GM0742/GM1040 Mathematics

Lectures 1-6
MATRIX ALGEBRA

Charles Nadeau
E-mail: [email protected]
Office: E-506
Office Hours: by appointment

Department of Economics, University of Göteborg

Autumn 2023
Modeling Systems in Matrix Form
Simple linear input-output model:
x1 – βx2 = d1
x2 – γx3 = d2
x3 - αx1 = 0

where (x1 , x2 , x3) denote endogeneous variables, (β, γ, α) denote coefficients and (d1 , d2) denote
constant terms.

Matrix form: Ax = d

where
1 −β 0 x1 d1
A = 0 1 −γ x = x2 d = d2
−α 0 1 x3 0

𝑑1 + β𝑑2 αγ𝑑1 + 𝑑2 α𝑑1 + αβ𝑑2

Solving: x1* = x2* = x3* =
1−αβγ 1−αβγ 1−αβγ
Matrix Modeling

Linear models can be written in (compact) matrix form Ax = d where:

A denotes the matrix of coefficients (coefficient matrix) (A33)

matrix ≡ group of mathematical terms arranged in rows and columns ([ ])

dimension ≡ denotes the size of a matrix in terms of # of rows and columns
elements ≡ individual terms in a matrix

x denotes the (column) vector of variable (x31)

vector ≡ special matrix with only 1 row or 1 column (row vector, column vector)

d denotes the (column) vector of constant terms (d31)

Note: If the coefficient matrix (A) is a non-singular, square matrix then the model can be
solved (x1*, x2*, x3*, ….) and the solution set will be unique.
Modeling Systems in Matrix Form (cont)

Suppose we have a system of m simultaneous, linear equations:

a11x1 + a12x2 + …. + a1nxn = d1

a21x1 + a22x2 + …. + a2nxn = d2
…..
am1x1 + am2x2 + …. + amnxn = dm

This system can be written as:

Ax = d
where

A = x = d =
Modeling Systems in Matrix Form (cont)

Suppose:
2x1 - x2 = 0
-x1 + x2 = 4

Matrix form:
Ax = d

where
2 −1 x1 0
A = x = d =
−1 1 x2 4

x1
First equation: [2 -1] = 0
x2

x1
Second equation: [-1 1] = 4
x2
Vectors

Vectors are ordered arrays of elements (e.g. numbers, variables, ect …):

Two-dimensional vectors:

x´ = [x1 x2]

y1
y = y
2

which can be represented as points or directed segments on a 2-

dimensional coordinate plane (see textbook discussion).
Vectors (cont)

Three-dimensional vectors:

x´ = [x1 x2 x3]

y =

which can be represented as points or directed segments on a 3-

dimensional coordinate space (see textbook discussion).
Vectors (cont)

n-dimensional vectors:

x´ = [x1 x2 x3 … xn]

y =

which cannot be represented in graphical form.

Equality of vectors: x´ = z´ iff xj = zj ∀ j

y = z iff yi = zi ∀ i
Matrices

Matrices are rectangular arrays of elements consisting of m rows and n

columns:

m x n matrix:

Amn =

where aij denotes the element of the matrix on the intersection of row i and
column j, considered as a single entity (e.g. number, parameter, variable).

Equality of matrices: A = B iff aij = bij ∀ i, j

Matrices (cont)

Special matrices:

Identity matrix:
1 0
I2 =
0 1

I3 =

There also exists I4 , I5 , I6 , ect …. all necessarily square matrices.

Matrix counterpart to scalar 1: IA = AI = A
Matrices (cont)

Special matrices:

Null matrix:
0 0
022 =
0 0

032 =

Null matrices are not necessarily square matrices.

Matrix counterpart to scalar 0: (A + 0) = (0 + A) = A
(A0) = (0A) = 0
Vector Operations (conformability conditions)

Addition/Subtraction:
Vectors must have the same dimension

Scalar Multiplication (α):

Multiply every element of the vector by the scalar (α)

Multiplication:

Column dimension of the lead vector must be same as the row dimension of lag vector
(e.g. 1x2 lead vector and 2x1 lag vector)

Division:
Impossible
Vector Addition and Subtraction
Suppose:
x´ = [8 9 -3]
z´ = [4 -5 6]
Then:
x´ + z´ = [12 4 3]
x´ - z´ = [4 14 -9]
Suppose:
−2
y = 7
−1

0
z = 13
4
Then:
−2 −2
y + z = 20 and y - z = −6
3 −5
Vector Addition and Subtraction (cont)

Generically:
x´ = [x1 x2 x3]
z´ = [z1 z2 z3]

x´ + z´ = [x1 + z1 x2 + z2 x3 + z3]
x´ - z´ = [x1 - z1 x2 - z2 x3 - z3]

y1 z1
y = y2 z = z2
y3 z3

y1 + z1 y1 − z1
y + z = y2 + z2 y - z = y2 − z2
y3 + z3 y3 − z3
Scalar Multiplication (α)

Suppose:
x´ = [x1 x2 x3]

y =

Then:

αx´ = [αx1 αx2 αx3]

αy =
Vector Multiplication (cont)

Generically:
x´ = [x1 x2 x3]

y =

Then:

x´y = [x1 x2 x3]

= x1y1 + x2y2 + x3y3

Requirement: (row vector) x (column vector) with equal number of elements

Result: scalar (calculated as the inner product of the vector elements)
Linear Dependence

A system of vectors (x, y, z) is linearly dependent if some non-trivial linear combination of them is
equal to the null vector:
α1x + α2y + α3z = 0

where the scalars (α1 , α2 , α3) are not all equal to zero.

Examples:

(1) Two vectors (x, y) are linearly dependent if they are proportional: x = αy
(2) Three vectors (x, y, z) are linearly dependent if one of the vectors is a linear combination of the
other two vectors: z = α1x + α2y

Textbook concepts: vector space, spanning vectors

Linear Dependence (cont)

Example:
v1´ = [5 12] v2´ = [10 24]

Row vectors v1´ and v2´ are linearly dependent because

2v1´ = v2´
2v1´ - v2´ = 0
Example:
2 1 4
v1 = v2 = v3 =
7 8 5

Column vectors v1 , v2 and v3 are linearly dependent because

3v1 – 2v2 = v3
3v1 – 2v2 – v3 = 0

Question: Is it even possible for v1 , v2 and v3 to be linearly independent?

Matrix Operations (conformability conditions)
Addition/Subtraction:
Matrices must have the same dimension
Commutative Law: A + B = B + A
Associative Law: (A + B) + C = A + (B + C)

Scalar Multiplication (α):

Multiply every element of the matrix by the scalar (α)

Commutative Law: αA = Aα

Multiplication:

Column dimension of the lead matrix must be the same as the row dimension of lag matrix
Associative Law: (AB)C = A(BC) = ABC
Distributive Law: A(B + C) = AB + AC

Division:
Impossible
Matrix Addition

Requirement: conformable for addition (i.e. same dimension)

a11 a12 b11 b12

A = and B =
a21 a22 b21 b22

a11 + b11 a12 + b12

A+B =
a21 + b21 a22 + b22

Rules:
(1) Commutative: A + B = B + A
(2) Associative: (A + B) + C = A + (B + C)
(3) Null matrix (0): A + 0 = A
(4) Opposite matrix (-A): A + (-A) = 0
Matrix Addition/Subtraction

Example:
4 9 2 0
A = B =
2 1 0 7

6 9
A+B =
2 8

Example:
19 3 6 8
A = B =
2 0 1 3

13 −5
A-B =
1 −3
Scalar Multiplication

αa11 αa12
αA =
αa21 αa22

Rules:
(1) α1(α2 A) = (α1 α2) A
(2) α (A + B) = αA + αB
(3) (α1 + α2) A = α1 A + α2 A
Matrix Multiplication

Conformability Requirement:
Amk Bkn = Cmn
(column dimension of lead matrix) = (row dimension of lag matrix) = k
Dimension of product matrix: (m x n)

Multiplication Rule:
cij = σ𝑘𝑝=1(𝑎ip)(𝑏pj )
(cij is calculated as the inner product of the ith row of A and the jth column of B)
Result: scalar

Rules: (1) AB ≠ BA (i.e. not commutative in general)

(2) Associative: (AB)C = A(BC) (assuming conformability holds)
(3) Distributive: A(B+C) = AB + AC
(B+C)A = BA + CA (assuming …)
Transpose of a Matrix
In order to transpose a matrix you interchange the rows and columns of the matrix:

Suppose:
1 2 3
A = 4 5 6
7 8 9
Then:
1 4 7
A´ = 2 5 8
3 6 9

Properties: (A´)´ = A
(A + B)´ = A´ + B´
(AB)´ = B´A´
Transpose of a Matrix (cont)

Suppose:
1 2 3 4
A =
5 6 7 8
Then:

A´ =
Matrix Inversion

The inverse of square matrix A is denoted as

A-1
and is defined such that

AA-1 = A-1 A = I

Note: (1) unlike a transpose matrix, an inverse matrix may not exist.
(2) if square matrix A has an inverse, then A is called nonsingular.
(3) if square matrix A has no inverse, then A is called singular.
(4) A and A-1 will always have the same dimension.
(5) if A-1 exists, then it is unique.

Issue: What does singularity or nonsingularity imply about a square matrix?

Matrix Inversion (cont)

Suppose matrix A and matrix B are both non-singular nxn matrices.

Properties:
(A-1)-1 = A
(AB)-1 = B-1 A-1
(A´)-1 = (A-1)´
Determinant of a Matrix

The determinant of a square matrix is a unique scalar computed in a particular way:

Suppose:
a11 a12
A =
a21 a22
Then:
a11 a12
A ≡ det A = = (a11a22 - a12a21) (scalar)
a21 a22

where the value of A is a crucial test for the singularity of the matrix (discussed earlier)
and is useful for calculating A-1 (also discussed earlier).
Determinant of a Matrix (cont)

Suppose:
a11 a12 a13
A = a21 a22 a23
a31 a32 a33
Then:
a22 a23 a a23 a a
A = a11 - a12 21 + a13 21 22
a32 a33 a31 a33 a31 a32

= a11 M11 - a12 M12 + a13 M13

where the subdeterminant Mij denotes the minor of element aij. The cofactor ( Cij ) of
element aij is defined as:

Cij ≡ (-1)i+j Mij

Note: In the example above, A was calculated via expansion along the first row …
Determinant of a Matrix (cont)

The value of the determinant of a matrix can be calculated by expansion using any row i
or column j in the matrix; thus, the formula for the value of the determinant of matrix A
can be written as either:

det A = σnj=1 aij Cij

if expanding using row i or, equivalently

det A = σni=1 aij Cij

if expanding using column j.

Determinant of a Matrix (cont)

Properties:

(1) If matrix A is singular (i.e. linearly dependent rows/columns) then:

A = 0

(2) A = A′
(3) If matrix A has a 0 row or column, then A = 0
(4) Interchanging rows (or columns) in matrix A does not affect A
(5) Multiplying 1 row/column in matrix A by scalar α yields: α A

Textbook concepts: rank, full rank

Determinant of a Matrix (cont)

Given:
2 0
A =
0 2
Calculating:
𝐴 = (2)(2) – (0)(0) = +4

Given:
−1 5
A =
2 −10
Calculating:
𝐴 = (-1)(-10) – (5)(2) = 0

Note: -2v1´ = v2´ (i.e. linear dependent rows)

Note: -5v1 = v2 (i.e. linear dependent columns)
Determinant of a Matrix (cont)

Suppose:

A =

B =

Question: Is there an easy way to calculate determinants here?

Verify: A = +72 B = -81

Calculating the Inverse of a Matrix

Given square matrix A, the inverse of A (i.e. A-1) is calculated as follows:

Steps:

(1) calculate A (assume: A ≠ 0)

(2) construct the cofactor matrix (C) for A

(3) take the transpose of C (i.e. C´ ≡ adjoint of A)

Calculating:
adj A
A-1 = A
Constructing a Cofactor Matrix

To construct a cofactor matrix (C) for matrix A we simply replace each element in matrix A
(aij) with its cofactor ( Cij ), where

Cij ≡ (-1)i+j Mij

Example:
a11 a12
A =
a21 a22

C11 C12 a22 −a21

C = =
C21 C22 −a12 a11

a22 −a12
adj A ≡ C´ =
−a21 a11
Constructing a Cofactor Matrix (cont)

Example:
a11 a12 a13
A = a21 a22 a23
a31 a32 a33

C =

adjA ≡ C´ =
Calculating the Inverse of a Matrix (cont)

Example:
1 0 0
I = 0 1 0
0 0 1

Calculating: I = (1)(1) + (0)(0) + (0)(0) = +1

1 0 0 1 0 0
C = 0 1 0 → adj I = 0 1 0
0 0 1 0 0 1

1 0 0
adj I
Thus: I-1 = I
= 0 1 0
0 0 1

Note: Identity matrix (I) is its own inverse (i.e. I = I-1)

Solving a Square System

Suppose:
Ax = d

where the coefficient matrix A is an nxn square matrix and d is an nx1 column vector
where d ≠ 0. If A ≠ 0 (i.e. matrix A is non-singular and thus full rank) then A-1 exists and
the unique solution to the system is:

x* = A-1 d
Solutions of a Square System

Suppose:
Ax = d

where the coefficient matrix A is an nxn square matrix (i.e. Ax = d denotes a square
system with n-equations and n-unknown variables).

A = 0: (1) if d = 0 then many solutions exist, including: x* = 0

(2) if d ≠ 0 then either no solution exists or many solutions exist

A ≠ 0: (1) if d = 0 then a unique solution exists where: x* = 0

(2) if d ≠ 0 then a unique solution exists where: x* = A-1 d
Solutions of a Non-Square System

Suppose:
Ax = d

where the coefficient matrix A is an mxn non-square matrix (i.e. Ax = d denotes a non-
square system with m-equations and n-unknown variables).

m < n: underdetermined system; generally many solutions possible

m > n: overdetermined system; generally no solution possible

Cramer’s Rule

Given:
Ax = d

As we have already seen, if A ≠ 0 and d ≠ 0 then a unique solution exists where

x* = A-1 d

According to Cramer’s rule

Aj
xj* =
A

where Aj denotes the matrix A with the jth column replaced by the column vector d.