Addis Ababa University

College of Business and

Economics
Econ 605: Mathematics for
Economists
2020/21 AY

Tewodros Negash Kahsay (PhD)

Main Results on Linear Systems
• Consider the general linear system of 𝑚 simultaneous equations in n
unknowns:
𝑎11 𝑥1 + 𝑎12 𝑥2 + ⋯ + 𝑎1𝑛 𝑥𝑛 = 𝑏1
𝑎21 𝑥1 + 𝑎22 𝑥2 + ⋯ + 𝑎2𝑛 𝑥𝑛 = 𝑏2
𝑜𝑟 𝐴𝑥 = 𝑏 (𝟐)
⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯
𝑎𝑚1 𝑥1 + 𝑎𝑚2 𝑥2 + ⋯ + 𝑎𝑚𝑛 𝑥𝑛 = 𝑏𝑚
Where 𝐴 is the 𝑚 × 𝑛 coefficient matrix .
• Define a new 𝑚 × (𝑛 + 1) matrix 𝐴𝑏 that contains 𝐴 in the first 𝑛 columns
and 𝑏 in column 𝑛 + 1. That is,
𝑎11 𝑎12 ⋯ 𝑎1𝑛 𝑎11 𝑎12 ⋯ 𝑎1𝑛 𝑏1
𝑎21 𝑎22 ⋯ 𝑎2𝑛 𝑎21 𝑎22 ⋯ 𝑎2𝑛 𝑏2
𝐴= and 𝐴𝑏 =
⋮ ⋮ ⋱ ⋮ ⋮ ⋮ ⋱ ⋮ ⋮
𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 𝑏𝑚
Tewodros Negash (PhD), AAU, Department of Economics 2
• 𝐴𝑏 is called the augmented matrix of the system (𝟐)
• The relationship between the ranks of 𝐴 and 𝐴𝑏 is crucial in determining
whether system (𝟐) has a solution
• Since all the columns in 𝐴 occur in 𝐴𝑏 , 𝑟(𝐴𝑏 ) ≥ 𝑟(𝐴) and because 𝐴𝑏
contains only one more column than 𝐴, 𝑟(𝐴𝑏 ) cannot be greater than
𝑟 𝐴 + 1.
• A necessary and sufficient condition for a linear system of equations to be
consistent (have at least one solution) is that the rank of the coefficient
matrix is equal to the rank of the augmented matrix. That is,
𝐴𝑥 = 𝑏 has a solution ⇔ 𝑟 𝐴 = 𝑟(𝐴𝑏 )

Tewodros Negash (PhD), AAU, Department of Economics 3

• Suppose that system (2)has solutions with 𝑟 𝐴 = 𝑟 𝐴𝑏 = 𝑘
a) If 𝑘 < 𝑚, i.e. the common rank 𝑘 is less than the number of
equations 𝑚, then 𝑚 − 𝑘 equations are superfluous in the sense
that if we choose any subsystem of equations corresponding to 𝑘
linearly independent rows, then any solution of these 𝑘 equations
also satisfies the remaining 𝑚 − 𝑘 equations.
b) If 𝑘 < 𝑛, i.e. the common rank 𝑘 is less than the number of
unknows 𝑛, then there exist 𝑛 − 𝑘 variables that can be chosen
freely, whereas the remaining 𝑘 variables are uniquely determined
by the choices of these 𝑛 − 𝑘 free variables. The system then has
𝑛 − 𝑘 degrees if freedom.

Tewodros Negash (PhD), AAU, Department of Economics 4

Exercise: Determine whether the following system of equations has any
solutions and, if it has, find the number of degrees of freedom and solve the
system
𝑥1 + 𝑥2 − 2𝑥3 + 𝑥4 + 3𝑥5 = 1
2𝑥1 − 𝑥2 + 2𝑥3 + 2𝑥4 + 6𝑥5 = 2
3𝑥1 + 5𝑥2 −10𝑥3 −3𝑥4 − 9𝑥5 = 3
3𝑥1 + 2𝑥2 − 4𝑥3 −3𝑥4 − 9𝑥5 = 3
Solution:
1 1 −2 1 3 1 1 −2 1 3 1
2 −1 2 2 6 2 −1 2 2 6 2
𝐴= 𝐴𝑏 =
3 5 −10 −3 −9 3 5 −10 −3 −9 3
3 2 −4 −3 −9 3 2 −4 −3 −9 3

Tewodros Negash (PhD), AAU, Department of Economics 5

• All minors of order 4 in 𝐴𝑏 are equal to 0 (note that several pairs of columns
are proportional – 1 & 6; 2 & 3; 4 & 5 )
• There are minors of order 3 in 𝐴 that ae different from 0.
• For example, the minor formed by the first, third and fourth columns and by
the first, second and fourth rows is different from 0, i.e.
1 −2 1
2 2 2 = −36 ≠ 0 (∗)
3 −4 −3
• Thus, 𝑟 𝐴 = 𝑟 𝐴𝑏 = 3 so the system has solutions
• There is one superfluous equation. Since the first, second and fourth rows in
𝐴𝑏 are linearly independent, the third equation can be dropped.
• The number of variables is 5 and 𝑟 𝐴 = 𝑟 𝐴𝑏 = 3, there are 2 degrees of
freedom
Tewodros Negash (PhD), AAU, Department of Economics 6
• In the next step we find all the solutions to the system of equations.
• Since the determinant of (∗) is different from 0, we rewrite the subsystem of
there independent equations as
𝑥1 − 2𝑥3 + 𝑥4 + 𝑥2 + 3𝑥5 = 1
2𝑥1 + 2𝑥3 + 2𝑥4 − 𝑥2 + 6𝑥5 = 2 (∗∗)
3𝑥1 − 4𝑥3 −3𝑥4 + 2𝑥2 − 9𝑥5 = 3
• In matrix form
1 −2 1 𝑥1 1 3 1
𝑥2
𝑥3 + −1 6
2 2 2 𝑥5 = 2
3 −4 −3 𝑥4 2 −9 3
• The 3 × 3 coefficient matrix corresponding to 𝑥1 , 𝑥3 and 𝑥4 in (∗∗) has a
determinant different form 0, so it has an inverse.

Tewodros Negash (PhD), AAU, Department of Economics 7

 𝑥1 1 −2 1 −1 1 1 −2 1 −1
1 3 𝑥2
• 𝑥3 = 2 2 2 2 − 2 2 2 −1 6 𝑥5 (∗∗∗)
𝑥4 3 −4 −3 3 3 −4 −3 2 −9
1 −2 1 −1 −1 5 3
1
• 2 2 2 = −6 3 0 (∗∗∗∗)
18
3 −4 −3 7 1 −3
• Making use of (∗∗∗) and ∗∗∗∗ we have
𝑥1 1 0 1
1 1
• 𝑥3 = 0 − −2 𝑥2 = 2 𝑥2
𝑥4 0 3𝑥5 −3𝑥5
• Thus, if 𝑥2 , and 𝑥5 take any real values, there is a solution for 𝑥1 , 𝑥3 and 𝑥4
• This confirms that there are two degrees of freedom

Tewodros Negash (PhD), AAU, Department of Economics 8

1.3 Eigenvalues and eigenvectors
• Many applied problems, especially in dynamic economics, involve the
powers 𝑨𝑛 , 𝑛 = 1,2, ⋯ of a square matrix 𝑨
• If the dimension of 𝑨 is large and 𝒙 is a nonzero vector, computing 𝑨𝑛 𝒙 for
large 𝑛 (e.g., 𝐴100 𝑥 ) is a problem.
• But suppose there exist a scalar 𝜆 with the special property that
𝑨𝒙 = 𝜆𝒙 (⋆)
In this case we would have 𝑨2 𝒙 = 𝑨 𝑨𝒙 = 𝑨 𝜆𝒙 = 𝜆𝑨𝒙 = 𝜆𝜆𝒙 = 𝜆2 𝒙
• In general, 𝑨𝑛 𝒙 = 𝜆𝑛 𝒙
• Many of the properties of 𝑨 and 𝑨𝑛 can be deduced by finding the pairs
𝜆, 𝑥 , 𝑥 ≠ 0 that satisfies (⋆). Zero solutions are not interesting because
𝑨𝟎 = 𝜆𝟎 for every scalar 𝜆
• A nonzero vector 𝒙 that solves (⋆) is called an eigenvector (characteristic
vector), and the associated 𝜆 is called an eigenvalue (characteristic root).
Tewodros Negash (PhD), AAU, Department of Economics 9
Eigenvalues for Matrices of Order 2
𝑎11 𝑎12 𝑥1
• When 𝑛 = 2, we have 𝑨 = 𝑎 𝑎22 and 𝒙 = 𝑥2 . Then ⋆ reduces to
21

𝑎11 𝑎12 𝑥1 𝑥1 𝑎11 𝑥1 + 𝑎12 𝑥2 = 𝜆𝑥1

𝑎21 𝑎22 𝑥2 = 𝜆 𝑥2 or ቊ
𝑎21 𝑥1 + 𝑎22 𝑥2 = 𝜆𝑥2
• This system can be written as
(𝑎11 −𝜆)𝑥1 + 𝑎12 𝑥2 = 0
ൠ or in matrix form (𝑨 − 𝜆𝐈)𝒙 = 𝟎 (𝟏)
𝑎21 𝑥1 + (𝑎22 −𝜆)𝑥2 = 0
1 0
Where 𝐈 is an identity matrix of order 2, i.e. 𝐈 =
0 1
• The homogeneous system (𝟏) has a solution 𝒙 ≠ 𝟎 iff the coefficient matrix
has determinant equal to 0, i.e. iff 𝑨 − 𝜆𝐈 = 0
Tewodros Negash (PhD), AAU, Department of Economics 10
• Note: If the right hand side of an equation system consist of only zeros, so that it
can be written in matrix form as 𝑨𝒙 = 𝟎, the system is called homogeneous. A
homogeneous system will always have the trivial solution 𝑥1 = 𝑥2 = ⋯ = 𝑥𝑛 = 0
𝑨𝒙 = 𝟎 (where 𝑨 is square) has nontrivial solutions (𝒙 ≠ 𝟎) ⇔ 𝑨 = 𝟎
• 𝑨 − 𝜆𝐈 = 0 is known as eigenvalue equation (characteristic equation)
• Evaluating the 2 × 2 determinant, we get
𝑎11 − 𝜆 𝑎12
𝑨 − 𝜆𝐈 =
𝑎21 𝑎22 − 𝜆
= 𝜆2 − 𝑎11 + 𝑎22 𝜆 + 𝑎11 𝑎22 − 𝑎12 𝑎21 = 0 (𝟐)
= 𝜆2 − 𝑇𝑟 𝐴 𝜆 + 𝐴 = 0
• The eigenvalues are the solutions of this quadratic equation and the
𝑥1
• Eigenvectors are the nonzero vectors 𝑥 that satisfy system (𝟏)
2

Tewodros Negash (PhD), AAU, Department of Economics 11

Example: Find the eigenvalues and the associated eigenvectors of the matrices
1 2 −1 3
𝑎 A= 𝑏 𝐵=
3 0 2 0
1−𝜆 2
Solution: 𝑎 𝑨 − 𝜆𝐈 = = 𝜆2 − 𝜆 − 6 = 0. This yields the
3 −𝜆
eigenvalues 𝜆1 = −2 and 𝜆2 = 3 .
• Use (𝑨 − 𝜆𝐈)𝒙 = 𝟎 to find corresponding eigenvectors.
1 2 1 0 𝑥1 0
• For 𝜆 = 𝜆1 = −2, −𝜆 𝑥 = yields two same
3 0 0 1 2 0
equations, 3𝑥1 + 2𝑥2 = 0 and we take 𝑥1 = 2 and 𝑥2 = −3
1 2 1 0 𝑥1 0
• For 𝜆 = 𝜆2 = 3 , −𝜆 𝑥 = results in equations
3 0 0 1 2 0
−2𝑥1 + 2𝑥2 = 0 and 3𝑥1 − 3𝑥2 = 0 so that 𝑥1 = 1 and 𝑥2 = 1. The
2 1
corresponding eigenvectors for𝜆1 = −2 𝑎𝑛𝑑 𝜆2 = 3 are therefore, and ,
−3 1
respectively.
Tewodros Negash (PhD), AAU, Department of Economics 12
 𝑏 The eigenvalue equation (the characteristic polynomial) is
−1 − 𝜆 3
𝑨 − 𝜆𝐈 = = 𝜆2 + 𝜆 − 6 = 0.
2 −𝜆
This yields the eigenvalues 𝜆1 = 2 and 𝜆2 = −3 .
• To find the corresponding eigenvectors use (𝑨 − 𝜆𝐈)𝒙 = 𝟎
−1 3 2 0 𝑥1 0
• For 𝜆 = 𝜆1 = 2, − 𝑥 = yields two
2 0 0 2 2 0
equations, −3𝑥1 + 3𝑥2 = 0 and 2𝑥1 − 2𝑥2 = 0. In both cases we have
1
𝑥1 = 1 and 𝑥2 = 1 and the eigenvector corresponding to 𝜆1 = 2 is
1
−1 3 −3 0 𝑥1 0
• For 𝜆 = 𝜆2 = −3 , − = results in two
2 0 0 −3 𝑥2 0
• same equations 2𝑥1 + 3𝑥2 = 0 and we take 𝑥1 = 3 and 𝑥2 = −2.
3
and conclude the eigenvectors for 𝜆2 = −3 to be
−2
Tewodros Negash (PhD), AAU, Department of Economics 13
The general case
• In the general case where A is an 𝑛 × 𝑛 matrix:
If 𝐀 is an 𝑛 × 𝑛 matrix, then a scalar 𝜆 is an eigen value of A if there is a
nonzero vector 𝒙 in ℝ𝑛 such that
𝑨𝒙 = 𝜆𝒙 (𝟑)
Then 𝒙 is an eigenvector of 𝐀 associated with 𝜆
• How to find eigenvalues
• The eigenvalue equation (𝟑) can be written as
(𝑨 − 𝜆𝐈)𝒙 = 𝟎 (𝟒)
where 𝐈 denotes the identity matrix of order 𝑛
• The homogeneous linear system of equations has a solution 𝒙 ≠ 𝟎 iff the
determinant of the coefficient matrix is equal to 0, i.e. iff 𝑨 − 𝜆𝐈 = 0
Tewodros Negash (PhD), AAU, Department of Economics 14
• Letting 𝑝 𝜆 = A − λI where 𝐴 = 𝑎𝑖𝑗 , we have the equation
𝑛×𝑛
𝑎11 − 𝜆 𝑎12 ⋯ 𝑎1𝑛
𝑎21 𝑎22 − 𝜆 ⋯ 𝑎2𝑛
• 𝑝 𝜆 = 𝐴 − 𝜆𝐼 = =0 (5)
⋮ ⋮ ⋱ ⋮
𝑎𝑛1 𝑎𝑛2 ⋯ 𝑎𝑛𝑛 − 𝜆
• (5) is called the characteristic equation (eigenvalue equation) of A.
• The polynomial 𝑝 𝜆 is called the characteristic polynomial of A.
• Equation (4) can be written as
𝑎11 − 𝜆 𝑥1 + 𝑎12 𝑥2 + ⋯ + 𝑎1𝑛 𝑥𝑛 = 0
𝑎21 𝑥1 + 𝑎22 − 𝜆 𝑥2 + ⋯ + 𝑎2𝑛 𝑥𝑛 = 0 (6)
⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯
𝑎𝑛1 𝑥1 + 𝑎𝑛2 𝑥2 + ⋯ + 𝑎𝑛𝑛 − 𝜆 𝑥𝑛 = 0
• An eigenvector associated with 𝜆 is a nontrivial solution 𝑥1 , ⋯ , 𝑥𝑛 of 6 .
Tewodros Negash (PhD), AAU, Department of Economics 15
Example: Find all the eigenvalues of the matrices and also the eigenvectors
associated with the real eigenvalues
0 0 −2
𝐵= 0 7 0
1 0 −3
Solution:
0−𝜆 0 −2
𝐴 − 𝜆𝐼 = 0 7−𝜆 0 = (7 − 𝜆)(𝜆2 + 3𝜆 + 2)
1 0 −3 − 𝜆
Thus, 𝜆1 = 7, 𝜆2 = −1 and 𝜆3 = −2 are the eigenvalues.
0 0
• The eigenvectors are for 𝜆1 = 7 , 𝑥 = 1 ; for 𝜆2 = −1, 𝑥 = 1 , and
0 0
1
• For 𝜆3 = −2, 𝑥 = 0
1 Tewodros Negash (PhD), AAU, Department of Economics 16
• If 𝐴 is an 𝑛 × 𝑛 matrix with eigenvalues 𝜆1 , 𝜆2 , ⋯ , 𝜆𝑛 then
a) 𝐴 = 𝜆1 𝜆2 ⋯ 𝜆𝑛
b) 𝑡𝑟 𝐴 = 𝑎11 + 𝑎22 + ⋯ + 𝑎𝑛𝑛 = 𝜆1 + 𝜆2 + ⋯ 𝜆𝑛
• That is, if 𝐴 is an 𝑛 × 𝑛 matrix, the product of all the eigenvalues is equal
to the determinant of 𝐴 while the sum of all the eigenvalues is equal to the
trace of 𝐴.
• Exercise: For the following matrices, find the eigenvalues and the
eigenvectors that correspond to the eigenvalues.
1 −1 0
2 −7
a) b) −1 2 −1
3 −8
0 −1 1
2 0 0
c) 0 3 0
0 0 4
Tewodros Negash (PhD), AAU, Department of Economics 17
Solutions
2−𝜆 −7
a) = 0 → 𝜆2 + 6𝜆 + 5 → 𝜆1 = −1 𝑎𝑛𝑑 𝜆2 = −5
3 −8 − 𝜆
Using the eigenvalue equation (𝑨 − 𝜆𝐈)𝒙 = 𝟎
2 − (−1) −7 𝑥1 0 3 −7 𝑥1 0
For 𝜆1 = −1, 𝑥 = → 𝑥 =
3 −8 − (−1) 2 0 3 −7 2 0
7
This yields the corresponding eigen vector
3
1
For 𝜆2 = −5, the associate eigenvector is
1
b)The eigenvalues are 𝜆1 = 0, 𝜆2 = 1 𝑎𝑛𝑑 𝜆3 = 3. The corresponding
1 −1 1
eigenvectors are 1 , 0 and −2 , respectively
1 1 1
Tewodros Negash (PhD), AAU, Department of Economics 18
c)The eigenvalues are 𝜆1 = 2, 𝜆2 = 3 𝑎𝑛𝑑 𝜆3 = 4.
Note: the eigenvalues of a diagonal matrix are the diagonal
elements.
1 0 0
The corresponding eigenvectors are 0 , 1 and 0 , respectively.
0 0 1

Tewodros Negash (PhD), AAU, Department of Economics 19

1.4 Diagonalization
• Let 𝐴 and 𝑃 be 𝑛 × 𝑛 matrices with 𝑃 invertible. Then
𝐴 and𝑃−1 𝐴𝑃 have the same eigenvalues (1)
• This is because the two matrices have the same characteristic polynomial:
𝑃−1 𝐴𝑃 − 𝜆𝐼 = 𝑃−1 𝐴 − 𝜆𝐼 𝑃 = 𝑃−1 𝐴 − 𝜆𝐼 𝑃 = 𝐴 − 𝜆𝐼
−1 1
→ 𝑃 = ⃒
𝑃
An 𝑛 × 𝑛 matrix 𝐴 is diagonalizable if there exists an invertible 𝑛 × 𝑛 matrix
𝑃 and a diagonal matrix 𝐷 such that
𝑃−1 𝐴𝑃 = D (2)
• Since the eigenvalues of a diagonal matrix are the diagonal elements, if 𝐴 is
diagonalizable, so that (2) holds, then 𝑃−1 𝐴𝑃 = diag(𝜆1 , ⋯ , 𝜆𝑛 ) where
𝜆1 , ⋯ , 𝜆𝑛 are the eigenvalues of 𝐴.
Tewodros Negash (PhD), AAU, Department of Economics 20
• Any square matrix hat has only nonzero entries on the main diagonal and
zeros everywhere else is known as a diagonal matrix.
• Two questions:
• Which square matrices are diagonalizable?
• If 𝐴 is diagonalizable, how do we find the matrix 𝑃 in (2)?
• The answer is:
An 𝑛 × 𝑛 matrix 𝐴 is diagonalizable iff it has a set of 𝑛 linearly
independent eigenvectors 𝑥1 , 𝑥2 , ⋯ , 𝑥𝑛 . In that case,
𝑃−1 𝐴𝑃 = diag(𝜆1 , ⋯ , 𝜆𝑛 ) (3)
where 𝑃 is the matrix with 𝑥1 , 𝑥2 , ⋯ , 𝑥𝑛 as its columns and 𝜆1 , ⋯ , 𝜆𝑛
are the corresponding eigenvalues
1 2
Example: (a) Verify (3) for A =
3 0
Tewodros Negash (PhD), AAU, Department of Economics 21
Solution: as shown in the solution to example (a) in section 1.3, the
eigenvalues of A are 𝜆1 = −2 and 𝜆2 = 3
2 1
• And the corresponding eigenvectors are and
−3 1
• These eigenvectors constitute the column vectors of for the matrix P, i.e.
1Τ − 1Τ
2 1
P= so that 𝑃−1 = 3 5 2 5
−3 1 Τ5 Τ5
• Direct multiplications shows that 𝑃−1 𝐴𝑃 = diag(𝜆1 , 𝜆2 )
1Τ − 1Τ5
5 1 2 2 1 −2 0
3Τ 2Τ =
5 5 3 0 −3 1 0 3
• Thus, (3) is verified

Tewodros Negash (PhD), AAU, Department of Economics 22

• Exercise: Verify 3 i.e. 𝑃−1 𝐴𝑃 = diag(𝜆1 , ⋯ , 𝜆𝑛 ) for
−1 3
a) 𝐴 =
2 0

Tewodros Negash (PhD), AAU, Department of Economics 23

1.5 Quadratic Forms
• Many applications of mathematics make use of a special kind of function
called quadratic form
• A general quadratic form in two variables 𝑥1 and 𝑥2 is
𝑄 𝑥1 , 𝑥2 = 𝑎11 𝑥12 + 𝑎12 𝑥1 𝑥2 + 𝑎21 𝑥2 𝑥1 + 𝑎22 𝑥22
• The above expression can be written as
𝑄 𝑥1 , 𝑥2 = 𝑎11 𝑥12 + 𝑎21 𝑥2 𝑥1 + 𝑎12 𝑥1 𝑥2 + 𝑎22 𝑥22 (1)
𝑎11 𝑥1 𝑎12 𝑥2
𝑥1 𝑥2 + 𝑥1 𝑥2
𝑎21 𝑥1 𝑎22 𝑥2
𝑎11 𝑎12
𝑥1 𝑥2
𝑎21 𝑥1 + 𝑎22 𝑥2
Tewodros Negash (PhD), AAU, Department of Economics 24
• Finally the expression in (1) can be written as
𝑎11 𝑎12 𝑥1
Q 𝑥1 , 𝑥2 = 𝑥1 𝑥2 𝑎 (2)
21 𝑎22 𝑥2
• Since 𝑥1 𝑥2 = 𝑥2 𝑥1 , the cross product term in (1) can be written as
𝑎12 𝑥1 𝑥2 + 𝑎21 𝑥2 𝑥1 = 𝑎12 + 𝑎21 𝑥1 𝑥2
• Assuming 𝑎12 = 𝑎21 , the matrix 𝑎𝑖𝑗 in (2) becomes symmetric and
2×2
(1) can be written as
𝑄 𝑥1 , 𝑥2 = 𝑎11 𝑥12 + 2𝑎12 𝑥1 𝑥2 + 𝑎22 𝑥22 (3)
• We are often interested in conditions on the coefficient 𝑎11 , 𝑎12 and 𝑎22
ensuring that Q 𝑥1 , 𝑥2 in (3) has the same sign for all 𝑥1 , 𝑥2 .

Tewodros Negash (PhD), AAU, Department of Economics 25

• For all 𝑥1 , 𝑥2 ≠ 0,0 , the quadratic form 𝑄 𝑥1 , 𝑥2 and its associated
symmetric matrix in (2) are called:
2
• Positive definite 𝑄 𝑥1 , 𝑥2 > 0 ⇔ 𝑎11 > 0 𝑎𝑛𝑑 𝑎11 𝑎22 − 𝑎12 >0
• Positive semidefinite 𝑄 𝑥1 , 𝑥2 ≥ 0 ⇔ 𝑎11 ≥ 0
2
𝑎𝑛𝑑 𝑎11 𝑎22 − 𝑎12 ≥0
2
• Negative definite 𝑄 𝑥1 , 𝑥2 < 0 ⇔ 𝑎11 < 0 𝑎𝑛𝑑 𝑎11 𝑎22 − 𝑎12 >0
• Negative semidefinite 𝑄 𝑥1 , 𝑥2 ≤ 0 ⇔ 𝑎11 ≤ 0
2
𝑎𝑛𝑑 𝑎11 𝑎22 − 𝑎12 ≥0
• The quadratic form 𝑄 𝑥1 , 𝑥2 is indefinite if there exist vectors 𝑥1∗ , 𝑥2∗ and
𝑦1∗ , 𝑦2∗ such that 𝑥1∗ , 𝑥2∗ < 0 and 𝑦1∗ , 𝑦2∗ > 0
• Thus an indefinite quadratic form assumes both negative and positive values

Tewodros Negash (PhD), AAU, Department of Economics 26

Example: Determine the definiteness of the following quadratic forms:
a) 𝑄 𝑥1 , 𝑥2 = 5𝑥12 − 2𝑥1 𝑥2 + 𝑥22
b) 𝑄 𝑥1 , 𝑥2 = −4𝑥12 + 12𝑥1 𝑥2 − 9𝑥22
Solution:
a) 𝑎11 = 5; 𝑎12 = −1 (𝑛𝑜𝑡 − 2) and 𝑎22 = 1
2
Thus, 𝑎11 > 0 and 𝑎11 𝑎22 − 𝑎12 =5−1=4>0
⇒ 𝑄 𝑥1 , 𝑥2 is positive definite
𝑏) 𝑎11 = −4; 𝑎12 = 6 (𝑛𝑜𝑡 12) and 𝑎22 = −9
2
Thus, 𝑎11 ≤ 0 and 𝑎11 𝑎22 − 𝑎12 = 36 − 36 = 0 ≥ 0
⇒ 𝑄 𝑥1 , 𝑥2 is negative semidefinite
Exercise: Determine the definiteness of the following quadratic forms
(a) 𝑄1 𝑥1 , 𝑥2 = 𝑥12 + 𝑥22 (b) 𝑄2 𝑥1 , 𝑥2 = −𝑥12 − 𝑥22
(c) 𝑄3 𝑥1 , 𝑥2 = 𝑥1 − 𝑥2 2 (d) 𝑄4 𝑥1 , 𝑥2 = − 𝑥1 − 𝑥2 2
(e) 𝑄5 𝑥1 , 𝑥2 = 𝑥12 − 𝑥22

Tewodros Negash (PhD), AAU, Department of Economics 27

The General Case
• A quadratic form in 𝒏 variables is a function 𝑸 of the form
𝑛 𝑛

𝑄 𝑥1 , ⋯ , 𝑥𝑛 = ෍ ෍ 𝑎𝑖𝑗 𝑥𝑖 𝑥𝑗 = 𝑎11 𝑥12 + 𝑎12 𝑥1 𝑥2 + ⋯ + 𝑎𝑖𝑗 𝑥𝑖 𝑥𝑗 + ⋯ + 𝑎𝑛𝑛 𝑥𝑛2 (4)

𝑖=1 𝑗=1
• Where the 𝑎𝑖𝑗 are constants.
• 𝑄 is homogeneous of degree two.
• Each term either contains the square of a variable or a product of
exactly two variables.
• We can see the structure of the quadratic form better if we write it as
follows
𝑄 𝑥1 , ⋯ , 𝑥𝑛 = 𝑎11 𝑥12 + 𝑎12 𝑥1 𝑥2 + ⋯ + 𝑎1𝑛 𝑥1 𝑥𝑛
+𝑎21 𝑥2 𝑥1 + 𝑎22 𝑥22 + ⋯ + 𝑎2𝑛 𝑥2 𝑥𝑛 (5)
⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯
+𝑎𝑛1 𝑥𝑛 𝑥1 + 𝑎𝑛2 𝑥𝑛 𝑥2 + ⋯ + 𝑎𝑛𝑛 𝑥𝑛2
Tewodros Negash (PhD), AAU, Department of Economics 28
• Putting𝒙 = 𝑥1 , 𝑥2 , ⋯ , 𝑥𝑛 𝑎𝑛𝑑 𝑨 = 𝑎𝑖𝑗
𝑛×𝑛
𝑄 𝑥1 , ⋯ , 𝑥𝑛 = 𝑄 𝒙 = 𝒙′ 𝑨𝒙 (6)
• Assuming 𝑎𝑖𝑗= 𝑎𝑗𝑖 for all 𝑖 𝑎𝑛𝑑 𝑗 as in the case of 𝑛 = 2, the matrix 𝑨 in (6)
becomes symmetric.
• Then 𝑨 is called the symmetric matrix associated with 𝑄 and 𝑄 is called a
symmetric quadratic form.
Example: Write 𝑄 𝑥1 , 𝑥2 , 𝑥3 = 3𝑥12 + 6𝑥1 𝑥3 + 𝑥22 − 4𝑥2 𝑥3 + 8𝑥32 in matrix
form with 𝑨 symmetric.
Solution: first write 𝑄 as follows:
𝑄 = 3𝑥12 + 0𝑥1 𝑥2 + 3𝑥1 𝑥3 + 0𝑥2 𝑥1 + 𝑥22 − 2𝑥2 𝑥3 + 3𝑥3 𝑥1 − 2𝑥3 𝑥2 + 8𝑥32

Tewodros Negash (PhD), AAU, Department of Economics 29

 3 0 3 𝑥1
• Then 𝑄 𝒙 = 𝒙′ 𝑨𝒙 = 𝑥1 𝑥2 𝑥3 0 1 −2 𝑥2
3 −2 8 𝑥3
3 0 3 𝑥1
where 𝑨 = 0 1 −2 and 𝒙 = 𝑥2
3 −2 8 𝑥3
Definiteness of a Quadratic Form
• A quadratic form 𝑄 𝒙 = 𝒙′ 𝑨𝒙 as well as its associated symmetric matrix 𝑨 are
said to be positive definite, positive semidefinite, negative definite or negative
semidefinite, respectively, if for all 𝒙 ≠ 𝟎
𝑄 𝒙 > 0, 𝑄 𝒙 ≥ 0, 𝑄 𝒙 < 0, and 𝑄 𝒙 ≤ 0 (7)
• The quadratic form 𝑄 𝒙 is indefinite if there exists vectors 𝒙∗ 𝑎𝑛𝑑 𝒚∗ and such
that 𝑄 𝒙∗ < 0 and 𝑄 𝒚∗ > 0.
• That is, an indefinite quadratic form assumes both negative and positive values.
Tewodros Negash (PhD), AAU, Department of Economics 30
Principal minor and leading principal minor
• To study the sign of quadratic forms we need some particular minors
• An arbitrary principal minor of order 𝑟 of an 𝑛 × 𝑛 matrix 𝐴 = 𝑎𝑖𝑗 is
𝑛×𝑛
the determinant of a matrix obtained by deleting 𝑛 − 𝑟 rows and the
corresponding 𝑛 − 𝑟 columns such that if the 𝑖 𝑡ℎ row (column) is selected so
is the 𝑖 𝑡ℎ column (row)
• In particular, a principal minor of order 𝑟 always includes exactly 𝑟
elements of the main (principal) diagonal
• The determinant 𝐴 itself with no rows or columns deleted is also a
principal minor.
• If the matrix 𝐴 is symmetric, so is each matrix whose determinant is a
principal minor.
• A principal minor is called a leading principal minor of order 𝒓(1 ≤ 𝑟 ≤ 𝑛) if
it consists of the first (leading) 𝒓 rows and columns of 𝐴 .
Tewodros Negash (PhD), AAU, Department of Economics 31
Example: Write down the principal and the leading principal minors of
𝑏11 𝑏12 𝑏13
𝑎11 𝑎12
a) 𝐴 = 𝑎 𝑏) 𝐵 = 𝑏21 𝑏 22 𝑏 23
21 𝑎22
𝑏31 𝑏32 𝑏33
Solution:
a) Deleting row 2 and column 2 in 𝐴 to yields (𝑎11 ), which has determinant 𝑎11 .
Deleting row 1 and column 1 in 𝐴 we get (𝑎22 ) which has determinant 𝑎22 .
The principal minors of 𝐴 are therefore 𝑎11 , 𝑎22 and 𝐴 . The leading principal
minors are 𝑎11 and 𝐴 .
b) The principal minors of 𝐵 are
𝑏11 𝑏12 𝑏11 𝑏13 𝑏22 𝑏23
𝑏11 , 𝑏22 , 𝑏33 , , , and 𝐵 .
𝑏21 𝑏22 𝑏31 𝑏33 𝑏32 𝑏33
𝑏11 𝑏12
The leading principal minors are 𝑏11 , and 𝐵
𝑏21 𝑏22
Tewodros Negash (PhD), AAU, Department of Economics 32
• Suppose 𝐴 is an arbitrary 𝑛 × 𝑛 matrix. The leading principal minors of 𝐴 are
𝑎11 𝑎12 ⋯ 𝑎1𝑘
𝑎21 𝑎22 ⋯ 𝑎2𝑘
𝐷𝑘 = , 𝑘 = 1, 2, ⋯ , 𝑛 (8)
⋮ ⋮ ⋱ ⋮
𝑎𝑘1 𝑎𝑘2 ⋯ 𝑎𝑘𝑘
• Consider the symmetric matrix 𝐴 and the associated quadratic form
𝑄 𝒙 = σ𝑛𝑖=1 σ𝑛𝑗=1 𝑎𝑖𝑗 𝑥𝑖 𝑥𝑗 𝑎𝑖𝑗 = 𝑎𝑗𝑖
• Let 𝐷𝑘 be the leading principal minor and let ∆𝑘 denote an arbitrary principal
minor of 𝐴 of order 𝒌. Then
• 𝑄 is positive definite ⇔ 𝐷𝑘 > 0 for 𝑘 = 1, ⋯ , 𝑛
∆𝑘 ≥ 0 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑝𝑟𝑖𝑛𝑐𝑖𝑝𝑎𝑙
• 𝑄 is positive semidefinite ⇔ ቊ
𝑚𝑖𝑛𝑜𝑟𝑠 𝑜𝑓 𝑜𝑟𝑑𝑒𝑟 𝑘 = 1, ⋯ , 𝑛
• 𝑄 is negative definite ⇔ −1 𝑘 𝐷𝑘 > 0 for 𝑘 = 1, ⋯ , 𝑛
−1 𝑘 ∆𝑘 ≥ 0 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑝𝑟𝑖𝑛𝑐𝑖𝑝𝑎𝑙
• 𝑄 is negative semidefinite ⇔ ൝
𝑚𝑖𝑛𝑜𝑟𝑠 𝑜𝑓 𝑜𝑟𝑑𝑒𝑟 𝑘 = 1, ⋯ , 𝑛
Tewodros Negash (PhD), AAU, Department of Economics 33
Example: Use leading principal minors to determine the definiteness of
a) 𝑄 = 3𝑥12 + 6𝑥1 𝑥3 + 𝑥22 − 4𝑥2 𝑥3 + 8𝑥32
b) 𝑄 = −𝑥12 + 6𝑥1 𝑥2 − 9𝑥22 − 2𝑥32
Solution: Rewrite Q as
𝑄 = 3𝑥12 + 0𝑥1 𝑥2 + 3𝑥1 𝑥3 + 0𝑥2 𝑥1 + 𝑥22 − 2𝑥2 𝑥3 + 3𝑥3 𝑥1 − 2𝑥3 𝑥2 + 8𝑥32
3 0 3
𝑨= 0 1 −2
3 −2 8
• Check the leading principal minors first, in case the matrix turns out to be
definite rather than semidefinite.
3 0 3
3 0
𝐷1 = 3, 𝐷2 = = 3 and 𝐷3 = 0 1 −2 = 3
0 1
3 −2 8
With 𝐷𝑘 > 0 for 𝑘 = 1, 2, 3, we conclude that 𝑄is positive definite.
Tewodros Negash (PhD), AAU, Department of Economics 34
b) Rewrite Q as
𝑄 = −𝑥12 + 3𝑥1 𝑥2 + 0𝑥1 𝑥3 + 3𝑥2 𝑥1 − 9𝑥22 + 0𝑥2 𝑥3 + 0𝑥3 𝑥1 + 0𝑥3 𝑥2 − 2𝑥32
−1 3 0
𝑨= 3 −9 0
0 0 −2
• Check the leading principal minors first
−1 3 0
−1 3
𝐷1 = −1, 𝐷2 = = 0 and 𝐷3 = 3 −9 0 =0
3 −9
0 0 −2
• The conditions for positive definite, positive semidefinite and negative
definite are not satisfied.
• To check the conditions for negative semidefinite, we examine all the principal
minors
Tewodros Negash (PhD), AAU, Department of Economics 35
 (1) (2)
• The 3 × 3 matrix 𝐴 has three principal minors of order 1, whose values ∆1 , ∆1 ,
(3)
and ∆1 are the diagonal elements.
1 (1) 1 (2)
(−1) ∆1 = −1 −1 = 1, (−1) ∆1 = −1 −9 = 9, and
(3)
(−1)1 ∆1 = −1 −2 = 2
• There are also three second – order principal minors
2 (1) −1 3 2 (2) −1 0
(−1) ∆2 = = 1 0 = 0, (−1) ∆2 = = 1 2 = 2 and
3 −9 0 −2
2 (3) −9 0
(−1) ∆2 = = 1 18 = 18
0 −2
−1 3 0
• Finally, (−1)3 ∆3 = 3 −9 0 = −1 0 = 0
0 0 −2
• Thus(−1)𝑘 ∆𝑘 ≥ 0 for all principal minors ∆𝑘 of order 𝑘(𝑘 = 1, 2, 3) in the matrix 𝑨
• Therefore, we conclude 𝑄 is negative semidefinite.
Tewodros Negash (PhD), AAU, Department of Economics 36
• The definiteness of a quadratic form can be determined easily from the
sighs of the eigenvalues of the associated matrix
• Let 𝑄 𝒙 = 𝒙′ 𝑨𝒙 be a quadratic form where the matrix 𝑨 is symmetric, and
let 𝜆1 , 𝜆2 , ⋯ , 𝜆𝑛 be the eigenvalues of 𝑨 . Then
a) 𝑄 is positive definite ⇔ 𝜆1 > 0, 𝜆2 > 0, ⋯ , 𝜆𝑛 > 0
b) 𝑄 is positive semidefinite ⇔ 𝜆1 ≥ 0, 𝜆2 ≥ 0, ⋯ , 𝜆𝑛 ≥ 0
c) 𝑄 is negative definite ⇔ 𝜆1 < 0, 𝜆2 < 0, ⋯ , 𝜆𝑛 < 0
d) 𝑄 is negative semidefinite ⇔ 𝜆1 ≤ 0, 𝜆2 ≤ 0, ⋯ , 𝜆𝑛 ≤ 0
e) 𝑄 is positive definite ⇔ 𝑨 has both positive and negative eigenvalues
Example: use the signs of eigenvalues to determine the definiteness of the
quadratic form in example b above.
Tewodros Negash (PhD), AAU, Department of Economics 37
• Given the quadratic form 𝑄 = −𝑥12 + 6𝑥1 𝑥2 − 9𝑥22 − 2𝑥32 , the
associated matrix 𝑨 is
−1 3 0
𝑨= 3 −9 0
0 0 −2
• Matrix 𝑨 has characteristic equation −𝜆 𝜆 + 2 𝜆 + 10 = 0
So the eigenvalues are 0, −2, 𝑎𝑛𝑑 − 10.
• With all the eigenvalues ≤ 0, the quadratic form is negative
semidefinite.
Exercise: use the signs of eigenvalues to determine the definiteness of
the quadratic form whose associated matrix 𝑨 is given by
−5 1 2
𝑨= 0 −2 0
4 2 −3
Tewodros Negash (PhD), AAU, Department of Economics 38