17 Multivariable

Calculus

W
e know (from Chapter 13) how to maximize a company’s profit when both
17.1 Partial Derivatives
revenue and cost are written as functions of a single quantity, namely the
17.2 Applications of Partial number of units produced. But of course the production level is itself
Derivatives determined by other factors—and, in general, no single variable can
represent them.
17.3 Implicit Partial The amount of oil pumped from an oil field each week, for example, depends on
Differentiation both the number of pumps and the number of hours that the pumps are operated. The
17.4 Higher-Order Partial number of pumps in the field will depend on the amount of capital originally available
Derivatives to build the pumps as well as the size and shape of the field. The number of hours
that the pumps can be operated depends on the labor available to run and maintain the
17.5 Chain Rule pumps. In addition, the amount of oil that the owner will be willing to have pumped
from the oil field will depend on the current demand for oil—which is related to the
17.6 Maxima and Minima
price of the oil.
for Functions of Two
Maximizing the weekly profit from an oil field will require a balance between the
Variables
number of pumps and the amount of time each pump can be operated. The maximum
17.7 Lagrange Multipliers profit will not be achieved by building more pumps than can be operated or by running
a few pumps full time.
17.8 Lines of Regression This is an example of the general problem of maximizing profit when production
17.9 Multiple Integrals depends on several factors. The solution involves an analysis of the production func-
tion, which relates production output to resources allocated for production. Because,
Chapter 17 Review in general, several variables are needed to describe the resource allocation, the most
profitable allocation cannot be found by differentiation with respect to a single variable,
as in preceding chapters. The more advanced techniques necessary to do the job will
Data Analysis to be covered in this chapter.
Model Cooling

From Introductory Mathematical Analysis For Business, Economics, and the Life and Social Sciences, Thirteenth Edition.
Ernest F. Haeussler, Jr., Richard S. Paul, Richard J.Wood. Copyright © 2011 by Pearson Education, Inc. All rights reserved.
755
 750 Chapter 17 Multivariable Calculus

Objective 17.1 Partial Derivatives

To compute partial derivatives. Throughout this book we have encountered many examples of functions of several
variables. We recall, from Section 2.8, that the graph of a function of two variables is
a surface. Figure 17.1 shows the surface z = f (x, y) and a plane that is parallel to the
x,z-plane and that passes through the point (a, b, f (a, b)) on the surface. The equation
TO REVIEW functions of several of this plane is y = b. Hence, any point on the curve that is the intersection of the
variables, see Section 2.8. surface z = f (x, y) with the plane y = b must have the form (x, b, f (x, b)). Thus, the
curve can be described by the equation z = f (x, b). Since b is constant, z = f (x, b)
can be considered a function of one variable, x. When the derivative of this function is
evaluated at a, it gives the slope of the tangent line to this curve at the point (a, b, f (a, b)).
(See Figure 17.1.) This slope is called the partial derivative of f with respect to x at
(a, b) and is denoted fx (a, b). In terms of limits,

f (a + h, b) − f (a, b)
fx (a, b) = lim (1)
h→0 h

z
z
Tangent line
Tangent line
(x, b, f (x, b))

(a, b, f (a, b)) (a, b, f (a, b))

z  f(x, b)
z  f(x, y)

z  f (a, y)
(a, y, f(a, y))

b
a b
y a
y
x (a, b, 0) (a, b, 0)
x

FIGURE 17.1 Geometric interpretation of fx (a, b). FIGURE 17.2 Geometric interpretation of fy (a, b).

On the other hand, in Figure 17.2, the plane x = a is parallel to the y,z-plane and
cuts the surface z = f (x, y) in a curve given by z = f (a, y), a function of y. When the
derivative of this function is evaluated at b, it gives the slope of the tangent line to this
curve at the point (a, b, f (a, b)). This slope is called the partial derivative of f with
respect to y at (a, b) and is denoted fy (a, b). In terms of limits,

f (a, b + h) − f (a, b)
fy (a, b) = lim (2)
h→0 h

This gives us a geometric interpretation We say that fx (a, b) is the slope of the tangent line to the graph of f at (a, b, f (a, b))
of a partial derivative. in the x-direction; similarly, fy (a, b) is the slope of the tangent line in the y-direction.
For generality, by replacing a and b in Equations (1) and (2) by x and y, respectively,
we get the following definition.

756
 Section 17.1 Partial Derivatives 751

Definition
If z = f (x, y), the partial derivative of f with respect to x, denoted fx , is the function,
of two variables, given by
f (x + h, y) − f (x, y)
fx (x, y) = lim
h→0 h
provided that the limit exists.
The partial derivative of f with respect to y, denoted fy , is the function, of two
variables, given by
f (x, y + h) − f (x, y)
fy (x, y) = lim
h→0 h
provided that the limit exists.

By analyzing the foregoing definition, we can state the following procedure to find
fx and fy :
This gives us a mechanical way to find
partial derivatives. Procedure to Find fx (x, y) and fy (x, y)
To find fx , treat y as a constant, and differentiate f with respect to x in the usual way.
To find fy , treat x as a constant, and differentiate f with respect to y in the
usual way.

EXAMPLE 1 Finding Partial Derivatives

If f (x, y) = xy2 + x 2 y, find fx (x, y) and fy (x, y). Also, find fx (3, 4) and fy (3, 4).
Solution: To find fx (x, y), we treat y as a constant and differentiate f with respect to x:
fx (x, y) = (1)y2 + (2x)y = y2 + 2xy
To find fy (x, y), we treat x as a constant and differentiate with respect to y:
fy (x, y) = x(2y) + x 2 (1) = 2xy + x 2
Note that fx (x, y) and fy (x, y) are each functions of the two variables x and y. To find
fx (3, 4), we evaluate fx (x, y) when x = 3 and y = 4:
fx (3, 4) = 42 + 2(3)(4) = 40
Similarly,
fy (3, 4) = 2(3)(4) + 32 = 33
Now Work Problem 1 
Notations for partial derivatives of z = f (x, y) are in Table 17.1. Table 17.2 gives
notations for partial derivatives evaluated at (a, b). Note that the symbol ∂ (not d) is
used to denote a partial derivative. The symbol ∂z/∂x is read “the partial derivative of
z with respect to x.”

EXAMPLE 2 Finding Partial Derivatives

 
3 3 2 2 ∂z ∂z ∂z  ∂z 
a. If z = 3x y − 9x y + xy + 4y, find , , and .
∂x ∂y ∂x (1,0) ∂y (1.0)
Solution: To find ∂z/∂x, we differentiate z with respect to x while treating y as a
constant:
∂z
= 3(3x 2 )y3 − 9(2x)y + (1)y2 + 0
∂x
= 9x 2 y3 − 18xy + y2

757
 752 Chapter 17 Multivariable Calculus

Table 17.1 Table 17.2

Partial Derivative of f Partial Derivative of f Partial Derivative of f Partial Derivative of f
(or z) with Respect to x (or z) with Respect to y (or z) with Respect to x (or z) with Respect to y
Evaluated at (a, b) Evaluated at (a, b)
fx (x, y) fy (x, y)
fx (a, b) fy (a, b)
∂ ∂
( f (x, y)) ( f (x, y))  
∂x ∂y ∂f  ∂f 
∂z ∂z ∂x  (a,b) ∂y (a,b)
∂x ∂y  
∂z  ∂z 
∂x  x=a ∂y  x=a
y=b y=b

Evaluating the latter equation at (1, 0), we obtain


∂z 
= 9(1)2 (0)3 − 18(1)(0) + 02 = 0
∂x (1,0)
To find ∂z/∂y, we differentiate z with respect to y while treating x as a constant:
∂z
= 3x 3 (3y2 ) − 9x 2 (1) + x(2y) + 4(1)
∂y
= 9x 3 y2 − 9x 2 + 2xy + 4

Thus,

∂z 
= 9(1)3 (0)2 − 9(1)2 + 2(1)(0) + 4 = −5
∂y (1,0)

b. If w = x 2 e2x+3y , find ∂w/∂x and ∂w/∂y.

Solution: To find ∂w/∂x, we treat y as a constant and differentiate with respect to x.
Since x 2 e2x+3y is a product of two functions, each involving x, we use the product rule:
∂w ∂ ∂
= x 2 (e2x+3y ) + e2x+3y (x 2 )
∂x ∂x ∂x
= x 2 (2e2x+3y ) + e2x+3y (2x)
= 2x(x + 1)e2x+3y
To find ∂w/∂y, we treat x as a constant and differentiate with respect to y:
∂w ∂
= x 2 (e2x+3y ) = 3x 2 e2x+3y
∂y ∂y
Now Work Problem 27 
We have seen that, for a function of two variables, two partial derivatives can be
considered. Actually, the concept of partial derivatives can be extended to functions
of more than two variables. For example, with w = f (x, y, z) we have three partial
derivatives:
the partial with respect to x, denoted fx (x, y, z), ∂w/∂x, and so on;
the partial with respect to y, denoted fy (x, y, z), ∂w/∂y, and so on;
and
the partial with respect to z, denoted fz (x, y, z), ∂w/∂z, and so on
To determine ∂w/∂x, treat y and z as constants, and differentiate w with respect to x.
For ∂w/∂y, treat x and z as constants, and differentiate with respect to y. For ∂w/∂z, treat
x and y as constants, and differentiate with respect to z. For a function of n variables,
we have n partial derivatives, which are determined in an analogous way.

758
 Section 17.1 Partial Derivatives 753

EXAMPLE 3 Partial Derivatives of a Function of Three Variables

If f (x, y, z) = x 2 + y2 z + z3 , find fx (x, y, z), fy (x, y, z), and fz (x, y, z).

Solution: To find fx (x, y, z), we treat y and z as constants and differentiate f with respect
to x:
fx (x, y, z) = 2x
Treating x and z as constants and differentiating with respect to y, we have
fy (x, y, z) = 2yz
Treating x and y as constants and differentiating with respect to z, we have
fz (x, y, z) = y2 + 3z2
Now Work Problem 23 
EXAMPLE 4 Partial Derivatives of a Function of Four Variables

rsu ∂p ∂p ∂p 
If p = g(r, s, t, u) = 2 , find , , and .
rt + s2 t ∂s ∂t ∂t (0,1,1,1)
Solution: To find ∂p/∂s, first note that p is a quotient of two functions, each involving
the variable s. Thus, we use the quotient rule and treat r, t, and u as constants:
∂ ∂
∂p (rt 2 + s2 t) (rsu) − rsu (rt 2 + s2 t)
= ∂s ∂s
∂s (rt 2 + s2 t)2
(rt 2 + s2 t)(ru) − (rsu)(2st)
=
(rt 2 + s2 t)2
Simplification gives
∂p ru(rt − s2 )
= a factor of t cancels
∂s t(rt + s2 )2
To find ∂p/∂t, we can first write p as
p = rsu(rt 2 + s2 t)−1
Next, we use the power rule and treat r, s, and u as constants:
∂p ∂
= rsu( − 1)(rt 2 + s2 t)−2 (rt 2 + s2 t)
∂t ∂t
2 −2
= −rsu(rt + s t) (2rt + s2 )
2

so that
∂p rsu(2rt + s2 )
=− 2
∂s (rt + s2 t)2
Letting r = 0, s = 1, t = 1, and u = 1 gives

∂p  0(1)(1)(2(0)(1) + (1)2 )
= − =0
∂t  (0(1)2 + (1)2 (1))2
(0,1,1,1)

Now Work Problem 31 

PROBLEMS 17.1
In Problems 1–26, a function of two or more variables is given. 2. f (x, y) = 2x 2 + 3xy
Find the partial derivative of the function with respect to each of 3. f (x, y) = 2y + 1 4. f (x, y) = ln 2
the variables. 5. g(x, y) = 3x 4 y + 2xy2 − 5xy + 8x − 9y
1. f (x, y) = 2x 2 + 3xy + 4y2 + 5x + 6y − 7 6. g(x, y) = (x 2 + 1)2 + (y3 − 3)3 + 5xy3 − 2x 2 y2

759
 754 Chapter 17 Multivariable Calculus

√ √
7. g(p, q) = pq 8. g(w, z) = 3
w2 + z 2 function f given by
s2 + 4 8uv2 (1 + r)1−z ln (1 + r)
9. h(s, t) = 10. h(u, v) = 2 u = f (t, r, z) =
t−3 u + v2 (1 + r)1−z − t
√ √ where u is the instantaneous rate of ask-price appreciation, r is an
11. u(q1 , q2 ) = ln q1 + 2 + ln 3 q2 + 5
annual opportunity rate of return, z is the fraction of a dividend
12. Q(l, k) = 2l0.38 k 1.79 − 3l 1.03 + 2k 0.13 cycle over which a share of stock is held by a midcycle seller, and
√ t is the effective rate of capital gains tax. They claim that
x 2 + 3xy + y2 x+9
13. h(x, y) = 14. h(x, y) = ∂u t(1 + r)1−z ln2 (1 + r)
x 2 + y2 x 2 y + y2 x =
∂z [(1 + r)1−z − t]2
15. z = e5xy 16. z = (x 3 + y3 )exy+3x+3y
Verify this.
17. z = 5x ln (x 2 + y) 18. z = ln (5x 3 y2 + 2y4 )4
√ 37. Money Demand In a discussion of inventory theory of
19. f (r, s) = r + 2s(r 3 − 2rs + s2 ) money demand, Swanson2 considers the function
√
20. f (r, s) = rs e2+r 21. f (r, s) = e3−r ln (7 − s) bT iC
F(b, C, T , i) = +
22. f (r, s) = (5r 2 + 3s3 )(2r − 5s) C 2
23. g(x, y, z) = 2x 3 y2 + 2xy3 z + 4z2 ∂F bT i
and determines that = − 2 + . Verify this partial derivative.
2 6
24. g(x, y, z) = 2xy z − 4x y z + 3xyz 2 3 2 ∂C C 2
25. g(r, s, t) = es+t (r 2 + 7s3 ) 26. g(r, s, t, u) = rs ln (t)eu 38. Interest Rate Deregulation In an article on interest rate
deregulation, Christofi and Agapos3 arrive at the equation
In Problems 27–34, evaluate the given partial derivatives.
∂r dC
27. f (x, y) = x 3 y + 7x 2 y2 ;
fx (1, −2) rL = r + D + (3)
 ∂D dD
∂z 
28. z = 3 2
2x + 5xy + 2y ; where r is the deposit rate paid by commercial banks, rL is the rate
∂x  x=0 earned by commercial banks, C is the administrative cost of
y=1
√ transforming deposits into return-earning assets, and D is the
x
29. g(x, y, z) = e y + 2z; gz (0, 6, 4) savings deposit level. Christofi and Agapos state that
3x 2 y2 + 2xy + x − y  
30. g(x, y, z) = , gy (1, 1, 5) 1+η dC
xy − yz + xz rL = r + (4)
η dD
31. h(r, s, t, u) = (rst 2 u) ln (1 + rstu); ht (1, 1, 0, 1)
r/D
7r + 3s2 u2 where η = is the deposit elasticity with respect to the
32. h(r, s, t, u) = ; ht (4, 3, 2, 1) ∂r/∂D
s deposit rate. Express Equation (3) in terms of η to verify
33. f (r, s, t) = rst(r 2 + s3 + t 4 ); fs (1, −1, 2) Equation (4).
 
x 2 + y2 ∂z  ∂z  39. Advertising and Profitability In an analysis of advertising
34. z = x2 +y2 ; ,
e ∂x  x=0 ∂y  x=1 and profitability, Swales4 considers a function f given by
y=0 y=1
r
35. If z = xe x−y
+ ye y−x
, show that R = f (r, a, n) =  
n−1
1+a
∂z ∂z 2
+ = ex−y + ey−x
∂x ∂y where R is the adjusted rate of profit, r is the accounting rate of
profit, a is a measure of advertising expenditures, and n is the
36. Stock Prices of a Dividend Cycle In a discussion of stock number of years that advertising fully depreciates. In the analysis,
prices of a dividend cycle, Palmon and Yaari1 consider the Swales determines ∂R/∂n. Find this partial derivative.

Objective 17.2 Applications of Partial Derivatives

To develop the notions of partial From Section 17.1, we know that if z = f (x, y), then ∂z/∂x and ∂z/∂y can be geomet-
marginal cost, marginal productivity,
and competitive and complementary rically interpreted as giving the slopes of the tangent lines to the surface z = f (x, y) in
products. the x- and y-directions, respectively. There are other interpretations: Because ∂z/∂x is

1 D. Palmon and U. Yaari, “Taxation of Capital Gains and the Behavior of Stock Prices over the Dividend Cycle,”
The American Economist, XXVII, no. 1 (1983), 13–22.
2 P. E. Swanson, “Integer Constraints on the Inventory Theory of Money Demand,” Quarterly Journal of Business
and Economics, 23, no. 1 (1984), 32–37.
3A. Christofi and A. Agapos, “Interest Rate Deregulation: An Empirical Justification,” Review of Business and
Economic Research, XX (1984), 39–49.
4 J. K. Swales, “Advertising as an Intangible Asset: Profitability and Entry Barriers: A Comment on Reekie and
Bhoyrub,” Applied Economics, 17, no. 4 (1985), 603–17.

760
 Section 17.2 Applications of Partial Derivatives 755

the derivative of z with respect to x when y is held fixed, and because a derivative is a
rate of change, we have

∂z
Here we have “rate of change” is the rate of change of z with respect to x when y is held fixed.
interpretations of partial derivatives. ∂x

Similarly,

∂z
is the rate of change of z with respect to y when x is held fixed.
∂y

We will now look at some applications in which the “rate of change” notion of a partial
derivative is very useful.
Suppose a manufacturer produces x units of product X and y units of product Y.
Then the total cost c of these units is a function of x and y and is called a joint-cost
function. If such a function is c = f (x, y), then ∂c/∂x is called the (partial) marginal
cost with respect to x and is the rate of change of c with respect to x when y is held
fixed. Similarly, ∂c/∂y is the (partial) marginal cost with respect to y and is the rate
of change of c with respect to y when x is held fixed. It also follows that ∂c/∂x(x, y) is
approximately the cost of producing one more unit of X when x units of X and y units
of Y are produced. Similarly, ∂c/∂y(x, y) is approximately the cost of producing one
more unit of Y when x units of X and y units of Y are produced.
For example, if c is expressed in dollars and ∂c/∂y = 2, then the cost of producing
an extra unit of Y when the level of production of X is fixed is approximately two
dollars.
If a manufacturer produces n products, the joint-cost function is a function of n
variables, and there are n (partial) marginal-cost functions.

EXAMPLE 1 Marginal Costs

A company manufactures two types of skis, the Lightning and the Alpine models.
Suppose the joint-cost function for producing x pairs of the Lightning model and y
pairs of the Alpine model per week is
c = f (x, y) = 0.07x 2 + 75x + 85y + 6000
where c is expressed in dollars. Determine the marginal costs ∂c/∂x and ∂c/∂y when
x = 100 and y = 50, and interpret the results.
Solution: The marginal costs are
∂c ∂c
= 0.14x + 75 and = 85
∂x ∂y
Thus,

∂c 
= 0.14(100) + 75 = 89 (1)
∂x (100,50)
and

∂c 
= 85 (2)
∂y (100,50)
Equation (1) means that increasing the output of the Lightning model from 100 to 101
while maintaining production of the Alpine model at 50 increases costs by approxi-
mately $89. Equation (2) means that increasing the output of the Alpine model from
50 to 51 and holding production of the Lightning model at 100 will increase costs by
approximately $85. In fact, since ∂c/∂y is a constant function, the marginal cost with
respect to y is $85 at all levels of production.
Now Work Problem 1 
761
 756 Chapter 17 Multivariable Calculus

EXAMPLE 2 Loss of Body Heat

On a cold day, a person may feel colder when the wind is blowing than when the wind
is calm because the rate of heat loss is a function of both temperature and wind speed.
The equation
√
H = (10.45 + 10 w − w)(33 − t)

indicates the rate of heat loss H (in kilocalories per square meter per hour) when the air
temperature is t (in degrees Celsius) and the wind speed is w (in meters per second).
For H = 2000, exposed flesh will freeze in one minute.5

a. Evaluate H when t = 0 and w = 4.

Solution: When t = 0 and w = 4,
√
H = (10.45 + 10 4 − 4)(33 − 0) = 872.85

b. Evaluate ∂H/∂w and ∂H/∂t when t = 0 and w = 4, and interpret the results.
  
∂H 5 ∂H 
Solution: = √ − 1 (33 − t), = 49.5
∂w w ∂w  t=0
w=4

∂H √ ∂H 
= (10.45 + 10 w − w)( − 1), = −26.45
∂t ∂t  t=0
w=4

These equations mean that when t = 0 and w = 4, increasing w by a small amount

while keeping t fixed will make H increase approximately 49.5 times as much as w
increases. Increasing t by a small amount while keeping w fixed will make H decrease
approximately 26.45 times as much as t increases.
c. When t = 0 and w = 4, which has a greater effect on H: a change in wind speed of
1 m/s or a change in temperature of 1◦ C?
Solution: Since the partial derivative of H with respect to w is greater in magnitude
than the partial with respect to t when t = 0 and w = 4, a change in wind speed of
1 m/s has a greater effect on H.
Now Work Problem 13 
The output of a product depends on many factors of production. Among these may
be labor, capital, land, machinery, and so on. For simplicity, let us suppose that output
depends only on labor and capital. If the function P = f (l, k) gives the output P when
the producer uses l units of labor and k units of capital, then this function is called a
production function. We define the marginal productivity with respect to l to be
∂P/∂l. This is the rate of change of P with respect to l when k is held fixed. Likewise,
the marginal productivity with respect to k is ∂P/∂k and is the rate of change of P
with respect to k when l is held fixed.

EXAMPLE 3 Marginal Productivity

A manufacturer
√ of a popular toy has determined that the production function is
P = lk, where l is the number of labor-hours per week and k is the capital
(expressed in hundreds of dollars per week) required for a weekly production of P
gross of the toy. (One gross is 144 units.) Determine the marginal productivity func-
tions, and evaluate them when l = 400 and k = 16. Interpret the results.

5 G. E. Folk, Jr., Textbook of Environmental Physiology, 2nd ed. (Philadelphia: Lea & Febiger, 1974).

762
 Section 17.2 Applications of Partial Derivatives 757

Solution: Since P = (lk)1/2 ,

∂P 1 k
= (lk)−1/2 k = √
∂l 2 2 lk
and
∂P 1 l
= (lk)−1/2 l = √
∂k 2 2 lk
Evaluating these equations when l = 400 and k = 16, we obtain

∂P  16 1
 = √ =
∂l l=400 2 400(16) 10
k=16

and

∂P  400 5
= √ =
∂k  l=400 2 400(16) 2
k=16

Thus, if l = 400 and k = 16, increasing l to 401 and holding k at 16 will increase
1
output by approximately 10 gross. But if k is increased to 17 while l is held at 400, the
output increases by approximately 25 gross.
Now Work Problem 5 
Competitive and Complementary Products
Sometimes two products may be related such that changes in the price of one of them
affect the demand for the other. A typical example is that of butter and margarine. If
such a relationship exists between products A and B, then the demand for each product
is dependent on the prices of both. Suppose qA and qB are the quantities demanded for
A and B, respectively, and pA and pB are their respective prices. Then both qA and qB
are functions of pA and pB :
qA = f (pA , pB ) demand function for A
qB = g(pA , pB ) demand function for B
We can find four partial derivatives:
∂qA
the marginal demand for A with respect to pA
∂pA
∂qA
the marginal demand for A with respect to pB
∂pB
∂qB
the marginal demand for B with respect to pA
∂pA
∂qB
the marginal demand for B with respect to pB
∂pB
Under typical conditions, if the price of B is fixed and the price of A increases, then
the quantity of A demanded will decrease. Thus, ∂qA /∂pA < 0. Similarly, ∂qB /∂pB < 0.
However, ∂qA /∂pB and ∂qB /∂pA may be either positive or negative. If
∂qA ∂qB
>0 and > 0
∂pB ∂pA
then A and B are said to be competitive products or substitutes. In this situation,
an increase in the price of B causes an increase in the demand for A, if it is assumed
that the price of A does not change. Similarly, an increase in the price of A causes an
increase in the demand for B when the price of B is held fixed. Butter and margarine
are examples of substitutes.

763
 758 Chapter 17 Multivariable Calculus

Proceeding to a different situation, we say that if

∂qA ∂qB
< 0 and < 0
∂pB ∂pA
then A and B are complementary products. In this case, an increase in the price of B
causes a decrease in the demand for A if the price of A does not change. Similarly, an
increase in the price of A causes a decrease in the demand for B when the price of B
is held fixed. For example, cars and gasoline are complementary products. An increase
in the price of gasoline will make driving more expensive. Hence, the demand for cars
will decrease. And an increase in the price of cars will reduce the demand for gasoline.

EXAMPLE 4 Determining Whether Products Are Competitive

or Complementary

The demand functions for products A and B are each a function of the prices of A and
B and are given by
√
50 3 pB 75pA
qA = √ and qB =
pA 3 2
pB
respectively. Find the four marginal-demand functions, and determine whether A and
B are competitive products, complementary products, or neither.
−1/2 1/3 −2/3
Solution: Writing qA = 50pA pB and qB = 75pA pB , we have
 
∂qA 1 −3/2 1/3 −3/2 1/3
= 50 − p pB = −25pA pB
∂pA 2 A
 
∂qA −1/2 1 −2/3 50 −1/2 −2/3
= 50pA pB = p pB
∂pB 3 3 A
∂qB −2/3 −2/3
= 75(1)pB = 75pB
∂pA
 
∂qB 2 −5/3 −5/3
= 75pA − pB = −50pA pB
∂pB 3
Since pA and pB represent prices, they are both positive. Hence, ∂qA /∂pB > 0 and
∂qB /∂pA > 0. We conclude that A and B are competitive products.
Now Work Problem 19 

PROBLEMS 17.2
For the joint-cost functions in Problems 1–3, find the indicated form P = Alα k β , where A, α, and β are constants and α + β = 1.
marginal cost at the given production level. For such a function, show that
∂c (a) ∂P/∂l = αP/l (b) ∂P/∂k = βP/k
1. c = 7x + 0.3y2 + 2y + 900; , x = 20, y = 30
∂y ∂P ∂P
(c) l +k = P. This means that summing the products of the
√ ∂c ∂l ∂k
2. c = 2x x + y + 6000; , x = 70, y = 74 marginal productivity of each factor and the amount of that factor
∂x
results in the total product P.
3. c = 0.03(x + y)3 − 0.6(x + y)2 + 9.5(x + y) + 7700;
∂c In Problems 7–9, qA and qB are demand functions for products
, x = 50, y = 80 A and B, respectively. In each case, find ∂qA /∂pA , ∂qA /∂pB ,
∂x
For the production functions in Problems 4 and 5, find the ∂qB /∂pA , ∂qB /∂pB and determine whether A and B are
marginal productivity functions ∂P/∂k and ∂P/∂l. competitive, complementary, or neither.

4. P = 15lk − 3l2 + 5k 2 + 500 7. qA = 1500 − 40pA + 3pB ; qB = 900 + 5pA − 20pB

8. qA = 20 − pA − 2pB ; qB = 50 − 2pA − 3pB
5. P = 2.314l0.357 k 0.643
100 500
6. Cobb–Douglas Production Function In economics, a 9. qA = √ ; qB = √
Cobb–Douglas production function is a production function of the pA pB pB 3 pA

764
 Section 17.2 Applications of Partial Derivatives 759

10. Canadian Manufacturing The production function for the Which is “easier” to read: an article for which w = w0 and s = s0 ,
Canadian manufacturing industries for 1927 is estimated by6 or one for which w = w0 + 1 and s = s0 ?
P = 33.0l0.46 k 0.52 , where P is product, l is labor, and k is capital. 16. Model for Voice The study of frequency of vibrations of a
Find the marginal productivities for labor and capital, and evaluate taut wire is useful in considering such things as an individual’s
when l = 1 and k = 1. voice. Suppose
11. Dairy Farming An estimate of the production function for 1 τ
dairy farming in Iowa (1939) is given by7 ω=
bL πρ
P = A0.27 B0.01 C 0.01 D0.23 E 0.09 F 0.27
where ω (a Greek letter read “omega”) is frequency, b is diameter,
where P is product, A is land, B is labor, C is improvements, D is L is length, ρ (a Greek letter read “rho”) is density, and τ (a Greek
liquid assets, E is working assets, and F is cash operating letter read “tau”) is tension.11 Find ∂ω/∂b, ∂ω/∂L, ∂ω/∂ρ, and
expenses. Find the marginal productivities for labor and ∂ω/∂τ.
improvements.
17. Traffic Flow Consider the following traffic-flow situation.
12. Production Function Suppose a production function is On a highway where two lanes of traffic flow in the same
kl direction, there is a maintenance vehicle blocking the left lane.
given by P = . (See Figure 17.3.) Two vehicles (lead and following) are in the
3k + 5l
(a) Determine the marginal productivity functions. right lane with a gap between them. The subject vehicle can
1 choose either to fill or not to fill the gap. That decision may be
(b) Show that when k = l, the marginal productivities sum to .
8 based not only on the distance x shown in the diagram, but also on
13. M.B.A. Compensation In a study of success among other factors (such as the velocity of the following vehicle). A gap
graduates with master of business administration (M.B.A.) index g has been used in analyzing such a decision.12,13 The
degrees, it was estimated that for staff managers (which include greater the g-value, the greater is the propensity for the subject
accountants, analysts, etc.), current annual compensation (in vehicle to fill the gap. Suppose
dollars) was given by  
x VF − V S
g= − 0.75 +
z = 43,960 + 4480x + 3492y VF 19.2
where x and y are the number of years of work experience before where x (in feet) is as before, VF is the velocity of the following
and after receiving the M.B.A. degree, respectively.8 Find ∂z/∂x vehicle (in feet per second), and VS is the velocity of the subject
and interpret your result. vehicle (in feet per second). From the diagram, it seems
14. Status A person’s general status Sg is believed to be a reasonable that if both VF and VS are fixed and x increases, then g
function of status attributable to education, Se , and status should increase. Show that this is true by applying calculus to the
attributable to income, Si , where Sg , Se , and Si are represented function g. Assume that x, VF , and VS are positive.
numerically. If
Subject Maintenance
Sg = 7 3 Se Si vehicle vehicle

determine ∂Sg /∂Se and ∂Sg /∂Si when Se = 125 and Si = 100, and Left lane
x
interpret your results.9
15. Reading Ease Sometimes we want to evaluate the degree Right lane
Following Lead
of readability of a piece of writing. Rudolf Flesch10 developed a vehicle vehicle
function of two variables that will do this, namely,
FIGURE 17.3
R = f (w, s) = 206.835 − (1.015w + 0.846s)
where R is called the reading ease score, w is the average number
18. Demand Suppose the demand equations for related
of words per sentence in 100-word samples, and s is the average
products A and B are
number of syllables in such samples. Flesch says that an article for
which R = 0 is “practically unreadable,” but one with R = 100 is 16
qA = e−(pA +pB ) and qB = 2 2
“easy for any literate person.” (a) Find ∂R/∂w and ∂R/∂s. (b) pA pB

6
P. Daly and P. Douglas, “The Production Function for Canadian Manufactures,”
Journal of the American Statistical Association, 38 (1943), 178–86.
7
G. Tintner and O. H. Brownlee, “Production Functions Derived from Farm
Records,” American Journal of Agricultural Economics, 26 (1944), 566–71. 11
R. M. Thrall, J. A. Mortimer, K. R. Rebman, and R. F. Baum, eds., Some
8Adapted from A. G. Weinstein and V. Srinivasen, “Predicting Managerial Mathematical Models in Biology, rev. ed., Report No. 40241-R-7. Prepared at
Success of Master of Business Administration (M.B.A.) Graduates,” Journal of University of Michigan, 1967.
Applied Psychology, 59, no. 2 (1974), 207–12. 12
P. M. Hurst, K. Perchonok, and E. L. Seguin, “Vehicle Kinematics and Gap
9
Adapted from R. K. Leik and B. F. Meeker, Mathematical Sociology Acceptance,” Journal of Applied Psychology, 52, no. 4 (1968), 321–24.
(Englewood Cliffs, NJ: Prentice-Hall, Inc., 1975). 13 K. Perchonok and P. M. Hurst, “Effect of Lane-Closure Signals upon Driver
10 R. Flesch, The Art of Readable Writing (New York: Harper & Row Publishers, Decision Making and Traffic Flow,” Journal of Applied Psychology, 52, no. 5
Inc., 1949). (1968), 410–13.

765
 760 Chapter 17 Multivariable Calculus

where qA and qB are the number of units of A and B demanded are the number of terms served in Congress, plus one, for the
when the unit prices (in thousands of dollars) are pA and pB , Republican and Democratic candidates, respectively; and N is the
respectively. percentage of the two-party presidential vote that Richard Nixon
(a) Classify A and B as competitive, complementary, or neither. received in the district for 1968. The variable N gives a measure of
(b) If the unit prices of A and B are $1000 and $2000, Republican strength in the district.
respectively, estimate the change in the demand for A when the (a) In the Federal Election Campaign Act of 1974, Congress set a
price of B is decreased by $20 and the price of A is held constant. limit of $188,000 on campaign expenditures. By analyzing
19. Demand The demand equations for related products A and ∂R/∂Er , would you have advised a Republican candidate who
B are given by served nine terms in Congress to spend $188,000 on his or her
campaign?
pB pA (b) Find the percentage above which the Nixon vote had a
qA = 10 and qB = 3 3
pA pB negative effect on R; that is, find N when ∂R/∂N < 0. Give your
where qA and qB are the quantities of A and B demanded and pA answer to the nearest percent.
and pB are the corresponding prices (in dollars) per unit. 22. Sales After a new product has been launched onto
(a) Find the values of the two marginal demands for product A the market, its sales volume (in thousands of units) is given by
when pA = 9 and pB = 16.
AT + 450
(b) If pB were reduced to 14 from 16, with pA fixed at 9, use part S= √
(a) to estimate the corresponding change in demand for product A. A+T2
20. Joint-Cost Function A manufacturer’s joint-cost function where T is the time (in months) since the product was first
for producing qA units of product A and qB units of product B is introduced and A is the amount (in hundreds of dollars) spent each
given by month on advertising.
q2 (q3 + qA )1/2 1/3 (a) Verify that the partial derivative of sales volume with respect to
c= A B + qA qB + 600 time is given by
17
where c is in dollars.
∂S A2 − 450T
(a) Find the marginal-cost functions with respect to qA =
∂T (A + T 2 )3/2
and qB .
(b) Evaluate the marginal-cost function with respect to qA (b) Use the result in part (a) to predict the number of months that
when qA = 17 and qB = 8. Round your answer to two will elapse before the sales volume begins to decrease if the
decimal places. amount allocated to advertising is held fixed at $9000 per month.
(c) Use your answer to part (a) to estimate the change in Let f be a demand function for product A and qA = f (pA , pB ),
cost if production of product A is decreased from 17 to where qA is the quantity of A demanded when the price per unit of
16 units, while production of product B is held constant A is pA and the price per unit of product B is pB . The partial
at 8 units. elasticity of demand for A with respect to pA , denoted ηpA , is
21. Elections For the congressional elections of 1974, the defined as ηpA = (pA /qA )(∂qA /∂pA ). The partial elasticity of
Republican percentage, R, of the Republican–Democratic vote in demand for A with respect to pB , denoted ηpB , is defined as
a district is given (approximately) by14 ηpB = (pB /qA )(∂qA /∂pB ). Loosely speaking, ηpA is the ratio of a
percentage change in the quantity of A demanded to a percentage
R = f (Er , Ed , Ir , Id , N)
change in the price of A when the price of B is fixed. Similarly,
= 15.4725 + 2.5945Er − 0.0804Er2 − 2.3648Ed ηpB can be loosely interpreted as the ratio of a percentage change
in the quantity of A demanded to a percentage change in the price
+ 0.0687Ed2 + 2.1914Ir − 0.0912Ir2 of B when the price of A is fixed. In Problems 23–25, find ηpA and
− 0.8096Id + 0.0081Id2 − 0.0277Er Ir ηpB for the given values of pA and pB .
+ 0.0493Ed Id + 0.8579N − 0.0061N 2 23. qA = 1000 − 50pA + 2pB ; pA = 2, pB = 10
24. qA = 60 − 3pA − 2pB ; pA = 5, pB = 3
Here Er and Ed are the campaign expenditures (in units of √
$10,000) by Republicans and Democrats, respectively; Ir and Id 25. qA = 100/(pA pB ); pA = 1, pB = 4

Objective 17.3 Implicit Partial Differentiation15

To find partial derivatives of a function An equation in x, y, and z does not necessarily define z as a function of x and y. For
defined implicitly.
example, in the equation
z 2 − x 2 − y2 = 0 (1)

14
J. Silberman and G.Yochum, “The Role of Money in Determining Election Outcomes,” Social Science Quarterly,
58, no. 4 (1978), 671–82.
15 This section can be omitted without loss of continuity.

766
 Section 17.3 Implicit Partial Differentiation 761
√
if x = 1 and y = 1, then z2 − 1 − 1 = 0, so z = ± 2. Thus, Equation (1) does not
define z as a function of x and y. However, solving Equation (1) for z gives

z= x 2 + y2 or z = − x 2 + y2

each of which defines z as a function of x and y.Although Equation (1) does not explicitly
express z as a function of x and y, it can be thought of as expressing z implicitly as one
of two different functions of x and y. Note that the equation z2 − x 2 − y2 = 0 has the
form F(x, y, z) = 0, where F is a function of three variables. Any equation of the form
F(x, y, z) = 0 can be thought of as expressing z implicitly as one of a set of possible
functions of x and y. Moreover, we can find ∂z/∂x and ∂z/∂y directly from the form
F(x, y, z) = 0.
To find ∂z/∂x for

z 2 − x 2 − y2 = 0 (2)

we first differentiate both sides of Equation (2) with respect to x while treating z as a
function of x and y and treating y as a constant:
∂ 2 ∂
(z − x 2 − y2 ) = (0)
∂x ∂x
∂ 2 ∂ 2 ∂ 2
(z ) − (x ) − (y ) = 0
∂x ∂x ∂x
Because y is treated as a constant,
∂z
∂y
= 0. 2z − 2x − 0 = 0
∂x ∂x
Solving for ∂z/∂x, we obtain
∂z
2z = 2x
∂x
∂z x
=
∂x z
To find ∂z/∂y, we differentiate both sides of Equation (2) with respect to y while treating
z as a function of x and y and treating x as a constant:
∂ 2 ∂
(z − x 2 − y2 ) = (0)
∂y ∂y
∂z ∂x
2z − 0 − 2y = 0 =0
∂y ∂y
∂z
2z = 2y
∂y
Hence,
∂z y
=
∂y z
The method we used to find ∂z/∂x and ∂z/∂y is called implicit partial differentiation.

EXAMPLE 1 Implicit Partial Differentiation

xz2 ∂z
If + y2 = 0, evaluate when x = −1, y = 2, and z = 2.
x+y ∂x
Solution: We treat z as a function of x and y and differentiate both sides of the equation
with respect to x:
 2 
∂ xz ∂ ∂
+ (y2 ) = (0)
∂x x + y ∂x ∂x

767
 762 Chapter 17 Multivariable Calculus

Using the quotient rule for the first term on the left, we have
∂ ∂
(x + y) (xz2 ) − xz2 (x + y)
∂x ∂x +0=0
(x + y)2
∂
Using the product rule for (xz2 ) gives
∂x
   
∂z
(x + y) x 2z + z (1) − xz2 (1)
2
∂x
=0
(x + y)2
Solving for ∂z/∂x, we obtain
∂z
2xz(x + y) + z2 (x + y) − xz2 = 0
∂x
∂z xz2 − z2 (x + y) yz
= =− z=
 0
∂x 2xz(x + y) 2x(x + y)
Thus,

∂z 
=2
∂x  (−1,2,2)

Now Work Problem 13 

EXAMPLE 2 Implicit Partial Differentiation
2 +u2
If ser = u ln (t 2 + 1), determine ∂t/∂u.
Solution: We consider t as a function of r, s, and u. By differentiating both sides with
respect to u while treating r and s as constants, we get
∂ 2 2 ∂
(ser +u ) = (u ln (t 2 + 1))
∂u ∂u
r 2 +u2 ∂ ∂
2sue = u ( ln (t 2 + 1)) + ln (t 2 + 1) (u) product rule
∂u ∂u
2 2 2t ∂t
2suer +u = u 2 + ln (t 2 + 1)
t + 1 ∂u
Therefore,
2 2
∂t (t 2 + 1)(2suer +u − ln (t 2 + 1))
=
∂u 2ut
Now Work Problem 1 
PROBLEMS 17.3
In Problems 1–11, find the indicated partial derivatives by the In Problems 12–20, evaluate the indicated partial derivatives for
method of implicit partial differentiation. the given values of the variables.
1. 2x 2 + 3y2 + 5z2 = 900; ∂z/∂x 12. xz + xyz − 5 = 0; ∂z/∂x, x = 1, y = 4, z = 1
2. z2 − 5x 2 + y2 = 0; ∂z/∂x 13. xz2 + yz2 − x 2 y = 1; ∂z/∂x, x = 1, y = 0, z = 1
3. 3z2 − 5x 2 − 7y2 = 0; ∂z/∂y zx
14. e = xyz; ∂z/∂y, x = 1, y = −e−1 , z = −1
4. 3x 2 + y2 + 2z3 = 9; ∂z/∂y 15. eyz = −xyz; ∂z/∂x, x = −e2 /2, y = 1, z = 2
5. x 2 − 2y − z2 + x 2 yz2 = 20; ∂z/∂x 16. xz + y2 − xy = 0; ∂z/∂y, x = 2, y = 2, z = 6
6. z3 + 2x 2 z2 − xy = 0; ∂z/∂x 17. ln z = 4x + y; ∂z/∂x, x = 5, y = −20, z = 1
7. ex + ey + ez = 10; ∂z/∂y r s 2 2
t 2

8. xyz + xy2 z3 − ln z4 = 0; ∂z/∂y 18. = ; ∂r/∂t, r = 1, s = 1, t = 1

s2 + t 2 2
9. ln (z) + 9z − xy = 1; ∂z/∂x 2
s +t 2

10. ln x + ln y − ln z = ey ; ∂z/∂x 19. = 10; ∂t/∂r, r = 1, s = 2, t = 4

√ rs
11. (z2 + 6xy) x 3 + 5 = 2; ∂z/∂y 20. ln (x + y + z) + xyz = zex+y+z ; ∂z/∂x, x = 0, y = 1, z = 0

768
 Section 17.4 Higher-Order Partial Derivatives 763

21. Joint-Cost Function A joint-cost function is defined where c denotes the total cost (in dollars) for producing qA units
implicitly by the equation of product A and qB units of product B.
(a) If qA = 6 and qB = 4, find the corresponding value of c.
√ (b) Determine the marginal costs with respect to qA and qB when
c+ c = 12 + qA 9 + qB2 qA = 6 and qB = 4.

Objective 17.4 Higher-Order Partial Derivatives

To compute higher-order partial If z = f (x, y), then not only is z a function of x and y, but also fx and fy are each functions
derivatives.
of x and y, which may themselves have partial derivatives. If we can differentiate fx and
fy , we obtain second-order partial derivatives of f . Symbolically,
fxx means (fx )x fxy means (fx )y
fyx means (fy )x fyy means (fy )y
In terms of ∂-notation,
   
∂2 z ∂ ∂z ∂2 z ∂ ∂z
2
means means
∂x ∂x ∂x ∂y ∂x ∂y ∂x
   
∂2 z ∂ ∂z ∂2 z ∂ ∂z
means means
∂x ∂y ∂x ∂y ∂y2 ∂y ∂y
Note that to find fxy , we first differentiate f with respect to x. For ∂2 z/∂x ∂y, we first
CAUTION differentiate with respect to y.
For z = f (x, y), fxy = ∂2 z/∂y ∂x. We can extend our notation beyond second-order partial derivatives. For example,
fxxy ( = ∂3 z/∂y ∂x 2 ) is a third-order partial derivative of f , namely, the partial derivative of
fxx ( = ∂2 z/∂x 2 ) with respect to y. The generalization of higher-order partial derivatives
to functions of more than two variables should be obvious.

EXAMPLE 1 Second-Order Partial Derivatives

Find the four second-order partial derivatives of f (x, y) = x 2 y + x 2 y2 .

Solution: Since
fx (x, y) = 2xy + 2xy2
we have
∂
fxx (x, y) = (2xy + 2xy2 ) = 2y + 2y2
∂x
and
∂
fxy (x, y) = (2xy + 2xy2 ) = 2x + 4xy
∂y
Also, since
fy (x, y) = x 2 + 2x 2 y
we have
∂ 2
fyy (x, y) = (x + 2x 2 y) = 2x 2
∂y
and
∂ 2
fyx (x, y) = (x + 2x 2 y) = 2x + 4xy
∂x
Now Work Problem 1 
The derivatives fxy and fyx are called mixed partial derivatives. Observe in Exam-
ple 1 that fxy (x, y) = fyx (x, y). Under suitable conditions, mixed partial derivatives of a
function are equal; that is, the order of differentiation is of no concern. You may assume
that this is the case for all the functions that we consider.

769
 764 Chapter 17 Multivariable Calculus

EXAMPLE 2 Mixed Partial Derivative


∂3 w 
Find the value of if w = (2x + 3y + 4z)3 .
∂z ∂y ∂x (1,2,3)
Solution: ∂w ∂
= 3(2x + 3y + 4z)2 (2x + 3y + 4z)
∂x ∂x
= 6(2x + 3y + 4z)2
∂2 w ∂
= 6 · 2(2x + 3y + 4z) (2x + 3y + 4z)
∂y ∂x ∂y
= 36(2x + 3y + 4z)
3
∂ w
= 36 · 4 = 144
∂z ∂y ∂x
Thus,

∂3 w 
= 144
∂z ∂y ∂x (1,2,3)
Now Work Problem 3 
EXAMPLE 3 Second-Order Partial Derivative of an Implicit Function16

∂2 z
Determine if z2 = xy.
∂x 2
Solution: By implicit differentiation, we first determine ∂z/∂x:
∂ 2 ∂
(z ) = (xy)
∂x ∂x
∂z
2z = y
∂x
∂z y
= z= 0
∂x 2z
Differentiating both sides with respect to x, we obtain
   
∂ ∂z ∂ 1 −1
= yz
∂x ∂x ∂x 2
∂2 z 1 ∂z
2
= − yz−2
∂x 2 ∂x
Substituting y/(2z) for ∂z/∂x, we have
 
∂2 z 1 −2 y y2
= − yz = − z=
 0
∂x 2 2 2z 4z3
Now Work Problem 23 
PROBLEMS 17.4
In Problems 1–10, find the indicated partial derivatives. 8. f (x, y, z) = x 2 y3 z4 ;
fx (x, y, z), fxz (x, y, z), fzx (x, y, z)
1. f (x, y) = 6xy ;2
fx (x, y), fxy (x, y), fyx (x, y) ∂z ∂2 z
9. z = ln x 2 + y2 ; , 2
2. f (x, y) = 2x 3 y2 + 6x 2 y3 − 3xy; fx (x, y), fxx (x, y) ∂y ∂y
2
3. f (x, y) = 7x 2 + 3y; fy (x, y), fyy (x, y), fyyx (x, y) ln (x + 5) ∂z ∂2 z
10. z = ; ,
y ∂x ∂y ∂x
4. f (x, y) = (x 2 + xy + y2 )(xy + x + y); fx (x, y), fxy (x, y)
In Problems 11–16, find the indicated value.
5. f (x, y) = 9e2xy ; fy (x, y), fyx (x, y), fyxy (x, y)
11. If f (x, y, z) = 7, find fyxx (4, 3, −2).
6. f (x, y) = ln (x 2 + y2 ) + 2; fx (x, y), fxx (x, y), fxy (x, y)
12. If f (x, y, z) = z2 (3x 2 − 4xy3 ), find fxyz (1, 2, 3).
7. f (x, y) = (x + y)2 (xy); fx (x, y), fy (x, y), fxx (x, y), fyy (x, y)

16
Omit if Section 17.3 was not covered.

770
 Section 17.5 Chain Rule 765

13. If f (l, k) = 3l3 k 6 − 2l 2 k 7 , find fk lk (2, 1). 18. For f (x, y) = x 4 y4 + 3x 3 y2 − 7x + 4, show that
14. If f (x, y) = x 3 y2 + x 2 y − x 2 y2 , find fxxy (2, 3) and fxyx (2, 3). fxyx (x, y) = fxxy (x, y)
15. If f (x, y) = y2 ex + ln (xy), find fxyy (1, 1). 19. For f (x, y) = e x 2 +xy+y2
, show that
16. If f (x, y) = x 3 − 6xy2 + x 2 − y3 , find fxy (1, −1).
fxy (x, y) = fyx (x, y)
17. Cost Function Suppose the cost c of producing qA units of
xy
product A and qB units of product B is given by 20. For f (x, y) = e , show that
c= (3qA2 + qB3 + 4) 1/3
fxx (x, y) + fxy (x, y) + fyx (x, y) + fyy (x, y)
and the coupled demand functions for the products are given by = f (x, y)((x + y)2 + 2)
qA = 10 − pA + p2B ∂2 z ∂2 z
21. For z = ln (x 2 + y2 ), show that 2
+ 2 = 0.
and ∂x ∂y
17 ∂2 z
qB = 20 + pA − 11pB 22. If z3 − x 3 − x 2 y − xy2 − y3 = 0, find 2 .
∂x
Find the value of ∂ 2
z
17
23. If z2 − 3x 2 + y2 = 0, find 2 .
∂2 c ∂y
∂qA ∂qB
17 ∂2 z
24. If 2z2 = x 2 + 2xy + xz, find .
when pA = 25 and pB = 4. ∂x ∂y

Objective 17.5 Chain Rule18

To show how to find partial derivatives Suppose a manufacturer of two related products A and B has a joint-cost function
of composite functions by using the
chain rule. given by
c = f (qA , qB )
where c is the total cost of producing quantities qA and qB of A and B, respectively.
Furthermore, suppose the demand functions for the products are
qA = g(pA , pB ) and qB = h(pA , pB )
where pA and pB are the prices per unit of A and B, respectively. Since c is a function
of qA and qB , and since both qA and qB are themselves functions of pA and pB , c can be
viewed as a function of pA and pB . (Appropriately, the variables qA and qB are called
intermediate variables of c.) Consequently, we should be able to determine ∂c/∂pA ,
the rate of change of total cost with respect to the price of A. One way to do this is
to substitute the expressions g(pA , pB ) and h(pA , pB ) for qA and qB , respectively, into
c = f (qA , qB ). Then c is a function of pA and pB , and we can differentiate c with respect
to pA directly. This approach has some drawbacks—especially when f , g, or h is given
by a complicated expression. Another way to approach the problem would be to use
the chain rule (actually a chain rule), which we now state without proof.

Chain Rule
Let z = f (x, y), where both x and y are functions of r and s given by x = x(r, s) and
y = y(r, s). If f , x, and y have continuous partial derivatives, then z is a function of
r and s, and
∂z ∂z ∂x ∂z ∂y
= +
∂r ∂x ∂r ∂y ∂r
and
∂z ∂z ∂x ∂z ∂y
= +
∂s ∂x ∂s ∂y ∂s

17 Omit if Section 17.3 was not covered.

18 This section can be omitted without loss of continuity.

771
 766 Chapter 17 Multivariable Calculus

Note that in the chain rule, the number of intermediate variables of z (two) is the
same as the number of terms that compose each of ∂z/∂r and ∂z/∂s.
Returning to the original situation concerning the manufacturer, we see that if f ,
qA , and qB have continuous partial derivatives, then, by the chain rule,
∂c ∂c ∂qA ∂c ∂qB
= +
∂pA ∂qA ∂pA ∂qB ∂pA

EXAMPLE 1 Rate of Change of Cost

For a manufacturer of cameras and film, the total cost c of producing qC cameras and
qF units of film is given by
c = 30qC + 0.015qC qF + qF + 900
The demand functions for the cameras and film are given by
9000
qC = √ and qF = 2000 − pC − 400pF
pC pF
where pC is the price per camera and pF is the price per unit of film. Find the rate of
change of total cost with respect to the price of the camera when pC = 50 and pF = 2.
Solution: We must first determine ∂c/∂pC . By the chain rule,
∂c ∂c ∂qC ∂c ∂qF
= +
∂pC ∂qC ∂pC ∂qF ∂pC
 
−9000
= (30 + 0.015qF ) 2 √ + (0.015qC + 1)(−1)
pC pF
√
When pC = 50 and pF = 2, then qC = 90 2 and qF = 1150. Substituting these values
into ∂c/∂pC and simplifying, we have

∂c 
≈ −123.2
∂p  pC =50
C
pF =2

Now Work Problem 1 

The chain rule can be extended. For example, suppose z = f (v, w, x, y) and v, w,
x, and y are all functions of r, s, and t. Then, if certain conditions of continuity are
assumed, z is a function of r, s, and t, and we have
∂z ∂z ∂v ∂z ∂w ∂z ∂x ∂z ∂y
= + + +
∂r ∂v ∂r ∂w ∂r ∂x ∂r ∂y ∂r
∂z ∂z ∂v ∂z ∂w ∂z ∂x ∂z ∂y
= + + +
∂s ∂v ∂s ∂w ∂s ∂x ∂s ∂y ∂s
and
∂z ∂z ∂v ∂z ∂w ∂z ∂x ∂z ∂y
= + + +
∂t ∂v ∂t ∂w ∂t ∂x ∂t ∂y ∂t
Observe that the number of intermediate variables of z (four) is the same as the number
of terms that form each of ∂z/∂r, ∂z/∂s, and ∂z/∂t.
Now consider the situation where z = f (x, y) such that x = x(t) and y = y(t).
Then
dz ∂z dx ∂z dy
= +
dt ∂x dt ∂y dt
Use the partial derivative symbols and Here we use the symbol dz/dt rather than ∂z/∂t, since z can be considered a function
the ordinary derivative symbols of one variable t. Likewise, the symbols dx/dt and dy/dt are used rather than ∂x/∂t
appropriately. and ∂y/∂t. As is typical, the number of terms that compose dz/dt equals the number of
intermediate variables of z. Other situations would be treated in a similar way.

772
 Section 17.5 Chain Rule 767

EXAMPLE 2 Chain Rule

a. If w = f (x, y, z) = 3x 2 y + xyz − 4y2 z3 , where

x = 2r − 3s y = 6r + s z =r−s
determine ∂w/∂r and ∂w/∂s.
Solution: Since x, y, and z are functions of r and s, then, by the chain rule,
∂w ∂w ∂x ∂w ∂y ∂w ∂z
= + +
∂r ∂x ∂r ∂y ∂r ∂z ∂r

= (6xy + yz)(2) + (3x 2 + xz − 8yz3 )(6) + (xy − 12y2 z2 )(1)

= x(18x + 13y + 6z) + 2yz(1 − 24z2 − 6yz)

Also,
∂w ∂w ∂x ∂w ∂y ∂w ∂z
= + +
∂s ∂x ∂s ∂y ∂s ∂z ∂s

= (6xy + yz)(−3) + (3x 2 + xz − 8yz3 )(1) + (xy − 12y2 z2 )(−1)

= x(3x − 19y + z) − yz(3 + 8z2 − 12yz)

x + ey
b. If z = , where x = rs + sert and y = 9 + rt, evaluate ∂z/∂s when r = −2,
y
s = 5, and t = 4.
Solution: Since x and y are functions of r, s, and t (note that we can write y =
9 + rt + 0 · s), by the chain rule,
∂z ∂z ∂x ∂z ∂y
= +
∂s ∂x ∂s ∂y ∂s
 
1 ∂z r + ert
= (r + ert ) + · (0) =
y ∂y y
If r = −2, s = 5, and t = 4, then y = 1. Thus,

∂z  −2 + e−8
= = −2 + e−8
∂s  r=−2
s=5
1
t=4

Now Work Problem 13 

EXAMPLE 3 Chain Rule

a. Determine ∂y/∂r if y = x 2 ln (x 4 + 6) and x = (r + 3s)6 .

Solution: By the chain rule,
∂y dy ∂x
=
∂r dx ∂r
 
4x 3
= x2 · 4 + 2x · ln (x 4 + 6) [6(r + 3s)5 ]
x +6
 
5 2x 4 4
= 12x(r + 3s) + ln (x + 6)
x4 + 6

773
 768 Chapter 17 Multivariable Calculus

b. Given that z = exy , x = r − 4s, and y = r − s, find ∂z/∂r in terms of r and s.

Solution: ∂z ∂z ∂x ∂z ∂y
= +
∂r ∂x ∂r ∂y ∂r
= (yexy )(1) + (xexy )(1)
= (x + y)exy
Since x = r − 4s and y = r − s,
∂z
= [(r − 4s) + (r − s)]e(r−4s)(r−s)
∂r
2 −5rs+4s2
= (2r − 5s)er
Now Work Problem 15 
PROBLEMS 17.5
In Problems 1–12, find the indicated derivatives by using the and the coupled demand functions for the products are given by
chain rule.
qA = 10 − pA + p2B
1. z = 5x + 3y, x = 2r + 3s, y = r − 2s; ∂z/∂r, ∂z/∂s
and
2. z = 2x 2 + 3xy + 2y2 , x = r 2 − s2 , y = r 2 + s2 ; ∂z/∂r, ∂z/∂s
√
3. z = ex+y , x = t 2 + 3, y = t 3 ; dz/dt qB = 20 + pA − 11pB
√ ∂c ∂c
4. z = 8x + y, x = t 2 + 3t + 4, y = t 3 + 4; dz/dt Use a chain rule to evaluate and when pA = 25 and
∂pA ∂pB
5. w = x 2 yz + xy2 z + xyz2 , x = et , y = tet , z = t 2 et ; dw/dt pB = 4.
2 2 2 2
6. w = ln (x + y + z ), x = 2 − 3t, y = t + 3, z = 4 − t; dw/dt 18. Suppose w = f (x, y), where x = g(t) and y = h(t).
2 2 3
7. z = (x + xy ) , x = r + s + t, y = 2r − 3s + 8t; ∂z/∂t (a) State a chain rule that gives dw/dt.
2 (b) Suppose h(t) = t, so that w = f (x, t), where x = g(t). Use part
8. z = x2 + y2 , x = r + s − t, y = r − s + t; ∂z/∂r (a) to find dw/dt and simplify your answer.
9. w = x + xyz + z , x = r − s , y = rs, z = r + s2 ;
2 2 2 2 2
∂w/∂s 19. (a) Suppose w is a function of x and y, where both x and y are
2
10. w = ln (xyz), x = r s, y = rs, z = rs ; ∂w/∂r 2 functions of s and t. State a chain rule that expresses ∂w/∂t in
terms of derivatives of these functions. √
11. y = x 2 − 7x + 5, x = 19rs + 2s2 t 2 ; ∂y/∂r (b) Let w = 2x 2 ln |3x − 5y|, where x = s t 2 + 2 and
12. y = 4 − x 2 , x = 2r + 3s − 4t; ∂y/∂t y = t − 3e2−s . Use part (a) to evaluate ∂w/∂t when s = 1 and
t = 0.
13. If z = (4x + 3y)3 , where x = r 2 s and y = r − 2s, evaluate
∂z/∂r when r = 0 and s = 1. 20. Production Function In considering a production function
√ P = f (l, k), where l is labor input and k is capital input, Fon,
14. If z = 2x + 3y, where x = 3t + 5 and y = t 2 + 2t + 1, Boulier, and Goldfarb19 assume that l = Lg(h), where L is the
evaluate dz/dt when t = 1. number of workers, h is the number of hours per day per worker,
15. If w = ex+y+z (x 2 + y2 + z2 ), where x = (r − s)2 , y = (r + s)2 , and g(h) is a labor effectiveness function. In maximizing profit p
and z = (s − r)2 , evaluate ∂w/∂s when r = 1 and s = 1. given by
16. If y = x/(x − 5), where x = 2t 2 − 3rs − r 2 t, evaluate ∂y/∂t p = aP − whL
when r = 0, s = 2, and t = −1.
where a is the price per unit of output and w is the hourly wage
17. Cost Function Suppose the cost c of producing qA units of per worker, Fon, Boulier, and Goldfarb determine ∂p/∂L and
product A and qB units of product B is given by ∂p/∂h. Assume that k is independent of L and h, and determine
c = (3qA2 + qB3 + 4)1/3 these partial derivatives.

Objective 17.6 Maxima and Minima for Functions

To discuss relative maxima and relative of Two Variables
minima, to find critical points, and to
apply the second-derivative test for a We now extend the notion of relative maxima and minima (or relative extrema) to
function of two variables. functions of two variables.

19V. Fon, B. L. Boulier, and R. S. Goldfarb, “The Firm’s Demand for Daily Hours of Work: Some Implications,”
Atlantic Economic Journal, XIII, no. 1 (1985), 36–42.

774
 Section 17.6 Maxima and Minima for Functions of Two Variables 769

Definition
A function z = f (x, y) is said to have a relative maximum at the point (a, b) if, for
all points (x, y) in the plane that are sufficiently close to (a, b), we have
f (a, b) ≥ f (x, y) (1)
For a relative minimum, we replace ≥ by ≤ in Equation (1).

To say that z = f (x, y) has a relative maximum at (a, b) means, geometrically,

that the point (a, b, f (a, b)) on the graph of f is higher than (or is as high as) all other
points on the surface that are “near” (a, b, f (a, b)). In Figure 17.4(a), f has a relative
maximum at (a, b). Similarly, the function f in Figure 17.4(b) has a relative minimum
when x = y = 0, which corresponds to a low point on the surface.

z z

Relative
maximum
point

Graph of f
Graph of f

a b (0, 0, 0)
y y
x x Relative
minimum
(a, b, 0) point
(a) (b)

FIGURE 17.4 Relative extrema.

Recall that in locating extrema for a function y = f (x) of one variable, we examine
those values of x in the domain of f for which f  (x) = 0 or f  (x) does not exist. For
functions of two (or more) variables, a similar procedure is followed. However, for the
functions that concern us, extrema will not occur where a derivative does not exist, and
such situations will be excluded from our consideration.
z z

Surface
Tangent (a, b, f (a, b))
Tangent
Surface

Plane y  b

a a
b b
y y
x (a, b, 0) x Plane x  a (a, b, 0)

(a) (b)

FIGURE 17.5 At relative extremum, fx (x, y) = 0 and fy (x, y) = 0.

Suppose z = f (x, y) has a relative maximum at (a, b), as indicated in Figure 17.5(a).
Then the curve where the plane y = b intersects the surface must have a relative
maximum when x = a. Hence, the slope of the tangent line to the surface in the
x-direction must be 0 at (a, b). Equivalently, fx (x, y) = 0 at (a, b). Similarly, on the

775
 770 Chapter 17 Multivariable Calculus

curve where the plane x = a intersects the surface [Figure 17.5(b)], there must be a
relative maximum when y = b. Thus, in the y-direction, the slope of the tangent to the
surface must be 0 at (a, b). Equivalently, fy (x, y) = 0 at (a, b). Since a similar discussion
applies to a relative minimum, we can combine these results as follows:

Rule 1
If z = f (x, y) has a relative maximum or minimum at (a, b), and if both fx and fy are
defined for all points close to (a, b), it is necessary that (a, b) be a solution of the
system
fx (x, y) = 0
fy (x, y) = 0

CAUTION A point (a, b) for which fx (a, b) = fy (a, b) = 0 is called a critical point of f . Thus,
Rule 1 does not imply that there must be
from Rule 1, we infer that, to locate relative extrema for a function, we should examine
an extremum at a critical point. Just as in its critical points.
the case of functions of one variable, a Two additional comments are in order: First, Rule 1, as well as the notion of a
critical point can give rise to a relative critical point, can be extended to functions of more than two variables. For example,
maximum, a relative minimum, or to locate possible extrema for w = f (x, y, z), we would examine those points for which
neither. A critical point is only a
candidate for a relative extremum.
wx = wy = wz = 0. Second, for a function whose domain is restricted, a thorough
examination for absolute extrema would include a consideration of boundary points.

EXAMPLE 1 Finding Critical Points

Find the critical points of the following functions.

a. f (x, y) = 2x 2 + y2 − 2xy + 5x − 3y + 1.
Solution: Since fx (x, y) = 4x − 2y + 5 and fy (x, y) = 2y − 2x − 3, we solve the system
4x − 2y + 5 = 0
−2x + 2y − 3 = 0
 
This gives x = −1 and y = 21 . Thus, −1, 21 is the only critical point.
b. f (l, k) = l3 + k 3 − lk.

Solution: fl (l, k) = 3l2 − k = 0 (2)
fk (l, k) = 3k 2 − l = 0 (3)
From Equation (2), k = 3l2 . Substituting for k in Equation (3) gives
0 = 27l4 − l = l(27l3 − 1)
Hence, either l = 0 or l = 13 . If l = 0, then k = 0; if l = 13 , then k = 13 . The critical
points are therefore (0, 0) and 13 , 13 .
c. f (x, y, z) = 2x 2 + xy + y2 + 100 − z(x + y − 100).
Solution: Solving the system

 fx (x, y, z) = 4x + y − z = 0
fy (x, y, z) = x + 2y − z = 0
 f (x, y, z) = −x − y + 100 = 0
z

gives the critical point (25, 75, 175), as the reader should verify.
Now Work Problem 1 
EXAMPLE 2 Finding Critical Points

Find the critical points of

f (x, y) = x 2 − 4x + 2y2 + 4y + 7

776
 Section 17.6 Maxima and Minima for Functions of Two Variables 771

Solution: We have fx (x, y) = 2x − 4 and fy (x, y) = 4y + 4. The system

2x − 4 = 0
4y + 4 = 0
gives the critical point (2, −1). Observe that we can write the given function as
f (x, y) = x 2 − 4x + 4 + 2(y2 + 2y + 1) + 1
= (x − 2)2 + 2(y + 1)2 + 1
and f (2, −1) = 1. Clearly, if (x, y) =  (2, −1), then f (x, y) > 1. Hence, a relative
minimum occurs at (2, −1). Moreover, there is an absolute minimum at (2, −1), since
f (x, y) > f (2, −1) for all (x, y) =
 (2, −1).
Now Work Problem 3 
Although in Example 2 we were able to show that the critical point gave rise
to a relative extremum, in many cases this is not so easy to do. There is, however, a
second-derivative test that gives conditions under which a critical point will be a relative
maximum or minimum. We state it now, omitting the proof.

Rule 2 Second-Derivative Test for Functions of Two Variables

Suppose z = f (x, y) has continuous partial derivatives fxx , fyy , and fxy at all points
(x, y) near a critical point (a, b). Let D be the function defined by
D(x, y) = fxx (x, y)fyy (x, y) − (fxy (x, y))2
Then
1. if D(a, b) > 0 and fxx (a, b) < 0, then f has a relative maximum at (a, b);
2. if D(a, b) > 0 and fxx (a, b) > 0, then f has a relative minimum at (a, b);
3. if D(a, b) < 0, then f has a saddle point at (a, b) (see Example 4);
4. if D(a, b) = 0, then no conclusion about an extremum at (a, b) can be drawn, and
further analysis is required.

We remark that when D(a, b) > 0, the sign of fxx (a, b) is necessarily the same as
the sign of fyy (a, b). Thus, when D(a, b) > 0 we can test either fxx (a, b) or fyy (a, b),
whichever is easiest, to make the determination required in parts 1 and 2 of the second
derivative test.

EXAMPLE 3 Applying the Second-Derivative Test

Examine f (x, y) = x 3 + y3 − xy for relative maxima or minima by using the second-

derivative test.
Solution: First we find critical points:
fx (x, y) = 3x 2 − y fy (x, y) = 3y2 − x

 asin Example 1(b), solving fx (x, y) = fy (x, y) = 0 gives the critical

In the same manner
points (0, 0) and 13 , 13 . Now,
fxx (x, y) = 6x fyy (x, y) = 6y fxy (x, y) = −1
Thus,
D(x, y) = (6x)(6y) − (−1)2 = 36xy−1

Since D(0, 0) =  −
 36(0)(0)  1 = −1 < 0, there is no
 relative
 extremum
  at (0, 0).
Also, since D 13 , 13 = 36 13 13 − 1 = 3 > 0 and fxx 13 , 13 = 6 13 = 2 > 0, there

777
 772 Chapter 17 Multivariable Calculus
 
is a relative minimum at 13 , 13 . At this point, the value of the function is
   3  3    
f 13 , 13 = 13 + 13 − 13 13 = − 27 1

Now Work Problem 7 

EXAMPLE 4 A Saddle Point

Examine f (x, y) = y2 − x 2 for relative extrema.

The surface in Figure 17.6 is called a
hyperbolic paraboloid.
Solution: Solving
fx (x, y) = −2x = 0 and fy (x, y) = 2y = 0
we get the critical point (0, 0). Now we apply the second-derivative test. At (0, 0), and
indeed at any point,
fxx (x, y) = −2 fyy (x, y) = 2 fxy (x, y) = 0
2
Because D(0, 0) = (−2)(2) − (0) = −4 < 0, no relative extremum exists at (0, 0). A
sketch of z = f (x, y) = y2 − x 2 appears in Figure 17.6. Note that, for the surface curve
cut by the plane y = 0, there is a maximum at (0, 0); but for the surface curve cut by the
plane x = 0, there is a minimum at (0, 0). Thus, on the surface, no relative extremum
can exist at the origin, although (0, 0) is a critical point. Around the origin the curve is
saddle shaped, and (0, 0) is called a saddle point of f .
Now Work Problem 11 
z

z  f (x, y)  y2  x2

Saddle point
at (0, 0)

(0, 0)

x fx(0, 0)  fy(0, 0)  0

FIGURE 17.6 Saddle point.

EXAMPLE 5 Finding Relative Extrema

Examine f (x, y) = x 4 + (x − y)4 for relative extrema.

Solution: If we set
fx (x, y) = 4x 3 + 4(x − y)3 = 0 (4)
and
fy (x, y) = −4(x − y)3 = 0 (5)
then, from Equation (5), we have x − y = 0, or x = y. Substituting into Equation (4)
gives 4x 3 = 0, or x = 0. Thus, x = y = 0, and (0, 0) is the only critical point. At (0, 0),
fxx (x, y) = 12x 2 + 12(x − y)2 = 0
fyy (x, y) = 12(x − y)2 = 0

778
 Section 17.6 Maxima and Minima for Functions of Two Variables 773

and
fxy (x, y) = −12(x − y)2 = 0
Hence, D(0, 0) = 0, and the second-derivative test gives no information. However, for
all (x, y) =
 (0, 0), we have f (x, y) > 0, whereas f (0, 0) = 0. Therefore, at (0, 0) the
graph of f has a low point, and we conclude that f has a relative (and absolute) minimum
at (0, 0).
Now Work Problem 13 
Applications
In many situations involving functions of two variables, and especially in their appli-
cations, the nature of the given problem is an indicator of whether a critical point is
in fact a relative (or absolute) maximum or a relative (or absolute) minimum. In such
cases, the second-derivative test is not needed. Often, in mathematical studies of applied
problems, the appropriate second-order conditions are assumed to hold.

EXAMPLE 6 Maximizing Output

Let P be a production function given by

P = f (l, k) = 0.54l2 − 0.02l 3 + 1.89k 2 − 0.09k 3
where l and k are the amounts of labor and capital, respectively, and P is the quantity
of output produced. Find the values of l and k that maximize P.
Solution: To find the critical points, we solve the system Pl = 0 and Pk = 0:
Pl = 1.08l − 0.06l 2 Pk = 3.78k − 0.27k 2
= 0.06l(18 − l) = 0 = 0.27k(14 − k) = 0
l = 0, l = 18 k = 0, k = 14
There are four critical points: (0, 0), (0, 14), (18, 0), and (18, 14).
Now we apply the second-derivative test to each critical point. We have
Pll = 1.08 − 0.12l Pkk = 3.78 − 0.54k Plk = 0
Thus,
D(l, k) = Pll Pkk − [Plk ]2
= (1.08 − 0.12l)(3.78 − 0.54k)
At (0, 0),
D(0, 0) = 1.08(3.78) > 0
Since D(0, 0) > 0 and Pll = 1.08 > 0, there is a relative minimum at (0, 0). At (0, 14),

D(0, 14) = 1.08(−3.78) < 0

Because D(0, 14) < 0, there is no relative extremum at (0, 14). At (18, 0),

D(18, 0) = (−1.08)(3.78) < 0

Since D(18, 0) < 0, there is no relative extremum at (18, 0). At (18, 14),

D(18, 14) = (−1.08)(−3.78) > 0

Because D(18, 14) > 0 and Pll = −1.08 < 0, there is a relative maximum at (18, 14).
Hence, the maximum output is obtained when l = 18 and k = 14.
Now Work Problem 21 
779
 774 Chapter 17 Multivariable Calculus

EXAMPLE 7 Profit Maximization

A candy company produces two types of candy, A and B, for which the average costs
of production are constant at $2 and $3 per pound, respectively. The quantities qA , qB (in
pounds) of A and B that can be sold each week are given by the joint-demand functions
qA = 400(pB − pA )
and
qB = 400(9 + pA − 2pB )
where pA and pB are the selling prices (in dollars per pound) of A and B, respectively.
Determine the selling prices that will maximize the company’s profit P.
Solution: The total profit is given by
     
profit pounds profit pounds
P =  per pound   of A  +  per pound   of B 
of A sold of B sold
For A and B, the profits per pound are pA − 2 and pB − 3, respectively. Thus,

P = (pA − 2)qA + (pB − 3)qB

= (pA − 2)[400(pB − pA )] + (pB − 3)[400(9 + pA − 2pB )]

Notice that P is expressed as a function of two variables, pA and pB . To maximize P,

we set its partial derivatives equal to 0:
∂P
= (pA − 2)[400(−1)] + [400(pB − pA )](1) + (pB − 3)[400(1)]
∂pA
=0
∂P
= (pA − 2)[400(1)] + (pB − 3)[400(−2)] + 400(9 + pA − 2pB )](1)
∂pB
=0

Simplifying the preceding two equations gives

−2pA + 2pB − 1 = 0
2pA − 4pB + 13 = 0
whose solution is pA = 5.5 and pB = 6. Moreover, we find that
∂2 P ∂2 P ∂2 P
= −800 = −1600 = 800
∂p2A ∂p2B ∂pB ∂pA
Therefore,

D(5.5, 6) = (−800)(−1600) − (800)2 > 0

Since ∂2 P/∂p2A < 0, we indeed have a maximum, and the company should sell candy
A at $5.50 per pound and B at $6.00 per pound.
Now Work Problem 23 
EXAMPLE 8 Profit Maximization for a Monopolist20

Suppose a monopolist is practicing price discrimination by selling the same product

in two separate markets at different prices. Let qA be the number of units sold in market

20 Omit if Section 17.5 was not covered.

780
 Section 17.6 Maxima and Minima for Functions of Two Variables 775

A, where the demand function is pA = f (qA ), and let qB be the number of units sold in
market B, where the demand function is pB = g(qB ). Then the revenue functions for
the two markets are
rA = qA f (qA ) and rB = qB g(qB )
Assume that all units are produced at one plant, and let the cost function for producing
q = qA + qB units be c = c(q). Keep in mind that rA is a function of qA and rB is a
function of qB . The monopolist’s profit P is
P = rA + rB − c
To maximize P with respect to outputs qA and qB , we set its partial derivatives equal to
zero. To begin with,
∂P drA ∂c
= +0−
∂qA dqA ∂qA
drA dc ∂q
= − =0 chain rule
dqA dq ∂qA
Because
∂q ∂
= (qA + qB ) = 1
∂qA ∂qA
we have
∂P drA dc
= − =0 (6)
∂qA dqA dq
Similarly,
∂P drB dc
= − =0 (7)
∂qB dqB dq
From Equations (6) and (7), we get
drA dc drB
= =
dqA dq dqB
But drA /dqA and drB /dqB are marginal revenues, and dc/dq is marginal cost. Hence,
to maximize profit, it is necessary to charge prices (and distribute output) so that the
marginal revenues in both markets will be the same and, loosely speaking, will also be
equal to the cost of the last unit produced in the plant.
Now Work Problem 25 
PROBLEMS 17.6
In Problems 1–6, find the critical points of the functions. 3
10. f (x, y) = 2x 2 + y2 + 3xy − 10x − 9y + 2
2 2
1. f (x, y) = x − 3y − 8x + 9y + 3xy 2
11. f (x, y) = x 2 + 3xy + y2 − 9x − 11y + 3
2. f (x, y) = x 2 + 4y2 − 6x + 16y
5 2 15 x3
3. f (x, y) = x 3 + y3 − x 2 + y2 − 4y + 7 12. f (x, y) = + y2 − 2x + 2y − 2xy
3 3 2 3
4. f (x, y) = xy − x + y 1
13. f (x, y) = (x 3 + 8y3 ) − 2(x 2 + y2 ) + 1
3
5. f (x, y, z) = 2x 2 + xy + y2 + 100 − z(x + y − 200)
14. f (x, y) = x 2 + y2 − xy + x 3
6. f (x, y, z, w) = x 2 + y2 + z2 + w(x + y + z − 3)
l2
In Problems 7–20, find the critical points of the functions. For 15. f (l, k) = + 2lk + 3k 2 − 69l − 164k + 17
each critical point, determine, by the second-derivative test, 2
1 1
whether it corresponds to a relative maximum, to a relative 16. f (l, k) = l2 + 4k 2 − 4lk 17. f (p, q) = pq − −
p q
minimum, or to neither, or whether the test gives no information.
7. f (x, y) = x 2 + 3y2 + 4x − 9y + 3 18. f (x, y) = (x − 3)(y − 3)(x + y − 3)

8. f (x, y) = −2x 2 + 8x − 3y2 + 24y + 7 19. f (x, y) = (y2 − 4)(ex − 1)

9. f (x, y) = y − y2 − 3x − 6x 2 20. f (x, y) = ln (xy) + 2x 2 − xy − 6x

781
 776 Chapter 17 Multivariable Calculus

21. Maximizing Output Suppose that maximize profit. What are the quantities of A and B that
correspond to these prices? What is the total profit?
P = f (l, k) = 2.18l2 − 0.02l 3 + 1.97k 2 − 0.03k 3
29. Cost An open-top rectangular box is to have a volume of
is a production function for a firm. Find the quantities of inputs l 6 ft3 . The cost per square foot of materials is $3 for the bottom, $1
and k that maximize output P. for the front and back, and $0.50 for the other two sides. Find the
22. Maximizing Output In a certain office, computers C and dimensions of the box so that the cost of materials is minimized.
D are utilized for c and d hours, respectively. If daily output Q (See Figure 17.7.)
is a function of c and d, namely,
Q = 18c + 20d − 2c2 − 4d 2 − cd
find the values of c and d that maximize Q.
x  width
In Problems 23–35, unless otherwise indicated, the variables pA y  length
y
and pB denote selling prices of products A and B, respectively. z
z  height
Similarly, qA and qB denote quantities of A and B that are
produced and sold during some time period. In all cases, the x
variables employed will be assumed to be units of output, input, Front
money, and so on.
23. Profit A candy company produces two varieties of candy, FIGURE 17.7
A and B, for which the constant average costs of production are 60
and 70 (cents per lb), respectively. The demand functions for 30. Collusion Suppose A and B are the only two firms in the
A and B are given by market selling the same product. (We say that they are duopolists.)
The industry demand function for the product is
qA = 5(pB − pA ) and qB = 500 + 5(pA − 2pB )
p = 92 − qA − qB
Find the selling prices pA and pB that maximize the company’s
profit. where qA and qB denote the output produced and sold by A and B,
24. Profit Repeat Problem 23 if the constant costs of respectively. For A, the cost function is cA = 10qA ; for B, it is
production of A and B are a and b (cents per lb), respectively. cB = 0.5qB2 . Suppose the firms decide to enter into an agreement
on output and price control by jointly acting as a monopoly. In this
25. Price Discrimination Suppose a monopolist is practicing
case, we say they enter into collusion. Show that the profit
price discrimination in the sale of a product by charging different
function for the monopoly is given by
prices in two separate markets. In market A the demand function is
pA = 100 − qA P = pqA − cA + pqB − cB

and in B it is Express P as a function of qA and qB , and determine how output

should be allocated so as to maximize the profit of the monopoly.
pB = 84 − qB
31. Suppose f (x, y) = x 2 + 3y2 + 9, where x and y must satisfy
where qA and qB are the quantities sold per week in A and B, and the equation x + y = 2. Find the relative extrema of f , subject to
pA and pB are the respective prices per unit. If the monopolist’s the given condition on x and y, by first solving the second
cost function is equation for y (or x). Substitute the result in the first equation.
Thus, f is expressed as a function of one variable. Now find where
c = 600 + 4(qA + qB )
relative extrema for f occur.
how much should be sold in each market to maximize profit? What 32. Repeat Problem 31 if f (x, y) = x 2 + 4y2 + 6, subject to the
selling prices give this maximum profit? Find the maximum profit. condition that 2x − 8y = 20.
26. Profit A monopolist sells two competitive products, A and 33. Suppose the joint-cost function
B, for which the demand functions are
c = qA2 + 3qB2 + 2qA qB + aqA + bqB + d
qA = 16 − pA + pB and qB = 24 + 2pA − 4pB
If the constant average cost of producing a unit of A is 2 and a unit has a relative minimum value of 15 when qA = 3 and qB = 1.
of B is 4, how many units of A and B should be sold to maximize Determine the values of the constants a, b, and d.
the monopolist’s profit? 34. Suppose that the function f (x, y) has continuous partial
27. Profit For products A and B, the joint-cost function for a derivatives fxx , fyy , and fxy at all points (x, y) near a critical point
manufacturer is (a, b). Let D(x, y) = fxx (x, y)fyy (x, y) − (fxy (x, y))2 and suppose that
D(a, b) > 0.
3
c = qA2 + 3qB2 (a) Show that fxx (a, b) < 0 if and only if fyy (a, b) < 0.
2
(b) Show that fxx (a, b) > 0 if and only if fyy (a, b) > 0.
and the demand functions are pA = 60 − qA2 and pB = 72 − 2qB2 .
35. Profit from Competitive Products A monopolist sells two
Find the level of production that maximizes profit.
competitive products, A and B, for which the demand equations
28. Profit For a monopolist’s products A and B, the joint-cost are
function is c = 2(qA + qB + qA qB ), and the demand functions are
qA = 20 − 2pA and qB = 10 − pB . Find the values of pA and pB pA = 35 − 2qA2 + qB

782
 Section 17.7 Lagrange Multipliers 777

and Find the values of x and y for which the profit is a relative
maximum. Use the second-derivative test to verify that your
pB = 20 − qB + qA answer corresponds to a relative maximum profit.
The joint-cost function is 37. Profit from Tomato Crop The revenue (in dollars per
square meter of ground) obtained from the sale of a crop of
1
c = −8 − 2qA3 + 3qA qB + 30qA + 12qB + qA2 tomatoes grown in an artificially heated greenhouse is given by
2
(a) How many units of A and B should be sold to obtain a relative r = 5T (1 − e−x )
maximum profit for the monopolist? Use the second-derivative
test to justify your answer. where T is the temperature (in ◦ C) maintained in the greenhouse
(b) Determine the selling prices required to realize the relative and x is the amount of fertilizer applied per square meter. The cost
maximum profit. Also, find this relative maximum profit. of fertilizer is 20x dollars per square meter, and the cost of heating
is given by 0.1T 2 dollars per square meter.
36. Profit and Advertising A retailer has determined that the
number of TV sets he can sell per week is (a) Find an expression, in terms of T and x, for the profit per
square meter obtained from the sale of the crop of tomatoes.
7x 4y (b) Verify that the pairs
+
2+x 5+y
(T , x) = (20, ln 5) and (T , x) = (5, ln 45 )
where x and y represent his weekly expenditures (in dollars) on
newspaper and radio advertising, respectively. The profit is $300 are critical points of the profit function in part (a). (Note: You need
per sale, less the cost of advertising, so the weekly profit is given not derive the pairs.)
by the formula (c) The points in part (b) are the only critical points of the profit
  function in part (a). Use the second-derivative test to determine
7x 4y whether either of these points corresponds to a relative maximum
P = 300 + −x−y
2+x 5+y profit per square meter.

Objective 17.7 Lagrange Multipliers

To find critical points for a function, We will now find relative maxima and minima for a function on which certain constraints
subject to constraints, by applying the
method of Lagrange multipliers. are imposed. Such a situation could arise if a manufacturer wished to minimize a joint-
cost function and yet obtain a particular production level.
Suppose we want to find the relative extrema of
w = x 2 + y2 + z 2 (1)
subject to the constraint that x, y, and z must satisfy
x − y + 2z = 6 (2)
We can transform w, which is a function of three variables, into a function of two
variables such that the new function reflects constraint (2). Solving Equation (2) for x,
we get
x = y − 2z + 6 (3)
which, when substituted for x in Equation (1), gives
w = (y − 2z + 6)2 + y2 + z2 (4)
Since w is now expressed as a function of two variables, to find relative extrema we
follow the usual procedure of setting the partial derivatives of w equal to 0:
∂w
= 2(y − 2z + 6) + 2y = 4y − 4z + 12 = 0 (5)
∂y
∂w
= −4(y − 2z + 6) + 2z = −4y + 10z − 24 = 0 (6)
∂z
Solving Equations (5) and (6) simultaneously gives y = −1 and z = 2. Substituting
into Equation (3), we get x = 1. Hence, the only critical point of Equation (1) subject to
the constraint represented by Equation (2) is (1, −1, 2). By using the second-derivative
test on Equation (4) when y = −1 and z = 2, we have
∂2 w ∂2 w ∂2 w
=4 = 10 = −4
∂y2 ∂z2 ∂z ∂y
D(−1, 2) = 4(10) − (−4)2 = 24 > 0

783
 778 Chapter 17 Multivariable Calculus

Thus w, subject to the constraint, has a relative minimum at (1, −1, 2).
This solution was found by using the constraint to express one of the variables in
the original function in terms of the other variables. Often this is not practical, but there
is another technique, called the method of Lagrange multipliers,21 that avoids this
step and yet allows us to obtain critical points.
The method is as follows. Suppose we have a function f (x, y, z) subject to the
constraint g(x, y, z) = 0. We construct a new function F of four variables defined by
the following (where λ is a Greek letter read “lambda”):
F(x, y, z, λ) = f (x, y, z) − λg(x, y, z)
It can be shown that if (a, b, c) is a critical point of f , subject to the constraint g(x, y, z) =
0, there exists a value of λ, say, λ0 , such that (a, b, c, λ0 ) is a critical point of F. The
number λ0 is called a Lagrange multiplier. Also, if (a, b, c, λ0 ) is a critical point of F,
then (a, b, c) is a critical point of f , subject to the constraint. Thus, to find critical points
of f , subject to g(x, y, z) = 0, we instead find critical points of F. These are obtained
by solving the simultaneous equations

 F (x, y, z, λ) = 0
 x
Fy (x, y, z, λ) = 0
 Fz (x, y, z, λ) = 0

Fλ (x, y, z, λ) = 0
At times, ingenuity must be used to solve the equations. Once we obtain a critical point
(a, b, c, λ0 ) of F, we can conclude that (a, b, c) is a critical point of f , subject to the
constraint g(x, y, z) = 0. Although f and g are functions of three variables, the method
of Lagrange multipliers can be extended to n variables.
Let us illustrate the method of Lagrange multipliers for the original situation,
namely,
f (x, y, z) = x2 + y2 + z2 subject to x − y + 2z = 6
First, we write the constraint as g(x, y, z) = x − y + 2z − 6 = 0. Second, we form the
function
F(x, y, z, λ) = f (x, y, z) − λg(x, y, z)
= x2 + y2 + z2 − λ(x − y + 2z − 6)
Next, we set each partial derivative of F equal to 0. For convenience, we will write
Fx (x, y, z, λ) as Fx , and so on:


 Fx = 2x − λ = 0 (7)



 Fy = 2y + λ = 0 (8)


 Fz = 2z − 2λ = 0 (9)


Fλ = −x + y − 2z + 6 = 0 (10)
From Equations (7)–(9), we see immediately that
λ λ
x= y=− z=λ (11)
2 2
Substituting these values into Equation (10), we obtain
λ λ
− − − 2λ + 6 = 0
2 2
−3λ + 6 = 0
λ=2
Thus, from Equation (11),
x=1 y = −1 z=2

21After the French mathematician Joseph-Louis Lagrange (1736–1813).

784
 Section 17.7 Lagrange Multipliers 779

Hence, the only critical point of f , subject to the constraint, is (1, −1, 2), at which there
may exist a relative maximum, a relative minimum, or neither of these. The method
of Lagrange multipliers does not directly indicate which of these possibilities occurs,
although from our previous work, we saw that (1, −1, 2) is indeed a relative mini-
mum. In applied problems, the nature of the problem itself may give a clue as to how
a critical point is to be regarded. Often the existence of either a relative minimum or
a relative maximum is assumed, and a critical point is treated accordingly. Actually,
sufficient second-order conditions for relative extrema are available, but we will not
consider them.

EXAMPLE 1 Method of Lagrange Multipliers

Find the critical points for z = f (x, y) = 3x − y + 6, subject to the constraint

x2 + y2 = 4.
Solution: We write the constraint as g(x, y) = x 2 + y2 − 4 = 0 and construct the
function
F(x, y, λ) = f (x, y) − λg(x, y) = 3x − y + 6 − λ(x 2 + y2 − 4)
Setting Fx = Fy = Fλ = 0, we have


 3 − 2xλ = 0 (12)


−1 − 2yλ = 0 (13)



−x − y2 + 4 = 0
2
(14)
From Equations (12) and (13), we can express x and y in terms of λ. Then we will
substitute for x and y in Equation (14) and solve for λ. Knowing λ, we can find x and
y. To begin, from Equations (12) and (13), we have
3 1
x= and y = −
2λ 2λ
Substituting into Equation (14), we obtain
9 1
− 2 − 2 +4=0
4λ 4λ
10
− 2 +4=0
4λ
√
10
λ=±
4
√
With these λ-values, we can find x and y. If λ = 10/4, then
√ √
3 3 10 1 10
x = √  = y = − √  = −
10 5 10 5
2 2
4 4
√
Similarly, if λ = − 10/4,
√ √
3 10 10
x=− y=
5 5
√ √
Thus,
√ the critical
√ points of f , subject to the constraint, are (3 10/5, − 10/5) and
(−3 10/5, 10/5). Note that the values of λ do not appear in the answer; they are
simply a means to obtain the solution.
Now Work Problem 1 
EXAMPLE 2 Method of Lagrange Multipliers

Find critical points for f (x, y, z) = xyz, where xyz =

 0, subject to the constraint
x + 2y + 3z = 36.

785
 780 Chapter 17 Multivariable Calculus

Solution: We have
F(x, y, z, λ) = xyz − λ(x + 2y + 3z − 36)
Setting Fx = Fy = Fz = Fλ = 0 gives, respectively,

 yz − λ = 0

xz − 2λ = 0
xy − 3λ = 0

−x − 2y − 3z + 36 = 0
Because we cannot directly express x, y, and z in terms of λ only, we cannot follow the
procedure in Example 1. However, observe that we can express the products yz, xz, and
xy as multiples of λ. This suggests that, by looking at quotients of equations, we can
obtain a relation between two variables that does not involve λ. (The λ’s will cancel.)
Proceeding to do this, we write the foregoing system as
 yz = λ (15)




 xz = 2λ (16)


 xy = 3λ (17)


x + 2y + 3z − 36 = 0 (18)
Dividing each side of Equation (15) by the corresponding side of Equation (16), we get
yz λ x
= so y =
xz 2λ 2
This division is valid, since xyz =
 0. Similarly, from Equations (15) and (17), we get
yz λ x
= so z =
xy 3λ 3
Now that we have y and z expressed in terms of x only, we can substitute into Equa-
tion (18) and solve for x:
x x
x+2 +3 − 36 = 0
2 3
x = 12
Thus, y = 6 and z = 4. Hence, (12, 6, 4) is the only critical point satisfying the given
conditions. Note that in this situation, we found the critical point without having to find
the value for λ.
Now Work Problem 7 
EXAMPLE 3 Minimizing Costs

Suppose a firm has an order for 200 units of its product and wishes to distribute its
manufacture between two of its plants, plant 1 and plant 2. Let q1 and q2 denote the
outputs of plants 1 and 2, respectively, and suppose the total-cost function is given by
c = f (q1 , q2 ) = 2q12 + q1 q2 + q22 + 200. How should the output be distributed in order
to minimize costs?
Solution: We minimize c = f (q1 , q2 ), given the constraint q1 + q2 = 200. We have
F(q1 , q2 , λ) = 2q12 + q1 q2 + q22 + 200 − λ(q1 + q2 − 200)

 ∂F


 = 4q1 + q2 − λ = 0 (19)

 ∂q1



 ∂F
= q1 + 2q2 − λ = 0 (20)

 ∂q2





 ∂F

 = −q1 − q2 + 200 = 0 (21)
∂λ
786
 Section 17.7 Lagrange Multipliers 781

We can eliminate λ from Equations (19) and (20) and obtain a relation between q1 and
q2 . Then, solving this equation for q2 in terms of q1 and substituting into Equation (21),
we can find q1 . We begin by subtracting Equation (20) from Equation (19), which gives
3q1 − q2 = 0 so q2 = 3q1
Substituting into Equation (21), we have
−q1 − 3q1 + 200 = 0
−4q1 = −200
q1 = 50
Thus, q2 = 150. Accordingly, plant 1 should produce 50 units and plant 2 should
produce 150 units in order to minimize costs.
Now Work Problem 13 
An interesting observation can be made concerning Example 3. From Equation (19),
λ = 4q1 + q2 = ∂c/∂q1 , the marginal cost of plant 1. From Equation (20), λ =
q1 + 2q2 = ∂c/∂q2 , the marginal cost of plant 2. Hence, ∂c/∂q1 = ∂c/∂q2 , and we
conclude that, to minimize cost, it is necessary that the marginal costs of each plant be
equal to each other.

EXAMPLE 4 Least-Cost Input Combination

Suppose a firm must produce a given quantity P0 of output in the cheapest possible
manner. If there are two input factors l and k, and their prices per unit are fixed at pl
and pk , respectively, discuss the economic significance of combining input to achieve
least cost. That is, describe the least-cost input combination.
Solution: Let P = f (l, k) be the production function. Then we must minimize the cost
function
c = lpl + kpk
subject to
P0 = f (l, k)
We construct
F(l, k, λ) = lpl + kpk − λ[f (l, k) − P0 ]
We have


 ∂F ∂

 = pl − λ [f (l, k)] = 0 (22)

 ∂l ∂l

 ∂F ∂
= pk − λ [f (l, k)] = 0 (23)

 ∂k ∂k



 ∂F

 = −f (l, k) + P0 = 0
∂λ
From Equations (22) and (23),
pl pk
λ= = (24)
∂ ∂
[f (l, k)] [f (l, k)]
∂l ∂k
Hence,
∂
[f (l, k)]
pl
= ∂l
pk ∂
[f (l, k)]
∂k
787
 782 Chapter 17 Multivariable Calculus

We conclude that when the least-cost combination of factors is used, the ratio of the
marginal productivities of the input factors must be equal to the ratio of their corre-
sponding unit prices.
Now Work Problem 15 

Multiple Constraints
The method of Lagrange multipliers is by no means restricted to problems involving
a single constraint. For example, suppose f (x, y, z, w) were subject to constraints
g1 (x, y, z, w) = 0 and g2 (x, y, z, w) = 0. Then there would be two lambdas, λ1 and
λ2 (one corresponding to each constraint), and we would construct the function F =
f − λ1 g1 − λ2 g2 . We would then solve the system
Fx = Fy = Fz = Fw = Fλ1 = Fλ2 = 0

EXAMPLE 5 Method of Lagrange Multipliers with Two Constraints

Find critical points for f (x, y, z) = xy + yz, subject to the constraints x 2 + y2 = 8 and
yz = 8.
Solution: Set
F(x, y, z, λ1 , λ2 ) = xy + yz − λ1 (x 2 + y2 − 8) − λ2 (yz − 8)
Then


 Fx
 = y − 2xλ1 = 0 (25)



 F = x + z − 2yλ1 − zλ2 = 0 (26)

 y
Fz = y − yλ2 = 0 (27)



Fλ1 = −x 2 − y2 + 8 = 0 (28)




Fλ2 = −yz + 8 = 0 (29)

This appears to be a challenging system to solve. Some ingenuity will come into play.
Here is one sequence of operations that will allow us to find the critical points. We can
write the system as
 y

 = λ1 (30)

 2x



 x + z − 2yλ1 − zλ2 = 0
 (31)
λ2 = 1 (32)



 2
x +y =82
(33)




 8
z= (34)
y
In deriving Equation (30) we assumed x = 0. This is permissible because if x = 0,
then by Equation (25) we have also y = 0, which is impossible because the second
constraint, yz = 8, provides y  = 0. We also used y  = 0 to derive Equations (32) and (34).
Substituting λ2 = 1 from Equation (32) into Equation (31) and simplifying gives
the equation x − 2yλ1 = 0, so
x
λ1 =
2y
Substituting into Equation (30) gives
y x
=
2x 2y
y2 = x 2 (35)

788
 Section 17.7 Lagrange Multipliers 783

Substituting into Equation (33) gives x 2 + x 2 = 8, from which it follows that x = ±2. If
x = 2, then, from Equation (35), we have y = ±2. Similarly, if x = −2, then y = ±2.
Thus, if x = 2 and y = 2, then, from Equation (34), we obtain z = 4. Continuing in
this manner, we obtain four critical points:
(2, 2, 4) (2, −2, −4) (−2, 2, 4) (−2, −2, −4)
Now Work Problem 9 
PROBLEMS 17.7
In Problems 1–12, find, by the method of Lagrange multipliers, the is 10% of sales, less the advertising cost, determine how to
critical points of the functions, subject to the given constraints. allocate the advertising budget in order to maximize the monthly
1. f (x, y) = x 2 + 4y2 + 6; 2x − 8y = 20 profit. (You may assume that the critical point obtained does
2 2
correspond to the maximum profit.)
2. f (x, y) = 3x − 2y + 9; x + y = 1
18. Maximizing Production When l units of labor and k units
3. f (x, y, z) = x 2 + y2 + z2 ; 2x + y − z = 9 of capital are invested, a manufacturer’s total production q is given
4. f (x, y, z) = x + y + z; xyz = 8 by the Cobb–Douglas production function q = 6l2/5 k 3/5 . Each
2 2 unit of labor costs $25 and each unit of capital costs $69. If
5. f (x, y, z) = 2x + xy + y + z; x + 2y + 4z = 3
exactly $25,875 is to be spent on production, determine the
2
6. f (x, y, z) = xyz ; x − y + z = 20 (xyz2 =
 0) numbers of units of labor and capital that should be invested to
7. f (x, y, z) = xyz; x + y + z = 1 (xyz  = 0) maximize production. (You may assume that the maximum occurs
at the critical point obtained.)
8. f (x, y, z) = x 2 + y2 + z2 ; x+y+z =3
2 2 19. Political Advertising Newspaper advertisements for
9. f (x, y, z) = x + 2y − z ; 2x − y = 0, y + z = 0
political parties always have some negative effects. The recently
10. f (x, y, z) = x 2 + y2 + z2 ; x + y + z = 4, x − y + z = 4 elected party assumed that the three most important election
11. f (x, y, z) = xy2 z; x + y + z = 1, x − y + z = 0 (xyz =
 0) issues, X, Y , and Z, had to be mentioned in each ad, with space x,
y, and z units, respectively, allotted to each. The combined bad
12. f (x, y, z, w) = x 2 + 2y2 + 3z2 − w2 ; 4x + 3y + 2z + w = 10 effect of this coverage was estimated by the party’s backroom
13. Production Allocation To fill an order for 100 units of its operative as
product, a firm wishes to distribute production between its two
plants, plant 1 and plant 2. The total-cost function is given by B(x, y, z) = x 2 + y2 + 2z2

c = f (q1 , q2 ) = 0.1q12 + 7q1 + 15q2 + 1000 Aesthetics dictated that the total space for X and Y together must
be 20, and realism suggested that the total space allotted to Y and
where q1 and q2 are the numbers of units produced at plants 1 and Z together must also be 20 units. What values of x, y, and z in each
2, respectively. How should the output be distributed in order to ad would produce the lowest negative effect? (You may assume
minimize costs? (You may assume that the critical point obtained that any critical point obtained provides the minimum effect.)
does correspond to the minimum cost.) 20. Maximizing Profit Suppose a manufacturer’s production
14. Production Allocation Repeat Problem 13 if the cost function is given by
function is
16q = 65 − 4(l − 4)2 − 2(k − 5)2
c= 3q12 + q1 q2 + 2q22
and the cost to the manufacturer is $8 per unit of labor and $16 per
and a total of 200 units are to be produced. unit of capital, so that the total cost (in dollars) is 8l + 16k. The
selling price of the product is $64 per unit.
15. Maximizing Output The production function for a firm is
(a) Express the profit as a function of l and k. Give your answer in
f (l, k) = 12l + 20k − l2 − 2k 2 expanded form.
(b) Find all critical points of the profit function obtained in part
The cost to the firm of l and k is 4 and 8 per unit, respectively. If (a). Apply the second-derivative test at each critical point. If the
the firm wants the total cost of input to be 88, find the greatest profit is a relative maximum at a critical point, compute the
output possible, subject to this budget constraint. (You may corresponding relative maximum profit.
assume that the critical point obtained does correspond to the (c) The profit may be considered a function of l, k, and q (that is,
maximum output.) P = 64q − 8l − 16k), subject to the constraint
16. Maximizing Output Repeat Problem 15, given that
16q = 65 − 4(l − 4)2 − 2(k − 5)2
f (l, k) = 20l + 25k − l2 − 3k 2
Use the method of Lagrange multipliers to find all critical points
and the budget constraint is 2l + 4k = 50. of P = 64q − 8l − 16k, subject to the constraint.
17. Advertising Budget A computer company has a monthly Problems 21–24 refer to the following definition. A utility function
advertising budget of $20,000. Its marketing department estimates is a function that attaches a measure to the satisfaction or utility a
that if x dollars are spent each month on advertising in newspapers consumer gets from the consumption of products per unit of time.
and y dollars per month on television advertising, then the Suppose U = f (x, y) is such a function, where x and y are the
monthly sales will be given by S = 80x 1/4 y3/4 dollars. If the profit amounts of two products, X and Y. The marginal utility of X is

789
 784 Chapter 17 Multivariable Calculus

∂U/∂x and approximately represents the change in total utility 23. U = f (x, y, z) = xyz; pX = pY = pZ = 1, I = 100 (xyz =
 0)
resulting from a one-unit change in consumption of product X per 24. Let U = f (x, y) be a utility function subject to the budget
unit of time. We define the marginal utility of Y in similar fashion. constraint xpX + ypY = I, where pX , pY , and I are constants. Show
If the prices of X and Y are pX and pY , respectively, and the that, to maximize satisfaction, it is necessary that
consumer has an income or budget of I to spend, then the budget
constraint is fx (x, y) fy (x, y)
λ= =
pX pY
xpX + ypY = I
where fx (x, y) and fy (x, y) are the marginal utilities of X and Y ,
In Problems 21–23, find the quantities of each product that the
respectively. Show that fx (x, y)/pX is the marginal utility of one
consumer should buy, subject to the budget, that will allow
dollar’s worth of X. Hence, maximum satisfaction is obtained
maximum satisfaction. That is, in Problems 21 and 22, find values
when the consumer allocates the budget so that the marginal
of x and y that maximize U = f (x, y), subject to xpX + ypY = I.
utility of a dollar’s worth of X is equal to the marginal utility per
Perform a similar procedure for Problem 23. Assume that such a
dollar’s worth of Y . Performing the same procedure as that for
maximum exists.
U = f (x, y), verify that this is true for U = f (x, y, z, w), subject to
21. U = x 3 y3 ; pX = 2, pY = 3, I = 48 (x 3 y3 =
 0) the corresponding budget equation. In each case, λ is called the
22. U = 40x − 8x 2 + 2y − y2 ; pX = 4, pY = 6, I = 100 marginal utility of income.

Objective 17.8 Lines of Regression22

To develop the method of least To study the influence of advertising on sales, a firm compiled the data in Table 17.3.
squares and introduce index numbers.
The variable x denotes advertising expenditures in hundreds of dollars, and the variable
y denotes the resulting sales revenue in thousands of dollars. If each pair (x, y) of data
is plotted, the result is called a scatter diagram [Figure 17.8(a)].
Table 17.3
From an observation of the distribution of the points, it is reasonable to assume that
Expenditures x 2 3 4.5 5.5 7 a relationship exists between x and y and that it is approximately linear. On this basis,
we may fit “by eye” a straight line that approximates the given data [Figure 17.8(b)]
Revenue y 3 6 8 10 11
and, from this line, predict a value of y for a given value of x. The line seems consistent
with the trend of the data, although other lines could be drawn as well. Unfortunately,
determining a line “by eye” is not very objective. We want to apply criteria in specifying
what we will call a line of “best fit.” A frequently used technique is called the method
of least squares.

y y

10 10

5 5

x x
5 5
(a) (b)

FIGURE 17.8 Scatter diagram and straight line approximation to data points.

To apply the method of least squares to the data in Table 17.3, we first assume that
x and y are approximately linearly related and that we can fit a straight line
y = a + bx
4 (1)

22 This section can be omitted without loss of continuity.

790
 Section 17.8 Lines of Regression 785

that approximates the given points by a suitable objective choice of the constants a and
b. For a given value of x in Equation (1), 4 y is the corresponding predicted value of y,
and (x,4y) will be on the line. Our aim is that 4
y be near y.
When x = 2, the observed value of y is 3. Our predicted value of y is obtained by
substituting x = 2 in Equation (1), which yields 4 y = a + 2b. The error of estimation,
y − y, which is
or vertical deviation of the point (2, 3) from the line, is 4
a + 2b − 3
This vertical deviation is indicated (although exaggerated for clarity) in Figure 17.9.
Similarly, the vertical deviation of (3, 6) from the line is a +3b −6, as is also illustrated.
To avoid possible difficulties associated with positive and negative deviations, we will
consider the squares of the deviations and will form the sum S of all such squares for
the given data:
S = (a + 2b − 3)2 + (a + 3b − 6)2 + (a + 4.5b − 8)2
+ (a + 5.5b − 10)2 + (a + 7b − 11)2

^
y

^
y  a  bx
6 (3, 6)

a  3b  6

(3, a  3b)

^
(2, y)
^
y – y  a  2b  3

3 (2, 3)

x
1 2 3

FIGURE 17.9 Vertical deviation of data points from straight line approximation.

The method of least squares requires that we choose as the line of “best fit” the one
obtained by selecting a and b so as to minimize S. We can minimize S with respect to
a and b by solving the system
 ∂S

 =0
∂a

 ∂S = 0
∂b
We have
∂S
= 2(a + 2b − 3) + 2(a + 3b − 6) + 2(a + 4.5b − 8)
∂a
+ 2(a + 5.5b − 10) + 2(a + 7b − 11) = 0
∂S
= 4(a + 2b − 3) + 6(a + 3b − 6) + 9(a + 4.5b − 8)
∂b
+ 11(a + 5.5b − 10) + 14(a + 7b − 11) = 0
which, when simplified, give
10a + 44b = 76
44a + 225b = 384

791
 786 Chapter 17 Multivariable Calculus

Solving for a and b, we obtain

102 248
a= ≈ 0.65 b = ≈ 1.58
157 157
From our calculations of ∂S/∂a and ∂S/∂b, we see that Saa = 10 > 0, Sbb = 225, and
Sab = 44. Thus D = Saa Sbb − (Sab )2 = 10 · 225 − 442 = 314 > 0. It follows from the
second-derivative test of Section 17.6 that S has a minimum value at the critical point.
y = a + bx is
Hence, in the sense of least squares, the line of best fit 4
y = 0.65 + 1.58x
4 (2)

This is, in fact, the line shown in Figure 17.8(b). It is called the least squares line of
y on x or the linear regression line of y on x. The constants a and b are called linear
regression coefficients. With Equation (2), we would predict that when x = 5, the
corresponding value of y is 4 y = 0.65 + 1.58(5) = 8.55.
More generally, suppose we are given the following n pairs of observations:
(x1 , y1 ), (x2 , y2 ), . . . , (xn , yn )
If we assume that x and y are approximately linearly related and that we can fit a
straight line
y = a + bx
4
y − y is
that approximates the data, the sum of the squares of the errors 4
S = (a + bx1 − y1 )2 + (a + bx2 − y2 )2 + · · · + (a + bxn − yn )2
Since S must be minimized with respect to a and b,

 ∂S

 = 2(a + bx1 − y1 ) + 2(a + bx2 − y2 ) + · · · + 2(a + bxn − yn ) = 0
∂a

 ∂S = 2x1 (a + bx1 − y1 ) + 2x2 (a + bx2 − y2 ) + · · · + 2xn (a + bxn − yn ) = 0

∂b
Dividing both equations by 2 and using summation notation, we have
  n 
 n



 na + x i b − yi = 0


i=1 i=1
 n   n  n

   

 xi a + 2
xi b − xi yi = 0


i=1 i=1 i=1

which is a system of two linear equations in a and b, the so-called normal equations:
  n 
  n



 na + x i b = yi

 i=1 i=1

  n   n  n

   

 xi a + 2
xi b = xi yi

i=1 i=1 i=1

The coefficients are of course no more than simple sums of values obtained from the
observed data. The solution is obtained easily using the techniques of Section 3.4.
 n  n   n  n 
   
xi2 yi − xi xi yi
i=1 i=1 i=1 i=1
a=  2 (3)
n
 n

n xi2 − xi
i=1 i=1

792
 Section 17.8 Lines of Regression 787
n
 n
 n 
  
n xi yi − xi yi
i=1 i=1 i=1
b=  2 (4)
n
 n

n xi2 − xi
i=1 i=1

 
Now we have Saa = 2n > 0 and D = Saa Sbb − (Sab )2 = (2n)(2 xi2 ) − (2 xi )2 ,
independent of (a, b). It can be shown that for distinct xi and n ≥ 2 that D > 0 so a and
b, given by Equations (3) and (4), do indeed minimize S. [For example, when n = 2,
D > 0 is provably equivalent to (x1 − x2 )2 > 0, which is true for distinct x1 and x2 .]
Computing the linear regression coefficients a and b by the formulas of Equa-
tions (3) and (4) gives the linear regression line of y on x, namely, 4y = a + bx, which
can be used to estimate y for any given value of x.
In the next example, as well as in the problems, you will encounter index numbers.
They are used to relate a variable in one period of time to the same variable in another
period, called the base period. An index number is a relative number that describes
data that are changing over time. Such data are referred to as time series.
For example, consider the time-series data of total production of widgets in the
United States for 2002–2006, given in Table 17.4. If we choose 2003 as the base year
and assign the index number 100 to it, then the other index numbers are obtained by
dividing each year’s production by the 2003 production of 900 and multiplying the
result by 100. We can, for example, interpret the index 106 for 2006 as meaning that
production for that year was 106% of the production in 2003.

Table 17.4
Production Index
Year (in thousands) (based on 2003)

2002 828 92
2003 900 100
2004 936 104
2005 891 99
2006 954 106

In time-series analysis, index numbers are obviously of great advantage if the

data involve numbers of great magnitude. But regardless of the magnitude of the
data, index numbers simplify the task of comparing changes in data over periods
of time.

EXAMPLE 1 Determining a Linear Regression Line

By means of the linear regression line, use the following table to represent the trend for
the index of total U.S. government revenue from 1995 to 2000 (1995 = 100).

Year 1995 1996 1997 1998 1999 2000

Index 100 107 117 127 135 150

Source: Economic Report of the President, 2001, U.S. Government Printing Office,
Washington, DC, 2001.

Solution: We will let x denote time and y denote the index and treat y as a linear
function of x. Also, we will designate 1995 by x = 1, 1996 by x = 2, and so on. There

793
 788 Chapter 17 Multivariable Calculus

are n = 6 pairs of measurements. To determine the linear regression coefficients by

using Equations (3) and (4), we first perform the arithmetic:

Year xi yi xi yi x2i

1995 1 100 100 1

1996 2 107 214 4
1997 3 117 351 9
1998 4 127 508 16
1999 5 135 675 25
2000 6 150 900 36
Total 21 736 2748 91
6
 6
 6
 6

= xi = yi = xi yi = xi2
i=1 i=1 i=1 i=1
y
150 Hence, by Equation (3),
140 91(736) − 21(2748)
130 a= ≈ 88.3
120 6(91) − (21)2
110 and, by Equation (4),
100
90 6(2748) − 21(736)
80 ˆ b= ≈ 9.83
6(91) − (21)2
x
1 2 3 4 5 6 Thus, the regression line of y on x is
FIGURE 17.10 Linear regression line y = 88.3 + 9.83x
4
for government revenue. whose graph, as well as a scatter diagram, appears in Figure 17.10.
Now Work Problem 1 

The TI-83 Plus has a utility that computes the equation of Next, we press STAT and move to CALC. Finally, pressing
the least squares line for a set of data. We will illustrate 8 and ENTER gives the result shown in Figure 17.12. (The
by giving the procedure for the six data points (xi , yi ) of number r ≈ 0.99448 is called the coefficient of correlation
Example 1. After pressing STAT and ENTER, we enter and is a measure of the degree to which the given data are
all the x-values and then the y-values. (See Figure 17.11.) linearly related.)

FIGURE 17.11 Data of Example 1. FIGURE 17.12 Equation of least squares line.

PROBLEMS 17.8
In this problem set, use a graphing calculator if your instructor 5. Demand A firm finds that when the price of its product
permits you to do so. is p dollars per unit, the number of units sold is q, as indicated in
In Problems 1–4, find an equation of the least squares linear the following table:
regression line of y on x for the given data, and sketch both the
line and the data. Predict the value of y corresponding to x = 3.5. Price, p 10 20 40 50 60 70
1. x 1 2 3 4 5 6 2. x 1 2 3 4 5 6 7
y 1.5 2.3 2.6 3.7 4.0 4.5 y 1 1.8 2 4 4.5 7 9 Demand, q 75 65 56 50 42 34

3. x 2 3 4.5 5.5 7 4. x 2 3 4 5 6 7
y 3 5 8 10 11 y 2.4 2.9 3.3 3.8 4.3 4.9 Find an equation of the regression line of q on p.

794
 Section 17.9 Multiple Integrals 789

6. Water and Crop Yield On a farm, an agronomist finds that 10. Industrial Production In the following table, let 1975
the amount of water applied (in inches) and the corresponding correspond to x = 1, 1977 correspond to x = 3, and so on:
yield of a certain crop (in tons per acre) are as given in the
Index of Industrial Production---Electrical
following table:
Machinery (based on 1977)
Water, x 8 16 24 32 Year Index
1975 77
Yield, y 5.2 5.7 6.3 6.7
1977 100
1979 126
Find an equation of the regression line of y on x. Predict y when
x = 20. 1981 134

7. Virus A rabbit was injected with a virus, and x hours after Source: Economic Report of the President, 1988, U.S.
the injection the temperature y (in degrees Fahrenheit) of the Government Printing Office, Washington, DC, 1988.
rabbit was measured.23 The data are given in the following table:
11. Computer Shipments
(a) Find an equation of the least squares line of y on x for the
Elapsed Time, x 24 32 48 56
following data (refer to 2002 as year x = 1, and so on):
Temperature, y 102.8 104.5 106.5 107.0
Overseas Shipments of Computers by
Acme Computer Co. (in thousands)
Find an equation of the regression line of y on x, and estimate the
rabbit’s temperature 40 hours after the injection. Year Quantity
8. Psychology In a psychological experiment, four persons 2002 35
were subjected to a stimulus. Both before and after the stimulus,
2003 31
the systolic blood pressure (in millimeters of mercury) of each
subject was measured. The data are given in the following table: 2004 26
2005 24
Blood Pressure 2006 26
Before Stimulus, x 131 132 135 141 (b) For the data in part (a), refer to 2002 as year x 
= −2, 2003 as
After Stimulus, y 139 139 142 149 year x = −1, 2004 as year x = 0 and so on. Then 5i=1 xi = 0. Fit
a least squares line and observe how the calculation is simplified.
Find an equation of the regression line of y on x, where x and y are 12. Medical Care For the following time series, find an
as defined in the table. equation of the linear regression line that best fits the data (refer to
For the time series in Problems 9 and 10, fit a linear regression 1983 as year x = −2, 1984 as year x = −1, and so on):
line by least squares; that is, find an equation of the linear
Consumer Price Index---Medical Care,
regression line of y on x. In each case, let the first year in the table
1983--1987 (based on 1967)
correspond to x = 1.
Year Index
9. Production of Product A, 2002--2006 (in 1983 357
thousands of units) 1984 380
Year Production 1985 403
2002 10 1986 434
2003 15 1987 462
2004 16 Source: Economic Report of the President, 1988, U.S.
2005 18 Government Printing Office, Washington, DC, 1988.
2006 21

Objective 17.9 Multiple Integrals

To compute double and triple integrals. Recall that the definite integral of a function of one variable is concerned with integration
over an interval. There are also definite integrals of functions of two variables, called
(definite) double integrals. These involve integration over a region in the plane.

23 R. R. Sokal and F. J. Rohlf, Introduction to Biostatistics (San Francisco: W. H. Freeman & Company, Publishers,
1973).

795
 790 Chapter 17 Multivariable Calculus

y For example, the symbol

3 23 4 3 2 3 4 
xy dx dy = xy dx dy
0 3 0 3
2
is the double integral of f (x, y) = xy over a region determined by the limits of integration.
That region consists of all points (x, y) in the x,y-plane such that 3 ≤ x ≤ 4 and
x 0 ≤ y ≤ 2. (See Figure 17.13.) 
3 4 A double integral is a limit of a sum of the form f (x, y) dx dy, where, in this
FIGURE
2224 17.13 Region over which example, the points (x, y) are in the shaded region. A geometric interpretation of a
0 3
xy dx dy is evaluated. double integral will be given later.
To evaluate
3 23 4 3 2 3 4 
xy dx dy = xy dx dy
0 3 0 3
we use successive integrations starting with the innermost integral. First, we evaluate
3 4
xy dx
3
by treating y as a constant and integrating with respect to x between the limits 3 and 4:
3 4 4
x 2 y 
xy dx =
3 2 3
Substituting the limits for the variable x, we have
4 2 · y 32 · y 16y 9y 7
− = − = y
2 2 2 2 2
Now we integrate this result with respect to y between the limits 0 and 2:
3 2 2
7 7y2  7 · 22
y dy =  = −0=7
0 2 4 0 4
y
Thus,
3 2 3 4
xy dx dy = 7
1 0 3
y  x2 Now consider the double integral
3 1 3 x2 3 3 
1 x2
3 3
y  x3 (x − xy) dy dx = (x − xy) dy dx
0 x3 0 x3
x
1 Here we integrate first with respect to y and then with respect to x. The region over which
FIGURE 17.14 Region over which
the integration takes places is all points (x, y) such that x 3 ≤ y ≤ x 2 and 0 ≤ x ≤ 1.
2 1 2 x2 3
(x − xy) dy dx is evaluated.
(See Figure 17.14.) This double integral is evaluated by first treating x as a constant
0 x3
and integrating x 3 − xy with respect to y between x 3 and x 2 , and then integrating the
result with respect to x between 0 and 1:
3 3 3 3  3  x2
1 x2 1 x2 1
3 3 3xy2 
(x − xy) dy dx = (x − xy) dy dx = x y− dx
0 x3 0 x3 0 2 x 3

3 1    
3 x(x 2 )2
2 3 x(x 3 )2
3
= x (x ) − − x (x ) − dx
0 2 2

3 1   3 1  
5x5 x7 x5 x7
= x − − x6 + dx = − x6 + dx
0 2 2 0 2 2

 1  
x6 x7 x 8  1 1 1 1
= − +  = − + −0=
12 7 16 0 12 7 16 336

796
 Section 17.9 Multiple Integrals 791

EXAMPLE 1 Evaluating a Double Integral

3 1 3 1−x
Find (2x + 1) dy dx.
−1 0
Solution: Here we first integrate with respect to y and then integrate the result with
respect to x: 3 
3 1 3 1−x 3 1 1−x
(2x + 1) dy dx = (2x + 1) dy dx
−1 0 −1 0
3 1 1−x 3 1

= 
(2xy + y) dx = ((2x(1 − x) + (1 − x)) − 0) dx
−1 0 −1
3 1 1 
2x 3
2 x2 
= ( − 2x + x + 1) dx = − + + x 
3 2
−1    −1
2 1 2 1 2
= − + +1 − + −1 =
3 2 3 2 3
Now Work Problem 9 
EXAMPLE 2 Evaluating a Double Integral
3 ln 2 3 2
Find dx dy.
1 ey
Solution: Here we first integrate with respect to x and then integrate the result with
respect to y:
3 ln 2 3 2 3 ln 2 3 2  3 ln 2
2

dx dy = dx dy = x  dy
1 ey 1 ey 1 e y

3 ln 2
ln 2
y

y 
= (2 − e ) dy = (2y − e )
1 1
= (2 ln 2 − 2) − (2 − e) = 2 ln 2 − 4 + e
= ln 4 − 4 + e
Now Work Problem 13 
z

z  f (x, y)

a c
d
b
y
dx
dy

x
2b2d
FIGURE 17.15 Interpreting a c
f (x, y) dy dx in terms of volume, where f (x, y) ≥ 0.

A double integral can be interpreted in terms of the volume of a region between

the x,y-plane and a surface z = f (x, y) if z ≥ 0. In Figure 17.15 is a region whose

797
 792 Chapter 17 Multivariable Calculus

volume we will consider. The element of volume for this region is a vertical column
with height approximately z = f (x, y) and base area dy dx. Thus, its volume is approx-
imately f (x, y) dy dx. The volume of the entire region can be found by summing the
volumes of all such elements for a ≤ x ≤ b and c ≤ y ≤ d via a double integral:
3 b3 d
volume = f (x, y) dy dx
a c
Triple integrals are handled by successively evaluating three integrals, as the next
example shows.

EXAMPLE 3 Evaluating a Triple Integral

3 1 3 x 3 x−y
Find x dz dy dx.
0 0 0
Solution: 3 1 3 x 3 x−y 3 1 3 x 3 x−y 
x dz dy dx = x dz dy dx
0 0 0 0 0 0
3 x−y
1 3 x 3 3 x 1

= (xz) dy dx = (x(x − y) − 0) dy dx
0 0 0 0 0
3 13 x 3 1 3 x 
2 2
= (x − xy) dy dx = (x − xy) dy dx
0 0 0 0
3 1 x 3 1   
xy2  x3
= x2 y − dx = x 3
− − 0 dx
0 2 0 0 2
3 1 3 1
x x4  1
= dx =  =
0 2 8 0 8
Now Work Problem 21 
PROBLEMS 17.9
In Problems 1–22, evaluate the multiple integrals. 23. Statistics In the study of statistics, a joint density function
3 33 4 3 43 3 z = f (x, y) defined on a region in the x,y-plane is represented by a
1. x dy dx 2. y dy dx surface in space. The probability that
0 0 1 0
3 1 3 1 3 1 3 1 a≤x≤b and c ≤ y ≤ d
3. xy dx dy 4. x 2 y2 dy dx is given by
0 0 0 0 3 d 3 b
3 3 3 2 3 3 3 2
2 2 P(a ≤ x ≤ b, c ≤ y ≤ d) = f (x, y) dx dy
5. (x − y) dx dy 6. (y − 2xy) dy dx c a
1 1 −2 0
3 3 3 3 and is represented by the volume between the graph of f and the
1 2 3 x
2 2 rectangular region given by
7. (x + y) dy dx 8. (x + y ) dy dx
0 0 0 0 a≤x≤b and c ≤ y ≤ d
3 3 3 2x 3 2 3 x−1
9. y dy dx 10. 2y dy dx If f (x, y) = e−(x+y) is a joint density function, where x ≥ 0 and
2 0 1 0 y ≥ 0, find
3 1 3 x2 3 2 3 x2
11. 14x 2 y dy dx 12. xy dy dx P(0 ≤ x ≤ 2, 1 ≤ y ≤ 2)
0 3x 0 0
3 3 3 √ 9−x 2
3 1 3 y and give your answer in terms of e.
13. y dy dx 14. x dx dy 24. Statistics In Problem 23, let f (x, y) = 6e−(2x+3y) for
0 0 0 y2
3 3 3 3 x, y ≥ 0. Find
1 1−x 3 3y
15. 3(x + y) dy dx 16. 5x dx dy P(1 ≤ x ≤ 3, 2 ≤ y ≤ 4)
−1 x 0 y2
3 1 3 y 3 1 3 1 and give your answer in terms of e.
17. ex+y dx dy 18. ey−x dx dy 25. Statistics In Problem 23, let f (x, y) = 1, where 0 ≤ x ≤ 1
0 0 0 0
3 1 3 2 3 3 3 1 3 x 3 x+y and 0 ≤ y ≤ 1. Find P(x ≥ 1/2, y ≥ 1/3).
19. xy2 z3 dx dy dz 20. x 2 dz dy dx 26. Statistics In Problem 23, let f be the uniform density
0 0 0 0 0 0
3 13 x3 xy
3 e3 x 3 y function f (x, y) = 1/8 defined over the rectangle
21. dz dy dx 22. dz dy dx 0 ≤ x ≤ 4, 0 ≤ y ≤ 2. Determine the probability that 0 ≤ x ≤ 1
0 x2 0 1 ln x 0 and 0 ≤ y ≤ 1.

798
 Chapter 17 Review 793

Chapter 17 Review
Important Terms and Symbols Examples
Section 17.1 Partial Derivatives 
∂z ∂z 
partial derivative = fx (x, y) = fx (a, b) Ex. 2, p. 751
∂x ∂x (a,b)
Section 17.2 Applications of Partial Derivatives
joint-cost function production function marginal productivity Ex. 3, p. 756
competitive products complementary products Ex. 4, p. 758
Section 17.3 Implicit Partial Differentiation
implicit partial differentiation Ex. 1, p. 761
Section 17.4 Higher-Order Partial Derivatives
∂2 z ∂2 z ∂2 z ∂2 z
= fxy = fyx = fxx = fyy Ex. 1, p. 763
∂y∂x ∂x∂y ∂x 2 ∂y2
Section 17.5 Chain Rule
chain rule intermediate variable Ex. 1, p. 766
Section 17.6 Maxima and Minima for Functions of Two Variables
relative maxima and minima critical point Ex. 1, p. 770
second-derivative test for functions of two variables Ex. 3, p. 771
Section 17.7 Lagrange Multipliers
Lagrange multipliers Ex. 1, p. 779
Section 17.8 Lines of Regression
method of least squares linear regression line of y on x index numbers Ex. 1, p. 787
Section 17.9 Multiple Integrals
double integral triple integral Ex. 3, p. 792

Summary
For a function of n variables, we can consider n partial deriva- Suppose two products, A and B, are such that the quan-
tives. For example, if w = f (x, y, z), we have the partial tity demanded of each is dependent on the prices of both. If
derivatives of f with respect to x, with respect to y, and with qA and qB are the quantities of A and B demanded when the
respect to z, denoted fx , fy , and fz , or ∂w/∂x, ∂w/∂y, and ∂w/∂z, prices of A and B are pA and pB , respectively, then qA and
respectively. To find fx (x, y, z), we treat y and z as constants qB are each functions of pA and pB . When ∂qA /∂pB > 0 and
and differentiate f with respect to x in the usual way. The ∂qB /∂pA > 0, then A and B are called competitive products
other partial derivatives are found in a similar fashion. We (or substitutes). When ∂qA /∂pB < 0 and ∂qB /∂pA < 0, then
can interpret fx (x, y, z) as the approximate change in w that A and B are called complementary products.
results from a one-unit change in x when y and z are held If z = f (x, y), where x = x(r, s) and y = y(r, s), then
fixed. There are similar interpretations for the other partial z can be considered as a function of r and s. To find, for
derivatives. A function of several variables may be defined example, ∂z/∂r, a chain rule can be used:
implicitly. In this case, its partial derivatives are found by ∂z ∂z ∂x ∂z ∂y
implicit partial differentiation. = +
Functions of several variables occur frequently in busi- ∂r ∂x ∂r ∂y ∂r
ness and economic analysis, as well as in other areas of study. A partial derivative of a function of n variables is itself a
If a manufacturer produces x units of product X and y units function of n variables. By successively taking partial deriva-
of product Y, then the total cost c of these units is a func- tives of partial derivatives, we obtain higher-order partial
tion of x and y and is called a joint-cost function. The partial derivatives. For example, if f is a function of x and y, then
derivatives ∂c/∂x and ∂c/∂y are called the marginal costs fxy denotes the partial derivative of fx with respect to y; fxy is
with respect to x and y, respectively. We can interpret, for called the second-partial derivative of f , first with respect to
example, ∂c/∂x as the approximate cost of producing an x and then with respect to y.
extra unit of X while the level of production of Y is held If the function f (x, y) has a relative extremum at (a, b),
fixed. then (a, b) must be a solution of the system
If l units of labor and k units of capital are used to pro-
duce P units of a product, then the function P = f (l, k) is fx (x, y) = 0
called a production function. The partial derivatives of P are fy (x, y) = 0
called marginal productivity functions.

799
 794 Chapter 17 Multivariable Calculus

Any solution of this system is called a critical point of f . Sometimes sample data for two variables, say, x and
Thus, critical points are the candidates at which a relative y, may be related in such a way that the relationship is
extremum may occur. The second-derivative test for func- approximately linear. When such data points (xi , yi ), where
tions of two variables gives conditions under which a critical i = 1, 2, 3, . . . , n, are given to us, we can fit a straight line
point corresponds to a relative maximum or a relative mini- that approximates them. Such a line is the linear regression
mum. The test states that if (a, b) is a critical point of f and line of y on x and is given by

D(x, y) = fxx (x, y)fyy (x, y) − [fxy (x, y)]2 y = a + bx

4
then where

1. if D(a, b) > 0 and fxx (a, b) < 0, then f has a relative  n  n   n  n 

   
maximum at (a, b); xi2 yi − xi xi yi
2. if D(a, b) > 0 and fxx (a, b) > 0, then f has a relative i=1 i=1 i=1 i=1
a=  2
minimum at (a, b); n
 n

3. if D(a, b) < 0, then f has a saddle point at (a, b); n xi2 − xi
4. if D(a, b) = 0, no conclusion about an extremum at i=1 i=1

(a, b) can yet be drawn, and further analysis is required.

and
To find critical points of a function of several variables,   n 
n
 n
 
subject to a constraint, we can sometimes use the method of
Lagrange multipliers. For example, to find the critical points n xi yi − xi yi
i=1 i=1 i=1
of f (x, y, z), subject to the constraint g(x, y, z) = 0, we first b=  2
n n
form the function  
n xi2 − xi
F(x, y, z, λ) = f (x, y, z) − λg(x, y, z) i=1 i=1

By solving the system

The4 y-values can be used to predict y-values for given values

F =0 of x.

 x When working with functions of several variables, we
Fy =0
=0 can consider their multiple integrals. These are determined
 Fz

by successive integration. For example, the double integral
Fλ =0
3 2 3 y
we obtain the critical points of F. If (a, b, c, λ0 ) is such a crit-
(x + y) dx dy
ical point, then (a, b, c) is a critical point of f , subject to the 1 0
constraint. It is important to write the constraint in the form
g(x, y, z) = 0. For example, if the constraint is 2x+3y−z = 4, is determined by first treating y as a constant and integrating
then g(x, y, z) = 2x + 3y − z − 4. If f (x, y, z) is subject to two x + y with respect to x. After evaluating between the limits
constraints, g1 (x, y, z) = 0 and g2 (x, y, z) = 0, then we would 0 and y, we integrate that result with respect to y from y = 1
form the function F = f − λ1 g1 − λ2 g2 and solve the system to y = 2. Thus,
 3 3 3 3 
 Fx =0 2 y 2 y

 Fy
 =0 (x + y) dx dy = (x + y) dx dy
Fz =0 1 0 1 0

 F =0

 λ1
Fλ2 =0 Triple integrals involve functions of three variables and are
also evaluated by successive integration.

Review Problems
In Problems 1–12, find the indicated partial derivatives. 4. f (pA , pB ) = 4(pA − 10) + 5(pB − 15); fpB (pA , pB )
2 2
1. f (x, y) = ln (x + y ); fx (x, y), fy (x, y)
∂
3 3 5. f (x, y) = ln x 2 + y2 ; [f (x, y)]
2. P = l + k − lk; ∂P/∂l, ∂P/∂k ∂y
x ∂z ∂z ∂w 7. w = ex yz ;
2
wxy (x, y, z)
3. z = ; , 6. w = x 2 + y2 ;
x+y ∂x ∂y ∂y

800
 Chapter 17 Review 795

8. f (x, y) = xy ln (xy); fxy (x, y) has a critical point at (x, y) = (0, 1), and the second-derivative test
∂ 2 is inconclusive at this point. Determine the values of the constants
9. f (x, y, z) = (x + y + z)(x 2 + y2 + z2 ); (f (x, y, z)) a, b, and c.
∂z2
10. z = (x 2 − y)(y2 − 2xy); ∂2 z/∂y2 27. Maximizing Profit A dairy produces two types of cheese,
A and B, at constant average costs of 50 cents and 60 cents per
11. w = ex+y+z ln (xyz); ∂3 w/∂z∂y∂x pound, respectively. When the selling price per pound of A is pA
12. P = 100l0.11 k 0.89 ; ∂2 P/∂k∂l cents and of B is pB cents, the demands (in pounds) for A and B,
x+y are, respectively,
13. If f (x, y, z) = , find fxyz (2, 7, 4)
xz qA = 250(pB − pA )
2
14. If f (x, y, z) = (6x + 1)ey ln (z+1) , find fxyz (0, 1, 0) and
15. If w = x 2 + 2xy + 3y2 , x = er , and y = ln (r + s), find ∂w/∂r qB = 32,000 + 250(pA − 2pB )
and ∂w/∂s. Find the selling prices that yield a relative maximum profit. Verify
16. If z = ln (x/y), x = r 2 + s2 , and y = (r + s)2 , find that the profit has a relative maximum at these prices.
∂z/∂r − ∂z/∂s. 28. Find all critical points of f (x, y, z) = xy2 z, subject to the
17. If x 2 + 2xy − 2z2 + xz + 2 = 0, find ∂z/∂x. condition that
x + y + z − 1 = 0 (xyz =  0)
18. If z2 + ln (yz) + ln z + x + z = 0, find ∂z/∂y.
24
19. Production Function If a manufacturer’s production 29. Find all critical points of f (x, y, z) = x 2 + y2 + z2 , subject to
function is defined by P = 20l0.7 k 0.3 , determine the marginal the constraint 3x + 2y + z = 14.
productivity functions. 30. Surviving Infection In an experiment,26 a group of fish
24
20. Joint-Cost Function A manufacturer’s cost for was injected with living bacteria. Of those fish maintained at
producing x units of product X and y units of product Y is given by 28◦ C, the percentage p that survived the infection t hours after the
injection is given in the following table:
c = 3x + 0.05xy + 9y + 500
t 8 10 18 20 48
Determine the (partial) marginal cost with respect to x when
x = 50 and y = 100. p 82 79 78 78 64
24
21. Competitive/Complementary Products If
qA = 100 − pA + 2pB and qB = 150 − 3pA − 2pB , where qA and Find the linear regression line of p on t.
qB are the number of units demanded of products A and B, 31. Equipment Expenditures Find the least squares linear
respectively, and pA and pB are their respective prices per unit, regression line of y on x for the data given in the following table
determine whether A and B are competitive products or (refer to year 1993 as year x = 1, etc.):
complementary products or neither.
24
22. Innovation For industry, the following model describes Equipment Expenditures of Allied
the rate α (a Greek letter read “alpha”) at which an innovation Computer Company, 1993--1998 (in
substitutes for an established process:25 millions of dollars)

α = Z + 0.530P − 0.027S Year Expenditures

1993 15
Here, Z is a constant that depends on the particular industry, P is
an index of profitability of the innovation, and S is an index of the 1994 22
extent of the investment necessary to make use of the innovation. 1995 21
Find ∂α/∂P and ∂α/∂S. 1996 26
23. Examine f (x, y) = x 2 + 2y2 − 2xy − 4y + 3 for relative 1997 27
extrema.
1998 34
24. Examine f (w, z) = 2w3 + 2z3 − 6wz + 7 for relative extrema.
25. Minimizing Material An open-top rectangular cardboard In Problems 32–35, evaluate the double integrals.
3 23 y 3 1 3 y2
box is to have a volume of 32 cubic feet. Find the dimensions of
the box so that the amount of cardboard used is minimized. 32. x 2 y2 dx dy 33. xy dx dy
1 0 0 0
26. The function 3 4 3 2x 3 1 3 x2
2 2
34. y dy dx 35. 7(x 2 + 2xy − 3y2 ) dy dx
f (x, y) = ax + by + cxy − x + y 1 x2 0
√
x

24
Problems 19–22 refer to Section 17.3 or Section 17.5.
25
A. P. Hurter, Jr., A. H. Rubenstein, et al., “Market Penetration by New
Innovations: The Technological Literature,” Technological Forecasting and 26
J. B. Covert and W. W. Reynolds, “Survival Value of Fever in Fish,”
Social Change, 11 (1978), 197–221. Nature, 267, no. 5606 (1977), 43–45.

801
 796 Chapter 17 Multivariable Calculus

Data Analysis to Model Cooling1

I
n Chapter 15 you worked with Newton’s law of cool- T
ing, which can be used to describe the temperature of a
cooling body with respect to time. Here you will deter-
mine that relationship in an empirical way via data analysis. 120
This will illustrate how mathematical models are designed
in many real-world situations. 100
Suppose that you want to create a mathematical model
of the cooling of hot tea after it is placed in a refrigera- 80
tor. To do this you place a pitcher containing hot tea and a
thermometer in the refrigerator and periodically read and
60
record the temperature of the tea. Table 17.5 gives the col-
lected data, where T is the Fahrenheit temperature t minutes 45
after the tea is placed in the refrigerator. Initially, that is, 40
at t = 0, the temperature is 124◦ F; when t = 391, then
T = 47◦ F. After being in the refrigerator overnight, the 20
temperature is 45◦ F. Figure 17.16 gives a graph of the data
points (t, T ) for t = 0 to t = 391. t
100 200 300 400
Table 17.5
Time Temperature Time Temperature FIGURE 17.16 Data points and exponential approximation.
t T t T
and then taking the natural logarithm of each side, we
0 min 124◦ F 128 min 64◦ F obtain a linear form:
5 118 144 62
10 114 178 59
4 − 45) = ln (Ceat )
ln (T
16 109 208 55 4 − 45) = ln C + ln eat
ln (T
20 106 244 51 4 − 45) = ln C + at
ln (T (2)
35 97 299 50
50 89 331 49 4l = ln (T
Letting T 4 − 45), Equation (2) becomes
65 82 391 47
85 74 Overnight 45
4l = at + ln C
T (3)

Because a and ln C are constants, Equation (3) is a linear

The pattern of these points strongly suggests that they equation in T4l and t. This means that for the original data,
nearly lie on the graph of a decreasing exponential func- if you plot the points (t, ln (T − 45)), they should nearly lie
tion, such as the one shown in Figure 17.16. In particular, on a straight line. These points are shown in Figure 17.17,
because the overnight temperature is 45◦ F, this exponential where Tl represents ln (T − 45). Thus for the line given
function should have T = 45 as a horizontal asymptote. by Equation (3) that predicts Tl = ln (T − 45), you can
Such a function has the form assume that it is the linear regression line of Tl on t. That
is, a and ln C are linear regression coefficients. By using
4 = Ceat + 45
T (1) the formulas for these coefficients and a calculator, you can
where T 4 gives the predicted temperature at time t, and C determine that
and a are constants with a < 0. (Note that since a < 0,  17   17   17 
  
then as t → ∞, you have Ceat → 0, so Ceat + 45 → 45.) 17 ti Tli − ti Tli
Now the problem is to find the values of C and a so i=1 i=1 i=1
that the curve given by Equation (1) best fits the data. By a=  17   17 2 ≈ −0.00921
writing Equation (1) as  
17 ti2 − ti
4 − 45 = Ceat
T i=1 i=1

1Adapted from Gloria Barrett, Dot Doyle, and Dan Teague, “Using Data Analysis in Precalculus to Model Cooling,”
The Mathematics Teacher, 81, no. 8 (November 1988), 680–84. By permission of the National Council of Teachers of
Mathematics.

802
 Chapter 17 Explore & Extend 797

Tl which is a model that predicts the temperature of the cool-

ing tea. The graph of this function is the curve shown in
Figure 17.16.

5 Problems
1. Plot the following data points on an x,y-coordinate
4 plane:

3 x 0 1 4 7 10

2
y 15 12 9 7 6

1 Suppose that these points nearly lie on the graph of a

decreasing exponential function with horizontal asymp-
100 200 300 400
t tote y = 5. Use the technique discussed in this Explore
and Extend to determine the function.
FIGURE 17.17 The points (t, Tl ), where Tl = ln (T − 45), 2. Suppose that observed data follow a relation given by
nearly lie on a straight line. y = C/x r , where x, y, C > 0. By taking the natural log-
arithm of each side, show that ln x and ln y are linearly
and related. Thus, the points (ln x, ln y) lie on a straight line.
 17   17   17   17 
    3. Use Newton’s law of cooling (see Section 15.6) and the
2
ti Tli − ti ti Tli data points (0, 124) and (128, 64) to determine the temper-
ln C =
i=1 i=1 i=1 i=1 ature T of the tea discussed in this Explore and Extend at
 17   17 2
  time t. Assume that the ambient temperature is 45◦ F.
17 ti2 − ti 4. Try obtaining the final regression equation in this
i=1 i=1 Explore and Extend using the regression capability of a
≈ 4.260074 graphing calculator. First use linear regression. How does
your result compare with the one in this Explore and
Since ln C ≈ 4.260074, then C ≈ e4.260074 ≈ 70.82.
Extend Then try skipping the linear-form transformation
Thus from Equation (1),
and perform an exponential regression. What difficulty do
4 = 70.82e−0.00921t + 45
T you encounter, if any? How could it be overcome?

803