Advanced Microeconomics

Negative/positive (semi-)definite matrix and

bordered Hessian matrix

1.A. Negative/positive (semi-)definite matrix

The definiteness of matrices are related to the second order condition for the uncon-
strained problems. We have also encountered the definiteness of matrices for the proper-
ties of the Slutsky matrix.

Definition 1.A.1 (Negative Definite). A (symmetric) N × N matrix M is negative

definite if
yT M y < 0 (1)

for all non-zero y ∈ RN .

Definition 1.A.2 (Negative Semi-definite). A (symmetric) N × N matrix M is negative

semi-definite if
yT M y ≤ 0 (2)

for all y ∈ RN .
! $ ! $
"−2 1 0% " y1 %
" % " %
" % " %
Example 1. M = " 1
" −2 1 %
% is negative definite since for any non-zero y = "y %,
" 2%
" % " %
# & # &
0 1 −2 y3
we have
! $! $ ! $
"−2 1 0 % " y1 % " y1 %
' (" %" % ' (" %
" %" % " %
y T M y = y1 y2 " 1
y3 " −2 1 % "y %
% " 2% = −2y1 + y2 y1 − 2y2 + y3 y2 − 2y3 " %
" y2 %
" %" % " %
# &# & # &
0 1 −2 y3 y3
) *
= − y12 + (y1 − y2 )2 + (y2 − y3 )2 + y32 < 0.

This result is the negative of sum of squares, and therefore non-positive. Furthermore,
the result is zero only if y1 = y2 = y3 = 0 that is, when y is the zero vector. Therefore,
for any non-zero vector y, the result is always negative.

1
 Advanced Microeconomics

! $ ! $
"−1 1% " y1 %
Example 2. M = "
#
%
&is negative semi-definite since for any y = "
# &, we have
%
1 −1 y2
! $! $ ! $
' (
"−1 1 % " y1 % '
"y1 %
(
2
y T M y = y1 y2 "
#
%"
&#
%
& = −y1 + y2 y1 − y2 "
# & = −(y1 + y2 ) ≤ 0.
%
1 −1 y2 y2

This result is the!negative

$ of sum of squares, and therefore non-positive. When y1 = −y2 ,
" 1%
for example y = "
#
%,
& the result is 0.
−1

Note that a matrix M with all negative entries may not be negative definite. Example 3
illustrates the case where all entries in M is negative whereas M is not negative definite.
! $ ! $
"−1 −2% "−1%
Example 3. M = "
#
%
& is not negative definite since for y = "
#
%
& we have
−2 −1 1
! $! $ ! $
' ( ' (
"−1 −2% "−1% "−1%
y T M y = −1 1 "
#
%"
&#
%
& = −1 1 "
#
%
& = 2 > 0.
−2 −1 1 1

Similarly, we could define positive (semi-)definite matrices analogously.

Definition 1.A.3 (Positive Definite). A (symmetric) N ×N matrix M is positive definite

if
yT M y > 0 (3)

for all non-zero y ∈ RN .

Definition 1.A.4 (Positive Semi-definite). A (symmetric) N × N matrix M is positive

semi-definite if
yT M y ≥ 0 (4)

for all y ∈ RN .

Remark. A matrix that is not positive semi-definite and not negative semi-definite is
called indefinite.

There are various ways to check the definiteness of matrices. In Examples 1, 2 and
3, we have used the definition to check the definiteness. Below, we will introduce the

2
 Advanced Microeconomics

determinantal test for definiteness. Before discussing the general theorem, we need to
learn some new concepts.

Definition 1.A.5 (Principal Submatrix and Principal Minor). Let M be a N ×N matrix.

A k × k submatrix of M formed by deleting n − k rows and the same n − k columns of
M is called the k th order principal submatrix of M . The determinant of a principal
submatrix is called the k th order principal minor of M .
! $
"a11 a12 a13 %
" %
" %
Example 4. For a general 3 × 3 matrix M = "a
" 21 a22 %.
a23 %
" %
# &
a31 a32 a33

1. There is one 3rd order principal minor, namely, det M ;

2. There are three 2nd order principal minors, namely,

! $
"a11 a12 %
a) det "
#
%,
& formed by deleting the 3rd row and the 3rd column;
a21 a22
! $
"a11 a13 %
b) det "
#
%,
& formed by deleting the 2nd row and the 2nd column;
a31 a33
! $
"a22 a23 %
c) det "
#
%,
& formed by deleting the 1st row and the 1st column.
a32 a33

3. There are three 1st order principal minors, namely,

' (
a) det a11 , formed by deleting the 2nd and 3rd rows and colomns;
' (
b) det a22 , formed by deleting the 1st and 3rd rows and colomns;
' (
c) det a33 , formed by deleting the 1st and 2nd rows and colomns.

Definition 1.A.6 (Leading Principal Submatrix and Leading Principal Minor). Let M
be a N × N matrix. The k th order prinipal submatrix of M obtained by deleting the last
n − k rows and column of M is called the k th order leading principal submatrix of
M ; and its determinant is called the k th order leading principal minor of M .

3
 Advanced Microeconomics

Example 5. For the general 3 × 3 matrix in Example 4,

1. The 3rd order leading principal minor is det M ;

! $
"a11 a12 %
2. The 2nd order leading principal minor is det "
#
%;
&
a21 a22
' (
3. The 1 order leading principal minor is det a11 .
st

The following two theorems provide the algorithm for testing the definiteness of a sym-
metric matrix.

Theorem 1.A.1. Let M be an N × N symmetric matrix. Then

1. M is positive definite if and only if all its leading principal minors are positive;

2. M is negative definite if and only if all its leading principal minors of odd order are
negative; and all its leading principal minors of even order are positive.

Theorem 1.A.2. Let M be an N × N symmetric matrix. Then

1. M is positive semi-definite if and only if all its principal minors are non-negative;

2. M is negative semi-definite if and only if all its principal minors of odd order are
non-positive ; and all all its principal minors of even order are non-negative.

Remark. Please note that to check the semi-definiteness of matrices, we must unfortu-
nately check not only the leading principal minors, but all principal minors.

Remark. The second-order partial derivative matrix, Fxx , is called Hessian Matrix.

1.B. Bordered Hessian matrix

As is mentioned in class, we could use bordered Hessian matrix to check the second-order
condition.

Definition 1.B.1. The matrix

! $
" 0 −Gx %
" %
# &
−Gx T Fxx − λGxx

is called Bordered Hessian Matrix.

4
 Advanced Microeconomics

To check the second-order sufficient condition, we need to look at n − m of the bordered

Hessian’s leading principal minors. Intuitively, we can think of the m constraints as
reducing the problem to one with n − m free variables.1 The smallest minor we consider
consisting of the truncated first 2m + 1 rows and columns, the next consisting of the
truncated first 2m + 2 rows and columns, and so on, with the last being the determinant
of the entire bordered Hessian. A sufficient condition for a local maximum of F is that
the smallest minor has the same sign as (−1)m+1 and that the rest of the principal minors
alternate in sign. The result is summarized in Theorem 1.B.1 below.

Theorem 1.B.1 (Second-order Sufficient Condition for Constrained Maximization Prob-

lem). If the last n − m leading principal minors of the bordered Hessian matrix at the
proposed optimum x∗ is such that the smallest minor (the (2m + 1)th minor) has the same
sign as (−1)m+1 and the rest of the principal minors alternate in sign, then x∗ is the local
maximum of the constrained maximization problem.

Example 6. Consider the following maximization problem with three variables (n = 3)

and two constraints (m = 2):
max F (x, y, z) ≡ z
x,y,z

s.t. G1 (x, y, z) ≡ x + y + z = 12

G2 (x, y, z) ≡ x2 + y 2 − z = 0

The Lagrangian is L(x, y, z, λ, µ) = z + λ(12 − x − y − z) + µ(−x2 − y 2 + z).

The first-order necessary conditions are

∂L/∂x = −λ − 2µx = 0

∂L/∂y = −λ − 2µy = 0

∂L/∂z = 1 − λ + µ = 0

∂L/∂λ = 12 − x − y − z = 0

∂L/∂µ = −x2 − y 2 + z = 0

The stationary points are (x∗ , y ∗ , z ∗ , λ, µ) = (2, 2, 8, 45 , − 15 ) and (−3, −3, 18, 65 , 15 ).
1
For example, the maximization problem: maxx,y,z x + y 2 + z subject to x + y + z = 1 can be reduced
to maxx,y x + y 2 + (1 − x − y) with no constraint.

5
 Advanced Microeconomics

The bordered Hessian matrix is

! $ ! $
" 0 0 −G1x −G1y −G1z % " 0 0 −1 −1 −1%
" % " %
" % " %
" 0 0 −G2x −G2y 2%
−Gz % " 0 0 −2x −2y 1 %
" " %
" % " %
" % " %
"−G1
" x −G2x L11 L12 L13 %
% = "−1
" −2x −2µ 0 0 %
%
" % " %
" % " %
"−G1
" y −G2y L21 L22 L23 %
%
"−1
" −2y 0 −2µ 0 %
%
" % " %
# & # &
−G1z −G2z L31 L32 L33 −1 1 0 0 0

We need to check n − m = 1 leading principal minors, i.e., we only need to check

the determinant of the bordered Hessian. For local maximum, the sign requirement is
(−1)m+1 = (−1)3 < 0.

1. For the first proposed optimum (x∗ , y ∗ , z ∗ , λ, µ) = (2, 2, 8, 45 , − 15 ), the determinant

of the bordered Hessian is 20;
2. For the second proposed optimum (x∗ , y ∗ , z ∗ , λ, µ) = (−3, −3, 18, 65 , 15 ), the deter-
minant of the bordered Hessian is −20.

Thus, the 2nd proposed optimum (x∗ , y ∗ , z ∗ , λ, µ) = (−3, −3, 18, 65 , 15 ) is a local maximum.

Example 7. Consider the following maximization problem with three variables (n = 3)

and one constraint (m = 1):
max F (x, y, z) ≡ x + y + z
x,y,z

s.t. G1 (x, y, z) ≡ x2 + y 2 + z 2 = 3

The Lagrangian is L(x, y, z, λ) = x + y + z + λ(3 − x2 − y 2 − z 2 ).

The first-order necessary conditions are

∂L/∂x = 1 − 2λx = 0

∂L/∂y = 1 − 2λy = 0

∂L/∂z = 1 − 2λz = 0

∂L/∂λ = 3 − x2 − y 2 − z 2 = 0

The stationary points are (x∗ , y ∗ , z ∗ , λ) = (−1, −1, −1, − 12 ) and (1, 1, 1, 12 ).

6
 Advanced Microeconomics

The bordered Hessian matrix is

! $ ! $
" 0 −G1x −G1y −G1z % " 0 −2x −2y −2z %
" % " %
" % " %
"−G1
" L11 L12 L13 %
%
"−2x
" −2λ 0 0 %
%
=
x
" % " %
" % " %
"−G1
" y L21 L22 L23 %
%
"−2y
" 0 −2λ 0 %
%
" % " %
# & # &
−G1z L31 L32 L33 −2z 0 0 −2λ

We need to check n − m = 2 leading principal minors, i.e., the 3rd order and the entire
bordered Hessian. For local maximum, the sign requirement is (−1)m+1 = (−1)2 > 0 for
the 3rd order leading principal minor and < 0 for the entired bordered Hessian.

1. For the first proposed optimum (x∗ , y ∗ , z ∗ , λ) = (−1, −1, −1, − 12 ), the 3rd order
leading principal minor is −8 < 0 and the determinant of the bordered Hessian is
−12 < 0;
2. For the second proposed optimum (x∗ , y ∗ , z ∗ , λ) = (1, 1, 1, 12 ), the 3rd order leading
principal minor is 8 > 0 and the determinant of the bordered Hessian is −12 < 0.

Thus, the 2nd proposed optimum (x∗ , y ∗ , z ∗ , λ) = (1, 1, 1, 12 ) is a local maximum.