a33 a34 | b3
a41 a42 a43 a44 | b4

You can solve this system (or see that it has no solutions) by adding multiples
of one row to another until you have simplified the thing as much as possible.
The first main thing you learn in M52 is an algorithmic way to do this, called

2
Gaussian elimination. This is really nothing more than a systematization of
what you already knew how to do.

Higher dimensional space: Here’s another view of our first example above.
Each of the two equations specifies a line in the plane, and then we are find-
ing the point where these lines intersect. What about the second example?
If there were just 3 variables, then each of the equations would specify a
plane in space, and we would be looking for the point where the 3 planes
intersect. Incidentally, one can see geometrically why such systems might
not have solutions. For instance, the lines and planes might be parallel. In
the actual second example, we could say that each of the equations in 4
variables specifies a copy of 3-dimensional space sitting inside 4-dimensional
space, and we are looking for the quadruple intersection point.
In M52 you learn that n-dimensional Euclidean space is nothing more
than the set of all n-tuples (x1 , ..., xn ) where these variables can take on any
real number. Some people find this to be quite exhilarating. Most people
are mystified by four dimensional space, and here you can think about a
billion dimensional space without really sweating. Otjer people find this to
be a letdown, because the concrete description we’ve just given completely
demystifies higher dimensional space.
Once you have the underlying space, you can talk about distances between
points. The distance between (x1 , ..., xn ) and (y1 , ..., yn ) is
q
(x1 − y1 )2 + ... + (xn − yn )2 .

This is really just a generalization of the Pythagorean theorem. Once you

have distance, you can talk about rotations of higher dimensional space, mir-
ror reflections, translations, and so on. This is all done in M52. So, this part
of M52 gives you a foundation for working with higher dimensional Euclidean
geometry.

Linear Mappings of Euclidean Space: A function of one variable takes

a number as input and spits out another number. The simplest kinds of
functions have the form f (x) = ax. These functions, called linear functions,
are characterized by the following two properties.
• f (x + y) = f (x) + f (y)

• f (bx) = bf (x). (On the right side, I mean b times f (x).)

3
They are known as linear functions.
n-dimensional Euclidean space is denoted by Rn . To generalize the notion
of a function, we could say that a map from Rn to Rm is a gadget that takes
a point in Rn as input and spits out a point in Rm . For instance

f (x1 , x2 , x3 ) = (x2 , 5x1 + 7x2 − x3 )

is a map from R3 to R2 . In general, we write f : Rn → Rm . Since it

makes sense to add and scale points in Rn , we call f linear if it satisfies
the two properties mentioned above, namely f (x + y) = f (x) + f (y) and
f (bx) = bf (x).
Note that we have suddenly made a jump in abstraction. In our last
equations, x and y represent points in Rn . That is, x = (x1 , ..., xn ) and
y = (y1 , ..., yn ) and x + y = (x1 + y1 , ..., xn + yn ) and bx = (bx1 , ..., bxn ). Just
as the single number a characterizes the linear function f (x) = ax, the linear
maps are characterized by m × n matrices. These are just like the arrays of
numbers we discussed above.
To facilitate the study of linear maps, one defines a way for matrices to
act on vectors, as follows:

x
 
a14  1 
   
a11 a12 a13 a11 x1 + a12 x2 + a13 x3 + a14 x4
x 
 a21 a22 a23 a24   2  =  a21 x1 + a22 x2 + a23 x3 + a24 x4  .
  
 x3 
a31 a32 a33 a34 a31 x1 + a32 x2 + a33 x3 + a34 x4
x4

We took m = 3 and n = 4 just to make the notation easier, but the general
case is essentially the same. Inspecting this last equation, we see that our
4 × 3 matrix sets up a map from R4 to R3 . The input is (x1 , x2 , x3 , x4 ) and
the output is the right hand side of the equation. This map turns out to be
linear, and all linear maps have this form.
In calculus, you learned about composing functions, and perhaps about
composing maps. (Remember the chain rule.) When linear maps are rep-
resented as matrices, the composition of linear maps corresponds to an op-
eration called matrix multiplication. The example we gave above illustrates
how the products of 3 × 4 matrix and a 4 × 1 matrix is a 3 × 1 matrix. In
general, the product of an m × n matrix and an n × p matrix is an m × p
matrix. The complete correspondence between linear maps and matrices is
worked out in M54.

4
Vector Spaces: A real vector space is a set V of elements called vectors.
This set has two operations on it, addition and scaling by real numbers. The
addition is denoted by (+) and the scaling is written without an intermedi-
ate symbol: rx means scale x by the real number r.. These rules obey the
following axioms.
• (v1 + v2 ) + v3 = v1 + (v2 + v3 ) for all v1 , v2 , v3 ∈ V .
• v1 + v2 = v2 + v1 for all v1 , v2 ∈ V .
• There exists an element, called 0, such that 0 + v = v + 0 for all v ∈ V .
• For any v ∈ V there exists w ∈ V such that v + w = 0.
• r(sx) = (rs)x.
• r(v1 + v2 ) = rv1 + rv2 .
Here are some examples of vector spaces.
• The space Rn is a vector space. When considered as a vector space,
the points in Rn are called vectors. The addition and scaling of vectors
are the obvious ones.
• Let Pn denote the space of real polynomials of degree n. Polynomials
in Pn are added and scaled in the obvious way, e.g.
3(1 + 2x + 7x2 ) = 3 + 6x + 21x2 .

• The space of functions f : R → R. Given f and g, the function

h = f + g has the property that h(x) + f (x) + g(x).
One point of introducing vector spaces is that a lot of basic properties of
Rn carry over to other examples. Sometimes the added abstraction clarifies
what is going on. For instance, Fourier series, a subject from analysis, can
be nicely formulated in terms of vector spaces.

Bases, Dimension, Isomorphism: Once vector spaces are defined in M52,

their basic properties are studied in a systematic way. I’ll give you a feel for
the sort of results you’ll see. A subset of vectors v1 , ..., vk ∈ V is called
dependent if there are constants a1 , ..., ak , not all zero, such that
a1 v1 + ... + ak vk = 0.

5
Otherwise, the set is called independent. A set of independent vectors is
called a basis if it is maximal: The inclusion of any new vector creates a
dependent set. One foundational result is this: If a vector space has a finite
basis, then all of the bases have the same number of elements. This common
number is called the dimension of the vector space.
Linear maps between vector spaces are defined just as for Rn . A linear
map T : V → W between vector spaces is called an isomorphism if T is
one-to-one and onto. In this case, V and W are said to be isomorphic. In-
tuitively, this means that V and W are really the same vector space, up to a
different convention for naming the elements. One standard result in M52 is
that an n-dimensional vector space is isomorphic to Rn . For instance, a basis
for the vector space of quadratic polynomials is given by {1, x, x2 }, and so P2
is isomorphic to R3 . An isomorphism is given by T (ax2 + bc + c) = (a, b, c).
Our third example of a vector space is infinite dimensional. It is not isomor-
phic to Rn for any n.

Quadratic Forms and Inner Products: A quadratic form on a real vec-

tor space V is an operation Q which maps pairs of vectors to the reals. Q
satisfies the following axioms.
• Q(v1 , v2 ) = Q(v2 , v1 ) for all v1 , v2 ∈ V .

• Q(v1 + v2 , v3 ) = Q(v1 , v3 ) + Q(v2 , v3 ) for all v1 , v2 , v3 ∈ V .

• Q(rv) = rQ(v) for all r ∈ R and all v ∈ V .

The dot product on Rn is a classic example of a quadratic form. The formula
is
Q(x, y) = x1 y1 + ... + xn yn .
The dot product is a special case of an inner product. The quadratic form
Q is an inner product if it is such that Q(v, v) > 0 for all nonzero v ∈ V .
Given a pair (V, Q), where V is a vector space and Q is a quadratic form,
one can consider O(V, Q), the set of linear isomorphisms from V to itself
such that Q(T (v), T (w)) = Q(v, w) for all v, w ∈ V . Then V = Rn and Q
is the dot product, we are just talking about the set of rotations of Rn that
fix the origin. Other examples of Q and V (when Q is not necessarily an
inner product) lead to very interesting geometrical objects, like hyperbolic
geometry and Minkowski spacetime. In particular, the linear algebra in M52
is a nice framework for special relativity.

6
Eigenvalues and Eigenvectors: Let T : V → V be a linear transfor-
mation from V to itself. An eigenvector of T is a vector v such that

T (v) = λv,

for some real λ. The number λ is called the eigenvalue associated to the
eigenvector. In M52 you learn the significance of eigenvalues and eigenvec-
tors, as well as methods for computing.
An example will illustrate some of the significance of these concepts.
Suppose that you have two kinds of cells, Type A and Type B. Each day, a
Type A cell mutates to a Type B cell. At the same time, each Type B cell
splits into a Type A cell and a Type B cell. Starting with, say, a single cell
of Type A, one might wonder roughly how many cells one has after N days.
One can think of the growth of these cells as a linear transformation on R2 .
The point (x, y) represents a population of x cells of Type A and y cells of
Type B. Then
T (x, y) = (y, x + y),
represents the population the next day.
Let √
1+ 5
φ= = 1.618...
2
be the famous golden ratio. It turns out that T has a basis of eigenvectors,
given by
v1 = (φ, 1); v2 = (−φ, 1),
with corresponding eigenvalues

λ1 = φ; λ2 = −1/φ.

Since {v1 , v2 } is a basis, and other vector will be a combination of these. For
such a combination, of the form a1 v1 + a2 v2 , we compute

T (a1 v1 + a2 v2 ) = a1 T (v1 ) + a2 T (v2 ) = a1 λ1 v1 + a2 λ2 v2 .

In general, we have

T n (a1 v2 + a2 v + 2) = a1 λn1 v1 + a2 λn2 v2 .

7
Note that |λ1 | > 1 and |λ2 | < 1. So, for large n, the second term is negligible.
That is
T n (a1 v1 + a2 v2 ) ≈ φn × (a1 v1 ).
This tells us that we can expect an exponential growth of cells, with the
growth rate being φ. That is, after many days have passed, the population
of cells just about grows by a factor of φ each day.
In general, studying the eigenvalues and eigenvectors of a linear transfor-
mation gives you a lot of information about what happens when this linear
transformation is applied over and over again. Our example is a prototypical
application of linear algebra to fields such as tumor growth or population
genetics.