Part IA Vectors and Matrices
Based on lectures by N. Peake
Notes taken by Dexter Chua
Michaelmas 2014
These notes are not endorsed by the lecturers, and I have modified them (often
significantly) after lectures. They are nowhere near accurate representations of what
was actually lectured, and in particular, all errors are almost surely mine.
Complex numbers
Review of complex numbers, including complex conjugate, inverse, modulus, argument
and Argand diagram. Informal treatment of complex logarithm,
n
-th roots and complex
powers. de Moivre’s theorem. [2]
Vectors
Review of elementary algebra of vectors in
R
3
, including scalar product. Brief discussion
of vectors in
R
n
and
C
n
; scalar product and the Cauchy-Schwarz inequality. Concepts
of linear span, linear independence, subspaces, basis and dimension.
Suffix notation: including summation convention,
δ
ij
and
ε
ijk
. Vector product and
triple product: definition and geometrical interpretation. Solution of linear vector
equations. Applications of vectors to geometry, including equations of lines, planes and
spheres. [5]
Matrices
Elementary algebra of 3
×
3 matrices, including determinants. Extension to
n × n
complex matrices. Trace, determinant, non-singular matrices and inverses. Matrices as
linear transformations; examples of geometrical actions including rotations, reflections,
dilations, shears; kernel and image. [4]
Simultaneous linear equations: matrix formulation; existence and uniqueness of solu-
tions, geometric interpretation; Gaussian elimination. [3]
Symmetric, anti-symmetric, orthogonal, hermitian and unitary matrices. Decomposition
of a general matrix into isotropic, symmetric trace-free and antisymmetric parts. [1]
Eigenvalues and Eigenvectors
Eigenvalues and eigenvectors; geometric significance. [2]
Proof that eigenvalues of hermitian matrix are real, and that distinct eigenvalues give
an orthogonal basis of eigenvectors. The effect of a general change of basis (similarity
transformations). Diagonalization of general matrices: sufficient conditions; examples
of matrices that cannot be diagonalized. Canonical forms for 2 × 2 matrices. [5]
Discussion of quadratic forms, including change of basis. Classification of conics,
cartesian and polar forms. [1]
Rotation matrices and Lorentz transformations as transformation groups. [1]
Contents
0 Introduction
1 Complex numbers
1.1 Basic properties
1.2 Complex exponential function
1.3 Roots of unity
1.4 Complex logarithm and power
1.5 De Moivre’s theorem
1.6 Lines and circles in C
2 Vectors
2.1 Definition and basic properties
2.2 Scalar product
2.2.1 Geometric picture (R
2
and R
3
only)
2.2.2 General algebraic definition
2.3 Cauchy-Schwarz inequality
2.4 Vector product
2.5 Scalar triple product
2.6 Spanning sets and bases
2.6.1 2D space
2.6.2 3D space
2.6.3 R
n
space
2.6.4 C
n
space
2.7 Vector subspaces
2.8 Suffix notation
2.9 Geometry
2.9.1 Lines
2.9.2 Plane
2.10 Vector equations
3 Linear maps
3.1 Examples
3.1.1 Rotation in R
3
3.1.2 Reflection in R
3
3.2 Linear Maps
3.3 Rank and nullity
3.4 Matrices
3.4.1 Examples
3.4.2 Matrix Algebra
3.4.3 Decomposition of an n × n matrix
3.4.4 Matrix inverse
3.5 Determinants
3.5.1 Permutations
3.5.2 Properties of determinants
3.5.3 Minors and Cofactors
4 Matrices and linear equations
4.1 Simple example, 2 × 2
4.2 Inverse of an n × n matrix
4.3 Homogeneous and inhomogeneous equations
4.3.1 Gaussian elimination
4.4 Matrix rank
4.5 Homogeneous problem Ax = 0
4.5.1 Geometrical interpretation
4.5.2 Linear mapping view of Ax = 0
4.6 General solution of Ax = d
5 Eigenvalues and eigenvectors
5.1 Preliminaries and definitions
5.2 Linearly independent eigenvectors
5.3 Transformation matrices
5.3.1 Transformation law for vectors
5.3.2 Transformation law for matrix
5.4 Similar matrices
5.5 Diagonalizable matrices
5.6 Canonical (Jordan normal) form
5.7 Cayley-Hamilton Theorem
5.8 Eigenvalues and eigenvectors of a Hermitian matrix
5.8.1 Eigenvalues and eigenvectors
5.8.2 Gram-Schmidt orthogonalization (non-examinable)
5.8.3 Unitary transformation
5.8.4 Diagonalization of n × n Hermitian matrices
5.8.5 Normal matrices
6 Quadratic forms and conics
6.1 Quadrics and conics
6.1.1 Quadrics
6.1.2 Conic sections (n = 2)
6.2 Focus-directrix property
7 Transformation groups
7.1 Groups of orthogonal matrices
7.2 Length preserving matrices
7.3 Lorentz transformations
0 Introduction
Vectors and matrices is the language in which a lot of mathematics is written
in. In physics, many variables such as position and momentum are expressed as
vectors. Heisenberg also formulated quantum mechanics in terms of vectors and
matrices. In statistics, one might pack all the results of all experiments into a
single vector, and work with a large vector instead of many small quantities. In
group theory, matrices are used to represent the symmetries of space (as well as
many other groups).
So what is a vector? Vectors are very general objects, and can in theory
represent very complex objects. However, in this course, our focus is on vectors
in
R
n
or
C
n
. We can think of each of these as an array of
n
real or complex
numbers. For example, (1
,
6
,
4) is a vector in
R
3
. These vectors are added in the
obvious way. For example, (1
,
6
,
4) + (3
,
5
,
2) = (4
,
11
,
6). We can also multiply
vectors by numbers, say 2(1
,
6
,
4) = (2
,
12
,
8). Often, these vectors represent
points in an n-dimensional space.
Matrices, on the other hand, represent functions between vectors, i.e. a
function that takes in a vector and outputs another vector. These, however, are
not arbitrary functions. Instead matrices represent linear functions. These are
functions that satisfy the equality
f
(
λx
+
µy
) =
λf
(
x
) +
µf
(
y
) for arbitrary
numbers
λ, µ
and vectors
x, y
. It is important to note that the function
x 7→ x
+
c
for some constant vector
c
is not linear according to this definition, even though
it might look linear.
It turns out that for each linear function from
R
n
to
R
m
, we can represent
the function uniquely by an
m × n
array of numbers, which is what we call the
matrix. Expressing a linear function as a matrix allows us to conveniently study
many of its properties, which is why we usually talk about matrices instead of
the function itself.
1 Complex numbers
In
R
, not every polynomial equation has a solution. For example, there does
not exist any
x
such that
x
2
+ 1 = 0, since for any
x
,
x
2
is non-negative, and
x
2
+ 1 can never be 0. To solve this problem, we introduce the “number”
i
that
satisfies
i
2
=
1. Then
i
is a solution to the equation
x
2
+ 1 = 0. Similarly,
i
is also a solution to the equation.
We can add and multiply numbers with
i
. For example, we can obtain
numbers 3 +
i
or 1 + 3
i
. These numbers are known as complex numbers. It turns
out that by adding this single number
i
, every polynomial equation will have a
root. In fact, for an
n
th order polynomial equation, we will later see that there
will always be
n
roots, if we account for multiplicity. We will go into details in
Chapter 5.
Apart from solving equations, complex numbers have a lot of rather important
applications. For example, they are used in electronics to represent alternating
currents, and form an integral part in the formulation of quantum mechanics.
1.1 Basic properties
Definition
(Complex number)
.
A complex number is a number
z C
of the
form
z
=
a
+
ib
with
a, b R
, where
i
2
=
1. We write
a
=
Re
(
z
) and
b
=
Im
(
z
).
We have
z
1
± z
2
= (a
1
+ ib
1
) ± (a
2
+ ib
2
)
= (a
1
± a
2
) + i(b
1
± b
2
)
z
1
z
2
= (a
1
+ ib
1
)(a
2
+ ib
2
)
= (a
1
a
2
b
1
b
2
) + i(b
1
a
2
+ a
1
b
2
)
z
1
=
1
a + ib
=
a ib
a
2
+ b
2
Definition
(Complex conjugate)
.
The complex conjugate of
z
=
a
+
ib
is
a ib
.
It is written as ¯z or z
.
It is often helpful to visualize complex numbers in a diagram:
Definition
(Argand diagram)
.
An Argand diagram is a diagram in which a
complex number
z
=
x
+
iy
is represented by a vector
p
=
x
y
. Addition of
vectors corresponds to vector addition and ¯z is the reflection of z in the x-axis.
Re
Im
z
1
z
2
¯z
2
z
1
+ z
2
Definition
(Modulus and argument of complex number)
.
The modulus of
z
=
x
+
iy
is
r
=
|z|
=
p
x
2
+ y
2
. The argument is
θ
=
arg z
=
tan
1
(
y/x
). The
modulus is the length of the vector in the Argand diagram, and the argument is
the angle between z and the real axis. We have
z = r(cos θ + i sin θ)
Clearly the pair (
r, θ
) uniquely describes a complex number
z
, but each complex
number
z C
can be described by many different
θ
since
sin
(2
π
+
θ
) =
sin θ
and cos(2π + θ) = cos θ. Often we take the principle value θ (π, π].
When writing z
i
= r
i
(cos θ
i
+ i sin θ
i
), we have
z
1
z
2
= r
1
r
2
[(cos θ
1
cos θ
2
sin θ
1
sin θ
2
) + i(sin θ
1
cos θ
2
+ sin θ
2
cos θ
1
)]
= r
1
r
2
[cos(θ
1
+ θ
2
) + i sin(θ
1
+ θ
2
)]
In other words, when multiplying complex numbers, the moduli multiply and
the arguments add.
Proposition. z¯z = a
2
+ b
2
= |z|
2
.
Proposition. z
1
= ¯z/|z|
2
.
Theorem (Triangle inequality). For all z
1
, z
2
C, we have
|z
1
+ z
2
| |z
1
| + |z
2
|.
Alternatively, we have |z
1
z
2
| ||z
1
| |z
2
||.
1.2 Complex exponential function
Exponentiation was originally defined for integer powers as repeated multiplica-
tion. This is then extended to rational powers using roots. We can also extend
this to any real number since real numbers can be approximated arbitrarily
accurately by rational numbers. However, what does it mean to take an exponent
of a complex number?
To do so, we use the Taylor series definition of the exponential function:
Definition (Exponential function). The exponential function is defined as
exp(z) = e
z
= 1 + z +
z
2
2!
+
z
3
3!
+ ··· =
X
n=0
z
n
n!
.
This automatically allows taking exponents of arbitrary complex numbers.
Having defined exponentiation this way, we want to check that it satisfies the
usual properties, such as
exp
(
z
+
w
) =
exp
(
z
)
exp
(
w
). To prove this, we will
first need a helpful lemma.
Lemma.
X
n=0
X
m=0
a
mn
=
X
r=0
r
X
m=0
a
rm,m
Proof.
X
n=0
X
m=0
a
mn
= a
00
+ a
01
+ a
02
+ ···
+ a
10
+ a
11
+ a
12
+ ···
+ a
20
+ a
21
+ a
22
+ ···
= (a
00
) + (a
10
+ a
01
) + (a
20
+ a
11
+ a
02
) + ···
=
X
r=0
r
X
m=0
a
rm,m
This is not exactly a rigorous proof, since we should not hand-wave about
infinite sums so casually. But in fact, we did not even show that the definition of
exp
(
z
) is well defined for all numbers
z
, since the sum might diverge. All these
will be done in that IA Analysis I course.
Theorem. exp(z
1
) exp(z
2
) = exp(z
1
+ z
2
)
Proof.
exp(z
1
) exp(z
2
) =
X
n=0
X
m=0
z
m
1
m!
z
n
2
n!
=
X
r=0
r
X
m=0
z
rm
1
(r m)!
z
m
2
m!
=
X
r=0
1
r!
r
X
m=0
r!
(r m)!m!
z
rm
1
z
m
2
=
X
r=0
(z
1
+ z
2
)
r
r!
Again, to define the sine and cosine functions, instead of referring to “angles”
(since it doesn’t make much sense to refer to complex “angles”), we again use a
series definition.
Definition (Sine and cosine functions). Define, for all z C,
sin z =
X
n=0
(1)
n
(2n + 1)!
z
2n+1
= z
1
3!
z
3
+
1
5!
z
5
+ ···
cos z =
X
n=0
(1)
n
(2n)!
z
2n
= 1
1
2!
z
2
+
1
4!
z
4
+ ···
One very important result is the relationship between exp, sin and cos.
Theorem. e
iz
= cos z + i sin z.
Alternatively, since sin(z) = sin z and cos(z) = cos z, we have
cos z =
e
iz
+ e
iz
2
,
sin z =
e
iz
e
iz
2i
.
Proof.
e
iz
=
X
n=0
i
n
n!
z
n
=
X
n=0
i
2n
(2n)!
z
2n
+
X
n=0
i
2n+1
(2n + 1)!
z
2n+1
=
X
n=0
(1)
n
(2n)!
z
2n
+ i
X
n=0
(1)
n
(2n + 1)!
z
2n+1
= cos z + i sin z
Thus we can write z = r(cos θ + i sin θ) = re
.
1.3 Roots of unity
Definition
(Roots of unity)
.
The
n
th roots of unity are the roots to the equation
z
n
= 1 for
n N
. Since this is a polynomial of order
n
, there are
n
roots of
unity. In fact, the nth roots of unity are exp
2πi
k
n
for k = 0, 1, 2, 3 ···n 1.
Proposition. If ω = exp
2πi
n
, then 1 + ω + ω
2
+ ··· + ω
n1
= 0
Proof. Two proofs are provided:
(i)
Consider the equation
z
n
= 1. The coefficient of
z
n1
is the sum of
all roots. Since the coefficient of
z
n1
is 0, then the sum of all roots
= 1 + ω + ω
2
+ ··· + ω
n1
= 0.
(ii)
Since
ω
n
1 = (
ω
1)(1 +
ω
+
···
+
ω
n1
) and
ω 6
= 1, dividing by (
ω
1),
we have 1 + ω + ··· + ω
n1
= (ω
n
1)/(ω 1) = 0.
1.4 Complex logarithm and power
Definition
(Complex logarithm)
.
The complex logarithm
w
=
log z
is a solution
to
e
ω
=
z
, i.e.
ω
=
log z
. Writing
z
=
re
, we have
log z
=
log
(
re
) =
log r
+
.
This can be multi-valued for different values of
θ
and, as above, we should select
the θ that satisfies π < θ π.
Example. log 2i = log 2 + i
π
2
Definition
(Complex power)
.
The complex power
z
α
for
z, α C
is defined as
z
α
=
e
α log z
. This, again, can be multi-valued, as
z
α
=
e
α log |z|
e
iαθ
e
2inπα
(there
are finitely many values if
α Q
, infinitely many otherwise). Nevertheless, we
make z
α
single-valued by insisting π < θ π.
1.5 De Moivre’s theorem
Theorem (De Moivre’s theorem).
cos + i sin = (cos θ + i sin θ)
n
.
Proof.
First prove for the
n
0 case by induction. The
n
= 0 case is true since
it merely reads 1 = 1. We then have
(cos θ + i sin θ)
n+1
= (cos θ + i sin θ)
n
(cos θ + i sin θ)
= (cos + i sin )(cos θ + i sin θ)
= cos(n + 1)θ + i sin(n + 1)θ
If n < 0, let m = n. Then m > 0 and
(cosθ + i sin θ)
m
= (cos + i sin )
1
=
cos i sin
(cos + i sin )(cos i sin )
=
cos() + i sin()
cos
2
+ sin
2
= cos() + i sin()
= cos + i sin
Note that
cos
+
i sin
=
e
inθ
= (
e
)
n
= (
cos θ
+
i sin θ
)
n
is not a valid
proof of De Moivre’s theorem, since we do not know yet that
e
inθ
= (
e
)
n
. In
fact, De Moivre’s theorem tells us that this is a valid rule to apply.
Example.
We have
cos
5
θ
+
i sin
5
θ
= (
cos θ
+
i sin θ
)
5
. By binomial expansion
of the RHS and taking real and imaginary parts, we have
cos 5θ = 5 cos θ 20 cos
3
θ + 16 cos
5
θ
sin 5θ = 5 sin θ 20 sin
3
θ + 16 sin
5
θ
1.6 Lines and circles in C
Since complex numbers can be regarded as points on the 2D plane, we can often
use complex numbers to represent two dimensional objects.
Suppose that we want to represent a straight line through
z
0
C
parallel to
w C
. The obvious way to do so is to let
z
=
z
0
+
λw
where
λ
can take any
real value. However, this is not an optimal way of doing so, since we are not
using the power of complex numbers fully. This is just the same as the vector
equation for straight lines, which you may or may not know from your A levels.
Instead, we arrange the equation to give
λ
=
zz
0
w
. We take the complex
conjugate of this expression to obtain
¯
λ
=
¯z ¯z
0
¯w
. The trick here is to realize that
λ is a real number. So we must have λ =
¯
λ. This means that we must have
z z
0
w
=
¯z ¯z
0
¯w
z ¯w ¯zw = z
0
¯w ¯z
0
w.
Theorem
(Equation of straight line)
.
The equation of a straight line through
z
0
and parallel to w is given by
z ¯w ¯zw = z
0
¯w ¯z
0
w.
The equation of a circle, on the other hand, is rather straightforward. Suppose
that we want a circle with center
c C
and radius
ρ R
+
. By definition of a
circle, a point
z
is on the circle iff its distance to
c
is
ρ
, i.e.
|z c|
=
ρ
. Recalling
that |z|
2
= z¯z, we obtain,
|z c| = ρ
|z c|
2
= ρ
2
(z c)(¯z ¯c) = ρ
2
z¯z ¯cz c¯z = ρ
2
c¯c
Theorem.
The general equation of a circle with center
c C
and radius
ρ R
+
can be given by
z¯z ¯cz c¯z = ρ
2
c¯c.
2 Vectors
We might have first learned vectors as arrays of numbers, and then defined
addition and multiplication in terms of the individual numbers in the vector.
This however, is not what we are going to do here. The array of numbers is just
a representation of the vector, instead of the vector itself.
Here, we will define vectors in terms of what they are, and then the various
operations are defined axiomatically according to their properties.
2.1 Definition and basic properties
Definition
(Vector)
.
A vector space over
R
or
C
is a collection of vectors
v V
,
together with two operations: addition of two vectors and multiplication of a
vector with a scalar (i.e. a number from R or C, respectively).
Vector addition has to satisfy the following axioms:
(i) a + b = b + a (commutativity)
(ii) (a + b) + c = a + (b + c) (associativity)
(iii) There is a vector 0 such that a + 0 = a. (identity)
(iv) For all vectors a, there is a vector (a) such that a + (a) = 0 (inverse)
Scalar multiplication has to satisfy the following axioms:
(i) λ(a + b) = λa + λb.
(ii) (λ + µ)a = λa + µa.
(iii) λ(µa) = (λµ)a.
(iv) 1a = a.
Often, vectors have a length and direction. The length is denoted by
|v|
. In
this case, we can think of a vector as an “arrow” in space. Note that
λa
is either
parallel (λ 0) to or anti-parallel (λ 0) to a.
Definition
(Unit vector)
.
A unit vector is a vector with length 1. We write a
unit vector as
ˆ
v.
Example. R
n
is a vector space with component-wise addition and scalar mul-
tiplication. Note that the vector space
R
is a line, but not all lines are vector
spaces. For example,
x
+
y
= 1 is not a vector space since it does not contain
0
.
2.2 Scalar product
In a vector space, we can define the scalar product of two vectors, which returns
a scalar (i.e. a real or complex number). We will first look at the usual scalar
product defined for R
n
, and then define the scalar product axiomatically.
2.2.1 Geometric picture (R
2
and R
3
only)
Definition
(Scalar/dot product)
. a · b
=
|a||b|cos θ
, where
θ
is the angle
between a and b. It satisfies the following properties:
(i) a · b = b · a
(ii) a · a = |a|
2
0
(iii) a · a = 0 iff a = 0
(iv) If a · b = 0 and a, b 6= 0, then a and b are perpendicular.
Intuitively, this is the product of the parts of a and b that are parallel.
b
a
|a|
|a|cos θ
Using the dot product, we can write the projection of
b
onto
a
as (
|b|cos θ
)
ˆ
a
=
(ˆa · b)ˆa.
The cosine rule can be derived as follows:
|
BC|
2
= |
AC
AB|
2
= (
AC
AB) · (
AC
AB)
= |
AB|
2
+ |
AC|
2
2|
AB||
AC|cos θ
We will later come up with a convenient algebraic way to evaluate this scalar
product.
2.2.2 General algebraic definition
Definition
(Inner/scalar product)
.
In a real vector space
V
, an inner product
or scalar product is a map
V × V R
that satisfies the following axioms. It is
written as x · y or hx | yi.
(i) x · y = y · x (symmetry)
(ii) x · (λy + µz) = λx · y + µx · z (linearity in 2nd argument)
(iii) x · x 0 with equality iff x = 0 (positive definite)
Note that this is a definition only for real vector spaces, where the scalars
are real. We will have a different set of definitions for complex vector spaces.
In particular, here we can use (i) and (ii) together to show linearity in 1st
argument. However, this is generally not true for complex vector spaces.
Definition. The norm of a vector, written as |a| or kak, is defined as
|a| =
a · a.
Example.
Instead of the usual
R
n
vector space, we can consider the set of all
real (integrable) functions as a vector space. We can define the following inner
product:
hf | gi =
Z
1
0
f(x)g(x) dx.
2.3 Cauchy-Schwarz inequality
Theorem (Cauchy-Schwarz inequality). For all x, y R
n
,
|x · y| |x||y|.
Proof. Consider the expression |x λy|
2
. We must have
|x λy|
2
0
(x λy) · (x λy) 0
λ
2
|y|
2
λ(2x · y) + |x|
2
0.
Viewing this as a quadratic in
λ
, we see that the quadratic is non-negative and
thus cannot have 2 real roots. Thus the discriminant 0. So
4(x · y)
2
4|y|
2
|x|
2
(x · y)
2
|x|
2
|y|
2
|x · y| |x||y|.
Note that we proved this using the axioms of the scalar product. So this
result holds for all possible scalar products on any (real) vector space.
Example.
Let
x
= (
α, β, γ
) and
y
= (1
,
1
,
1). Then by the Cauchy-Schwarz
inequality, we have
α + β + γ
3
p
α
2
+ β
2
+ γ
2
α
2
+ β
2
+ γ
2
αβ + βγ + γα,
with equality if α = β = γ.
Corollary (Triangle inequality).
|x + y| |x| + |y|.
Proof.
|x + y|
2
= (x + y) · (x + y)
= |x|
2
+ 2x · y + |y|
2
|x|
2
+ 2|x||y| + |y|
2
= (|x| + |y|)
2
.
So
|x + y| |x| + |y|.
2.4 Vector product
Apart from the scalar product, we can also define the vector product. However,
this is defined only for R
3
space, but not spaces in general.
Definition
(Vector/cross product)
.
Consider
a, b R
3
. Define the vector
product
a × b = |a||b|sin θ
ˆ
n,
where
ˆn
is a unit vector perpendicular to both
a
and
b
. Since there are two
(opposite) unit vectors that are perpendicular to both of them, we pick
ˆn
to be
the one that is perpendicular to a, b in a right-handed sense.
a
b
a × b
The vector product satisfies the following properties:
(i) a × b = b × a.
(ii) a × a = 0.
(iii) a × b = 0 a = λb for some λ R (or b = 0).
(iv) a × (λb) = λ(a × b).
(v) a × (b + c) = a × b + a × c.
If we have a triangle
OAB
, its area is given by
1
2
|
OA||
OB|sin θ
=
1
2
|
OA×
OB|
.
We define the vector area as
1
2
OA ×
OB
, which is often a helpful notion when
we want to do calculus with surfaces.
There is a convenient way of calculating vector products:
Proposition.
a × b = (a
1
ˆ
i + a
2
ˆ
j + a
3
ˆ
k) × (b
1
ˆ
i + b
2
ˆ
j + b
3
ˆ
k)
= (a
2
b
3
a
3
b
2
)
ˆ
i + ···
=
ˆ
i
ˆ
j
ˆ
k
a
1
a
2
a
3
b
1
b
2
b
3
2.5 Scalar triple product
Definition (Scalar triple product). The scalar triple product is defined as
[a, b, c] = a · (b × c).
Proposition.
If a parallelepiped has sides represented by vectors
a, b, c
that
form a right-handed system, then the volume of the parallelepiped is given by
[a, b, c].
b
c
a
Proof.
The area of the base of the parallelepiped is given by
|b||c|sin θ
=
|b × c|
.
Thus the volume=
|b × c||a|cos φ
=
|a · (b × c)|
, where
φ
is the angle between
a
and the normal to
b
and
c
. However, since
a, b, c
form a right-handed system,
we have a · (b × c) 0. Therefore the volume is a · (b × c).
Since the order of a, b, c doesn’t affect the volume, we know that
[a, b, c] = [b, c, a] = [c, a, b] = [b, a, c] = [a, c, b] = [c, b, a].
Theorem. a × (b + c) = a × b + a × c.
Proof. Let d = a × (b + c) a × b a × c. We have
d · d = d · [a × (b + c)] d · (a × b) d · (a × c)
= (b + c) · (d × a) b · (d × a) c · (d × a)
= 0
Thus d = 0.
2.6 Spanning sets and bases
2.6.1 2D space
Definition
(Spanning set)
.
A set of vectors
{a, b}
spans
R
2
if for all vectors
r R
2
, there exist some λ, µ R such that r = λa + µb.
In R
2
, two vectors span the space if a × b 6= 0.
Theorem. The coefficients λ, µ are unique.
Proof.
Suppose that
r
=
λa
+
µb
=
λ
0
a
+
µ
0
b
. Take the vector product with
a
on both sides to get (
µ µ
0
)
a × b
=
0
. Since
a × b 6
=
0
, then
µ
=
µ
0
. Similarly,
λ = λ
0
.
Definition
(Linearly independent vectors in
R
2
)
.
Two vectors
a
and
b
are
linearly independent if for
α, β R
,
αa
+
βb
=
0
iff
α
=
β
= 0. In
R
2
,
a
and
b
are linearly independent if a × b 6= 0.
Definition
(Basis of
R
2
)
.
A set of vectors is a basis of
R
2
if it spans
R
2
and
are linearly independent.
Example. {
ˆ
i,
ˆ
j}
=
{
(1
,
0)
,
(0
,
1)
}
is a basis of
R
2
. They are the standard basis
of R
2
.
2.6.2 3D space
We can extend the above definitions of spanning set and linear independent set
to R
3
. Here we have
Theorem.
If
a, b, c R
3
are non-coplanar, i.e.
a ·
(
b × c
)
6
= 0, then they form
a basis of R
3
.
Proof.
For any
r
, write
r
=
λa
+
µb
+
νc
. Performing the scalar product
with
b × c
on both sides, one obtains
r · (b × c)
=
λa · (b × c)
+
µb · (b × c)
+
νc · (b × c)
=
λ[a, b, c]
. Thus
λ
=
[r, b, c]/[a, b, c]
. The values of
µ
and
ν
can
be found similarly. Thus each
r
can be written as a linear combination of
a, b
and c.
By the formula derived above, it follows that if
αa
+
βb
+
γc
=
0
, then
α = β = γ = 0. Thus they are linearly independent.
Note that while we came up with formulas for
λ, µ
and
ν
, we did not actually
prove that these coefficients indeed work. This is rather unsatisfactory. We
could, of course, expand everything out and show that this indeed works, but
in IB Linear Algebra, we will prove a much more general result, saying that if
we have an
n
-dimensional space and a set of
n
linear independent vectors, then
they form a basis.
In R
3
, the standard basis is
ˆ
i,
ˆ
j,
ˆ
k, or (1, 0, 0), (0, 1, 0) and (0, 0, 1).
2.6.3 R
n
space
In general, we can define
Definition
(Linearly independent vectors)
.
A set of vectors
{v
1
, v
2
, v
3
···v
m
}
is linearly independent if
m
X
i=1
λ
i
v
i
= 0 (i) λ
i
= 0.
Definition
(Spanning set)
.
A set of vectors
{u
1
, u
2
, u
3
···u
m
} R
n
is a
spanning set of R
n
if
(x R
n
)(λ
i
)
m
X
i=1
λ
i
u
i
= x
Definition
(Basis vectors)
.
A basis of
R
n
is a linearly independent spanning
set. The standard basis of
R
n
is
e
1
= (1
,
0
,
0
, ···
0)
, e
2
= (0
,
1
,
0
, ···
0)
, ···e
n
=
(0, 0, 0, ··· , 1).
Definition
(Orthonormal basis)
.
A basis
{e
i
}
is orthonormal if
e
i
· e
j
= 0 if
i 6= j and e
i
· e
i
= 1 for all i, j.
Using the Kronecker Delta symbol, which we will define later, we can write
this condition as e
i
· e
j
= δ
ij
.
Definition
(Dimension of vector space)
.
The dimension of a vector space is
the number of vectors in its basis. (Exercise: show that this is well-defined)
We usually denote the components of a vector
x
by
x
i
. So we have
x
=
(x
1
, x
2
, ··· , x
n
).
Definition
(Scalar product)
.
The scalar product of
x, y R
n
is defined as
x · y =
P
x
i
y
i
.
The reader should check that this definition coincides with the
|x||y|cos θ
definition in the case of R
2
and R
3
.
2.6.4 C
n
space
C
n
is very similar to
R
n
, except that we have complex numbers. As a result, we
need a different definition of the scalar product. If we still defined
u ·v
=
P
u
i
v
i
,
then if we let
u
= (0
, i
), then
u · u
=
1
<
0. This would be bad if we want to
use the scalar product to define a norm.
Definition
(
C
n
)
. C
n
=
{
(
z
1
, z
2
, ··· , z
n
) :
z
i
C}
. It has the same standard
basis as
R
n
but the scalar product is defined differently. For
u, v C
n
,
u · v
=
P
u
i
v
i
. The scalar product has the following properties:
(i) u · v = (v · u)
(ii) u · (λv + µw) = λ(u · v) + µ(u · w)
(iii) u · u 0 and u · u = 0 iff u = 0
Instead of linearity in the first argument, here we have (
λu
+
µv
)
· w
=
λ
u · w + µ
v · w.
Example.
4
X
k=1
(i)
k
|x + i
k
y|
2
=
X
(i)
k
hx + i
k
y | x + i
k
yi
=
X
(i)
k
(hx + i
k
y | xi + i
k
hx + i
k
y | yi)
=
X
(i)
k
(hx | xi + (i)
k
hy | xi + i
k
hx | yi + i
k
(i)
k
hy | yi)
=
X
(i)
k
[(|x|
2
+ |y|
2
) + (1)
k
hy | xi + hx | yi]
= (|x|
2
+ |y|
2
)
X
(i)
k
+ hy | xi
X
(1)
k
+ hx | yi
X
1
= 4hx | yi.
We can prove the Cauchy-Schwarz inequality for complex vector spaces using
the same proof as the real case, except that this time we have to first multiply
y
by some
e
so that
x ·
(
e
y
) is a real number. The factor of
e
will drop off at
the end when we take the modulus signs.
2.7 Vector subspaces
Definition
(Vector subspace)
.
A vector subspace of a vector space
V
is a subset
of
V
that is also a vector space under the same operations. Both
V
and
{0}
are
subspaces of V . All others are proper subspaces.
A useful criterion is that a subset U V is a subspace iff
(i) x, y U (x + y) U.
(ii) x U λx U for all scalars λ.
(iii) 0 U .
This can be more concisely written as
U
is non-empty and for all
x, y U
,
(λx + µy) U”.
Example.
(i)
If
{a, b, c}
is a basis of
R
3
, then
{a + c, b + c}
is a basis of a 2D subspace.
Suppose x, y span{a + c, b + c}. Let
x = α
1
(a + c) + β
1
(b + c);
y = α
2
(a + c) + β
2
(b + c).
Then
λx + µy = (λα
1
+ µα
2
)(a + c) + (λβ
1
+ µβ
2
)(b + c) span{a + c, b + c}.
Thus this is a subspace of R
3
.
Now check that
a + c, b + c
is a basis. We only need to check linear
independence. If
α
(
a + c
) +
β
(
b + c
) =
0
, then
αa
+
βb
+ (
α
+
β
)
c
=
0
.
Since
{a, b, c}
is a basis of
R
3
, therefore
a, b, c
are linearly independent
and
α
=
β
= 0. Therefore
a + c, b + c
is a basis and the subspace has
dimension 2.
(ii)
Given a set of numbers
α
i
, let
U
=
{x R
n
:
P
n
i=1
α
i
x
i
= 0
}
. We show
that this is a vector subspace of
R
n
: Take
x, y U
, then consider
λx
+
µy
.
We have
P
α
i
(
λx
i
+
µy
i
) =
λ
P
α
i
x
i
+
µ
P
α
i
y
i
= 0. Thus
λx
+
µy U
.
The dimension of the subspace is
n
1 as we can freely choose
x
i
for
i = 1, ··· , n 1 and then x
n
is uniquely determined by the previous x
i
’s.
(iii)
Let
W
=
{x R
n
:
P
α
i
x
i
= 1
}
. Then
P
α
i
(
λx
i
+
µy
i
) =
λ
+
µ 6
= 1.
Therefore W is not a vector subspace.
2.8 Suffix notation
Here we are going to introduce a powerful notation that can help us simplify a
lot of things.
First of all, let
v R
3
. We can write
v
=
v
1
e
1
+
v
2
e
2
+
v
3
e
3
= (
v
1
, v
2
, v
3
).
So in general, the
i
th component of
v
is written as
v
i
. We can thus write
vector equations in component form. For example,
a
=
b a
i
=
b
i
or
c
=
αa
+
βb c
i
=
αa
i
+
βb
i
. A vector has one free suffix,
i
, while a scalar
has none.
Notation
(Einstein’s summation convention)
.
Consider a sum
x · y
=
P
x
i
y
i
.
The summation convention says that we can drop the
P
symbol and simply
write x · y = x
i
y
i
. If suffixes are repeated once, summation is understood.
Note that
i
is a dummy suffix and doesn’t matter what it’s called, i.e.
x
i
y
i
= x
j
y
j
= x
k
y
k
etc.
The rules of this convention are:
(i) Suffix appears once in a term: free suffix
(ii) Suffix appears twice in a term: dummy suffix and is summed over
(iii) Suffix appears three times or more: WRONG!
Example. [(a · b)c (a · c)b]
i
= a
j
b
j
c
i
a
j
c
j
b
i
summing over j understood.
It is possible for an item to have more than one index. These objects are
known as tensors, which will be studied in depth in the IA Vector Calculus
course.
Here we will define two important tensors:
Definition (Kronecker delta).
δ
ij
=
(
1 i = j
0 i 6= j
.
We have
δ
11
δ
12
δ
13
δ
21
δ
22
δ
23
δ
31
δ
32
δ
33
=
1 0 0
0 1 0
0 0 1
= I.
So the Kronecker delta represents an identity matrix.
Example.
(i) a
i
δ
i1
= a
1
. In general, a
i
δ
ij
= a
j
(i is dummy, j is free).
(ii) δ
ij
δ
jk
= δ
ik
(iii) δ
ii
= n if we are in R
n
.
(iv) a
p
δ
pq
b
q
= a
p
b
p
with p, q both dummy suffices and summed over.
Definition
(Alternating symbol
ε
ijk
)
.
Consider rearrangements of 1
,
2
,
3. We
can divide them into even and odd permutations. Even permutations include
(1
,
2
,
3), (2
,
3
,
1) and (3
,
1
,
2). These are permutations obtained by performing
two (or no) swaps of the elements of (1
,
2
,
3). (Alternatively, it is any “rotation”
of (1, 2, 3))
The odd permutations are (2
,
1
,
3), (1
,
3
,
2) and (3
,
2
,
1). They are the
permutations obtained by one swap only.
Define
ε
ijk
=
+1 ijk is even permutation
1 ijk is odd permutation
0 otherwise (i.e. repeated suffices)
ε
ijk
has 3 free suffices.
We have
ε
123
=
ε
231
=
ε
312
= +1 and
ε
213
=
ε
132
=
ε
321
=
1.
ε
112
=
ε
111
= ··· = 0.
We have
(i) ε
ijk
δ
jk
= ε
ijj
= 0
(ii)
If
a
jk
=
a
kj
(i.e.
a
ij
is symmetric), then
ε
ijk
a
jk
=
ε
ijk
a
kj
=
ε
ikj
a
kj
.
Since
ε
ijk
a
jk
=
ε
ikj
a
kj
(we simply renamed dummy suffices), we have
ε
ijk
a
jk
= 0.
Proposition. (a × b)
i
= ε
ijk
a
j
b
k
Proof. By expansion of formula
Theorem. ε
ijk
ε
ipq
= δ
jp
δ
kq
δ
jq
δ
kp
Proof. Proof by exhaustion:
RHS =
+1 if j = p and k = q
1 if j = q and k = p
0 otherwise
LHS: Summing over
i
, the only non-zero terms are when
j, k 6
=
i
and
p, q 6
=
i
.
If
j
=
p
and
k
=
q
, LHS is (
1)
2
or (+1)
2
= 1. If
j
=
q
and
k
=
p
, LHS is
(+1)(1) or (1)(+1) = 1. All other possibilities result in 0.
Equally, we have ε
ijk
ε
pqk
= δ
ip
δ
jq
δ
jp
δ
iq
and ε
ijk
ε
pjq
= δ
ip
δ
kq
δ
iq
δ
kp
.
Proposition.
a · (b × c) = b · (c × a)
Proof. In suffix notation, we have
a · (b × c) = a
i
(b × c)
i
= ε
ijk
b
j
c
k
a
i
= ε
jki
b
j
c
k
a
i
= b · (c × a).
Theorem (Vector triple product).
a × (b × c) = (a · c)b (a · b)c.
Proof.
[a × (b × c)]
i
= ε
ijk
a
j
(b × c)
k
= ε
ijk
ε
kpq
a
j
b
p
c
q
= ε
ijk
ε
pqk
a
j
b
p
c
q
= (δ
ip
δ
jq
δ
iq
δ
jp
)a
j
b
p
c
q
= a
j
b
i
c
j
a
j
c
i
b
j
= (a · c)b
i
(a · b)c
i
Similarly, (a × b) × c = (a · c)b (b · c)a.
Spherical trigonometry
Proposition. (a × b) · (a × c) = (a · a)(b · c) (a · b)(a · c).
Proof.
LHS = (a × b)
i
(a × c)
i
= ε
ijk
a
j
b
k
ε
ipq
a
p
c
q
= (δ
jp
δ
kq
δ
jq
δ
kp
)a
j
b
k
a
p
c
q
= a
j
b
k
a
j
c
k
a
j
b
k
a
k
c
j
= (a · a)(b · c) (a · b)(a · c)
Consider the unit sphere, center O, with a, b, c on the surface.
A
B C
δ(A, B)
α
Suppose we are living on the surface of the sphere. So the distance from
A
to
B
is
the arc length on the sphere. We can imagine this to be along the circumference
of the circle through
A
and
B
with center
O
. So the distance is
AOB
, which we
shall denote by
δ
(
A, B
). So
a · b
=
cos AOB
=
cos δ
(
A, B
). We obtain similar
expressions for other dot products. Similarly, we get |a × b| = sin δ(A, B).
cos α =
(a × b) · (a × c)
|a × b||a × c|
=
b · c (a · b)(a · c)
|a × b||a × c|
Putting in our expressions for the dot and cross products, we obtain
cos α sin δ(A, B) sin δ(A, C) = cos δ(B, C) cos δ(A, B) cos δ(A, C).
This is the spherical cosine rule that applies when we live on the surface of a
sphere. What does this spherical geometry look like?
Consider a spherical equilateral triangle. Using the spherical cosine rule,
cos α =
cos δ cos
2
δ
sin
2
δ
= 1
1
1 + cos δ
.
Since
cos δ
1, we have
cos α
1
2
and
α
60
. Equality holds iff
δ
= 0, i.e. the
triangle is simply a point. So on a sphere, each angle of an equilateral triangle is
greater than 60
, and the angle sum of a triangle is greater than 180
.
2.9 Geometry
2.9.1 Lines
Any line through a and parallel to t can be written as
x = a + λt.
By crossing both sides of the equation with t, we have
Theorem. The equation of a straight line through a and parallel to t is
(x a) × t = 0 or x × t = a × t.
2.9.2 Plane
To define a plane Π, we need a normal
n
to the plane and a fixed point
b
. For
any
x
Π, the vector
x b
is contained in the plane and is thus normal to
n
,
i.e. (x b) · n = 0.
Theorem. The equation of a plane through b with normal n is given by
x · n = b · n.
If
n = ˆn
is a unit normal, then
d
=
x · ˆn = b · ˆn
is the perpendicular distance
from the origin to Π.
Alternatively, if a, b, c lie in the plane, then the equation of the plane is
(x a) · [(b a) × (c a)] = 0.
Example.
(i)
Consider the intersection between a line
x × t = a × t
with the plane
x · n = b · n. Cross n on the right with the line equation to obtain
(x · n)t (t · n)x = (a × t) × n
Eliminate x · n using x · n = b · n
(t · n)x = (b · n)t (a × t) × n
Provided t · n is non-zero, the point of intersection is
x =
(b · n)t (a × t) × n
t · n
.
Exercise: what if t · n = 0?
(ii)
Shortest distance between two lines. Let
L
1
be (
x a
1
)
× t
1
=
0
and
L
2
be (x a
2
) × t
2
= 0.
The distance of closest approach
s
is along a line perpendicular to both
L
1
and
L
2
, i.e. the line of closest approach is perpendicular to both lines and
thus parallel to
t
1
× t
2
. The distance
s
can then be found by projecting
a
1
a
2
onto t
1
× t
2
. Thus s =
(a
1
a
2
) ·
t
1
×t
2
|t
1
×t
2
|
.
2.10 Vector equations
Example. x (x × a) × b = c
. Strategy: take the dot or cross of the equation
with suitable vectors. The equation can be expanded to form
x (x · b)a + (a · b)x = c.
Dot this with b to obtain
x · b (x · b)(a · b) + (a · b)(x · b) = c · b
x · b = c · b.
Substituting this into the original equation, we have
x(1 + a · b) = c + (c · b)a
If (1 + a · b) is non-zero, then
x =
c + (c · b)a
1 + a · b
Otherwise, when (1 +
a · b
) = 0, if
c + (c · b)a 6= 0
, then a contradiction is
reached. Otherwise,
x · b = c · b
is the most general solution, which is a plane
of solutions.
3 Linear maps
A linear map is a special type of function between vector spaces. In fact, most
of the time, these are the only functions we actually care about. They are maps
that satisfy the property f(λa + µb) = λf(a) + µf (b).
We will first look at two important examples of linear maps rotations and
reflections, and then study their properties formally.
3.1 Examples
3.1.1 Rotation in R
3
In
R
3
, first consider the simple cases where we rotate about the
z
axis by
θ
. We
call this rotation R and write x
0
= R(x).
Suppose that initially,
x
= (
x, y, z
) = (
r cos φ, r sin φ, z
). Then after a
rotation by θ, we get
x
0
= (r cos(φ + θ), r sin(φ + θ), z)
= (r cos φ cos θ r sin φ sin θ, r sin φ cos θ + r cos φ sin θ, z)
= (x cos θ y sin θ, x sin θ + y cos θ, z).
We can represent this by a matrix
R
such that
x
0
i
=
R
ij
x
j
. Using our formula
above, we obtain
R =
cos θ sin θ 0
sin θ cos θ 0
0 0 1
Now consider the general case where we rotate by θ about
ˆ
n.
O
ˆ
n
A
x
B
A
0
C
x
0
B A
A
0
C
θ
We have x
0
=
OB +
BC +
CA
0
. We know that
OB = (ˆn · x)ˆn
BC =
BA cos θ
= (
BO +
OA) cos θ
= ((ˆn · x)ˆn + x) cos θ
Finally, to get
CA
, we know that
|
CA
0
|
=
|
BA
0
|sin θ
=
|
BA|sin θ
=
|ˆn × x|sin θ
.
Also,
CA
0
is parallel to
ˆ
n × x. So we must have
CA
0
= (
ˆ
n × x) sin θ.
Thus x
0
= x cos θ + (1 cos θ)(ˆn · x)ˆn + ˆn × x sin θ. In components,
x
0
i
= x
i
cos θ + (1 cos θ)n
j
x
j
n
i
ε
ijk
x
j
n
k
sin θ.
We want to find an R such that x
0
i
= R
ij
x
j
. So
R
ij
= δ
ij
cos θ + (1 cos θ)n
i
n
j
ε
ijk
n
k
sin θ.
3.1.2 Reflection in R
3
Suppose we want to reflect through a plane through
O
with normal
ˆ
n
. First of
all the projection of
x
onto
ˆ
n
is given by (
x ·
ˆ
n
)
ˆ
n
. So we get
x
0
=
x
2
(x · ˆn)ˆn
.
In suffix notation, we have
x
0
i
=
x
i
2
x
j
n
j
n
i
. So our reflection matrix is
R
ij
= δ
ij
2n
i
n
j
.
x
0
ˆ
n x
3.2 Linear Maps
Definition
(Domain, codomain and image of map)
.
Consider sets
A
and
B
and mapping
T
:
A B
such that each
x A
is mapped into a unique
x
0
=
T
(
x
)
B
.
A
is the domain of
T
and
B
is the co-domain of
T
. Typically,
we have T : R
n
R
m
or T : C
n
C
m
.
Definition
(Linear map)
.
Let
V, W
be real (or complex) vector spaces, and
T : V W . Then T is a linear map if
(i) T (a + b) = T (a) + T (b) for all a, b V .
(ii) T (λa) = λT (a) for all λ R (or C).
Equivalently, we have T (λa + µb) = λT (a) + µT (b).
Example.
(i)
Consider a translation
T
:
R
3
R
3
with
T
(
x
) =
x + a
for some fixed,
given
a
. This is not a linear map since
T
(
λx
+
µy
)
6
=
λx
+
µy
+ (
λ
+
µ
)
a
.
(ii) Rotation, reflection and projection are linear transformations.
Definition
(Image and kernel of map)
.
The image of a map
f
:
U V
is the
subset of V {f(u) : u U}. The kernel is the subset of U {u U : f(u) = 0}.
Example.
(i)
Consider
S
:
R
3
R
2
with
S
(
x, y, z
) = (
x
+
y,
2
x z
). Simple yet
tedious algebra shows that this is linear. Now consider the effect of
S
on
the standard basis.
S
(1
,
0
,
0) = (1
,
2),
S
(0
,
1
,
0) = (1
,
0) and
S
(0
,
0
,
1) =
(0
,
1). Clearly these are linearly dependent, but they do span the whole
of R
2
. We can say S(R
3
) = R
2
. So the image is R
2
.
Now solve
S
(
x, y, z
) =
0
. We need
x
+
y
= 0 and 2
x z
= 0. Thus
x
=
(
x, x,
2
x
), i.e. it is parallel to (1
,
1
,
2). So the set
{λ
(1
,
1
,
2) :
λ R}
is the kernel of S.
(ii)
Consider a rotation in
R
3
. The kernel is the zero vector and the image is
R
3
.
(iii)
Consider a projection of
x
onto a plane with normal
ˆn
. The image is the
plane itself, and the kernel is any vector parallel to ˆn
Theorem.
Consider a linear map
f
:
U V
, where
U, V
are vector spaces.
Then im(f ) is a subspace of V , and ker(f) is a subspace of U.
Proof. Both are non-empty since f(0) = 0.
If
x, y im
(
f
), then
a, b U
such that
x
=
f
(
a
)
, y
=
f
(
b
). Then
λx
+
µy
=
λf
(
a
) +
µf
(
b
) =
f
(
λa
+
µb
). Now
λa
+
µb U
since
U
is a vector
space, so there is an element in
U
that maps to
λx
+
µy
. So
λx
+
µy im
(
f
)
and im(f ) is a subspace of V .
Suppose
x, y ker
(
f
), i.e.
f
(
x
) =
f
(
y
) =
0
. Then
f
(
λx
+
µy
) =
λf
(
x
) +
µf(y) = λ0 + µ0 = 0. Therefore λx + µy ker(f).
3.3 Rank and nullity
Definition
(Rank of linear map)
.
The rank of a linear map
f
:
U V
, denoted
by r(f ), is the dimension of the image of f.
Definition
(Nullity of linear map)
.
The nullity of
f
, denoted
n
(
f
) is the
dimension of the kernel of f.
Example.
For the projection onto a plane in
R
3
, the image is the whole plane
and the rank is 2. The kernel is a line so the nullity is 1.
Theorem (Rank-nullity theorem). For a linear map f : U V ,
r(f) + n(f) = dim(U ).
Proof.
(Non-examinable) Write
dim
(
U
) =
n
and
n
(
f
) =
m
. If
m
=
n
, then
f
is
the zero map, and the proof is trivial, since
r
(
f
) = 0. Otherwise, assume
m < n
.
Suppose
{e
1
, e
2
, ··· , e
m
}
is a basis of
ker f
, Extend this to a basis of the
whole of
U
to get
{e
1
, e
2
, ··· , e
m
, e
m+1
, ··· , e
n
}
. To prove the theorem, we
need to prove that {f(e
m+1
), f(e
m+2
), ···f(e
n
)} is a basis of im(f).
(i)
First show that it spans
im
(
f
). Take
y im
(
f
). Thus
x U
such that
y = f(x). Then
y = f(α
1
e
1
+ α
2
e
2
+ ··· + α
n
e
n
),
since e
1
, ···e
n
is a basis of U. Thus
y = α
1
f(e
1
) + α
2
f(e
2
) + ···+ α
m
f(e
m
) + α
m+1
f(e
m+1
) + ···+ α
n
f(e
n
).
The first
m
terms map to
0
, since
e
1
, ···e
m
is the basis of the kernel of
f
.
Thus
y = α
m+1
f(e
m+1
) + ··· + α
n
f(e
n
).
(ii) To show that they are linearly independent, suppose
α
m+1
f(e
m+1
) + ··· + α
n
f(e
n
) = 0.
Then
f(α
m+1
e
m+1
+ ··· + α
n
e
n
) = 0.
Thus
α
m+1
e
m+1
+
···
+
α
n
e
n
ker
(
f
). Since
{e
1
, ··· , e
m
}
span
ker
(
f
),
there exist some α
1
, α
2
, ···α
m
such that
α
m+1
e
m+1
+ ··· + α
n
e
n
= α
1
e
1
+ ··· + α
m
e
m
.
But
e
1
···e
n
is a basis of
U
and are linearly independent. So
α
i
= 0 for all
i
.
Then the only solution to the equation
α
m+1
f
(
e
m+1
) +
···
+
α
n
f
(
e
n
) =
0
is α
i
= 0, and they are linearly independent by definition.
Example.
Calculate the kernel and image of
f
:
R
3
R
3
, defined by
f(x, y, z) = (x + y + z, 2x y + 5z, x + 2z).
First find the kernel: we’ve got the system of equations:
x + y + z = 0
2x y + 5z = 0
x + 2z = 0
Note that the first and second equation add to give 3
x
+6
z
= 0, which is identical
to the third. Then using the first and third equation, we have
y
=
x z
=
z
.
So the kernel is any vector in the form (2z, z, z) and is the span of (2, 1, 1).
To find the image, extend the basis of
ker
(
f
) to a basis of the whole of
R
3
:
{
(
2
,
1
,
1)
,
(0
,
1
,
0)
,
(0
,
0
,
1)
}
. Apply
f
to this basis to obtain (0
,
0
,
0)
,
(1
,
1
,
0)
and (1
,
5
,
2). From the proof of the rank-nullity theorem, we know that
f
(0
,
1
,
0)
and f (0, 0, 1) is a basis of the image.
To get the standard form of the image, we know that the normal to the plane
is parallel to (1
,
1
,
0)
×
(1
,
5
,
2)
k
(1
,
1
,
3). Since
0 im
(
f
), the equation of
the plane is x + y 3z = 0.
3.4 Matrices
In the examples above, we have represented our linear maps by some object
R
such that
x
0
i
=
R
ij
x
j
. We call
R
the matrix for the linear map. In general, let
α : R
n
R
m
be a linear map, and x
0
= α(x).
Let {e
i
} be a basis of R
n
. Then x = x
j
e
j
for some x
j
. Then we get
x
0
= α(x
j
e
j
) = x
j
α(e
j
).
So we get that
x
0
i
= [α(e
j
)]
i
x
j
.
We now define A
ij
= [α(e
j
)]
i
. Then x
0
i
= A
ij
x
j
. We write
A = {A
ij
} =
A
11
··· A
1n
.
.
. A
ij
.
.
.
A
m1
··· A
mn
Here
A
ij
is the entry in the
i
th row of the
j
th column. We say that
A
is an
m × n matrix, and write x
0
= Ax.
We see that the columns of the matrix are the images of the standard basis
vectors under the mapping α.
Example.
3.4.1 Examples
(i)
In
R
2
, consider a reflection in a line with an angle
θ
to the
x
axis. We
know that
ˆ
i 7→ cos
2
θ
ˆ
i
+
sin
2
θ
ˆ
j
, with
ˆ
j 7→ cos
2
θ
ˆ
j
+
sin
2
θ
ˆ
i
. Then the
matrix is
cos 2θ sin 2θ
sin 2θ cos 2θ
.
(ii)
In
R
3
, as we’ve previously seen, a rotation by
θ
about the
z
axis is given
by
R =
cos θ sin θ 0
sin θ cos θ 0
0 0 1
(iii)
In
R
3
, a reflection in plane with normal
ˆ
n
is given by
R
ij
=
δ
ij
2
ˆn
i
ˆn
j
.
Written as a matrix, we have
1 2ˆn
2
1
2ˆn
1
ˆn
2
2ˆn
1
ˆn
3
2ˆn
2
ˆn
1
1 2ˆn
2
2
2ˆn
2
ˆn
3
2ˆn
3
ˆn
1
2ˆn
3
ˆn
2
1 2ˆn
2
3
(iv)
Dilation (“stretching”)
α
:
R
3
R
3
is given by a map (
x, y, z
)
7→
(λx, µy, νz) for some λ, µ, ν. The matrix is
λ 0 0
0 µ 0
0 0 ν
(v) Shear: Consider S : R
3
R
3
that sheers in the x direction:
x
y
x x
0
sheer in x direction
We have (x, y, z) 7→ (x + λy, y, z). Then
S =
1 λ 0
0 1 0
0 0 1
3.4.2 Matrix Algebra
This part is mostly on a whole lot of definitions, saying what we can do with
matrices and classifying them into different types.
Definition
(Addition of matrices)
.
Consider two linear maps
α, β
:
R
n
R
m
.
The sum of α and β is defined by
(α + β)(x) = α(x) + β(x)
In terms of the matrix, we have
(A + B)
ij
x
j
= A
ij
x
j
+ B
ij
x
j
,
or
(A + B)
ij
= A
ij
+ B
ij
.
Definition
(Scalar multiplication of matrices)
.
Define (
λα
)
x
=
λ
[
α
(
x
)]. So
(λA)
ij
= λA
ij
.
Definition
(Matrix multiplication)
.
Consider maps
α
:
R
`
R
n
and
β
:
R
n
R
m
. The composition is
βα
:
R
`
R
m
. Take
x R
`
7→ x
00
R
m
.
Then
x
00
= (
BA
)
x
=
Bx
0
, where
x
0
=
Ax
. Using suffix notation, we have
x
00
i
= (Bx
0
)
i
= b
ik
x
0
k
= B
ik
A
kj
x
j
. But x
00
i
= (BA)
ij
x
j
. So
(BA)
ij
= B
ik
A
kj
.
Generally, an
m ×n
matrix multiplied by an
n ×`
matrix gives an
m ×`
matrix.
(BA)
ij
is given by the ith row of B dotted with the jth column of A.
Note that the number of columns of
B
has to be equal to the number of rows
of
A
for multiplication to be defined. If
`
=
m
as well, then both
BA
and
AB
make sense, but
AB 6
=
BA
in general. In fact, they don’t even have to have the
same dimensions.
Also, since function composition is associative, we get A(BC) = (AB)C.
Definition
(Transpose of matrix)
.
If
A
is an
m × n
matrix, the transpose
A
T
is an n × m matrix defined by (A
T
)
ij
= A
ji
.
Proposition.
(i) (A
T
)
T
= A.
(ii) If x is a column vector
x
1
x
2
.
.
.
x
n
, x
T
is a row vector (x
1
x
2
···x
n
).
(iii) (AB)
T
= B
T
A
T
since (AB)
T
ij
= (AB)
ji
= A
jk
B
ki
= B
ki
A
jk
= (B
T
)
ik
(A
T
)
kj
= (B
T
A
T
)
ij
.
Definition
(Hermitian conjugate)
.
Define
A
= (
A
T
)
. Similarly, (
AB
)
=
B
A
.
Definition (Symmetric matrix). A matrix is symmetric if A
T
= A.
Definition
(Hermitian matrix)
.
A matrix is Hermitian if
A
=
A
. (The diagonal
of a Hermitian matrix must be real).
Definition
(Anti/skew symmetric matrix)
.
A matrix is anti-symmetric or skew
symmetric if A
T
= A. The diagonals are all zero.
Definition
(Skew-Hermitian matrix)
.
A matrix is skew-Hermitian if
A
=
A
.
The diagonals are pure imaginary.
Definition
(Trace of matrix)
.
The trace of an
n ×n
matrix
A
is the sum of the
diagonal. tr(A) = A
ii
.
Example.
Consider the reflection matrix
R
ij
=
δ
ij
2
ˆn
i
ˆn
j
. We have
tr
(
A
) =
R
ii
= 3 2ˆn · ˆn = 3 2 = 1.
Proposition. tr(BC) = tr(CB)
Proof. tr(BC) = B
ik
C
ki
= C
ki
B
ik
= (CB)
kk
= tr(CB)
Definition (Identity matrix). I = δ
ij
.
3.4.3 Decomposition of an n × n matrix
Any
n ×n
matrix
B
can be split as a sum of symmetric and antisymmetric parts.
Write
B
ij
=
1
2
(B
ij
+ B
ji
)
| {z }
S
ij
+
1
2
(B
ij
B
ji
)
| {z }
A
ij
.
We have
S
ij
=
S
ji
, so
S
is symmetric, while
A
ji
=
A
ij
, and
A
is antisymmetric.
So B = S + A.
Furthermore , we can decompose
S
into an isotropic part (a scalar multiple
of the identity) plus a trace-less part (i.e. sum of diagonal = 0). Write
S
ij
=
1
n
tr(S)δ
ij
| {z }
isotropic part
+ (S
ij
1
n
tr(S)δ
ij
)
| {z }
T
ij
.
We have tr(T ) = T
ii
= S
ii
1
n
tr(S)δ
ii
= tr(S)
1
n
tr(S)(n) = 0.
Putting all these together,
B =
1
n
tr(B)I +
1
2
(B + B
T
)
1
n
tr(B)I
+
1
2
(B B
T
).
In three dimensions, we can write the antisymmetric part
A
in terms of a single
vector: we have
A =
0 a b
a 0 c
b c 0
and we can consider
ε
ijk
ω
k
=
0 ω
3
ω
2
ω
3
0 ω
1
ω
2
ω
1
0
So if we have ω = (c, b, a), then A
ij
= ε
ijk
ω
k
.
This decomposition can be useful in certain physical applications. For
example, if the matrix represents the stress of a system, different parts of the
decomposition will correspond to different types of stresses.
3.4.4 Matrix inverse
Definition
(Inverse of matrix)
.
Consider an
m×n
matrix
A
and
n×m
matrices
B
and
C
. If
BA
=
I
, then we say
B
is the left inverse of
A
. If
AC
=
I
, then
we say
C
is the right inverse of
A
. If
A
is square (
n × n
), then
B
=
B
(
AC
) =
(
BA
)
C
=
C
, i.e. the left and right inverses coincide. Both are denoted by
A
1
,
the inverse of A. Therefore we have
AA
1
= A
1
A = I.
Note that not all square matrices have inverses. For example, the zero matrix
clearly has no inverse.
Definition (Invertible matrix). If A has an inverse, then A is invertible.
Proposition. (AB)
1
= B
1
A
1
Proof. (B
1
A
1
)(AB) = B
1
(A
1
A)B = B
1
B = I.
Definition
(Orthogonal and unitary matrices)
.
A real
n×n
matrix is orthogonal
if
A
T
A
=
AA
T
=
I
, i.e.
A
T
=
A
1
. A complex
n × n
matrix is unitary if
U
U = UU
= I, i.e. U
= U
1
.
Note that an orthogonal matrix
A
satisfies
A
ik
(
A
T
kj
) =
δ
ij
, i.e.
A
ik
A
jk
=
δ
ij
.
We can see this as saying “the scalar product of two distinct rows is 0, and the
scalar product of a row with itself is 1”. Alternatively, the rows (and columns
by considering A
T
) of an orthogonal matrix form an orthonormal set.
Similarly, for a unitary matrix,
U
ik
U
kj
=
δ
ij
, i.e.
u
ik
u
jk
=
u
ik
u
jk
=
δ
ij
. i.e.
the rows are orthonormal, using the definition of complex scalar product.
Example.
(i)
The reflection in a plane is an orthogonal matrix. Since
R
ij
=
δ
ij
2
n
i
n
j
,
We have
R
ik
R
jk
= (δ
ik
2n
i
n
k
)(δ
jk
2n
j
n
k
)
= δ
ik
δ
jk
2δ
jk
n
i
n
k
2δ
ik
n
j
n
k
+ 2n
i
n
k
n
j
n
k
= δ
ij
2n
i
n
j
2n
j
n
i
+ 4n
i
n
j
(n
k
n
k
)
= δ
ij
(ii)
The rotation is an orthogonal matrix. We could multiply out using suffix
notation, but it would be cumbersome to do so. Alternatively, denote
rotation matrix by
θ
about
ˆn
as
R
(
θ, ˆn
). Clearly,
R
(
θ, ˆn
)
1
=
R
(
θ, ˆn
).
We have
R
ij
(θ, ˆn) = (cos θ)δ
ij
+ n
i
n
j
(1 cos θ) + ε
ijk
n
k
sin θ
= (cos θ)δ
ji
+ n
j
n
i
(1 cos θ) ε
jik
n
k
sin θ
= R
ji
(θ, ˆn)
In other words, R(θ, ˆn) = R(θ, ˆn)
T
. So R(θ, ˆn)
1
= R(θ, ˆn)
T
.
3.5 Determinants
Consider a linear map
α
:
R
3
R
3
. The standard basis
e
1
, e
2
, e
3
is mapped to
e
0
1
, e
0
2
, e
0
3
with
e
0
i
=
Ae
i
. Thus the unit cube formed by
e
1
, e
2
, e
3
is mapped to
the parallelepiped with volume
[e
0
1
, e
0
2
, e
0
3
] = ε
ijk
(e
0
1
)
i
(e
0
2
)
j
(e
0
3
)
k
= ε
ijk
A
i`
(e
1
)
`
|{z}
δ
1`
A
jm
(e
2
)
m
|{z}
δ
2m
A
kn
(e
3
)
n
|{z}
δ
3n
= ε
ijk
A
i1
A
j2
A
k3
We call this the determinant and write as
det(A) =
A
11
A
12
A
13
A
21
A
22
A
23
A
31
A
32
A
33
3.5.1 Permutations
To define the determinant for square matrices of arbitrary size, we first have to
consider permutations.
Definition (Permutation). A permutation of a set S is a bijection ε : S S.
Notation.
Consider the set
S
n
of all permutations of 1
,
2
,
3
, ··· , n
.
S
n
contains
n! elements. Consider ρ S
n
with i 7→ ρ(i). We write
ρ =
1 2 ··· n
ρ(1) ρ(2) ··· ρ(n)
.
Definition
(Fixed point)
.
A fixed point of
ρ
is a
k
such that
ρ
(
k
) =
k
. e.g. in
1 2 3 4
4 1 3 2
, 3 is the fixed point. By convention, we can omit the fixed point
and write as
1 2 4
4 1 2
.
Definition (Disjoint permutation). Two permutations are disjoint if numbers
moved by one are fixed by the other, and vice versa. e.g.
1 2 4 5 6
5 6 1 4 2
=
2 6
6 2
1 4 5
5 1 4
, and the two cycles on the right hand side are disjoint.
Disjoint permutations commute, but in general non-disjoint permutations do
not.
Definition
(Transposition and
k
-cycle)
.
2 6
6 2
is a 2-cycle or a transposition,
and we can simply write (2 6).
1 4 5
5 1 4
is a 3-cycle, and we can simply write
(1 5 4). (1 is mapped to 5; 5 is mapped to 4; 4 is mapped to 1)
Proposition. Any q-cycle can be written as a product of 2-cycles.
Proof. (1 2 3 ··· n) = (1 2)(2 3)(3 4) ···(n 1 n).
Definition
(Sign of permutation)
.
The sign of a permutation
ε
(
ρ
) is (
1)
r
,
where
r
is the number of 2-cycles when
ρ
is written as a product of 2-cycles. If
ε
(
ρ
) = +1, it is an even permutation. Otherwise, it is an odd permutation. Note
that ε(ρσ) = ε(ρ)ε(σ) and ε(ρ
1
) = ε(ρ).
The proof that this is well-defined can be found in IA Groups.
Definition (Levi-Civita symbol). The Levi-Civita symbol is defined by
ε
j
1
j
2
···j
n
=
+1 if j
1
j
2
j
3
···j
n
is an even permutation of 1, 2, ···n
1 if it is an odd permutation
0 if any 2 of them are equal
Clearly, ε
ρ(1)ρ(2)···ρ(n)
= ε(ρ).
Definition
(Determinant)
.
The determinant of an
n ×n
matrix
A
is defined as:
det(A) =
X
σS
n
ε(σ)A
σ(1)1
A
σ(2)2
···A
σ(n)n
,
or equivalently,
det(A) = ε
j
1
j
2
···j
n
A
j
1
1
A
j
2
2
···A
j
n
n
.
Proposition.
a b
c d
= ad bc
3.5.2 Properties of determinants
Proposition. det(A) = det(A
T
).
Proof.
Take a single term
A
σ(1)1
A
σ(2)2
···A
σ(n)n
and let
ρ
be another permuta-
tion in S
n
. We have
A
σ(1)1
A
σ(2)2
···A
σ(n)n
= A
σ(ρ(1))ρ(1)
A
σ(ρ(2))ρ(2)
···A
σ(ρ(n))ρ(n)
since the right hand side is just re-ordering the order of multiplication. Choose
ρ = σ
1
and note that ε(σ) = ε(ρ). Then
det(A) =
X
ρS
n
ε(ρ)A
1ρ(1)
A
2ρ(2)
···A
(n)
= det(A
T
).
Proposition.
If matrix
B
is formed by multiplying every element in a single row
of
A
by a scalar
λ
, then
det
(
B
) =
λ det
(
A
). Consequently,
det
(
λA
) =
λ
n
det
(
A
).
Proof.
Each term in the sum is multiplied by
λ
, so the whole sum is multiplied
by λ
n
.
Proposition.
If 2 rows (or 2 columns) of
A
are identical, the determinant is 0.
Proof. wlog, suppose columns 1 and 2 are the same. Then
det(A) =
X
σS
n
ε(σ)A
σ(1)1
A
σ(2)2
···A
σ(n)n
.
Now write an arbitrary
σ
in the form
σ
=
ρ
(1 2). Then
ε
(
σ
) =
ε
(
ρ
)
ε
((1 2)) =
ε(ρ). So
det(A) =
X
ρS
n
ε(ρ)A
ρ(2)1
A
ρ(1)2
A
ρ(3)3
···A
ρ(n)n
.
But columns 1 and 2 are identical, so
A
ρ(2)1
=
A
ρ(2)2
and
A
ρ(1)2
=
A
ρ(1)1
. So
det(A) = det(A) and det(A) = 0.
Proposition.
If 2 rows or 2 columns of a matrix are linearly dependent, then
the determinant is zero.
Proof. Suppose in A, (column r) + λ(column s) = 0. Define
B
ij
=
(
A
ij
j 6= r
A
ij
+ λA
is
j = r
.
Then
det
(
B
) =
det
(
A
) +
λ det
(matrix with column
r
= column
s
) =
det
(
A
).
Then we can see that the
r
th column of
B
is all zeroes. So each term in the sum
contains one zero and det(A) = det(B) = 0.
Even if we don’t have linearly dependent rows or columns, we can still run
the exact same proof as above, and still get that
det
(
B
) =
det
(
A
). Linear
dependence is only required to show that
det
(
B
) = 0. So in general, we can add
a linear multiple of a column (or row) onto another column (or row) without
changing the determinant.
Proposition.
Given a matrix
A
, if
B
is a matrix obtained by adding a multiple
of a column (or row) of
A
to another column (or row) of
A
, then
det A
=
det B
.
Corollary.
Swapping two rows or columns of a matrix negates the determinant.
Proof. We do the column case only. Let A = (a
1
···a
i
···a
j
···a
n
). Then
det(a
1
···a
i
···a
j
···a
n
) = det(a
1
···a
i
+ a
j
···a
j
···a
n
)
= det(a
1
···a
i
+ a
j
···a
j
(a
i
+ a
j
) ···a
n
)
= det(a
1
···a
i
+ a
j
··· a
i
···a
n
)
= det(a
1
···a
j
··· a
i
···a
n
)
= det(a
1
···a
j
···a
i
···a
n
)
Alternatively, we can prove this from the definition directly, using the fact that
the sign of a transposition is 1 (and that the sign is multiplicative).
Proposition. det(AB) = det(A) det(B).
Proof.
First note that
P
σ
ε
(
σ
)
A
σ(1)ρ(1)
A
σ(2)ρ(2)
=
ε
(
ρ
)
det
(
A
), i.e. swapping
columns (or rows) an even/odd number of times gives a factor
±
1 respectively.
We can prove this by writing σ = µρ.
Now
det AB =
X
σ
ε(σ)(AB)
σ(1)1
(AB)
σ(2)2
···(AB)
σ(n)n
=
X
σ
ε(σ)
n
X
k
1
,k
2
,···,k
n
A
σ(1)k
1
B
k
1
1
···A
σ(n)k
n
B
k
n
n
=
X
k
1
,···,k
n
B
k
1
1
···B
k
n
n
X
σ
ε(σ)A
σ(1)k
1
A
σ(2)k
2
···A
σ(n)k
n
| {z }
S
Now consider the many different
S
’s. If in
S
, two of
k
1
and
k
n
are equal, then
S
is a determinant of a matrix with two columns the same, i.e.
S
= 0. So we only
have to consider the sum over distinct
k
i
s. Thus the
k
i
s are are a permutation
of 1, ···n, say k
i
= ρ(i). Then we can write
det AB =
X
ρ
B
ρ(1)1
···B
ρ(n)n
X
σ
ε(σ)A
σ(1)ρ(1)
···A
σ(n)ρ(n)
=
X
ρ
B
ρ(1)1
···B
ρ(n)n
(ε(ρ) det A)
= det A
X
ρ
ε(ρ)B
ρ(1)1
···B
ρ(n)n
= det A det B
Corollary. If A is orthogonal, det A = ±1.
Proof.
AA
T
= I
det AA
T
= det I
det A det A
T
= 1
(det A)
2
= 1
det A = ±1
Corollary. If U is unitary, |det U| = 1.
Proof.
We have
det U
= (
det U
T
)
=
det
(
U
)
. Since
UU
=
I
, we have
det(U) det(U)
= 1.
Proposition.
In
R
3
, orthogonal matrices represent either a rotation (
det
= 1)
or a reflection (det = 1).
3.5.3 Minors and Cofactors
Definition
(Minor and cofactor)
.
For an
n × n
matrix
A
, define
A
ij
to be the
(n 1) × (n 1) matrix in which row i and column j of A have been removed.
The minor of the ijth element of A is M
ij
= det A
ij
The cofactor of the ijth element of A is
ij
= (1)
i+j
M
ij
.
Notation.
We use
¯
to denote a symbol which has been missed out of a natural
sequence.
Example. 1, 2, 3, 5 = 1, 2, 3,
¯
4, 5.
The significance of these definitions is that we can use them to provide a
systematic way of evaluating determinants. We will also use them to find inverses
of matrices.
Theorem (Laplace expansion formula). For any particular fixed i,
det A =
n
X
j=1
A
ji
ji
.
Proof.
det A =
n
X
j
i
=1
A
j
i
i
n
X
j
1
,···,j
i
,···j
n
ε
j
1
j
2
···j
n
A
j
1
1
A
j
2
2
···A
j
i
i
···A
j
n
n
Let
σ S
n
be the permutation which moves
j
i
to the
i
th position, and leave
everything else in its natural order, i.e.
σ =
1 ··· i i + 1 i + 2 ··· j
i
1 j
i
j
i
+ 1 ··· n
1 ··· j
i
i i + 1 ··· j
i
2 j
i
1 j
i
+ 1 ··· n
if
j
i
> i
, and similarly for other cases. To perform this permutation,
|i j
i
|
transpositions are made. So ε(σ) = (1)
ij
i
.
Now consider the permutation ρ S
n
ρ =
1 ··· ···
¯
j
i
··· n
j
1
···
¯
j
i
··· ··· j
n
The composition
ρσ
reorders (1
, ··· , n
) to (
j
1
, j
2
, ··· , j
n
). So
ε
(
ρσ
) =
ε
j
1
···j
n
=
ε(ρ)ε(σ) = (1)
ij
i
ε
j
1
···
¯
j
i
···j
n
. Hence the original equation becomes
det A =
n
X
j
i
=1
A
j
i
i
X
j
1
···
¯
j
i
···j
n
(1)
ij
i
ε
j
1
···
¯
j
i
···j
n
A
j
1
1
···A
j
i
i
···A
j
n
n
=
n
X
j
i
=1
A
j
i
i
(1)
ij
i
M
j
i
i
=
n
X
j
i
=1
A
j
i
i
j
i
i
=
n
X
j=1
A
ji
ji
Example. det A =
2 4 2
3 2 1
2 0 1
. We can pick the first row and have
det A = 2
2 1
0 1
4
3 1
2 1
+ 2
3 2
2 0
= 2(2 0) 4(3 2) + 2(0 4)
= 8.
Alternatively, we can pick the second column and have
det A = 4
3 1
2 1
+ 2
2 2
2 1
0
2 2
3 1
= 4(3 2) + 2(2 4) 0
= 8.
In practical terms, we use a combination of properties of determinants with
a sensible choice of i to evaluate det(A).
Example. Consider
1 a a
2
1 b b
2
1 c c
2
. Row 1 - row 2 gives
0 a b a
2
b
2
1 b b
2
1 c c
2
= (a b)
0 1 a + b
1 b b
2
1 c c
2
.
Do row 2 - row 3. We obtain
(a b)(b c)
0 1 a + b
0 1 b + c
1 c c
2
.
Row 1 - row 2 gives
(a b)(b c)(a c)
0 0 1
0 1 b + c
1 c c
2
= (a b)(b c)(a c).
4 Matrices and linear equations
4.1 Simple example, 2 × 2
Consider the system of equations
A
11
x
1
+ A
12
x
2
= d
1
(a)
A
21
x
1
+ A
22
x
2
= d
2
. (b)
We can write this as
Ax = d.
If we do (a)×A
22
(b)×A
12
and similarly the other way round, we obtain
(A
11
A
22
A
12
A
21
)x
1
= A
22
d
1
A
12
d
2
(A
11
A
22
A
12
A
21
)
| {z }
det A
x
2
= A
11
d
2
A
21
d
1
Dividing by det A and writing in matrix form, we have
x
1
x
2
=
1
det A
A
22
A
12
A
21
A
11
d
1
d
2
On the other hand, given the equation
Ax
=
d
, if
A
1
exists, then by multiplying
both sides on the left by A
1
, we obtain x = A
1
d.
Hence, we have constructed
A
1
in the 2
×
2 case, and shown that the
condition for its existence is det A 6= 0, with
A
1
=
1
det A
A
22
A
12
A
21
A
11
4.2 Inverse of an n × n matrix
For larger matrices, the formula for the inverse is similar, but slightly more
complicated (and costly to evaluate). The key to finding the inverse is the
following:
Lemma.
P
A
ik
jk
= δ
ij
det A.
Proof.
If
i 6
=
j
, then consider an
n × n
matrix
B
, which is identical to
A
except
the
j
th row is replaced by the
i
th row of
A
. So
jk
of
B
= ∆
jk
of
A
, since
jk
does not depend on the elements in row
j
. Since
B
has a duplicate row, we know
that
0 = det B =
n
X
k=1
B
jk
jk
=
n
X
k=1
A
ik
jk
.
If i = j, then the expression is det A by the Laplace expansion formula.
Theorem. If det A 6= 0, then A
1
exists and is given by
(A
1
)
ij
=
ji
det A
.
Proof.
(A
1
)
ik
A
kj
=
ki
det A
A
kj
=
δ
ij
det A
det A
= δ
ij
.
So A
1
A = I.
The other direction is easy to prove. If
det A
= 0, then it has no inverse,
since for any matrix B, det AB = 0, and hence AB cannot be the identity.
Example.
Consider the shear matrix
S
λ
=
1 λ 0
0 1 0
0 0 1
. We have
det S
λ
= 1.
The cofactors are
11
= 1
12
= 0
13
= 0
21
λ
22
= 1
23
= 0
31
= 0
32
= 0
33
= 1
So S
1
λ
=
1 λ 0
0 1 0
0 0 1
.
How many arithmetic operations are involved in calculating the inverse of an
n × n matrix? We just count multiplication operations since they are the most
time-consuming. Suppose that calculating
det A
takes
f
n
multiplications. This
involves
n
(
n
1)
×
(
n
1) determinants, and you need
n
more multiplications to
put them together. So f
n
= nf
n1
+ n. So f
n
= O(n!) (in fact f
n
(1 + e)n!).
To find the inverse, we need to calculate
n
2
cofactors. Each is a
n
1
determinant, and each takes
O
((
n
1)!). So the time complexity is
O
(
n
2
(
n
1)!) =
O(n · n!).
This is incredibly slow. Hence while it is theoretically possible to solve
systems of linear equations by inverting a matrix, sane people do not do so
in general. Instead, we develop certain better methods to solve the equations.
In fact, the “usual” method people use to solve equations by hand only has
complexity O(n
3
), which is a much better complexity.
4.3 Homogeneous and inhomogeneous equations
Consider
Ax
=
b
where
A
is an
n ×n
matrix,
x
and
b
are
n ×
1 column vectors.
Definition
(Homogeneous equation)
.
If
b
=
0
, then the system is homogeneous.
Otherwise, it’s inhomogeneous.
Suppose
det A 6
= 0. Then there is a unique solution
x
=
A
1
b
(
x
=
0
for
homogeneous).
How can we understand this result? Recall that
det A 6
= 0 means that the
columns of
A
are linearly independent. The columns are the images of the stan-
dard basis,
e
0
i
=
Ae
i
. So
det A 6
= 0 means that
e
0
i
are linearly independent and
form a basis of
R
n
. Therefore the image is the whole of
R
n
. This automatically
ensures that b is in the image, i.e. there is a solution.
To show that there is exactly one solution, suppose
x
and
x
0
are both solutions.
Then
Ax
=
Ax
0
=
b
. So
A
(
x x
0
) =
0
. So
x x
0
is in the kernel of
A
. But
since the rank of
A
is
n
, by the rank-nullity theorem, the nullity is 0. So the
kernel is trivial. So x x
0
= 0, i.e. x = x
0
.
4.3.1 Gaussian elimination
Consider a general solution
A
11
x
1
+ A
12
x
2
+ ··· + A
1n
x
n
= d
1
A
21
x
1
+ A
22
x
2
+ ··· + A
2n
x
n
= d
2
.
.
.
A
m1
x
1
+ A
m2
x
2
+ ··· + A
mn
x
n
= d
m
So we have m equations and n unknowns.
Assume
A
11
6
= 0 (if not, we can re-order the equations). We can use the
first equation to eliminate
x
1
from the remaining (
m
1) equations. Then use
the second equation to eliminate
x
2
from the remaining (
m
2) equations (if
anything goes wrong, just re-order until things work). Repeat.
We are left with
A
11
x
1
+ A
12
x
2
+ A
13
x
3
+ ··· + A
1n
x
n
= d
1
A
(2)
22
x
2
+ A
(2)
23
x
3
+ ··· + A
(2)
2n
x
n
= d
2
.
.
.
A
(r)
rr
x
r
+ ··· + A
(r)
rn
x
n
= d
r
0 = d
(r)
r+1
.
.
.
0 = d
(r)
m
Here
A
(i)
ii
6
= 0 (which we can achieve by re-ordering), and the superfix (
i
) refers
to the “version number” of the coefficient, e.g.
A
(2)
22
is the second version of the
coefficient of x
2
in the second row.
Let’s consider the different possibilities:
(i) r < m
and at least one of
d
(r)
r+1
, ···d
(r)
m
6
= 0. Then a contradiction is
reached. The system is inconsistent and has no solution. We say it is
overdetermined.
Example. Consider the system
3x
1
+ 2x
2
+ x
3
= 3
6x
1
+ 3x
2
+ 3x
3
= 0
6x
1
+ 2x
2
+ 4x
3
= 6
This becomes
3x
1
+ 2x
2
+ x
3
= 3
0 x
2
+ x
3
= 6
0 2x
2
+ 2x
3
= 0
And then
3x
1
+ 2x
2
+ x
3
= 3
0 x
2
+ x
3
= 6
0 = 12
We have d
(3)
3
= 12 = 0 and there is no solution.
(ii)
If
r
=
n m
, and all
d
(r)
r+i
= 0. Then from the
n
th equation, there
is a unique solution for
x
n
=
d
(n)
n
/A
(n)
nn
, and hence for all
x
i
by back
substitution. This system is determined.
Example.
2x
1
+ 5x
2
= 2
4x
1
+ 3x
2
= 11
This becomes
2x
1
+ 5x
2
= 2
7x
2
= 7
So x
2
= 1 and thus x
1
= 7/2.
(iii)
If
r < n
and
d
(r)
r+i
= 0, then
x
r+1
, ···x
n
can be freely chosen, and there
are infinitely many solutions. System is under-determined. e.g.
x
1
+ x
2
= 1
2x
1
+ 2x
2
= 2
Which gives
x
1
+ x
2
= 1
0 = 0
So x
1
= 1 x
2
is a solution for any x
2
.
In the
n
=
m
case, there are
O
(
n
3
) operations involved, which is much less than
inverting the matrix. So this is an efficient way of solving equations.
This is also be related to the determinant. Consider the case where
m
=
n
and
A
is square. Since row operations do not change the determinant and
swapping rows give a factor of (1). So
det A = (1)
k
A
11
A
12
··· ··· ··· A
1n
0 A
(2)
22
··· ··· ··· A
(n)
2n
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 ··· A
(r)
rr
··· A
(n)
rn
0 0 ··· 0 0 ···
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
This determinant is an upper triangular one (all elements below diagonal are 0)
and the determinant is the product of its diagonal elements.
Hence if
r < n
(and
d
(r)
i
= 0 for
i > r
), then we have case (ii) and the
det A = 0. If r = n, then det A = (1)
k
A
11
A
(2)
22
···A
(n)
nn
6= 0.
4.4 Matrix rank
Consider a linear map
α
:
R
n
R
m
. Recall the rank
r
(
α
) is the dimension of
the image. Suppose that the matrix
A
is associated with the linear map. We
also call r(A) the rank of A.
Recall that if the standard basis is
e
1
, ···e
n
, then
Ae
1
, ··· , Ae
n
span the
image (but not necessarily linearly independent).
Further,
Ae
1
, ··· , Ae
n
are the columns of the matrix
A
. Hence
r
(
A
) is the
number of linearly independent columns.
Definition
(Column and row rank of linear map)
.
The column rank of a matrix
is the maximum number of linearly independent columns.
The row rank of a matrix is the maximum number of linearly independent
rows.
Theorem. The column rank and row rank are equal for any m × n matrix.
Proof.
Let
r
be the row rank of
A
. Write the biggest set of linearly independent
rows as
v
T
1
, v
T
2
, ···v
T
r
or in component form
v
T
k
= (
v
k1
, v
k2
, ··· , v
kn
) for
k
=
1, 2, ··· , r.
Now denote the ith row of A as r
T
i
= (A
i1
, A
i2
, ···A
in
).
Note that every row of
A
can be written as a linear combination of the
v
’s.
(If
r
i
cannot be written as a linear combination of the
v
’s, then it is independent
of the
v
’s and
v
is not the maximum collection of linearly independent rows)
Write
r
T
i
=
r
X
k=1
C
ik
v
T
k
.
For some coefficients C
ik
with 1 i m and 1 k r.
Now the elements of A are
A
ij
= (r
i
)
T
j
=
r
X
k=1
C
ik
(v
k
)
j
,
or
A
1j
A
2j
.
.
.
A
mj
=
r
X
k=1
v
k
j
C
1k
C
2k
.
.
.
C
mk
So every column of
A
can be written as a linear combination of the
r
column
vectors c
k
. Then the column rank of A r, the row rank of A.
Apply the same argument to
A
T
to see that the row rank is
the column
rank.
4.5 Homogeneous problem Ax = 0
We restrict our attention to the square case, i.e. number of unknowns = number
of equations. Here A is an n × n matrix. We want to solve Ax = 0.
First of all, if
det A 6
= 0, then
A
1
exists and
x
1
=
A
1
0
=
0
, which is the
unique solution. Hence if Ax = 0 with x 6= 0, then det A = 0.
4.5.1 Geometrical interpretation
We consider a 3 × 3 matrix
A =
r
T
1
r
T
2
r
T
3
Ax
=
0
means that
r
i
· x
= 0 for all
i
. Each equation
r
i
· x
= 0 represents a
plane through the origin. So the solution is the intersection of the three planes.
There are three possibilities:
(i)
If
det A
= [
r
1
, r
2
, r
3
]
6
= 0, span
{r
1
, r
2
, r
3
}
=
R
3
and thus
r
(
A
) = 3. By
the rank-nullity theorem,
n
(
A
) = 0 and the kernel is
{0}
. So
x
=
0
is the
unique solution.
(ii) If det A = 0, then dim(span{r
1
, r
2
, r
3
}) = 1 or 2.
(a)
If rank = 2, wlog assume
r
1
, r
2
are linearly independent. So
x
lies
on the intersection of two planes
x · r
1
= 0 and
x · r
2
= 0, which is
the line
{x R
3
:
x
=
λr
1
× r
2
}
(Since
x
lies on the intersection of
the two planes, it has to be normal to the normals of both planes).
All such points on this line also satisfy
x · r
3
= 0 since
r
3
is a linear
combination of r
1
and r
2
. The kernel is a line, n(A) = 1.
(b)
If rank = 1, then
r
1
, r
2
, r
3
are parallel. So
x·r
1
= 0
x·r
2
=
x·r
3
= 0.
So all
x
that satisfy
x ·r
1
= 0 are in the kernel, and the kernel now is
a plane. n(A) = 2.
(We also have the trivial case where
r
(
A
) = 0, we have the zero mapping and
the kernel is R
3
)
4.5.2 Linear mapping view of Ax = 0
In the general case, consider a linear map
α
:
R
n
R
n
x 7→ x
0
=
Ax
. The
kernel k(A) = {x R
n
: Ax = 0} has dimension n(A).
(i)
If
n
(
A
) = 0, then
A
(
e
1
)
, A
(
e
2
)
, ··· , A
(
e
n
) is a linearly independent set,
and r(A) = n.
(ii)
If
n
(
A
)
>
0, then the image is not the whole of
R
n
. Let
{u
i
}, i
=
1
, ··· , n
(
A
) be a basis of the kernel, i.e. so given any solution to
Ax
=
0
,
x
=
n(A)
X
i=1
λ
i
u
i
for some
λ
i
. Extend
{u
i
}
to be a basis of
R
n
by introducing
extra vectors
u
i
for
i
=
n
(
A
) + 1
, ··· , n
. The vectors
A
(
u
i
) for
i
=
n(A) + 1, ··· , n form a basis of the image.
4.6 General solution of Ax = d
Finally consider the general equation
Ax
=
d
, where
A
is an
n × n
matrix and
x, d are n × 1 column vectors. We can separate into two main cases.
(i) det
(
A
)
6
= 0. So
A
1
exists and
n
(
A
) = 0,
r
(
A
) =
n
. Then for any
d R
n
,
a unique solution must exists and it is x = A
1
d.
(ii) det
(
A
) = 0. Then
A
1
does not exist, and
n
(
A
)
>
0,
r
(
A
)
< n
. So the
image of A is not the whole of R
n
.
(a) If d 6∈ im A, then there is no solution (by definition of the image)
(b)
If
d im A
, then by definition there exists at least one
x
such that
Ax
=
d
. The general solution of
Ax
=
d
can be written as
x
=
x
0
+
y
,
where
x
0
is a particular solution (i.e.
Ax
0
=
d
), and
y
is any vector
in ker A (i.e. Ay = 0). (cf. Isomorphism theorem)
If
n
(
A
) = 0, then
y = 0
only, and then the solution is unique (i.e.
case (i)). If
n
(
A
)
>
0 , then
{u
i
}, i
= 1
, ··· , n
(
A
) is a basis of the
kernel. Hence
y =
n(A)
X
j=1
µ
j
u
j
,
so
x = x
0
+
n(A)
X
j=1
µ
j
u
j
for any µ
j
, i.e. there are infinitely many solutions.
Example.
1 1
a 1
x
1
x
2
=
1
b
We have det A = 1 a. If a 6= 1, then A
1
exists and
A
1
=
1
1 a
=
1
1 a
1 1
a 1
.
Then
x =
1
1 a
1 b
a + b
.
If a = 1, then
Ax =
x
1
+ x
2
x
1
+ x
2
= (x
1
+ x
2
)
1
1
.
So
im A
=
span

1
1

and
ker A
=
span

1
1

. If
b 6
= 1, then
1
b
6∈ im A
and there is no solution. If b = 1, then
1
b
im A.
We find a particular solution of
1
0
. So The general solution is
x =
1
0
+ λ
1
1
.
Example. Find the general solution of
a a b
b a a
a b a
x
y
z
=
1
c
1
We have
det A
= (
a b
)
2
(2
a
+
b
). If
a 6
=
b
and
b 6
=
2
a
, then the inverse exists
and there is a unique solution for any c. Otherwise, the possible cases are
(i) a
=
b, b 6
=
2
a
. So
a 6
= 0. The kernel is the plane
x
+
y
+
z
= 0 which is
span
1
1
0
,
1
0
1
We extend this basis to R
3
by adding
1
0
0
.
So the image is the span of
a
a
a
=
1
1
1
. Hence if
c 6
= 1, then
1
c
1
is not
in the image and there is no solution. If
c
= 1, then a particular solution
is
1
a
0
0
and the general solution is
x =
1
a
0
0
+ λ
1
1
0
+ µ
1
0
1
(ii) If a 6= b and b = 2a, then a 6= 0. The kernel satisfies
x + y 2z = 0
2x + y + z = 0
x 2y + z = 0
This can be solved to give
x
=
y
=
z
, and the kernel is
span
1
1
1
. We
add
1
0
0
and
0
0
1
to form a basis of
R
3
. So the image is the span of
1
2
1
,
2
1
1
.
If
1
c
1
is in the image, then
1
c
1
= λ
1
2
1
+ µ
2
1
1
.
Then the only solution is
µ
= 0
, λ
= 1
, c
=
2. Thus there is no solution if
c 6
=
2, and when
c
=
2, pick a particular solution
1
a
0
0
and the general
solution is
x =
1
a
0
0
+ λ
1
1
1
(iii)
If
a
=
b
and
b
=
2
a
, then
a
=
b
= 0 and
ker A
=
R
3
. So there is no
solution for any c.
5 Eigenvalues and eigenvectors
Given a matrix A, an eigenvector is a vector x that satisfies Ax = λx for some
λ
. We call
λ
the associated eigenvalue. In some sense, these vectors are not
modified by the matrix, and are just scaled up by the matrix. We will look
at the properties of eigenvectors and eigenvalues, and see their importance in
diagonalizing matrices.
5.1 Preliminaries and definitions
Theorem
(Fundamental theorem of algebra)
.
Let
p
(
z
) be a polynomial of degree
m 1, i.e.
p(z) =
m
X
j=0
c
j
z
j
,
where c
j
C and c
m
6= 0.
Then
p
(
z
) = 0 has precisely
m
(not necessarily distinct) roots in the complex
plane, accounting for multiplicity.
Note that we have the disclaimer “accounting for multiplicity”. For example,
x
2
2
x
+ 1 = 0 has only one distinct root, 1, but we say that this root has
multiplicity 2, and is thus counted twice. Formally, multiplicity is defined as
follows:
Definition
(Multiplicity of root)
.
The root
z
=
ω
has multiplicity
k
if (
z ω
)
k
is a factor of p(z) but (z ω)
k+1
is not.
Example.
Let
p
(
z
) =
z
3
z
2
z
+ 1 = (
z
1)
2
(
z
+ 1). So
p
(
z
) = 0 has roots
1, 1, 1, where z = 1 has multiplicity 2.
Definition
(Eigenvector and eigenvalue)
.
Let
α
:
C
n
C
n
be a linear map
with associated matrix A. Then x 6= 0 is an eigenvector of A if
Ax = λx
for some
λ
.
λ
is the associated eigenvalue. This means that the direction of the
eigenvector is preserved by the mapping, but is scaled up by λ.
There is a rather easy way of finding eigenvalues:
Theorem. λ is an eigenvalue of A iff
det(A λI) = 0.
Proof.
(
) Suppose that
λ
is an eigenvalue and
x
is the associated eigenvector.
We can rearrange the equation in the definition above to
(A λI)x = 0
and thus
x ker(A λI)
But
x 6
=
0
. So
ker
(
AλI
) is non-trivial and
det
(
AλI
) = 0. The (
) direction
is similar.
Definition
(Characteristic equation of matrix)
.
The characteristic equation of
A is
det(A λI) = 0.
Definition
(Characteristic polynomial of matrix)
.
The characteristic polynomial
of A is
p
A
(λ) = det(A λI).
From the definition of the determinant,
p
A
(λ) = det(A λI)
= ε
j
1
j
2
···j
n
(A
j
1
1
λδ
j
1
1
) ···(A
j
n
n
λδ
j
n
n
)
= c
0
+ c
1
λ + ··· + c
n
λ
n
for some constants c
0
, ··· , c
n
. From this, we see that
(i) p
A
(
λ
) has degree
n
and has
n
roots. So an
n ×n
matrix has
n
eigenvalues
(accounting for multiplicity).
(ii)
If
A
is real, then all
c
i
R
. So eigenvalues are either real or come in
complex conjugate pairs.
(iii) c
n
= (
1)
n
and
c
n1
= (
1)
n1
(
A
11
+
A
22
+
···
+
A
nn
) = (
1)
n1
tr
(
A
).
But c
n1
is the sum of roots, i.e. c
n1
= (1)
n1
(λ
1
+ λ
2
+ ···λ
n
), so
tr(A) = λ
1
+ λ
2
+ ··· + λ
n
.
Finally,
c
0
=
p
A
(0) =
det
(
A
). Also
c
0
is the product of all roots, i.e.
c
0
= λ
1
λ
2
···λ
n
. So
det A = λ
1
λ
2
···λ
n
.
The kernel of the matrix
A λI
is the set
{x
:
Ax
=
λx}
. This is a vector
subspace because the kernel of any map is always a subspace.
Definition
(Eigenspace)
.
The eigenspace denoted by
E
λ
is the kernel of the
matrix A λI, i.e. the set of eigenvectors with eigenvalue λ.
Definition
(Algebraic multiplicity of eigenvalue)
.
The algebraic multiplicity
M
(
λ
) or
M
λ
of an eigenvalue
λ
is the multiplicity of
λ
in
p
A
(
λ
) = 0. By the
fundamental theorem of algebra,
X
λ
M(λ) = n.
If M (λ) > 1, then the eigenvalue is degenerate.
Definition
(Geometric multiplicity of eigenvalue)
.
The geometric multiplicity
m
(
λ
) or
m
λ
of an eigenvalue
λ
is the dimension of the eigenspace, i.e. the
maximum number of linearly independent eigenvectors with eigenvalue λ.
Definition (Defect of eigenvalue). The defect
λ
of eigenvalue λ is
λ
= M(λ) m(λ).
It can be proven that
λ
0, i.e. the geometric multiplicity is never greater
than the algebraic multiplicity.
5.2 Linearly independent eigenvectors
Theorem.
Suppose
n×n
matrix
A
has distinct eigenvalues
λ
1
, λ
2
, ··· , λ
n
. Then
the corresponding eigenvectors x
1
, x
2
, ··· , x
n
are linearly independent.
Proof.
Proof by contradiction: Suppose
x
1
, x
2
, ··· , x
n
are linearly dependent.
Then we can find non-zero constants d
i
for i = 1, 2, ··· , r, such that
d
1
x
1
+ d
2
x
2
+ ··· + d
r
x
r
= 0.
Suppose that this is the shortest non-trivial linear combination that gives
0
(we
may need to re-order x
i
).
Now apply (A λ
1
I) to the whole equation to obtain
d
1
(λ
1
λ
1
)x
1
+ d
2
(λ
2
λ
1
)x
2
+ ··· + d
r
(λ
r
λ
1
)x
r
= 0.
We know that the first term is
0
, while the others are not (since we assumed
λ
i
6= λ
j
for i 6= j). So
d
2
(λ
2
λ
1
)x
2
+ ··· + d
r
(λ
r
λ
1
)x
r
= 0,
and we have found a shorter linear combination that gives
0
. Contradiction.
Example.
(i) A =
0 1
1 0
. Then p
A
(λ) = λ
2
+ 1 = 0. So λ
1
= i and λ
2
= i.
To solve (A λ
1
I)x = 0, we obtain
i 1
1 i
x
1
x
2
= 0.
So we obtain
x
1
x
2
=
1
i
to be an eigenvector. Clearly any scalar multiple of
1
i
is also a solution,
but still in the same eigenspace E
i
= span
1
i
Solving (A λ
2
I)x = 0 gives
x
1
x
2
=
1
i
.
So E
i
= span
1
i
.
Note that
M
(
±i
) =
m
(
±i
) = 1, so
±i
= 0. Also note that the two
eigenvectors are linearly independent and form a basis of C
2
.
(ii) Consider
A =
2 2 3
2 1 6
1 2 0
Then
det
(
A λI
) = 0 gives 45 + 21
λ λ
2
λ
3
. So
λ
1
= 5
, λ
2
=
λ
3
=
3.
The eigenvector with eigenvalue 5 is
x =
1
2
1
We can find that the eigenvectors with eigenvalue 3 are
x =
2x
2
+ 3x
3
x
2
x
3
for any
x
2
, x
3
. This gives two linearly independent eigenvectors, say
2
1
0
,
3
0
1
.
So
M
(5) =
m
(5) = 1 and
M
(
3) =
m
(
3) = 2, and there is no defect for
both of them. Note that these three eigenvectors form a basis of C
3
.
(iii) Let
A =
3 1 1
1 3 1
2 2 0
Then 0 =
p
A
(
λ
) =
(
λ
+ 2)
4
. So
λ
=
2
,
2
,
2. To find the eigenvectors,
we have
(A + 2I)x =
1 1 1
1 1 1
2 2 2
x
1
x
2
x
3
= 0
The general solution is thus
x
1
+
x
2
x
3
= 0, and the general solution is
thus x =
x
1
x
2
x
1
+ x
2
. The eigenspace E
2
= span
1
0
1
,
0
1
1
.
Hence
M
(
2) = 3 and
m
(
2) = 2. Thus the defect
2
= 1. So the
eigenvectors do not form a basis of C
3
.
(iv)
Consider the reflection
R
in the plane with normal
n
. Clearly
Rn
=
n
.
The eigenvalue is
1 and the eigenvector is
n
. Then
E
1
=
span{n}
. So
M(1) = m(1) = 1.
If
p
is any vector in the plane,
Rp
=
p
. So this has an eigenvalue of 1 and
eigenvectors being any vector in the plane. So M (1) = m(1) = 2.
So the eigenvectors form a basis of R
3
.
(v)
Consider a rotation
R
by
θ
about
n
. Since
Rn
=
n
, we have an eigenvalue
of 1 and eigenspace E
1
= span{n}.
We know that there are no other real eigenvalues since rotation changes
the direction of any other vector. The other eigenvalues turn out to be
e
±
. If
θ 6
= 0, there are 3 distinct eigenvalues and the eigenvectors form a
basis of C
3
.
(vi) Consider a shear
A =
1 µ
0 1
The characteristic equation is (1
λ
)
2
= 0 and
λ
= 1. The eigenvectors
corresponding to
λ
= 1 is
x
=
1
0
. We have
M
(1) = 2 and
m
(1) = 1. So
1
= 1.
If
n × n
matrix
A
has
n
distinct eigenvalues, and hence has
n
linearly
independent eigenvectors
v
1
, v
2
, ···v
n
, then with respect to this eigenvector
basis, A is diagonal.
In this basis,
v
1
= (1
,
0
, ··· ,
0) etc. We know that
Av
i
=
λ
i
v
i
(no summation).
So the image of the
i
th basis vector is
λ
i
times the
i
th basis. Since the columns
of A are simply the images of the basis,
λ
1
0 ··· 0
0 λ
2
··· 0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 ··· λ
n
The fact that
A
can be diagonalized by changing the basis is an important
observation. We will now look at how we can change bases and see how we can
make use of this.
5.3 Transformation matrices
How do the components of a vector or a matrix change when we change the
basis?
Let
{e
1
, e
2
, ··· , e
n
}
and
{
˜
e
1
,
˜
e
2
, ··· ,
˜
e
n
}
be 2 different bases of
R
n
or
C
n
.
Then we can write
˜
e
j
=
n
X
i=1
P
ij
e
i
i.e.
P
ij
is the
i
th component of
˜
e
j
with respect to the basis
{e
1
, e
2
, ··· , e
n
}
.
Note that the sum is made as
P
ij
e
i
, not
P
ij
e
j
. This is different from the formula
for matrix multiplication.
Matrix
P
has as its columns the vectors
˜
e
j
relative to
{e
1
, e
2
, ··· , e
n
}
. So
P = (
˜
e
1
˜
e
2
···
˜
e
n
) and
P (e
i
) =
˜
e
i
Similarly, we can write
e
i
=
n
X
k=1
Q
ki
˜
e
k
with Q = (e
1
e
2
··· e
n
).
Substituting this into the equation for
˜
e
j
, we have
˜
e
j
=
n
X
i=1
n
X
k=1
Q
ki
˜
e
k
!
P
ij
=
n
X
k=1
˜
e
k
n
X
i=1
Q
ki
P
ij
!
But
˜
e
1
,
˜
e
2
, ··· ,
˜
e
n
are linearly independent, so this is only possible if
n
X
i=1
Q
ki
P
ij
= δ
kj
,
which is just a fancy way of saying QP = I, or Q = P
1
.
5.3.1 Transformation law for vectors
With respect to basis
{e
i
}
,
u
=
P
n
i=1
u
i
e
i
. With respect to basis
{
˜
e
i
}
,
u
=
P
n
i=1
˜u
i
˜
e
i
. Note that this is the same vector
u
but has different components
with respect to different bases. Using the transformation matrix above for the
basis, we have
u =
n
X
j=1
˜u
j
n
X
i=1
P
ij
e
i
=
n
X
i=1
n
X
j=1
P
ij
˜u
j
e
i
By comparison, we know that
u
i
=
n
X
j=1
P
ij
˜u
j
Theorem.
Denote vector as
u
with respect to
{e
i
}
and
˜
u
with respect to
{
˜
e
i
}
.
Then
u = P ˜u and ˜u = P
1
u
Example.
Take the first basis as
{e
1
= (1
,
0)
, e
2
= (0
,
1)
}
and the second as
{
˜
e
1
= (1, 1),
˜
e
2
= (1, 1)}.
So
˜
e
1
= e
1
+ e
2
and
˜
e
2
= e
1
+ e
2
. We have
P =
1 1
1 1
.
Then for an arbitrary vector u, we have
u = u
1
e
1
+ u
2
e
2
= u
1
1
2
(
˜
e
1
˜
e
2
) + u
2
1
2
(
˜
e
1
+
˜
e
2
)
=
1
2
(u
1
+ u
2
)
˜
e
1
+
1
2
(u
1
+ u
2
)
˜
e
2
.
Alternatively, using the formula above, we obtain
˜u = P
1
u
=
1
2
1 1
1 1
u
1
u
2
=
1
2
(u
1
+ u
2
)
1
2
(u
1
+ u
2
)
Which agrees with the above direct expansion.
5.3.2 Transformation law for mat rix
Consider a linear map α : C
n
C
n
with associated n × n matrix A. We have
u
0
= α(u) = Au.
Denote
u
and
u
0
as being with respect to basis
{e
i
}
(i.e. same basis in both
spaces), and ˜u, ˜u
0
with respect to {
˜
e
i
}.
Using what we’ve got above, we have
u
0
= Au
P ˜u
0
= AP
˜
u
˜u
0
= P
1
AP ˜u
=
˜
A
˜
u
So
Theorem.
˜
A = P
1
AP.
Example.
Consider the shear
S
λ
=
1 λ 0
0 1 0
0 0 1
with respect to the standard
basis. Choose a new set of basis vectors by rotating by θ about the e
3
axis:
˜
e
1
= cos θe
1
+ sin θe
2
˜
e
2
= sin θe
1
+ cos θe
2
˜
e
3
= e
3
So we have
P =
cos θ sin θ 0
sin θ cos θ 0
0 0 1
, P
1
=
cos θ sin θ 0
sin θ cos θ 0
0 0 1
Now use the basis transformation laws to obtain
˜
S
λ
=
1 + λ sin θ cos θ λ cos
2
θ 0
λ sin
2
θ 1 λ sin θ cos θ 0
0 0 1
Clearly this is much more complicated than our original basis. This shows that
choosing a sensible basis is important.
More generally, given
α
:
C
m
C
n
, given
x C
m
,
x
0
C
n
with
x
0
=
Ax
.
We know that A is an n × m matrix.
Suppose
C
m
has a basis
{e
i
}
and
C
n
has a basis
{f
i
}
. Now change bases to
{
˜
e
i
} and {
˜
f
i
}.
We know that
x
=
P ˜x
with
P
being an
m × m
matrix, with
x
0
=
R
˜
x
0
with
R being an n × n matrix.
Combining both of these, we have
R
˜
x
0
= AP
˜
x
˜
x
0
= R
1
AP ˜x
Therefore
˜
A = R
1
AP .
Example.
Consider
α
:
R
3
R
2
, with respect to the standard bases in both
spaces,
A =
2 3 4
1 6 3
Use a new basis
2
1
,
1
5
in
R
2
and keep the standard basis in
R
3
. The basis
change matrix in R
3
is simply I, while
R =
2 1
1 5
, R
1
=
1
9
5 1
1 2
is the transformation matrix for R
2
. So
˜
A =
2 1
1 5
2 3 4
1 6 3
I
=
1
9
5 1
1 2
2 3 4
1 6 3
=
1 1 17/9
0 1 2/9
We can alternatively do it this way: we know that
˜
f
1
=
2
1
,
˜
f
2
=
1
5
Then
we know that
˜
e
1
= e
1
7→ 2f
1
+ f
2
= f
1
˜
e
2
= e
2
7→ 3f
1
+ 6f
2
=
˜
f
1
+
˜
f
2
˜
e
3
= e
3
7→ 4f
1
+ 3f
2
=
17
9
˜
f
1
+
2
9
˜
f
2
and we can construct the matrix correspondingly.
5.4 Similar matrices
Definition
(Similar matrices)
.
Two
n ×n
matrices
A
and
B
are similar if there
exists an invertible matrix P such that
B = P
1
AP,
i.e. they represent the same map under different bases. Alternatively, using the
language from IA Groups, we say that they are in the same conjugacy class.
Proposition. Similar matrices have the following properties:
(i) Similar matrices have the same determinant.
(ii) Similar matrices have the same trace.
(iii) Similar matrices have the same characteristic polynomial.
Note that (iii) implies (i) and (ii) since the determinant and trace are the
coefficients of the characteristic polynomial
Proof. They are proven as follows:
(i) det B = det(P
1
AP ) = (det A)(det P )
1
(det P ) = det A
(ii)
tr B = B
ii
= P
1
ij
A
jk
P
ki
= A
jk
P
ki
P
1
ij
= A
jk
(P P
1
)
kj
= A
jk
δ
kj
= A
jj
= tr A
(iii)
p
B
(λ) = det(B λI)
= det(P
1
AP λI)
= det(P
1
AP λP
1
IP )
= det(P
1
(A λI)P )
= det(A λI)
= p
A
(λ)
5.5 Diagonalizable matrices
Definition
(Diagonalizable matrices)
.
An
n × n
matrix
A
is diagonalizable if
it is similar to a diagonal matrix. We showed above that this is equivalent to
saying the eigenvectors form a basis of C
n
.
The requirement that matrix
A
has
n
distinct eigenvalues is a sufficient
condition for diagonalizability as shown above. However, it is not necessary.
Consider the second example in Section 5.2,
A =
2 2 3
2 1 6
1 2 0
We found three linear eigenvectors
˜
e
1
=
1
2
1
,
˜
e
2
=
2
1
0
,
˜
e
3
=
3
0
1
If we let
P =
1 2 3
2 1 0
1 0 1
, P
1
=
1
8
1 2 3
2 4 6
1 2 5
,
then
˜
A = P
1
AP =
5 0 0
0 3 0
0 0 3
,
so A is diagonalizable.
Theorem.
Let
λ
1
, λ
2
, ··· , λ
r
, with
r n
be the distinct eigenvalues of
A
. Let
B
1
, B
2
, ···B
r
be the bases of the eigenspaces
E
λ
1
, E
λ
2
, ··· , E
λ
r
correspondingly.
Then the set B =
r
[
i=1
B
i
is linearly independent.
This is similar to the proof we had for the case where the eigenvalues are
distinct. However, we are going to do it much concisely, and the actual meat of
the proof is actually just a single line.
Proof.
Write
B
1
=
{x
(1)
1
, x
(1)
2
, ···x
(1)
m(λ
1
)
}
. Then
m
(
λ
1
) =
dim
(
E
λ
1
), and simi-
larly for all B
i
.
Consider the following general linear combination of all elements in
B
. Con-
sider the equation
r
X
i=1
m(λ
i
)
X
j=1
α
ij
x
(i)
j
= 0.
The first sum is summing over all eigenspaces, and the second sum sums over
the basis vectors in B
i
. Now apply the matrix
Y
k=1,2,··· ,
¯
K,··· ,r
(A λ
k
I)
to the above sum, for some arbitrary K. We obtain
m(λ
K
)
X
j=1
α
Kj
Y
k=1,2,··· ,
¯
K,··· ,r
(λ
K
λ
k
)
x
(K)
j
= 0.
Since the
x
(K)
j
are linearly independent (
B
K
is a basis),
α
Kj
= 0 for all
j
. Since
K was arbitrary, all α
ij
must be zero. So B is linearly independent.
Proposition. A is diagonalizable iff all its eigenvalues have zero defect.
5.6 Canonical (Jordan normal) form
Given a matrix
A
, if its eigenvalues all have non-zero defect, then we can find
a basis in which it is diagonal. However, if some eigenvalue does have defect,
we can still put it into an almost-diagonal form. This is known as the Jordan
normal form.
Theorem. Any 2 × 2 complex matrix A is similar to exactly one of
λ
1
0
0 λ
2
,
λ 0
0 λ
,
λ 1
0 λ
Proof. For each case:
(i)
If
A
has two distinct eigenvalues, then eigenvectors are linearly independent.
Then we can use P formed from eigenvectors as its columns
(ii)
If
λ
1
=
λ
2
=
λ
and
dim E
λ
= 2, then write
E
λ
=
span{u, v}
, with
u, v
linearly independent. Now use
{u, v}
as a new basis of
C
2
and
˜
A = P
1
AP =
λ 0
0 λ
= λI
Note that since
P
1
AP
=
λI
, we have
A
=
P
(
λI
)
P
1
=
λI
. So
A
is
isotropic, i.e. the same with respect to any basis.
(iii)
If
λ
1
=
λ
2
=
λ
and
dim
(
E
λ
) = 1, then
E
λ
=
span{v}
. Now choose basis
of C
2
as {v, w}, where w C
2
\ E
λ
.
We know that
Aw C
2
. So
Aw
=
αv
+
βw
. Hence, if we change basis to
{v, w}, then
˜
A = P
1
AP =
λ α
0 β
.
However,
A
and
˜
A
both have eigenvalue
λ
with algebraic multiplicity 2.
So we must have
β
=
λ
. To make
α
= 1, let
u
= (
˜
A λI
)
w
. We know
u 6= 0 since w is not in the eigenspace. Then
(
˜
A λI)u = (
˜
A λI)
2
w =
0 α
0 0
0 α
0 0
w = 0.
So u is an eigenvector of
˜
A with eigenvalue λ.
We have u =
˜
Aw λw. So
˜
Aw = u + λw.
Change basis to {u, w}. Then A with respect to this basis is
λ 1
0 λ
.
This is a two-stage process:
P
sends basis to
{v, w}
and then matrix
Q
sends to basis
{u, w}
. So the similarity transformation is
Q
1
(
P
1
AP
)
Q
=
(P Q)
1
A(P Q).
Proposition.
(Without proof) The canonical form, or Jordan normal form,
exists for any
n × n
matrix
A
. Specifically, there exists a similarity transform
such that A is similar to a matrix to
˜
A that satisfies the following properties:
(i)
˜
A
αα
= λ
α
, i.e. the diagonal composes of the eigenvalues.
(ii)
˜
A
α,α+1
= 0 or 1.
(iii)
˜
A
ij
= 0 otherwise.
The actual theorem is actually stronger than this, and the Jordan normal
form satisfies some additional properties in addition to the above. However, we
shall not go into details, and this is left for the IB Linear Algebra course.
Example. Let
A =
3 1 1
1 3 1
2 2 0
The eigenvalues are
2
,
2
,
2 and the eigenvectors are
1
1
0
,
1
0
1
. Pick
w
=
1
0
0
. Write
u
= (
A λI
)
w
=
1 1 1
1 1 1
2 2 2
1
0
0
=
1
1
2
. Note that
Au
=
2
u
. We also have
Aw
=
u
2
w
. Form a basis
{u, w, v}
, where
v
is
another eigenvector linearly independent from u, say
1
0
1
.
Now change to this basis with
P
=
1 1 1
1 0 0
2 0 1
. Then the Jordan normal
form is P
1
AP =
2 1 0
0 2 0
0 0 2
5.7 Cayley-Hamilton Theorem
Theorem
(Cayley-Hamilton theorem)
.
Every
n × n
complex matrix satisfies
its own characteristic equation.
Proof.
We will only prove for diagonalizable matrices here. So suppose for our
matrix
A
, there is some
P
such that
D
=
diag
(
λ
1
, λ
2
, ··· , λ
n
) =
P
1
AP
. Note
that
D
i
= (P
1
AP )(P
1
AP ) ···(P
1
AP ) = P
1
A
i
P.
Hence
p
D
(D) = p
D
(P
1
AP ) = P
1
[p
D
(A)]P.
Since similar matrices have the same characteristic polynomial. So
p
A
(D) = P
1
[p
A
(A)]P.
However, we also know that D
i
= diag(λ
i
1
, λ
i
2
, ···λ
i
n
). So
p
A
(D) = diag(p
A
(λ
1
), p
A
(λ
2
), ··· , p
A
(λ
n
)) = diag(0, 0, ··· , 0)
since the eigenvalues are roots of
p
A
(
λ
) = 0. So 0 =
p
A
(
D
) =
P
1
p
A
(
A
)
P
and
thus p
A
(A) = 0.
There are a few things to note.
(i)
If
A
1
exists, then
A
1
p
A
(
A
) =
A
1
(
c
0
+
c
1
A
+
c
2
A
2
+
···
+
c
n
A
n
) = 0.
So
c
0
A
1
+
c
1
+
c
2
A
+
···
+
c
n
A
n1
. Since
A
1
exists,
c
0
=
±det A 6
= 0.
So
A
1
=
1
c
0
(c
1
+ c
2
A + ··· + c
n
A
n1
).
So we can calculate A
1
from positive powers of A.
(ii) We can define matrix exponentiation by
e
A
= I + A +
1
2!
A
2
+ ··· +
1
n!
A
n
+ ··· .
It is a fact that this always converges.
If
A
is diagonalizable with
P
with
D
=
P
1
AP
=
diag
(
λ
1
, λ
2
, ··· , λ
n
),
then
P
1
e
A
P = P
1
IP + P
1
AP +
1
2!
P
1
A
2
P + ···
= I + D +
1
2!
D
2
+ ···
= diag(e
λ
1
, e
λ
2
, ···e
λ
n
)
So
e
A
= P [diag(e
λ
1
, e
λ
2
, ··· , e
λ
n
)]P
1
.
(iii)
For 2
×
2 matrices which are similar to
B
=
λ 1
0 λ
We see that the
characteristic polynomial
p
B
(
z
) =
det
(
B zI
) = (
λ z
)
2
. Then
p
B
(
B
) =
(λI B)
2
=
0 1
0 0
2
=
0 0
0 0
.
Since we have proved for the diagonalizable matrices above, we now know
that any 2 × 2 matrix satisfies Cayley-Hamilton theorem.
In IB Linear Algebra, we will prove the Cayley Hamilton theorem properly for
all matrices without assuming diagonalizability.
5.8 Eigenvalues and eigenvectors of a Hermitian matrix
5.8.1 Eigenvalues and eigenvectors
Theorem. The eigenvalues of a Hermitian matrix H are real.
Proof. Suppose that H has eigenvalue λ with eigenvector v 6= 0. Then
Hv = λv.
We pre-multiply by v
, a 1 × n row vector, to obtain
v
Hv = λv
v ()
We take the Hermitian conjugate of both sides. The left hand side is
(v
Hv)
= v
H
v = v
Hv
since H is Hermitian. The right hand side is
(λv
v)
= λ
v
v
So we have
v
Hv = λ
v
v.
From (
), we know that
λv
v
=
λ
v
v
. Since
v 6
= 0, we know that
v
v
=
v · v 6= 0. So λ = λ
and λ is real.
Theorem.
The eigenvectors of a Hermitian matrix
H
corresponding to distinct
eigenvalues are orthogonal.
Proof. Let
Hv
i
= λ
i
v
i
(i)
Hv
j
= λ
j
v
j
. (ii)
Pre-multiply (i) by v
j
to obtain
v
j
Hv
i
= λ
i
v
j
v
i
. (iii)
Pre-multiply (ii) by v
i
and take the Hermitian conjugate to obtain
v
j
Hv
i
= λ
j
v
j
v
i
. (iv)
Equating (iii) and (iv) yields
λ
i
v
j
v
i
= λ
j
v
j
v
i
.
Since
λ
i
6
=
λ
j
, we must have
v
j
v
i
= 0. So their inner product is zero and are
orthogonal.
So we know that if a Hermitian matrix has
n
distinct eigenvalues, then
the eigenvectors form an orthonormal basis. However, if there are degenerate
eigenvalues, it is more difficult, and requires the Gram-Schmidt process.
5.8.2 Gram-Schmidt orthogonalization (non-examinable)
Suppose we have a set
B
=
{w
1
, w
2
, ··· , w
r
}
of linearly independent vectors.
We want to find an orthogonal set
˜
B = {v
1
, v
2
, ··· , v
r
}.
Define the projection of
w
onto
v
by
P
v
(
w
) =
hv|wi
hv|vi
v
. Now construct
˜
B
iteratively:
(i) v
1
= w
1
(ii) v
2
= w
2
P
v
1
(w)
Then we get that hv
1
| v
2
i = hv
1
| w
2
i
hv
1
|w
2
i
hv
1
|v
1
i
hv
1
| v
1
i = 0
(iii) v
3
= w
3
P
v
1
(w
3
) P
v
2
(w
3
)
(iv)
.
.
.
(v) v
r
= w
r
r1
X
j=1
P
v
j
(w
r
)
At each step, we subtract out the components of
v
i
that belong to the space
of
{v
1
, ··· , v
k1
}
. This ensures that all the vectors are orthogonal. Finally, we
normalize each basis vector individually to obtain an orthonormal basis.
5.8.3 Unitary transformation
Suppose
U
is the transformation between one orthonormal basis and a new
orthonormal basis {u
1
, u
2
, ··· , u
n
}, i.e. hu
i
| u
j
i = δ
ij
. Then
U =
(u
1
)
1
(u
2
)
1
··· (u
n
)
1
(u
1
)
2
(u
2
)
2
··· (u
n
)
2
.
.
.
.
.
.
.
.
.
.
.
.
(u
1
)
n
(u
2
)
n
··· (u
n
)
n
Then
(U
U)
ij
= (U
)
ik
U
kj
= U
ki
U
kj
= (u
i
)
k
(u
j
)
k
= hu
i
| u
j
i
= δ
ij
So U is a unitary matrix.
5.8.4 Diagonalization of n × n Hermitian matrices
Theorem.
An
n × n
Hermitian matrix has precisely
n
orthogonal eigenvectors.
Proof.
(Non-examinable) Let
λ
1
, λ
2
, ··· , λ
r
be the distinct eigenvalues of
H
(
r
n
), with a set of corresponding orthonormal eigenvectors
B
=
{v
1
, v
2
, ··· , v
r
}
.
Extend to a basis of the whole of C
n
B
0
= {v
1
, v
2
, ··· , v
r
, w
1
, w
2
, ··· , w
nr
}
Now use Gram-Schmidt to create an orthonormal basis
˜
B = {v
1
, v
2
, ··· , v
r
, u
1
, u
2
, ··· , u
nr
}.
Now write
P =
v
1
v
2
··· v
r
u
1
··· u
nr
We have shown above that this is a unitary matrix, i.e.
P
1
=
P
. So if we
change basis, we have
P
1
HP = P
HP
=
λ
1
0 ··· 0 0 0 ··· 0
0 λ
2
··· 0 0 0 ··· 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0
0 0 ··· λ
r
0 0 ··· 0
0 0 ··· 0 c
11
c
12
··· c
1,nr
0 0 ··· 0 c
21
c
22
··· c
2,nr
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 ··· 0 c
nr,1
c
nr,2
··· c
nr,nr
Here
C
is an (
n r
)
×
(
n r
) Hermitian matrix. The eigenvalues of
C
are also
eigenvalues of
H
because
det
(
H λI
) =
det
(
P
HP λI
) = (
λ
1
λ
)
···
(
λ
r
λ) det(C λI). So the eigenvalues of C are the eigenvalues of H.
We can keep repeating the process on
C
until we finish all rows. For example,
if the eigenvalues of
C
are all distinct, there are
n r
orthonormal eigenvectors
w
j
(for j = r + 1, ··· , n) of C. Let
Q =
1
1
.
.
.
1
w
r+1
w
r+2
··· w
n
with other entries 0. (where we have a
r × r
identity matrix block on the top
left corner and a (n r) × (n r) with columns formed by w
j
)
Since the columns of
Q
are orthonormal,
Q
is unitary. So
Q
P
HP Q
=
diag
(
λ
1
, λ
2
, ··· , λ
r
, λ
r+1
, ··· , λ
n
), where the first
r λ
s are distinct and the re-
maining ones are copies of previous ones.
The n linearly-independent eigenvectors are the columns of P Q.
So it now follows that
H
is diagonalizable via transformation
U
(=
P Q
).
U
is a unitary matrix because P and Q are. We have
D = U
HU
H = UDU
Note that a real symmetric matrix
S
is a special case of Hermitian matrices. So
we have
D = Q
T
SQ
S = QDQ
T
Example.
Find the orthogonal matrix which diagonalizes the following real
symmetric matrix: S =
1 β
β 1
with β 6= 0 R.
We find the eigenvalues by solving the characteristic equation:
det
(
SλI
) = 0,
and obtain λ = 1 ± β.
The corresponding eigenvectors satisfy (
S λI
)
x
= 0, which gives
x
=
1
2
1
±1
We change the basis from the standard basis to
1
2
1
1
,
1
2
1
1
(which
is just a rotation by π/4).
The transformation matrix is
Q
=
1/
2 1/
2
1/
2 1/
2
. Then we know that
S = QDQ
T
with D = diag(1, 1)
5.8.5 Normal matrices
We have seen that the eigenvalues and eigenvectors of Hermitian matrices satisfy
some nice properties. More generally, we can define the following:
Definition
(Normal matrix)
.
A normal matrix as a matrix that commutes with
its own Hermitian conjugate, i.e.
NN
= N
N
Hermitian, real symmetric, skew-Hermitian, real anti-symmetric, orthogonal,
unitary matrices are all special cases of normal matrices.
It can be shown that:
Proposition.
(i) If λ is an eigenvalue of N , then λ
is an eigenvalue of N
.
(ii) The eigenvectors of distinct eigenvalues are orthogonal.
(iii)
A normal matrix can always be diagonalized with an orthonormal basis of
eigenvectors.
6 Quadratic forms and conics
We want to study quantities like
x
2
1
+
x
2
2
and 3
x
2
1
+ 2
x
1
x
2
+ 4
x
2
2
. For example,
conic sections generally take this form. The common characteristic of these is
that each term has degree 2. Consequently, we can write it in the form
x
Ax
for some matrix A.
Definition
(Sesquilinear, Hermitian and quadratic forms)
.
A sesquilinear form
is a quantity
F
=
x
Ax
=
x
i
A
ij
x
j
. If
A
is Hermitian, then
F
is a Hermitian
form. If A is real symmetric, then F is a quadratic form.
Theorem. Hermitian forms are real.
Proof.
(
x
Hx
)
= (
x
Hx
)
=
x
H
x
=
x
Hx
. So (
x
Hx
)
=
x
Hx
and it is
real.
We know that any Hermitian matrix can be diagonalized with a unitary
transformation. So
F
(
x
) =
x
Hx
=
x
UDU
x
. Write
x
0
=
U
x
. So
F
=
(x
0
)
Dx
0
, where D = diag(λ
1
, ··· , λ
n
).
We know that x
0
is the vector x relative to the eigenvector basis. So
F (x) =
n
X
i=1
λ
i
|x
0
i
|
2
The eigenvectors are known as the principal axes.
Example.
Take
F
= 2
x
2
4
xy
+ 5
y
2
=
x
T
Sx
, where
x
=
x
y
and
S
=
2 2
2 5
.
Note that we can always choose the matrix to be symmetric. This is since
for any antisymmetric
A
, we have
x
Ax
= 0. So we can just take the symmetric
part.
The eigenvalues are 1
,
6 with corresponding eigenvectors
1
5
2
1
,
1
5
1
2
.
Now change basis with
Q =
1
5
2 1
1 2
Then x
0
= Q
T
x =
1
5
2x + y
x 2y
. Then F = (x
0
)
2
+ 6(y
0
)
2
.
So F = c is an ellipse.
6.1 Quadrics and conics
6.1.1 Quadrics
Definition
(Quadric)
.
A quadric is an
n
-dimensional surface defined by the
zero of a real quadratic polynomial, i.e.
x
T
Ax + b
T
x + c = 0,
where
A
is a real
n ×n
matrix,
x, b
are
n
-dimensional column vectors and
c
is a
constant scalar.
As noted in example, anti-symmetric matrix has
x
T
Ax
= 0, so for any
A
,
we can split it into symmetric and anti-symmetric parts, and just retain the
symmetric part S = (A + A
T
)/2. So we can have
x
T
Sx + b
T
x + c = 0
with S symmetric.
Since
S
is real and symmetric, we can diagonalize it using
S
=
QDQ
T
with
D diagonal. We write x
0
= Q
T
x and b
0
= Q
T
b. So we have
(x
0
)
T
Dx
0
+ (b
0
)
T
x
0
+ c = 0.
If
S
is invertible, i.e. with no zero eigenvalues, then write
x
00
=
x
0
+
1
2
D
1
b
0
which shifts the origin to eliminate the linear term (
b
0
)
T
x
0
and finally have
(dropping the prime superfixes)
x
T
Dx = k.
So through two transformations, we have ended up with a simple quadratic form.
6.1.2 Conic sections (n = 2)
From the equation above, we obtain
λ
1
x
2
1
+ λ
2
x
2
2
= k.
We have the following cases:
(i) λ
1
λ
2
>
0: we have ellipses with axes coinciding with eigenvectors of
S
.
(We require
sgn
(
k
) =
sgn
(
λ
1
, λ
2
), or else we would have no solutions at all)
(ii) λ
1
λ
2
< 0: say λ
1
= k/a
2
> 0, λ
2
= k/b
2
< 0. So we obtain
x
2
1
a
2
x
2
2
b
2
= 1,
which is a hyperbola.
(iii) λ
1
λ
2
= 0: Say
λ
2
= 0,
λ
1
6
= 0. Note that in this case, our symmetric
matrix S is not invertible and we cannot shift our origin using as above.
From our initial equation, we have
λ
1
(x
0
1
)
2
+ b
0
1
x
0
1
+ b
0
2
x
0
2
+ c = 0.
We perform the coordinate transform (which is simply completing the
square!)
x
00
1
= x
0
1
+
b
0
1
2λ
1
x
00
2
= x
0
2
+
c
b
0
2
(b
0
1
)
2
4λ
1
b
0
2
to remove the x
0
1
and constant term. Dropping the primes, we have
λ
1
x
2
1
+ b
2
x
2
= 0,
which is a parabola.
Note that above we assumed
b
0
2
6
= 0. If
b
0
2
= 0, we have
λ
1
(
x
0
1
)
2
+
b
0
1
x
0
1
+
c
=
0. If we solve this quadratic for
x
0
1
, we obtain 0, 1 or 2 solutions for
x
1
(and x
2
can be any value). So we have 0, 1 or 2 straight lines.
These are known as conic sections. As you will see in IA Dynamics and Relativity,
this are the trajectories of planets under the influence of gravity.
6.2 Focus-directrix property
Conic sections can be defined in a different way, in terms of
Definition
(Conic sections)
.
The eccentricity and scale are properties of a conic
section that satisfy the following:
Let the foci of a conic section be (±ae, 0) and the directrices be x = ±a/e.
A conic section is the set of points whose distance from focus is
e×
distance
from directrix which is closer to that of focus (unless
e
= 1, where we take the
distance to the other directrix).
Now consider the different cases of e:
(i) e < 1. By definition,
x
y
O
x = a/e
ae
(x, y)
p
(x ae)
2
+ y
2
= e
a
e
x
x
2
a
2
+
y
2
a
2
(1 e
2
)
= 1
Which is an ellipse with semi-major axis
a
and semi-minor axis
a
1 e
2
.
(if e = 0, then we have a circle)
(ii) e > 1. So
x
y
O
x = a/e
ae
(x, y)
p
(x ae)
2
+ y
2
= e
x
a
e
x
2
a
2
y
2
a
2
(e
2
1)
= 1
and we have a hyperbola.
(iii) e = 1: Then
x
y
O
x = a
a
(x, y)
p
(x a)
2
+ y
2
= (x + 1)
y
2
= 4ax
and we have a parabola.
Conics also work in polar coordinates. We introduce a new parameter
l
such
that l/e is the distance from the focus to the directrix. So
l = a|1 e
2
|.
We use polar coordinates (
r, θ
) centered on a focus. So the focus-directrix
property is
r = e
l
e
r cos θ
r =
l
1 + e cos θ
We see that
r
if
θ cos
1
(
1
/e
), which is only possible if
e
1, i.e.
hyperbola or parabola. But ellipses have e < 1. So r is bounded, as expected.
7 Transformation groups
We have previously seen that orthogonal matrices are used to transform between
orthonormal bases. Alternatively, we can see them as transformations of space
itself that preserve distances, which is something we will prove shortly.
Using this as the definition of an orthogonal matrix, we see that our definition
of orthogonal matrices is dependent on our choice of the notion of distance, or
metric. In special relativity, we will need to use a different metric, which will
lead to the Lorentz matrices, the matrices that conserve distances in special
relativity. We will have a brief look at these as well.
7.1 Groups of orthogonal matrices
Proposition.
The set of all
n × n
orthogonal matrices
P
forms a group under
matrix multiplication.
Proof.
0.
If
P, Q
are orthogonal, then consider
R
=
P Q
.
RR
T
= (
P Q
)(
P Q
)
T
=
P (QQ
T
)P
T
= P P
T
= I. So R is orthogonal.
1. I satisfies II
T
= I. So I is orthogonal and is an identity of the group.
2.
Inverse: if
P
is orthogonal, then
P
1
=
P
T
by definition, which is also
orthogonal.
3.
Matrix multiplication is associative since function composition is associative.
Definition
(Orthogonal group)
.
The orthogonal group
O
(
n
) is the group of
orthogonal matrices.
Definition
(Special orthogonal group)
.
The special orthogonal group is the
subgroup of O(n) that consists of all orthogonal matrices with determinant 1.
In general, we can show that any matrix in O(2) is of the form
cos θ sin θ
sin θ cos θ
or
cos θ sin θ
sin θ cos θ
7.2 Length preserving matrices
Theorem. Let P O(n). Then the following are equivalent:
(i) P is orthogonal
(ii) |P x| = |x|
(iii) (P x)
T
(P y) = x
T
y, i.e. (P x) · (P y) = x · y.
(iv) If (v
1
, v
2
, ··· , v
n
) are orthonormal, so are (Pv
1
, P v
2
, ··· , P v
n
)
(v) The columns of P are orthonormal.
Proof. We do them one by one:
(i) (ii): |P x|
2
= (P x)
T
(P x) = x
T
P
T
P x = x
T
x = |x|
2
(ii) (iii): |P (x + y)|
2
= |x + y|
2
. The right hand side is
(x
T
+ y
T
)(x + y) = x
T
x + y
T
y + y
T
x + x
T
y = |x|
2
+ |y|
2
+ 2x
T
y.
Similarly, the left hand side is
|P x + P y|
2
= |P x|
2
+ |P y| + 2(P x)
T
P y = |x|
2
+ |y|
2
+ 2(P x)
T
P y.
So (P x)
T
P y = x
T
y.
(iii) (iv): (P v
i
)
T
P v
j
= v
T
i
v
j
= δ
ij
. So P v
i
’s are also orthonormal.
(iv)
(v): Take the
v
i
’s to be the standard basis. So the columns of
P
, being
P e
i
, are orthonormal.
(v)
(i): The columns of
P
are orthonormal. Then (
P P
T
)
ij
=
P
ik
P
jk
=
(P
i
) · (P
j
) = δ
ij
, viewing P
i
as the ith column of P. So P P
T
= I.
Therefore the set of length-preserving matrices is precisely O(n).
7.3 Lorentz transformations
Consider the Minkowski 1 + 1 dimension spacetime (i.e. 1 space dimension and
1 time dimension)
Definition
(Minkowski inner product)
.
The Minkowski inner product of 2
vectors x and y is
hx | yi = x
T
Jy,
where
J =
1 0
0 1
Then hx | yi = x
1
y
1
x
2
y
2
.
This is to be compared to the usual Euclidean inner product of
x, y R
2
,
given by
hx | yi = x
T
y = x
T
Iy = x
1
y
1
+ x
2
y
2
.
Definition
(Preservation of inner product)
.
A transformation matrix
M
pre-
serves the Minkowski inner product if
hx|yi = hMx|Myi
for all x, y.
We know that
x
T
Jy
= (
Mx
)
T
JMy
=
x
T
M
T
JMy
. Since this has to be
true for all x and y, we must have
J = M
T
JM.
We can show that M takes the form of
H
α
=
cosh α sinh α
sinh α cosh α
or K
α/2
=
cosh α sinh α
sinh α cosh α
where H
α
is a hyperbolic rotation, and K
α/2
is a hyperbolic reflection.
This is technically all matrices that preserve the metric, since these only
include matrices with
M
11
>
0. In physics, these are the matrices we want, since
M
11
< 0 corresponds to inverting time, which is frowned upon.
Definition
(Lorentz matrix)
.
A Lorentz matrix or a Lorentz boost is a matrix
in the form
B
v
=
1
1 v
2
1 v
v 1
.
Here
|v| <
1, where we have chosen units in which the speed of light is equal to
1. We have B
v
= H
tanh
1
v
Definition
(Lorentz group)
.
The Lorentz group is a group of all Lorentz matrices
under matrix multiplication.
It is easy to prove that this is a group. For the closure axiom, we have
B
v
1
B
v
2
= B
v
3
, where
v
3
= tanh(tanh
1
v
1
+ tanh
1
v
2
) =
v
1
+ v
2
1 + v
1
v
2
The set of all
B
v
is a group of transformations which preserve the Minkowski
inner product.