1 June 25th

1.1 Some Review

1.1.1 Linear Algebra - Terms
The symbol R stands for real numbers and the symbol C stands for complex numbers. For
n
most of our purposes the vector spaces we will be dealing with will be R .
n n
R is n-dimensional space. We can view vectors v ∈ R as coordinate vectors, n-tuples of
real numbers:
v = (x1 , x2 , . . . , xn ), xi ∈ R
You can add vectors by adding their components:

(x1 , . . . , xn ) + (y1 , . . . , yn ) = (x1 + y1 , . . . , xn + yn )

You can multiply (or scale) vectors by real numbers:

c ⋅ (x1 , . . . , xn ) = (c ⋅ x1 , . . . , c ⋅ xn ), c∈R

A real number c in R scaling vectors in this way is called a scalar.

n
Definition 1. A linear combination of vectors v1 , . . . , vk in R is a sum:

a1 v1 + ⋅ ⋅ ⋅ + ak vk

with ai ∈ R scalars.
Definition 2. A set of vectors v1 , . . . , vk in a vector space V is linearly dependent if there
exists scalars ai such that:
a1 v1 + ⋅ ⋅ ⋅ + ak vk = 0
with not all ai = 0.
Another way to phrase this is that the set vi are linearly dependent if a non-zero linear
combination of them is equal to 0.
Definition 3. A set of vectors v1 , . . . , vk is linearly independent if they are not linearly
dependent.
Definition 4. A basis for R is a set of n linearly independent vectors { v1 , . . . , vn }.
n

Theorem 1. A basis β = { v1 , . . . , vn } spans R : every vector v ∈ R can be written as a

n n

unique linear combination of the basis vectors v1 , . . . , vn .

n
Let v ∈ R be a vector given by the linear combination:

v = y1 v1 + ⋅ ⋅ ⋅ + yn vn

where yi are scalars. Then yi are the coordinates of v in the basis β. The coordinate vector
for v in the basis β is denoted:
[v]β = (y1 , . . . , yn )

1
Example 1. If we view R as the space of n-tuples (x1 , . . . , xn ) then we have a standard
n

basis { e1 , . . . , en }. This is the basis:

ei = (0, . . . , 0, 1, 0, . . . , 0)

where ei has 1 in its ith component.

A vector v = (x1 , . . . , xn ) in R can then be written as the linear combination:
n

v = x1 e1 + ⋅ ⋅ ⋅ + xn en
n m
Definition 5. A linear transformation T ∶ R → R is a map between vector spaces such
that:
T (v1 + v2 ) = T (v1 ) + T (v2 ), T (c ⋅ v1 ) = c ⋅ T (v1 )
n
For v1 , v2 vectors in R and c ∈ R a scalar.
If you choose a basis α = { v1 , . . . , vn } for R and a basis β for R , then you can write
n m

T as an n × m matrix given by:

[[T (v1 )]β . . . [T (vn )]β ]

The columns are the images of the basis vectors in α, T (vi ), written as coordinate vectors in
the basis β.
Example 2. The space of smooth functions on R is a vector space V . Differentation is a
linear transformation V → V .
n n
Example 3. Let I ∶ R → R be the identity transformation:

I(v) = v

Then in any basis, I is given by the matrix:

⎡
⎢ 1 0 ... 0⎤
⎥
⎢
⎢ ⎥
⎢ 0⎥
⎥
I=⎢ ⎥
⎢ 0 1 ... ⎥
⎢
⎢ ⋮⎥
⎥
⎢
⎢ ⋮ ⋮ ... ⎥
⎥
⎢
⎢ 1⎥
⎥
⎣0 0 ... ⎦
with 1-s along the diagonal and all other entries 0.
Example 4. Let e1 = (1, 0) and e2 = (0, 1) be the standard basis vectors for R . The linear
2

transformation:
T (e1 ) = 2e1 , T (e2 ) = e1 − e2
corresponds to the matrix:
2 1
[ ]
0 −1
If v = (x, y) in the standard basis then:

2 1 x 2x + y
T (v) = [ ][ ] = [ ]
0 −1 y −y

2
Definition 6. Let α = { v1 , . . . , vn }, β be two basis for R . The change of basis matrix from
n

α to β is the matrix:
B = [[v1 ]β . . . [vn ]β ]
n
If v ∈ R is a vector then B sends its α coordinates to β coordinates:

B[v]α = [v]β

This changes the coordinate vector of v but not the actual vector itself. Note that if B is the
−1
change of basis α → β then B is the change of basis matrix β → α.
n n
Theorem 2. Let T ∶ R → R be a linear transformation and α, β two basis. Then:

[T ]β = B[T ]α B
−1

where B is the change of basis matrix from α to β.

1.1.2 Finding Eigenvalues

n n
Definition 7. Let T be a linear operator T ∶ R → R . Then an eigenvalue is a scalar
n
λ ∈ R such that there exists a non-zero vector v ∈ R such that:

T (v) = λ ⋅ v

That is T scales v by λ. If v exists then it is called an eigenvector of T with eigenvalue λ.

n n
Theorem 3. Let T ∶ R → R be a linear operator. The eigenvalues of T are the roots of its
characteristic polynomial:
χ(T ) = det(T − tI)
This is a polynomial of degree n in t.
n n
Theorem 4. Eigenvalues of T ∶ R → R are either real or complex. If they are complex
they come in conjugate pairs:
a + bi, a − bi
Example 5 (Diagonalizable). Let’s find the eigenvalues of the matrix:
2 −1
A=[ ]
−1 2
Then:
2 −1 1 0 2 − t −1
A − tI = [ ] − t[ ]=[ ]
−1 2 0 1 −1 2 − t
Then:
2 2 2
det(A − tI) = (2 − t) − (−1) = 4 − 4t + t − 1
2
= 3 − 4t + t
= (t − 3)(t − 1)

The roots of this polynomial are 3, 1 which are the eigenvalues of A.

3
Example 6 (Complex Eigenvalues). Let’s find the eigenvalues of the matrix:

0 1
A=[ ]
−1 0

Then:
−t 1
det(A − tI) = det ([ ]) = t + 1
2
−1 −t
This factor over C as:
2
t + 1 = (t − i)(t + i)
√ 2
since ±i = ± −1 is a root of t + 1.
Therefore the eigenvalues of A are ±i.

1.1.3 Finding Eigenvectors

n m
Definition 8. The nullspace or kernel of a linear transformation T ∶ R → R is the set of
n
vectors v ∈ R such that:
T (v) = 0
Writing T as a matrix, this amounts to the solution space of a system of linear equations.
n n
Theorem 5. Let T ∶ R → R be a linear operator. Then the set of eigenvectors with
eigenvalue λ form a subspace called the eigenspace of T for the eigenvalue λ.
n n
Theorem 6. Let T ∶ R → R be a linear operator. Let λ be an eigenvalue of T . Then v is
an eigenvector of T with eigenvalue λ if and only if v is in the kernel (nullspace) of T − λI.
In otherwords the eigenspace for λ is equal to ker(T − λI)

Example 7. Let’s find the eigenvectors for the matrix

2 −1
A=[ ]
−1 2

We found two eigenvalues λ = 3, 1 for this matrix before.

To find the eigenvectors with eigenvalue 3 we consider the kernel of:

−1 −1
A − 3I = [ ]
−1 −1

To find the kernel we consider the system of equations:

0 −1 −1 x −x − y
[ ]=[ ]⋅[ ]=[ ]
0 −1 −1 y −x − y

Therefore solutions to this system are vectors (x, y) such that:

0 = −x − y ⟹ y = −x

4
That is vectors of the form (x, −x). Pulling out x as a scalar, every solution can be written
as x ⋅ (1, −1). Therefore:
ker(A − 3I) = span { (1, −1) }
and (1, −1) is a basis for this eigenspace.
Similarly eigenvectors with eigenvalue 1 are in the kernel of:

1 −1
A−I =[ ]
−1 1

The corresponding system of equations is:

0 1 −1 x x−y
[ ]=[ ]⋅[ ]=[ ]
0 −1 1 y −x + y

Solutions are vectors (x, y) such that:

0=x−y ⟹ x=y

Therefore:
ker(A − I) = span { (1, 1) }
is the eigenspace and (1, 1) is a basis vector.

Example 8. Let’s find the eigenvectors of the matrix:

0 1
A=[ ]
−1 0

A has eigenvalues ±i. The eigenspace for i is the kernel:

−i 1
A − iI = [ ]
−1 −i

so solutions of the equation:

0 −i 1 x −ix + y
[ ]=[ ][ ] = [ ]
0 −1 −i y −x − iy

Note that:
2
(−i)(−ix + y) = i x − iy = −x − iy
so the first and second equation are linearly dependent. This means that solutions over C
are vectors (x, y) such that:
0 = −ix + y ⟹ ix = y
so vectors (x, ix). Pulling out x as a scalar, we have:

ker(A − iI) = span { (1, i) }

is the eigenspace for i which has basis (1, i).

5
 The eigenspace for −i is the kernel of:

i 1
A + iI = [ ]
−1 i

so vectors (x, y) such that:

0 = ix + y ⟹ −ix = y
Therefore:
ker(A + iI) = span { (1, −i) }
is the eigenspace of −i which has basis (1, −i).

1.1.4 Diagonalizable matrices / The point of eigenvectors

What do eigenvalues and eigenvectors tell us about a matrix T ? Eigenvectors give us directions
or axes in which the linear transformation T just scales vectors. This makes understanding T
much simpler. The more eigenvectors we have, the simpler we can make T . The best version
of this is diagonalizable matrices:
n n n
Definition 9. A matrix T ∶ R → R is diagonalizable if there is a basis for R consisting
of eigenvectors of T .
In other words, T is diagonalizable if it has n linearly independent eigenvectors. This
means that there is a basis in which T is a diagonal matrix.
n n
Example 9. Let T ∶ R → R . Suppose we can find v1 , . . . , vn be n-linearly independent
eigenvectors of T with corresponding eigenvalues λ1 , . . . , λn . Then β = { v1 , . . . , vn } forms a
n
basis for R which is often called an eigenbasis.
As a matrix with coordinates in this basis, T is given by:

[T ]β = [[T (v1 )]β . . . [T (vn )]β ]

But:
T (vi ) = λi vi
Therefore:
[T (vi )]β = (0, . . . , λi , . . . , 0)
i.e. λi in position i. Therefore:
⎡
⎢ λ1 0 ... 0⎤ ⎥
⎢
⎢ ⎥
⎢ 0⎥ ⎥
[T ]β = ⎢ ⎥
⎢ 0 λ2 ... ⎥
⎢
⎢ ⋮⎥⎥
⎢
⎢ ⋮ ⋮ ... ⎥
⎥
⎢
⎢ λn ⎥
⎥
⎣0 0 ... ⎦
the diagonal matrix with entries λi along the diagonal.
So changing basis to an eigenbasis makes a potentially complicated transformation T very
simple: it just stretches each coordinate by scalars λi !

Theorem 7. Eigenvectors with different eigenvalues are linearly independent.

6
Theorem 8. If A is a n × n matrix with n-distinct eigenvalues then each of its eigenspaces
is 1-dimensional and A is diagonalizable.
Example 10. For the matrix:
2 −1
A=[ ]
−1 2
we calculated that it had eigenvectors and eigenvalues:
v1 = (1, −1), λ1 = 3
v2 = (1, 1), λ2 = 1
Therefore in the eigenbasis β = { v1 , v2 } A is given by the diagonal matrix:
3 0
[A]β = [ ]
0 1
Example 11. For the matrix:
0 1
A=[ ]
−1 0
we found eigenvectors and eigenvalues:
v1 = (1, i), λ1 = i
v2 = (1, −i), λ2 = −i
Therefore in the basis β = { v1 , v2 } for C , A is the diagonal matrix:
2

i 0
[A]β = [ ]
0 −i
A useful fact:
Theorem 9. Let ⟨v1 , v2 ⟩ be the inner product:
T
⟨v1 , v2 ⟩ = v1 v2
Then eigenvectors v1 , v2 with distinct eigenvalues are orthogonal with respect to this inner
product, i.e.:
⟨v1 , v2 ⟩ = 0
Theorems you have seen but may or may not be useful:
n n
Theorem 10 (Spectral Theorem). If A ∶ R → R is a real matrix that is symmetric,
T
A = A , then it is diagonalizable.
Theorem 11 (Perron-Frobenius). Let A be a real matrix whose entries are all positive
(non-zero). Then A has a real eigenvalue λ such that:
′
1. λ has maximal magnitude, meaning for all other eigenvalues ζ (including complex
ones):
∣λ∣ > ∣ζ ∣
′

2. The eigenspace for λ is 1-dimensional, so spanned by a single vector.

3. There is an eigenvector for λ which has all positive coordinates.

7
1.2 ODES
Some very basic review of ways to solve an ODE that you will see in the homework:

1.2.1 Separation of variables

Suppose you are given an ODE:
dx g(y)
=
dy f (x)
You can solve this by separating variables and integrating:

f (x)dx = g(y)dy

⟹ ∫ f (x)dx = ∫ g(y)dy

Example 12. Consider the ODE:

dx
= ax + b
dt
where a, b are constants. Then:
dx
= adt
x + b/a
dx
⟹ ∫ = ∫ adt
x + b/a
⟹ ln(x + b/a) = at + C
′ at
⟹ x + b/a = C e

Suppose x(0) = X0 then we have:

′
X0 + b/a = C
Therefore:
at
x(t) = (X0 + b/a)e − b/a
Note that if b = 0 then we have the solution:
at
x(t) = X0 e

as expected.

1.2.2 Integrating Factor

Given an ODE of the form:
dy
= f (x)y + g(x)
dx
Multiply by the equation by the integrating factor:
− ∫ f (x)
e

8
and rearrange to get:
− ∫ f (x) dy − ∫ f (x) − ∫ f (x)g(x)
e −e f (x)y = e
dx
By product rule and chain rule:
d − ∫ f (x) − ∫ f (x) dy − ∫ f (x)
(e y) = e − f (x)e y
dx dx
Therefore the equation can be written as:
d − ∫ f (x) − inf tf (x)
(e y) = e g(x)
dx
Integrating we have:
− ∫ f (x) − ∫ f (x)
e y=∫ e g(x)dx

So if the right side is integrable we have a solution for y.

Example 13. Consider the equation:

x
dy 2y e
+ x = 2
dx x

Then f (x) = −1/x and g(x) = e /x . The integrating factor is:

x 2

∫ 2/x 2 ln(x) 2
e =e =x

Following the process above we get:

2 2 x x
x y = ∫ x g(x)dx = ∫ e dx = e + C

Therefore: x
e +C
y=
x2

2 Population Growth
In this section we model systems in discrete time. That is we use a fixed time step, e.g. years,
days, months, and we try to model the change observed in the system when sampling at this
time step. This yields a difference equation.

Example 14. Let us try to model the growth of a population over time, given by the function
N (t). Fix a time step ∆t.
The growth rate with respect to ∆t is the percentage change in the population over the
time step ∆t, i.e.:
N (t + ∆t) − N (t)
R=
∆tN (t)

9
Assume that R is constant, so that it does not depend on t. Then:

N (t + ∆t) − N (t) = R∆tN (t)

Rearranging we get a difference equation:

N (t + ∆t) = (1 + R∆t)N (t)

Let N0 = N (0) be the initial value of the population. Then:

n
N (t + n∆t) = (1 + R∆t) N0

Example 15. Let N be a population which has a birth rate of 59 per 100 every year and a
death rate of 48 per 100 every year. With ∆t being a year, the change in population between
time steps is:
59 49 11
N (t + ∆t) − N (t) = N (t) − N (t) = N (t)
100 100 100
The resulting recurrence relation is:

N (t + ∆t) = (1 + 0.11)N (t)

Therefore the solution to the system is given by:

m
N (m∆t) = (1 + 0.11) N0

Example 16. With the same setup as the previous example, when does the population
double? If the population doubles in m years then:
m
2N0 = N (m∆t) = (1 + 0.11) N0

Then:
m
2 = (1 + 0.11) ⟹ ln(2) = m ln(1 + 0.11)
Solving for m:
ln(2)
m=
ln(1 + 0.11)
Example 17. With the same setup, suppose that every year 50 members of another population
migrate into this one. Then the resulting difference equation is:

N (t + ∆t) = (1 + 0.11)N (t) + 50

which is no longer a linear system.

10