Microeconomic Theory

Basic Principles and Extensions

Chapter 2
THE MATHEMATICS OF
OPTIMIZATION

Copyright ©2005 by South-Western, a division of Thomson Learning. All rights reserved. 1

The Mathematics of Optimization
• Many economic theories begin with the
assumption that an economic agent is
seeking to find the optimal value of some
function
– consumers seek to maximize utility
– firms seek to maximize profit
• This chapter introduces the mathematics
common to these problems
2
Maximization of a Function of
One Variable
• Simple example: Manager of a firm
wishes to maximize profits
  f (q )

Maximum profits of
* * occur at q*
 = f(q)

Quantity
q*
3
 Maximization of a Function of
One Variable
• The manager will likely try to vary q to see
where the maximum profit occurs
– an increase from q1 to q2 leads to a rise in 


0
*
2  = f(q)
q
1

Quantity
q1 q2 q*
4
 Maximization of a Function of
One Variable
• If output is increased beyond q*, profit will
decline
– an increase from q* to q3 leads to a drop in 



0
*

 = f(q)
q
3

Quantity
q* q3
5
 Derivatives
• The derivative of  = f(q) is the limit of
/q for very small changes in q
d df f ( q1  h)  f ( q1 )
  lim
dq dq h 0 h

• The value of this ratio depends on the

value of q1

6
Value of a Derivative at a Point
• The evaluation of the derivative at the
point q = q1 can be denoted
d
dq q q1

• In our previous example,

d d d
0 0 0
dq q q
1
dq q q
3
dq q q *
7
 First Order Condition for a
Maximum
• For a function of one variable to attain
its maximum value at some point, the
derivative at that point must be zero

df
0
dq q q *

8
 Second Order Conditions
• The first order condition (d/dq) is a
necessary condition for a maximum, but
it is not a sufficient condition

If the profit function was u-shaped,
the first order condition would result
in q* being chosen and  would
be minimized

*

Quantity
q*
9
 Second Order Conditions
• This must mean that, in order for q* to
be the optimum,
d d
 0 for q  q * and  0 for q  q *
dq dq
• Therefore, at q*, d/dq must be
decreasing

10
 Second Derivatives
• The derivative of a derivative is called a
second derivative
• The second derivative can be denoted
by
d 2 d 2f
2
or 2
or f " (q )
dq dq

11
 Second Order Condition
• The second order condition to represent
a (local) maximum is
2
d 
2
f " (q ) q q *  0
dq q q *

12
Rules for Finding Derivatives
db
1. If b is a constant, then 0
dx
d [ bf ( x )]
2. If b is a constant, then  bf ' ( x )
dx

dx b
3. If b is constant, then bx b  1
dx
d ln x 1
4. 
dx x 13
Rules for Finding Derivatives

da x x
5. a ln a for any constant a
dx
– a special case of this rule is dex/dx = ex

14
Rules for Finding Derivatives
• Suppose that f(x) and g(x) are two
functions of x and f’(x) and g’(x) exist
• Then
d [f ( x )  g ( x )]
6. f ' ( x )  g ' ( x )
dx

d [f ( x ) g ( x )]
7.  f ( x )g ' ( x )  f ' ( x )g ( x )
dx
15
Rules for Finding Derivatives
 f (x) 
d  
 g ( x )  f ' ( x )g ( x )  f ( x )g ' ( x )
8.  2
dx [g ( x )]
provided that g ( x ) 0

16
Rules for Finding Derivatives
• If y = f(x) and x = g(z) and if both f’(x)
and g’(x) exist, then:
dy dy dx df dg
9.    
dz dx dz dx dz
• This is called the chain rule. The chain
rule allows us to study how one variable
(z) affects another variable (y) through
its influence on some intermediate
variable (x) 17
 Rules for Finding Derivatives
• Some examples of the chain rule
include
de ax de ax d (ax )
10.   e ax a ae ax
dx d (ax ) dx
d [ln( ax )] d [ln( ax )] d (ax )
11.   ln( ax ) a a ln( ax )
dx d (ax ) dx
d [ln( x 2 )] d [ln( x 2 )] d ( x 2 ) 1 2
12.  2
  2 2 x 
dx d(x ) dx x x18
Example of Profit Maximization
• Suppose that the relationship between
profit and output is
 = 1,000q - 5q2
• The first order condition for a maximum is
d/dq = 1,000 - 10q = 0
q* = 100
• Since the second derivative is always
-10, q = 100 is a global maximum
19
Functions of Several Variables
• Most goals of economic agents depend
on several variables
– trade-offs must be made
• The dependence of one variable (y) on
a series of other variables (x1,x2,…,xn) is
denoted by
y  f ( x1 , x 2 ,..., x n )
20
 Partial Derivatives
• The partial derivative of y with respect
to x1 is denoted by
y f
or or fx or f1
x1 x1 1

• It is understood that in calculating the

partial derivative, all of the other x’s are
held constant
21
 Partial Derivatives
• A more formal definition of the partial
derivative is
f f ( x1  h, x 2 ,..., x n )  f ( x1, x 2 ,..., x n )
lim
x1 x 2 ,...,x n
h 0 h

22
Calculating Partial Derivatives
1. If y f ( x1, x 2 ) ax12  bx1 x 2  cx 22 , then
f
f1 2ax1  bx 2 and
x1
f
f2 bx1  2cx 2
x 2
ax1  bx 2
2. If y  f ( x1, x 2 )  e , then
f ax  bx f ax  bx
 f1  ae 1 2
and  f2  be 1 2

x1 x 2
23
Calculating Partial Derivatives
3. If y  f ( x1, x 2 )  a ln x1  b ln x 2 , then
f a f b
 f1  and  f2 
x1 x1 x 2 x2

24
 Partial Derivatives
• Partial derivatives are the mathematical
expression of the ceteris paribus
assumption
– show how changes in one variable affect
some outcome when other influences are
held constant

25
Second-Order Partial Derivatives
• The partial derivative of a partial
derivative is called a second-order
partial derivative

( f / x i )  2f
 fij
x j x j x i

26
 Young’s Theorem
• Under general conditions, the order in
which partial differentiation is conducted
to evaluate second-order partial
derivatives does not matter

fij  f ji

27
Use of Second-Order Partials
• Second-order partials play an important
role in many economic theories
• One of the most important is a
variable’s own second-order partial, fii
– shows how the marginal influence of xi on
y(y/xi) changes as the value of xi
increases
– a value of fii < 0 indicates diminishing
marginal effectiveness
28
 Total Differential
• Suppose that y = f(x1,x2,…,xn)
• If all x’s are varied by a small amount,
the total effect on y will be
f f f
dy  dx 1  dx 2  ...  dx n
x1 x 2 x n

dy  f1dx 1  f2 dx 2  ...  fn dx n

29
 First-Order Condition for a
Maximum (or Minimum)
• A necessary condition for a maximum (or
minimum) of the function f(x1,x2,…,xn) is
that dy = 0 for any combination of small
changes in the x’s
• The only way for this to be true is if
f1 f2 ... fn 0
• A point where this condition holds is
called a critical point 30
 Finding a Maximum
• y represents an individual’s health, x1
and x2 are daily dosages of two drugs
• Suppose that y is a function of x1 and x2
y = - (x1 - 1)2 - (x2 - 2)2 + 10
y = - x12 + 2x1 - x22 + 4x2 + 5
• First-order conditions imply that
y
  2 x1  2  0 x1* 1
x1 OR
y x 2*  2
  2 x2  4  0 31
x 2
 Implicit Function Theorem
• It may not always be possible to solve
implicit functions of the form g(x,y)=0 for
unique explicit functions of the form y = f(x)

32
 The Envelope Theorem
• The envelope theorem concerns how the
optimal value for a particular function
changes when a parameter of the function
changes
• This is easiest to see by using an example

33
 The Envelope Theorem
• Suppose that y is a function of x
y = -x2 + ax
• For different values of a, this function
represents a family of inverted parabolas
• If a is assigned a specific value, then y
becomes a function of x only and the value
of x that maximizes y can be calculated

34
 The Envelope Theorem
Optimal Values of x and y for alternative values of a

Value of a Value of x* Value of y*

0 0 0
1 1/2 1/4
2 1 1
3 3/2 9/4
4 2 4
5 5/2 25/4
6 3 9
35
 The Envelope Theorem
y*
10

9
As a increases,
8 the maximal value
7 for y (y*) increases
6

4
The relationship
3

2
between a and y
1
is quadratic
0 a
0 1 2 3 4 5 6 7

36
 The Envelope Theorem
• Suppose we are interested in how y*
changes as a changes
• There are two ways we can do this
– calculate the slope of y directly
– hold x constant at its optimal value and
calculate y/a directly

37
 The Envelope Theorem
• To calculate the slope of the function, we
must solve for the optimal value of x for
any value of a
dy/dx = -2x + a = 0
x* = a/2
• Substituting, we get
y* = -(x*)2 + a(x*) = -(a/2)2 + a(a/2)
y* = -a2/4 + a2/2 = a2/4
38
 The Envelope Theorem
• Therefore,
dy*/da = 2a/4 = a/2 = x*

• But, we can save time by using the

envelope theorem
– for small changes in a, dy*/da can be
computed by holding x at x* and calculating
y/ a directly from y

39
 The Envelope Theorem
y/ a = x
• Holding x = x*

y/ a = x* = a/2

• This is the same result found earlier

40
 The Envelope Theorem
• The envelope theorem states that the
change in the optimal value of a function
with respect to a parameter of that function
can be found by partially differentiating the
objective function while holding x (or
several x’s) at its optimal value

dy * y
 {x  x * ( a)}
da a
41
 The Envelope Theorem
• The envelope theorem can be extended to
the case where y is a function of several
variables
y = f(x1,…xn,a)

• Finding an optimal value for y would consist

of solving n first-order equations
y/xi = 0 (i = 1,…,n)

42
 The Envelope Theorem
• Optimal values for theses x’s would be
determined that are a function of a

x1* = x1*(a)
x2* = x2*(a)
.
.
.
xn*= xn*(a)

43
 The Envelope Theorem
• Substituting into the original objective
function yields an expression for the optimal
value of y (y*)
y* = f [x1*(a), x2*(a),…,xn*(a),a]
• Differentiating yields
dy * f dx 1 f dx 2 f dx n f
     ...   
da x1 da x 2 da x n da a

44
 The Envelope Theorem
• Because of first-order conditions, all terms
except f/a are equal to zero if the x’s are
at their optimal values

• Therefore,

dy * f
 {x  x * ( a)}
da a

45
 Constrained Maximization
• What if all values for the x’s are not
feasible?
– the values of x may all have to be positive
– a consumer’s choices are limited by the
amount of purchasing power available
• One method used to solve constrained
maximization problems is the Lagrangian
multiplier method

46
Lagrangian Multiplier Method
• Suppose that we wish to find the values
of x1, x2,…, xn that maximize
y = f(x1, x2,…, xn)
subject to a constraint that permits only
certain values of the x’s to be used
g(x1, x2,…, xn) = 0

47
Lagrangian Multiplier Method
• The Lagrangian multiplier method starts
with setting up the expression
L = f(x1, x2,…, xn ) + g(x1, x2,…, xn)
where  is an additional variable called
a Lagrangian multiplier
• When the constraint holds, L = f
because g(x1, x2,…, xn) = 0

48
Lagrangian Multiplier Method
• First-Order Conditions
L/x1 = f1 + g1 = 0
L/x2 = f2 + g2 = 0
.
.
.
L/xn = fn + gn = 0
L/ = g(x1, x2,…, xn) = 0
49
Lagrangian Multiplier Method
• The first-order conditions can generally
be solved for x1, x2,…, xn and 

• The solution will have two properties:

– the x’s will obey the constraint
– these x’s will make the value of L (and
therefore f) as large as possible

50
Lagrangian Multiplier Method
• The Lagrangian multiplier () has an
important economic interpretation
• The first-order conditions imply that
f1/-g1 = f2/-g2 =…= fn/-gn = 
– the numerators above measure the
marginal benefit that one more unit of xi will
have for the function f
– the denominators reflect the added burden
on the constraint of using more xi
51
Lagrangian Multiplier Method
• At the optimal choices for the x’s, the
ratio of the marginal benefit of increasing
xi to the marginal cost of increasing xi
should be the same for every x
•  is the common cost-benefit ratio for all
of the x’s
marginal benefit of x i

marginal cost of x i
52
Lagrangian Multiplier Method
• If the constraint was relaxed slightly, it
would not matter which x is changed
• The Lagrangian multiplier provides a
measure of how the relaxation in the
constraint will affect the value of y
•  provides a “shadow price” to the
constraint

53
Lagrangian Multiplier Method
• A high value of  indicates that y could
be increased substantially by relaxing
the constraint
– each x has a high cost-benefit ratio
• A low value of  indicates that there is
not much to be gained by relaxing the
constraint
• =0 implies that the constraint is not
binding
54
Lagrangian Multiplier Method
• Example:
y = - x12 + 2x1 - x22 + 4x2 + 5
x1+x2=1

55
 Duality
• Any constrained maximization problem
has associated with it a dual problem in
constrained minimization that focuses
attention on the constraints in the
original problem

56
 Duality
• Individuals maximize utility subject to a
budget constraint
– dual problem: individuals minimize the
expenditure needed to achieve a given level
of utility
• Firms minimize the cost of inputs to
produce a given level of output
– dual problem: firms maximize output for a
given cost of inputs purchased
57
 Constrained Maximization
• Suppose a farmer had a certain length of
fence (P) and wished to enclose the
largest possible rectangular shape
• Let x be the length of one side
• Let y be the length of the other side
• Problem: choose x and y so as to
maximize the area (A = x·y) subject to the
constraint that the perimeter is fixed at P =
2x + 2y
58
 Constrained Maximization
• Setting up the Lagrangian multiplier
L = x·y + (P - 2x - 2y)
• The first-order conditions for a maximum
are
L/x = y - 2 = 0
L/y = x - 2 = 0
L/ = P - 2x - 2y = 0

59
 Constrained Maximization
• Since y/2 = x/2 = , x must be equal to y
– the field should be square
– x and y should be chosen so that the ratio of
marginal benefits to marginal costs should be
the same
• Since x = y and y = 2, we can use the
constraint to show that
x = y = P/4
 = P/8
60
 Constrained Maximization
• Interpretation of the Lagrangian multiplier
– if the farmer was interested in knowing how
much more field could be fenced by adding an
extra yard of fence,  suggests that he could
find out by dividing the present perimeter (P)
by 8

– thus, the Lagrangian multiplier provides

information about the implicit value of the
constraint
61
 Constrained Maximization
• Dual problem: choose x and y to minimize
the amount of fence required to surround
the field
minimize P = 2x + 2y subject to A = x·y

• Setting up the Lagrangian:

LD = 2x + 2y + D(A - xy)

62
 Constrained Maximization
• First-order conditions:
LD/x = 2 - D·y = 0
LD/y = 2 - D·x = 0
LD/D = A - x·y = 0
• Solving, we get
x = y = A1/2
• The Lagrangian multiplier (D) = 2A-1/2

63
 Envelope Theorem &
Constrained Maximization
• Suppose that we want to maximize
y = f(x1,…,xn;a)
subject to the constraint
g(x1,…,xn;a) = 0
• One way to solve would be to set up the
Lagrangian expression and solve the first-
order conditions
64
 Envelope Theorem &
Constrained Maximization
• Alternatively, it can be shown that

dy*/da = L/a(x1*,…,xn*;a)
• The change in the maximal value of y that
results when a changes can be found by
partially differentiating L and evaluating
the partial derivative at the optimal point

65
 Inequality Constraints
• In some economic problems the
constraints need not hold exactly
• For example, suppose we seek to
maximize y = f(x1,x2) subject to
g(x1,x2)  0,
x1  0, and
x2  0
66
 Inequality Constraints
• One way to solve this problem is to
introduce three new variables (a, b, and
c) that convert the inequalities into
equalities
• To ensure that the inequalities continue
to hold, we will square these new
variables to ensure that their values are
positive
67
 Inequality Constraints

g(x1,x2) - a2 = 0;
x1 - b2 = 0; and
x2 - c2 = 0
• Any solution that obeys these three
equality constraints will also obey the
inequality constraints

68
 Inequality Constraints
• We can set up the Lagrangian

L = f(x1,x2) + 1[g(x1,x2) - a2] + 2[x1 - b2] + 3[x2 -

c2]

• This will lead to eight first-order

conditions

69
Inequality Constraints
L/x1 = f1 + 1g1 + 2 = 0
L/x2 = f1 + 1g2 + 3 = 0
L/a = -2a1 = 0
L/b = -2b2 = 0
L/c = -2c3 = 0
L/1 = g(x1,x2) - a2 = 0
L/2 = x1 - b2 = 0
70
L/3 = x2 - c = 0
2
 Inequality Constraints
• According to the third condition, either a
or 1 = 0
– if a = 0, the constraint g(x1,x2) holds exactly
– if 1 = 0, the availability of some slackness
of the constraint implies that its value to the
objective function is 0
• Similar complemetary slackness
relationships also hold for x1 and x2
71
 Inequality Constraints
• These results are sometimes called
Kuhn-Tucker conditions
– they show that solutions to optimization
problems involving inequality constraints
will differ from similar problems involving
equality constraints in rather simple ways
– we cannot go wrong by working primarily
with constraints involving equalities

72
 Second Order Conditions -
Functions of One Variable
• Let y = f(x)
• A necessary condition for a maximum is
that
dy/dx = f ’(x) = 0
• To ensure that the point is a maximum, y
must be decreasing for movements away
from it
73
 Second Order Conditions -
Functions of One Variable
• The total differential measures the change
in y
dy = f ’(x) dx
• To be at a maximum, dy must be
decreasing for small increases in x
• To see the changes in dy, we must use the
second derivative of y
74
 Second Order Conditions -
Functions of One Variable
2 d [f ' ( x )dx ] 2
d y dx  f " ( x )dx dx f " ( x )dx
dx
• Note that d 2y < 0 implies that f ’’(x)dx2 < 0
• Since dx2 must be positive, f ’’(x) < 0

• This means that the function f must have a

concave shape at the critical point
75
 Second Order Conditions -
Functions of Two Variables
• Suppose that y = f(x1, x2)
• First order conditions for a maximum are
y/x1 = f1 = 0
y/x2 = f2 = 0
• To ensure that the point is a maximum, y
must diminish for movements in any
direction away from the critical point
76
 Second Order Conditions -
Functions of Two Variables
• The slope in the x1 direction (f1) must be
diminishing at the critical point
• The slope in the x2 direction (f2) must be
diminishing at the critical point
• But, conditions must also be placed on the
cross-partial derivative (f12 = f21) to ensure
that dy is decreasing for all movements
through the critical point 77
 Second Order Conditions -
Functions of Two Variables
• The total differential of y is given by
dy = f1 dx1 + f2 dx2
• The differential of that function is
d 2y = (f11dx1 + f12dx2)dx1 + (f21dx1 + f22dx2)dx2
d 2y = f11dx12 + f12dx2dx1 + f21dx1 dx2 + f22dx22
• By Young’s theorem, f12 = f21 and
d 2y = f11dx12 + 2f12dx1dx2 + f22dx22 78
 Second Order Conditions -
Functions of Two Variables
d 2y = f11dx12 + 2f12dx1dx2 + f22dx22
• For this equation to be unambiguously
negative for any change in the x’s, f11 and
f22 must be negative
• If dx2 = 0, then d 2y = f11 dx12
– for d 2y < 0, f11 < 0
• If dx1 = 0, then d 2y = f22 dx22
– for d 2y < 0, f22 < 0 79
 Second Order Conditions -
Functions of Two Variables
d 2y = f11dx12 + 2f12dx1dx2 + f22dx22
• If neither dx1 nor dx2 is zero, then d 2y will
be unambiguously negative only if
f11 f22 - f122 > 0

80
 Constrained Maximization
• Suppose we want to choose x1 and x2 to
maximize
y = f(x1, x2)
• subject to the linear constraint
c - b1x1 - b2x2 = 0
• We can set up the Lagrangian
L = f(x1, x2) + (c - b1x1 - b2x2)

81
 Constrained Maximization
• The first-order conditions are
f1 - b1 = 0
f2 - b2 = 0
c - b1x1 - b2x2 = 0
• To ensure we have a maximum, we
must use the “second” total differential
d 2y = f11dx12 + 2f12dx1dx2 + f22dx22
82
 Constrained Maximization
• Only the values of x1 and x2 that satisfy
the constraint can be considered valid
alternatives to the critical point
• Thus, we must calculate the total
differential of the constraint
-b1 dx1 - b2 dx2 = 0
dx2 = -(b1/b2)dx1
• These are the allowable relative changes
in x1 and x2 83
 Constrained Maximization
• Because the first-order conditions imply
that f1/f2 = b1/b2, we can substitute and
get
dx2 = -(f1/f2) dx1
• Since
d 2y = f11dx12 + 2f12dx1dx2 + f22dx22

we can substitute for dx2 and get

d 2y = f11dx12 - 2f12(f1/f2)dx12 + f22(f12/f22)dx12
84
 Constrained Maximization
• Combining terms and rearranging
d 2y = (f11 f22 - 2f12f1f2 + f22f12 )[dx12/ f22]
• Therefore, for d 2y < 0, it must be true
that
f11 f22 - 2f12f1f2 + f22f12 < 0
• This equation characterizes a set of
functions termed quasi-concave
functions
– any two points within the set can be joined
85
by a line contained completely in the set
Constrained Maximization

86
 Concave and Quasi-
Concave Functions
• Concave: the Hassian is negative
definite

87
 Concave and Quasi-
Concave Functions
• Concave: the bordered Hassian is
negative definite

88
 Concave and Quasi-
Concave Functions
• The differences between concave and
quasi-concave functions can be
illustrated with the function
y = f(x1,x2) = (x1x2)k
where the x’s take on only positive
values and k can take on a variety of
positive values
89
 Concave and Quasi-
Concave Functions
• The differences between concave and
quasi-concave functions can be
illustrated with the function
y = f(x1,x2) = (x1x2)k
where the x’s take on only positive
values and k can take on a variety of
positive values
90
 Concave and Quasi-
Concave Functions
• No matter what value k takes, this
function is quasi-concave
• Whether or not the function is concave
depends on the value of k
– if k < 0.5, the function is concave
– if k > 0.5, the function is convex

91
 Concave and Quasi-
Concave Functions

k = 0.2 k=1
92
 Homogeneous Functions
• A function f(x1,x2,…xn) is said to be
homogeneous of degree k if
f(tx1,tx2,…txn) = tk f(x1,x2,…xn)
– when a function is homogeneous of degree
one, a doubling of all of its arguments
doubles the value of the function itself
– when a function is homogeneous of degree
zero, a doubling of all of its arguments
leaves the value of the function unchanged
93
 Homogeneous Functions
• If a function is homogeneous of degree
k, the partial derivatives of the function
will be homogeneous of degree k-1

94
 Euler’s Theorem
• If we differentiate the definition for
homogeneity with respect to the
proportionality factor t, we get
ktk-1f(x1,…,xn) = x1f1(tx1,…,txn) + … + xnfn(x1,…,xn)
• This relationship is called Euler’s theorem

95
 Euler’s Theorem
• Euler’s theorem shows that, for
homogeneous functions, there is a
definite relationship between the
values of the function and the values of
its partial derivatives

96
 Important Points to Note:
• Using mathematics provides a
convenient, short-hand way for
economists to develop their models
– implications of various economic
assumptions can be studied in a
simplified setting through the use of such
mathematical tools

97
 Important Points to Note:
• Derivatives are often used in economics
because economists are interested in
how marginal changes in one variable
affect another
– partial derivatives incorporate the ceteris
paribus assumption used in most economic
models

98
 Important Points to Note:
• The mathematics of optimization is an
important tool for the development of
models that assume that economic
agents rationally pursue some goal
– the first-order condition for a maximum
requires that all partial derivatives equal
zero

99
 Important Points to Note:
• Most economic optimization
problems involve constraints on the
choices that agents can make
– the first-order conditions for a
maximum suggest that each activity be
operated at a level at which the ratio of
the marginal benefit of the activity to its
marginal cost

100
 Important Points to Note:
• The Lagrangian multiplier is used to
help solve constrained maximization
problems
– the Lagrangian multiplier can be
interpreted as the implicit value (shadow
price) of the constraint

101
 Important Points to Note:
• The implicit function theorem
illustrates the dependence of the
choices that result from an
optimization problem on the
parameters of that problem

102
Important Points to Note:
• The envelope theorem examines
how optimal choices will change as
the problem’s parameters change
• Some optimization problems may
involve constraints that are
inequalities rather than equalities

103
Important Points to Note:
• First-order conditions are necessary
but not sufficient for ensuring a
maximum or minimum
– second-order conditions that describe
the curvature of the function must be
checked

104
Important Points to Note:
• Certain types of functions occur in
many economic problems
– quasi-concave functions obey the
second-order conditions of constrained
maximum or minimum problems when
the constraints are linear

105