MEA Solutions

8 CHAPTER 3 INTRODUCTORY TOPICS III: MISCELLANEOUS
Answers to Even-Numbered Problems

3.1

2. (a) 2 0 + 2 1 + 2 2 + 2 3 + 2 4 = 2(3 + 2 + 3)
(b) (x + 0)2 + (x + 2)2 + (x + 4)2 + (x + 6)2 = 4(x 2 + 6x + 14)
(c) a1i b2 + a2i b3 + a3i b4 + + ani bn+1 (d) f (x0 )1x0 + f (x1 )1x1 + f (x2 )1x2 + + f (xm )1xm
23+35+47 6 + 15 + 28 49
4. 100 = 100 = 100 144.12
13+25+37 3 + 10 + 21 34
6. (a) The total number of people moving from region i. (b) The total number of people moving to region j .
3.2
2. (a + b)6 = a 6 + 6a 5 b + 15a 4 b2 + 20a 3 b3 + 15a 2 b4 + 6ab5 + b6 . (The coefficients are those in the
seventh row of Pascals triangle in the text.)

5 543 54321 5! 5! m m(m 1) (m k + 1)
4. (a) = = = = . In general, =
3 123 12321 3! 2! 2! 3! k k!
m(m 1) (m k + 1) (m k)! m!
= = .
k!(m k)! (m k)! k!

8 876 8 8 87654 8 8
(b) = = 56. Also, = = = 56; + =
3 123 8 3 5 12345 3 3+1
8765 8+1 9 9876
56 + = 126 and = = = 126.
1234 3+1 4 1234

m m! m m m m! m!
(c) = = and + = +
k (m k)!k! mk k k+1 (m k)!k! (m k 1)!(k + 1)!

m!(k + 1 + m k) (m + 1)! m+1
= = =
(m k)!(k + 1)! (m k)!(k + 1)! k+1
3.3
2. (a) The total number of units of good i.
(b) The total number of units of all goods owned by person j .
(c) The total number of units of goods owned by the group as a whole.
4. Because arj a is independent of the summation index s, it is a common factor when we sum over s, so
Pm Pm
s=1 (arj a)(asj a) = (arj a) s=1 (asj a) for each r. Next, summing over r gives
m X
X m X
m X
m
(arj a)(asj a) = (arj a) (asj a) ()
r=1 s=1 r=1 s=1
Using the properties of sums and the definition of aj , we have

m
X m
X m
X
(arj a) = arj a = maj ma = m(aj a)
r=1 r=1 r=1
Pm
Similarly, replacing r with s as the index of summation, one also has s=1 (asj a) = m(aj a).
Substituting these values into () then confirms ().
Knut Sydster and Peter Hammond 2006

CHAPTER 3 INTRODUCTORY TOPICS III: MISCELLANEOUS 9
3.4
2. x = 2. (x = 1, 0, and 1 make the equation meaningless. Multiplying each term by the common
denominator x(x 1)(x + 1) yields 2x(x 2 3x + 2) = 0, or 2x(x 1)(x 2) = 0. Hence, x = 2 is
the only solution.)
4. (a) x 0 is necessary, but not sufficient. (b) x 50 is sufficient, but not necessary.
(c) x 4 is necessary and sufficient.

6. (a) No solutions. (Squaring each side yields x4 = x+518 x + 5+81, which reduces to x + 5 = 5,
with solution x = 20. But x = 20 does not satisfy the given equation.)
(b) x = 20
(i) (ii)
8. (a) x+ x + 4 = 2 x + 4 = 2x x+4 = 44x+x 2 x 2 5x = 0 x5 = 0 x = 5.
Here implication (i) is incorrect (x 2 5x = 0 x 5 = 0 or x = 0.) Implication (ii) is correct, but
it breaks the chain of implications. (b) x = 0. (After correcting implication (i), we see that the given

equation implies x = 5 or x = 0. But only x = 0 is a solution; x = 5 solves x x + 4 = 2.)
3.5
2. (a) Logically the two statements are equivalent. (b) Appending the second statement is still an expressive
poetic reinforcement.
3.6
2. F B C is the set of all

female biology students in the university choir; M F the female mathematics
students; (M B) \ C \ T the students who study both mathematics and biology but neither play tennis
nor belong to the university choir.
4. (a) B M (b) F B C 6= (c) T B = (d) F \ (T C) B
6. (b) and (c) are true, the others are wrong. (Counter example for (a), (d), and (f): A = {1, 2}, B = {1},
C = {1, 3}. As for (e), note in particular that A B = A C = A whenever B and C are subsets of A,
even if B 6= C.)
8. (a) Consider Fig. M3.6.8, and let nk denote the number of people in the set marked Sk , for k = 1, 2, . . . , 8.
Obviously n1 + n2 + + n8 = 1000. The responses imply that: n1 + n3 + n4 + n7 = 420; n1 + n2 +
n5 + n7 = 316; n2 + n3 + n6 + n7 = 160; n1 + n7 = 116; n3 + n7 = 100; n2 + n7 = 30; n7 = 16,
and n8 = 334. From these equations we easily find n1 = 100, n2 = 14, n3 = 84, n4 = 220, n5 = 186,
n6 = 46, n7 = 16. (i) n3 + n4 = 304 had read A but not B; (ii) n6 = 46; (iii) n8 = 334.
(b) We find n(A \ B) = n3 + n4 = 304, n(C \ (A B)) = n6 = 46, and n( \ (A B C)) = n8 = 334.
The last equality is a special case of n( \ D) = n() n(D). (The number of persons who are in ,
but not in D, is the number of persons in all of minus the number of those who are in D.)

10 CHAPTER 3 INTRODUCTORY TOPICS III: MISCELLANEOUS
A B
S5
S1
S4
S7
S2
S3
S6
S8
C
Figure M3.6.8
3.7
2. We prove only (3.2.6); the proof of (3.2.5) is very similar, but slightly easier. For n = 1 the LHS and the
RHS of (3.2.6) are both equal to 1. As the induction hypothesis, suppose (3.2.6) is true for n = k, so that
k
X
i 3 = 13 + 23 + + k 3 = [ 21 k(k + 1)]2
i=1
Then
k+1
X k
X
3
i = i 3 + (k + 1)3 = [ 21 k(k + 1)]2 + (k + 1)3 = (k + 1)2 ( 41 k 2 + k + 1)
i=1 i=1
But this last expression is equal to 41 (k + 1)2 (k 2 + 4k + 4) = [ 21 (k + 1)(k + 2)]2 , which proves that
(3.2.6) is true for n = k + 1. By induction, we have proved (3.2.6).
4. The claim is true for n = 1. As the induction hypothesis, suppose k 3 + (k + 1)3 + (k + 2)3 is divisible
by 9. Note that (k + 1)3 + (k + 2)3 + (k + 3)3 = (k + 1)3 + (k + 2)3 + k 3 + 9k 2 + 27k + 27 =
k 3 + (k + 1)3 + (k + 2)3 + 9(k 2 + 3k + 3). This is divisible by 9 because the induction hypothesis implies
that the sum of the first three terms is divisible by 9, whereas the last term is also obviously divisible by 9.
Review Problems for Chapter 3

2. (a) 12 4+22 5+32 6+42 7 = 4+20+54+112 = 190 (b) 1 61 = 5
6 (c) 12 +21 +30 +41 +52 +63 =
1 + 1/2 + 1 + 4 + 25 + 216 = 495/2
38
X 16
X 81
X
Pn 1 x 2j 1
4. (a) i(i + 2) (b) i
(c) (d) (1)k1
i=1 x 2j + 1 k
i=4 j =0 k=1
6. (a) true, false (b) false (because x 2 = 16 also has the solution x = 4), true (c) true,
false (x can be 3) (d) and both true
8. A B = ; A B = {1, 2, 4, 6, 11, 12}; \ B = {1, 3, 4, 5, 6, 7, 8, 9, 10}; CA = \ A =
{2, 3, 5, 7, 8, 9, 10, 11, 12}
10. For n = 1, the inequality is correct: 1 + x 1 + x. Suppose (1 + x)k 1 + kx for an arbitrary natural
number k. We have to prove that then (1 + x)k+1 1 + (k + 1)x. But because x 1 and k > 0, we
have (1 + x)k+1 = (1 + x)k (1 + x) (1 + kx)(1 + x) = 1 + (k + 1)x + kx 2 1 + (k + 1)x. Thus, by
induction, Bernoullis inequality is true for all natural numbers n.

CHAPTER 4 FUNCTIONS OF ONE VARIABLE 11
Chapter 4 Functions of One Variable

Functions of one real variable are the main subject of Chapters 4 to 9. Chapter 4 begins with a gentle
introduction. Section 4.1 presents a few simple examples, mostly drawn from economics, before Section 4.2
presents a definition, and introduces functional notation systematically. The domain and range of a function
are defined, as are increasing and decreasing functions.
In our experience, training students to master functional notation is very important, yet sometimes neg-
lected in introductory texts. The students should be encouraged to do many of the numerous problems in this
section, and throughout the chapter.
Next, Section 4.3 begins with a review of how functions can be represented by graphs in a two-dimensional
coordinate system. It displays the graphs of a number of important simple functions which students need to
learn as soon as possible.
The special case of linear functions is the subject of Section 4.4. The intercept is defined, and the formula
for the slope provided. Next come the pointslope formula for the straight line with a given slope through
a specified point, and the pointpoint formula for the line through two specified points with different x-
coordinates. The section concludes with the general equation for a straight line that includes the case when the
line is vertical, along with the graphical method of solving linear equations, and the graphical representation
of linear inequalities.
Section 4.5 considers a (dubious) linear model of population growth and the linear consumption function
as well as a simple model of market equilibrium of supply and demand.
Section 4.6 treats quadratic functions, and the possible shapes of their graphs. The solution of quadratic
equations is reviewed (see also Section 2.3) and the method of completing the square is used to identify the
maximum or the minimum point of a quadratic function. The CauchySchwarz inequality is discussed as a
harder problem.
In order to have students see mathematics applied as soon as possible to economic optimization problems,
we consider a monopolist with a quadratic cost function who faces a linear (inverse) demand function. The
perfectly competitive firm is treated as a special case. It is shown that, in order that the profit maximum for
a monopolist should coincide with that for a competitive firm, its output should be subsidized (rather than
taxed). (Alternatively, it should have its price regulated.)
Next, Section 4.7 covers more general polynomials, starting with cubic functions. An example shows
how they can represent a richer family of cost curves. The solutions of polynomial equations are discussed.
Techniques for finding integer roots are emphasized. Then division of polynomials is considered, especially
the remainder theorem. The correspondence between the roots and linear factors of a polynomial is pointed
out. Then a technique for dividing polynomials is presented. Finally, rational functionsboth proper and
improperare defined.
Power functions are the subject of Section 4.8. The need for fractional powers is illustrated, and rational
powers are then defined. The graphs of power functions are illustrated.
Section 4.9 treats exponential functions. The natural exponential function is introduced already at this
stage, although the reason for its importance will only be clear once students have mastered some calculus.
Chapter 4 concludes with a discussion of logarithmic functions, emphasizing the natural logarithmic function
which is now the most commonly used.

12 CHAPTER 4 FUNCTIONS OF ONE VARIABLE

4.2
2. F (0) = F (3) = 10, F (a + h) F (a) = 10 10 = 0

4. (a) f (1/10) = 10/101, f (0) = 0, f (1/ 2) = 2/3, f ( ) = /(1 + ), f (2) = 2/5
(b) f (x) = x/(1 + (x)2 ) = x/(1 + x 2 ) = f (x) and f (1/x) = (1/x)/[1 + (1/x)2 ] =
(1/x) x 2 /[1 + (1/x)2 ] x 2 = x/(1 + x 2 ) = f (x)
6. (a) C(0) = 1000, C(100) = 41 000, and C(101) C(100) = 501
(b) C(x + 1) C(x) = 2x + 301 = incremental cost of increasing production from x to x + 1.
8. (a) f (tx) = 100(tx)2 = 100t 2 x 2 = t 2 f (x) (b) P (tx) = (tx)1/2 = t 1/2 x 1/2 = t 1/2 P (x)
10. (a) No. f (2 + 1) = f (3) =
18, whereas f (2) + f (1) = 10. (b)
Yes. f (2 + 1) = f (2) + f (1) = 9
(c) No. f (2 + 1) = f (3) = 3 1.73, whereas f (2) + f (1) = 2 + 1 2.41.
12. See Figs. M4.2.12a and M4.2.12b.
1 x1 11 1
x x2 1x x
x 1 x
Figure M4.2.12a The area Figure M4.2.12b The area

2 2
is (x + 1) = x + 2x + 1 is x 2 + 1
14. (a) Defined for x 6 = 2, i.e. Df = (, 2) (2, ) (b) f (8) = 5

3x + 6
(c) f (x) = = 3 3x + 6 = 3(x 2) 6 = 6, which is impossible.
x2
4.3
2. (a) f (5) = 0, f (3) = 3, f (2) = 0, f (0) = 2, f (3) = 4, f (4) = 0
(b) Df = [5, 4], Rf = [3, 4]
x 2 1 0 1 2 3 4
4.
h(x) = x 2 2x 3 5 0 3 4 3 0 5
See Fig. M4.3.4.
x 2 1 0 1 2 3
6.
G(x) = 1 2x 3 1 0 1/2 3/4 7/8
See Fig. M4.3.6.

y
6
5 y
4 6
5
3 4
2 3
1 2
1
2 1 2 3 4 x
-2 -1 1 2 3 x
-1
-2
3 -3
4 -4
Figure M4.3.4 Figure M4.3.6
4.4
2. See Figs. M4.4.2a, M4.4.2b, M4.4.2c
y y y
1 4
4
1 2 3 4 5 6 7 8 10 x 3
3 1
2
2 2
1
3
1
4 1 2 3 4 5 x
1 2 3 4 5 1
x
Figure M4.4.2a Figure M4.4.2b Figure M4.4.2c
4. L1 is y = x + 2, with slope 1; L2 is y = 53 x + 3, with slope 3/5; L3 is y = 1, with slope 0;

L4 is y = 3x 14, with slope 3; L5 is y = 19 x + 2, with slope 1/9.
6. If P is the price of Q copies, then applying the pointpoint formula gives P 1400 = 30001400
500100 (Q100)
or P = 1000 + 4Q. The price of printing 300 copies is therefore P = 1000 + 4 300 = 2200.
8. (a) See Figs. M4.4.8a, M4.4.8b, and M4.4.8c.
y y y
1 1
1 x 1 1 x 1 x

10. See Fig. M4.4.10. Each arrow points to the side of the line where the relevant inequality is satisfied. The
shaded triangle is the required set.
y y
3
xy =1 2 f (x) = 1 x 2 x + 3
3
2 2
1
2
1 -4 -3 -2 -1 1 2 3 x
-1
x -2
2 3 4 -3
-4
3x + 4y = 12
3x + y = 3
4.5
2. (a) 75 3P e = 20 + 2P e , and hence P e = 11. (b) P e = 90
4. C = 0.8y +100. (The equation is C = ay +b. We are told that 900 = 1000a +b and a = 80/100 = 0.8.
So b = 100.)
6. (a) April 1960 corresponds to t = 9/4, when N (9/4) = 17 400 (9/4) + 151 000 = 111 850.
(b) 17 400 t + 151 000 = 0 implies t = 8.68, which corresponds roughly to September 1966.
4.6
x 4 3 2 1 0 1 2
2. (a)
f (x) = 21 x 2 x + 3
2 2.5 0 1.5 2 1.5 0 2.5
(b) See Fig. M4.6.2. (c) f (x) = 21 (x + 1)2 + 2. Maximum point (1, 2).
(d) x = 3 and x = 1. (e) f (x) > 0 in (3, 1), f (x) < 0 for x < 3 and for x > 1.
4. (a) x(x + 4). Zeros 0 and 4. (b) No factoring is possible.
No zeros.
(c) 3(x x1 )(x x2 ), where the zeros are x1 = 5 + 15 and x2 = 5
15.
(d) 9(x x1 )(x x2 ), where the zeros are x1 = 1/3 + 5 and x2 = 1/3 5.
(e) (x + 300)(x 100). Zeros 300 and 100. (f) (x + 200)(x 100). Zeros 200 and 100.
6. U (x) has maximum for x = 4(r 1)/(1 + r 2 ). (Use (5).)
8. (a) If PE = 1 13 Q denotes the selling price in England, and PG = 2 + 61 Q denotes the buying
price in Ghana, (Q) = (PE PG )Q = 21 Q2 + (1 2 )Q. (b) Using (5), we see that
Q = 1 2 maximizes profit if 1 2 > 0. If 1 2 0, then Q = 0.
(c) (Q) = 21 Q2 + (1 2 t)Q and Q = 1 2 t if 1 2 t > 0.
(d) T = tQ = t (1 2 t). Here t = 21 (1 2 ) maximizes export tax revenue.
4.7
2. (a) 1 and 2 (b) 1, 5, and 5 (c) 1
4. (a) y = 21 (x + 1)(x 3) (b) y = 2(x + 3)(x 1)(x 2) (c) y = 21 (x + 3)(x 2)2
6. c4 + 3c2 + 5 = (c2 + 3/2)2 + 11/4 is not 0 for any choice of c, so the division has to leave a remainder.

4.8
2. (a): C (b): D (c): E (d): B (e): A (f) F: y = 2 (1/2)x
4. (a) 23 = 8, so x = 3/2 (b) 3x + 1 = 4, so x = 5/3 (c) x 2 2x + 2 = 2, so x 2 2x = 0,
implying that x = 0 or x = 2.
2/3
V = (4/3) r 3 implies r 3 = 3V /4 and so r = (3V /4 )1/3 . Hence, S = 4 r 2 = 4 3V /4
6. =
3
36 V 2/3 .
4.9

2. P (t) = 1.22 1.034t . The doubling time t is given by the equation (1.034)t = 2, and we find t 20.7
(years).
4. The graphs are drawn in Fig. M4.9.4.
x 3 2 1 0 1 2 3
2x 1/8 1/4 1/2 1 2 4 8
2x 8 4 2 1 1/2 1/4 1/8
y
8
y = 2x 6 y = 2x
4
y f (x) = 1 + 4x
x 2 +4
2 2
1
3 2 1 1 2 3 x x
4 3 2 1 1 2 3 4 5
Figure M4.9.4 Figure M4.R.2
6. We find (1.035)t = 3.91 105 /5.1 0.7667 105 , and (taking ln of each side) we find t 327. So the
year is 1969 + 327 = 2296. This is when every Zimbabwean would have only 1 m2 of land on average.
8. (b) and (d) do not define exponential functions. (In (f): y = (1/2)x .)
10. Use the two points on each graph to determine A and b from the equation y = Abx .
You then get: (a) y = 2 2x (b) y = 2 3x (c) y = 4(1/2)x .
4.10
2. (a) ln 3x = x ln 3 = ln 8, so x = ln 8/ ln 3. (b) x = e3 (c)x 2 4x + 5 = 1 so (x 2)2 = 0. Hence,
x = 2. (d) x(x 2) = 1 or x 2 2x 1 = 0, so x = 1 2. (e) x = 0 or ln(x + 3) = 0, so x = 0

or x = 2. (f) x 5 = 1 so x = 36.

1 B 1 5.6
4. (a) t = ln (b) t = ln 22. It will take approximately 22 years before the two
r s A 0.07 1.2
countries have the same GNP.
6. (a) Wrong. (Let A = e.) (b) and (c) are right.

4
2. (a) F (0) = 1, F (2) = 0, F (2) = 2, and F (3) = 25/13 (b) F (x) = 1 + tends to 1 as x
x + 4/x
becomes large positive or negative. (c) See Fig. M4.R.2.
4. (a) |x| 1, i.e. x 1 or x 1 (b) x > 4 (c) 3 x 5
6. (a) Slope 4 (b) Slope 3/4 (c) Slope b/a
8. f (2) = 3 and f (1) = 3 give 2a + b = 3 and a + b = 3, so a = 2, b = 1. Hence f (x) = 2x 1
and f (3) = 7. (Or use the pointpoint formula.)
10. y = 2x 2 + x 6. (3 = a + b + c, 6 = c, and 15 = 9a + 3b + c)
12. The new profit is t = 100Q 25 Q2 tQ, which is maximized at Qt = 51 (100 t).
14. (a) p(x) = x(x 3)(x + 4) (b) q(x) = 2(x 2)(x + 4)(x 1/2)
16. (a) k = 4 (b) k = 3/2 and k = 1 (c) k = 26 (d) k = 1 and k = 4
18. (1 + p/100)15 = 2 gives p = 100(21/15 1) 4.7 as the percentage rate.
q
20. (a) t = (ln x b)/a (b) t = (ln 2)/a (c) t = ln 8/ 2

CHAPTER 5 PROPERTIES OF FUNCTIONS 17
Chapter 5 Properties of Functions

This chapter begins in Section 5.1 by examining how changes in a function relate to shifts in its graph. An
economic application is to shifting demand and supply curves. Section 5.2 considers how to construct new
functions from old ones, and includes a discussion of symmetry.
Next, Section 5.3 introduces inverse functions. By definition, if a function is one-to-one, there is an
inverse from its range or image space back to its domain. The obvious geometric properties of the graphs of
a function and its inverse are explained.
Any equation in two variables can be represented by a curve (or a set of points) in the xy-plane. Some
examples are given in Section 5.4. Also included is the vertical line test for a graph to represent a function,
and the effect on a graph of changing units. The section ends with some examples of compound functions.
The distance between points in a plane is defined in Section 5.5, and used to derive the equations whose
graph is a circle with a given centre and radius. Ellipses and other conics are briefly discussed.
Section 5.6 concludes the chapter with a discussion of the general concept of a function and the associated
terminologydomain, target, range, image. Definitions of one-to-one functions and of inverse functions are
also given.

5.1
2. (a) The graph of y = f (x) is moved 2 units to the right. See Fig. M5.1.2a. (b) The graph of y = f (x)
is moved downwards by 2 units. See Fig. M5.1.2b. (c) The graph of y = f (x) is reflected about the
y-axis. See Fig. M5.1.2c.
y y y
1 1 1
1 x 1 x 1 x
4. Move y = |x| two units to the left. Then reflect the graph about the x-axis. Finally, move the graph up 2
units. See Fig. M5.1.4. y
1 x
Figure M5.1.4
Bd 2 Ad + c
6. y =
2Bd

18 CHAPTER 5 PROPERTIES OF FUNCTIONS
5.2
2
2. See Figs. M5.2.2a to M5.2.2c. (Note that the graph of ex + x is not symmetric about (0, 1).)
y y y
x x
4. f (f (x)) = f (3x + 7) = 3(3x + 7) + 7 = 9x + 28. Then 9x + 28 = 100 for x = 8.
5.3
2. p = (157.8/D)10/3
4. (a) The domain of f 1 is {4, 2, 0, 2, 4, 6, 8}. f 1 (2) = 1 (b) f (x) = 2x + 4, f 1 (x) = 21 x 2
6. (a) f (x) = x/2 and g(x) = 2x are inverse functions. (b) f (x) = 3x 2 and g(x) = 13 (x + 2) are
inverse functions. (c) C = 95 (F 32) and F = 59 C + 32 are inverse functions.
8. (a) See Fig. M5.3.8a. (b) See Fig. M5.3.8b. Triangles OBA and OBC are congruent.
y y
y=x y=x y
C = (b, a)
D
(3, 5) 3
B
2
(1, 3) (5, 3)
1
A = (a, b)
(3, 1)
1 2 3 x
x E x
Figure M5.3.8a Figure M5.3.8b Figure M5.4.2
10. (a) x = ln y 4, y > 0 (b) x = ey+4 , y (, ) (c) x = 3 + ln(ey 2), y > ln 2
5.4

2. (a) See Fig. M5.4.2. (b) If a > 1, then g(a) = 1 a 1 and g(a) < 1 (see the figure), so that

g(g(a)) = g(1 a 1) = (1 a 1)2 2(1 a 1)+2 2
p = a. If a < 1, then g(a) = a 2a +2,
2 2
and g(a) > 1, so that g(g(a)) = 1 a 2a + 2 1 = 1 (a 1) = 1 + (a 1) = a. Moreover,
g(g(1)) = 1. It follows that g(g(a)) = a for all a, so g is its own inverse. (Alternatively, show that one
can solve y = g(x) for x in terms of y, in which case x = g(y).)

CHAPTER 5 PROPERTIES OF FUNCTIONS 19
4. F (100 000) = 4070. The graph is the thick line sketched in Fig. M5.4.4.
y
Y y
4070 2 A = (3, 2)
(2, 4)
2 x
2
P
RN
B = (5, 4)
2 5 x
7500 100 000
Figure M5.4.4 Figure M5.5.2 Figure M5.R.10
5.5
2. (5 2)2 + (y 4)2 = 13, or y 2 8y + 12
= 0, with solutions y = 2 and y = 6. Geometric explanation:
The circle with centre at (2, 4) and radius 13 intersects the line x = 5 at two points. See Fig. M5.5.2.
4. (a) (x 2)2 + (y 3)2 = 16 (b) (x 2)2 + (y 5)2 = 13
p p
6. The condition is that (x + 2)2 + y 2 = 2 (x 4)2 + y 2 , which reduces to (x 6)2 + y 2 = 42 .
5.6
2. The function in (b) is one-to-one and has an inverse: the rule mapping each youngest child alive today to
his/her mother. (Though the youngest child of a mother with several children will have been different at
different dates.) The function in (d) is one-to-one and has an inverse: the rule mapping the surface area to
the volume. The function in (e) is one-to-one and has an inverse: the rule that maps (u, v) to (u 3, v).
The other functions are many-to-one, in general, and so have no inverses.

2. (f + g)(x) = x 2 2, (f g)(x) = 2x 3 x 2 2, (fg)(x) = x 2 (1 x)(x 3 2), (f/g)(x) =
(x 3 2)/x 2 (1 x), f (g(1)) = f (0) = 2, and g(f (1)) = g(1) = 2.
4. p = (64 10D)/3

6. (a) x = 50 21 y (b) x = 5 y/2 (c) x = 13 [2 + ln(y/5)]
p
8. (a) 13 (b) 17 (c) (2 3a)2 = |23a| (Note that 23a is the correct answer only if 23a 0,
i.e. a 2/3. Ask students with the wrong answer to put, say, a = 3.)
10. (x 3)2 + (y 2)2 = (x 5)2 + (y + 4)2 , which reduces to x 3y = 7. See Fig. M5.R.10.

20 CHAPTER 6 DIFFERENTIATION
Chapter 6 Differentiation
Our treatment of the single variable calculus begins in Chapter 6. Section 6.1 is devoted to the slopes of curves
and their tangents, as well as the tangent definition of derivative. Section 6.2 discusses how the tangent is
the limit of a sequence of lines joining nearby points on the graph of a function. It proceeds to present the
formula for the Newton (or differential) quotient, whose limit is the derivative. The derivative is then used in
the general equation for the tangent to the graph of a function at any point where it is differentiable. A general
recipe for finding derivatives is provided, and Leibnizs differential notation is introduced.
Section 6.3 discusses increasing and decreasing functions and explains how the sign of the derivative
can help us to find the intervals of increase/decrease of a differentiable function. (Note 8.4.2 shows how the
mean-value theorem can be used to prove the main results of this section.)
Section 6.4 is devoted to the economic significance of the derivative of a function. The instantaneous
and proportional rates of change are distinguished. Marginal cost, revenue, profit, propensity to consume,
product, are all defined in terms of derivatives. Marginal cost is distinguished from incremental cost.
It is crucially important for economics students to understand the definition of the derivative. In fact, this
is more important than knowing how to differentiate very complicated functions. Example 6.4.3 and problems
like 6.4.5 should be emphasized. The procedure of smoothing a function to fit a set of discrete data points is
illustrated at the end of this section.
Limits were used to define derivatives informally in Section 6.2. The more careful discussion that is
really needed is the topic of Section 6.5, though even this remains rather informal. The discussion proceeds
mostly through examples. Some general rules for limits of sums, differences, products, ratios, and powers are
presented.
After these preliminaries, Section 6.6 is devoted to some simple rules for differentiation, including the
rule for differentiating a power function. Section 6.7 presents the rules for differentiating sums, differences,
products, and quotients. The product rule in particular is illustrated economically. The quotient rule is used
to show that average cost is increasing iff it is less than marginal cost.
Section 6.8 states the chain rule, first using Leibnizs notation in a suggestive way, then more carefully.
An incomplete proof of the chain rule is provided in small print; we judge that a complete proof belongs to
more rigorous courses in mathematical analysis.
The next section 6.9 introduces second-order derivatives. It also defines convex and concave functions
using the sign of the second derivative. In fact, much modern economic analysis has been built on specific
assumptions to the effect that certain functions are either concave or convex. The examples illustrated in
Figures 4 and 5, or similar examples, should be carefully discussed. In Example 5 it is shown that an increas-
ing concave transformation of a concave function is concave. Higher-order derivatives, found by repeated
differentiation, are discussed next.
Sections 4.9 and 4.10 introduced exponential and logarithmic functions, which are often used in econom-
ics, of course, as well as in statistics. Section 6.10 shows that if the special number e = 2.718281828 . . . is
used as the base for the exponential function, then dex /dx = ex for all x, so the derivative of the function is the
function itself. It also shows how to differentiate eg(x) using the chain rule. Finally, the rule for differentiating
the general exponential function a x is derived.
Section 6.11 shows how to differentiate the natural logarithmic function, d ln x/dx = 1/x, and also
the derivative d ln[h(x)]/dx for any positive valued differentiable function h(x). The section includes a
discussion of the useful technique of logarithmic differentiationi.e., taking the logarithm of an expression
before differentiating. As shown later in Section 7.7, this is often a natural way for economists to calculate
elasticities.

CHAPTER 6 DIFFERENTIATION 21
A subsection characterizes the number e as limh0 (1 + h)1/ h . Finally, when a is any real number and
x > 0, the power x a can be defined as exp(a ln x), and using the chain rule, the power rule then follows for
all real exponents a.
In fact, Chapter 6 almost completes the inventory of functions of a single variable used in this book, and in
most mathematical work in economics. The major omission is the family of trigonometric functions discussed
in FMEA. In economics, they are used almost exclusively to solve difference and differential equations, which
are topics in FMEA.

6.1
2. g(5) = 1, g (5) = 1
6.2
2. (a) f (x) = 6x + 2 (b) f (0) = 2, f (2) = 10, f (3) = 20. The tangent equation is y = 2x 1.
f (x + h) f (x) 1/(x + h) 1/x x (x + h) h 1 1
4. = = = = 2
h h hx(x + h) hx(x + h) x(x + h) h0 x
6. (a) f (x + h) f (x) = a(x + h)2 + b(x + h) + c (ax 2 + bx + c) = 2ahx + bh + ah2 , so
[f (x + h) f (x)]/ h = 2ax + b + ah 2ax + b as h 0. Thus f (x) = 2ax + b.
(b) f (x) = 0 for x = b/2a. The tangent is parallel to the x-axis at the minimum/maximum point.

8. (a) Use the difference of squares formula. (b) [f (x + h) f (x)]/ h = x + h x)/ h. The identity
in (a) yields the result. (c) Letting h 0, the formula follows. (The students may need reminding that

x = x 1/2 and 1/ x = x 1/2 .)
f (x + h) f (x) (x + h)1/3 x 1/3 1 1
10. = = 2/3 1/3 1/3 2/3
as h 0,
h h (x + h) + (x + h) x + x 3x 2/3
and 1/(3x 2/3 ) = 31 x 2/3 .
6.3

2. f (x) = 3x 2 + 8x 1 = 3(x x0 )(x x1 ), where x0 = 31 (4 13) 0.13 and x1 = 31 (4 + 13)
2.54. Then f (x) is increasing in [x0 , x1 ], decreasing in (, x0 ] and in [x1 , ).
6.4
2. I is the fixed cost, whereas k is the marginal cost, and also the (constant) incremental cost of producing
each additional unit.
4. T (y) = t, so the marginal tax rate is constant.
6. (a) C (x) = 3x 2 180x + 7500 (b) x = 30 (C (x) has a minimum at x = 180/6 = 30, using (4.6.5).)
6.5
2. (a) 0.6931 (b) 1.0986 (c) 0.4055 (Actually, using the result in Example 7.12.2, the precise values of
these limits are ln 2, ln 3, and ln(3/2), respectively.)

22 CHAPTER 6 DIFFERENTIATION
4. (a) 5 (b) 1/5 (c) 1 (d) 2 (e) 3x 2 (f) h2

6. (a) 4 (= f (1)) (b) 5 (= f (2) f (1)) (c) 6 (= f (2)) (d) 2a + 2 (= f (a)) (e) 2a + 2 (= f (a))
(f) [f (a+h)f (ah)]/ h = [(a+h)2 +2(a+h)(ah)2 2(ah)]/ h = 4a+4 4a+4 as h 0, so
f (a) = 4a+4. (Alternatively, [f (a+h)f (ah)]/ h = [f (a+h)f (a)]/ h+[f (a)f (ah)]/ h
f (a) + f (a) = 4a + 4 as h 0.)
6.6
2. (a) 2g (x) (b) 16 g (x) (c) 13 g (x)
4. (a) 8 r (b) A(b + 1)y b (c) (5/2)A7/2
x a+1
6. (a) F (x) = 13 x 3 + C (b) F (x) = x 2 + 3x + C (c) F (x) = + C, where in all three cases C is
a+1
an arbitrary constant.
6.7
2. (a) 56 x 14x 6 21 x 1/2 (b) 4x(3x 4 x 2 1) (c) 10x 9 + 5x 4 + 4x 3 x 2
3 x 4 + 5x 2 + 18x + 2 2(1 + 2x) 4x 2x 2 + 2
4. (a) (b) (c) (d) (e)
2 x( x + 1)2 (x 2 + 2)2 (x + 3)2 x3 (x 2 + 1)2 (x 2 x + 1)2
3 2
2(x x + x + 1)
(f)
3x 3 (x + 1)2

6. (a) [2, ) (b) 3, 0 and in 3, (c) 2, 2
x(x x1 )(x x2 )
(d) (, x1 ] and in [0, x2 ], where x1,2 = 21 21 5. (f (x) = )
(x + 1)2
3(x 1)(x + 1)
8. f (x) = . f (x) is increasing in (, 1], in 1, 2 + 3 , and in 2 + 3, .
(x 2 + 4x 1)2
10. Differentiating f (x) f (x) = x gives f (x) f (x) + f (x) f (x) = 1, so 2f (x) f (x) = 1. Hence,
1 1
f (x) = = .
2f (x) 2 x
6.8
2. (a) dY /dt = (dY /dV )(dV /dt) = (3)5(V + 1)4 t 2 = 15t 2 (t 3 /3 + 1)4
(b) dK/dt = (dK/dL)(dL/dt) = AaLa1 b = Aab(bt + c)a1
dx a
4. (dY /dt)t=t0 = (dY /dK)t=t0 (dK/dt)t=t0 = Y (K(t0 ))K (t0 ) 6. =
dp 2 ap c

8. b(t) is the total fuel consumption after t hours. b (t) = B s(t) s (t). So the rate of fuel consumption
per hour is equal to the rate per kilometre multiplied by the speed in kph.
10. (a) y = 5(x 4 )4 4x 3 = 20x 19(b) y = 3(1 x)2 (1) = 3 + 6x 3x 2
3 2
12. (a) 1 + f (x) (b) 2f (x)f (x) 1 (c) 4 f (x) f (x) (d) 2xf (x) + x 2 f (x) + 3 f (x) f (x)
f (x) 2xf (x) x 2 f (x) 2xf (x)f (x) 3[f (x)]2
(e) f (x) + xf (x) (f) (g) (h)
2 f (x) [f (x)]2 x4

CHAPTER 6 DIFFERENTIATION 23
6.9
2. d 2 y/dx 2 = (1 + x 2 )1/2 x 2 (1 + x 2 )3/2 = (1 + x 2 )3/2 4. g (2) = 2. (g (t) = 2(t 1)3 )
6. d 2 L/dt 2 = 3(2t 1)5/2 8. Because g(u) is not concave. 10. d 3 L/dt 3 > 0
6.10
qt 4 + 2qt 3 pt p
2. (a) x = et [a + b + t (b + 2c + ct)] (b) x =
t 2 et
2 t 2 2
2(at + bt )(a + 2bt)e (at + bt ) e t t (a + bt)(bt 2 + (4b a)t + 2a)
(c) x = =
(et )2 et

4. (a) (, ) (b) [0, 1/2] (c) (, 2/2] and in [0, 2/2]
x x +x et et 3 3
6. (a) ee ex = ee (b) 21 (et/2 et/2 ) (c) (d) z2 ez (ez 1)2/3
(et + et )2
6.11
x(2 ln x 1) 10(ln x)9 2 ln x
2. (a) x 2 ln x(3 ln x + 2) (b) (c) (d) + 6 ln x + 18x + 6
(ln x)2 x x
4. (a) x > 1 (b) 1/3 < x < 1 (c) x 6 = 0
6. (a) (2, 0] (y is defined only in (2, 2).) (b) [e1/3 , ) (y = x 2 (3 ln x + 1), x > 0)
(c) [e, e3 ] (y = (1 ln x)(ln x 3)/2x 2 , x > 0)
f (x) 2 f (x) f (x) 1 2x 4x 3
8. (a) = (b) = 2 ln x + 2 (c) = + 2 + 4
f (x) 3(x 2 1) f (x) f (x) 2x 4 x + 1 x + 6
10. ln y = v ln u, so y /y = v ln u + vu /u.

2. [f (x + h) f (x)]/ h = 6x 2 + 2x 6xh 2h2 + h 6x 2 + 2x as h 0, so f (x) = 6x 2 + 2x.
4. Because C (1000) C(1001) C(1000), if C (1000) = 25, the additional cost of producing 1 more
than 1000 units is approximately 25. It is profitable to increase production if each unit is sold for 30.
6. Because A (100) A(101) A(100), the additional cost of increasing the area from 100 to 101 m2 is
approximately 250 dollars.

8. (a) 2at (b) a 2 2t (c) 2x 1/2
10. (a) dZ/dt = (dZ/du)(du/dt) = 3(u2 1)2 2u3t 2 = 18t 5 (t 6 1)2

(b) dK/dt = (dK/dL)(dL/dt) = (1/2 L)(1/t 2 ) = 1/2t 2 1 + 1/t
12. dR/dt = (dR/dS)(dS/dK)(dK/dt) = S 1 K 1 Apt p1 = Apt p1 S 1 K 1
2
14. (a) y = 7ex (b) y = 6xe3x (c) y = xex (2 x) (d) y = ex [ln(x 2 + 2) + 2x/(x 2 + 2)]
3
(e) y = 15x 2 e5x (f) y = x 3 ex (x 4) (g) y = 10(ex + 2x)(ex + x 2 )9

(h) y = 1/2 x( x + 1)
16. (a) d/dQ = P (Q) + QP (Q) c (b) d/dL = P F (L) w

24 CHAPTER 7 DERIVATIVES IN USE
Chapter 7 Derivatives in Use

Implicit differentiation, often used in economic analysis, is the subject of the first three sections. Section
7.1 concentrates on theory, with purely mathematical examples and problems, while in Section 7.2 several
important economic examples are discussed. In Section 7.3 the formula for the derivative of an inverse function
is derived.
Linear approximations and differentials are treated in Section 7.4. It is important for students to learn
how differentials are manipulated, since economists often make use of them.
Then quadratic and general polynomial approximations are discussed in Section 7.5, and the nth order
Taylor polynomial is introduced.
Section 7.6 extends the discussion of Section 7.5 and introduces the Lagrange remainder formula. This
leads to the Taylor formula which is an important result in mathematical analysis.
Section 7.7 introduces elasticities, and their definition in terms of derivatives. The alternative character-
ization of elasticities as logarithmic derivatives is explained.
Continuity is the subject of Section 7.8. After some preliminary discussion, the need to consider continu-
ity is explained, then a formal definition in terms of limits is provided. It is pointed out that sums, differences,
products, powers, and composites of continuous functions are continuous. So are quotients, where the de-
nominator is not zero.
Section 7.9 continues the discussion of limits that was started in Section 6.5. In some sense, limits and
continuity are logically prior to the first results on differentiation contained in Chapter 6. Nevertheless, we
have chosen this order of presentation because the topics treated in Section 7.9 may seem rather abstract
unless one has first seen limits being used in connection with derivatives. One-sided limits are discussed, as
are one-sided derivatives for functions whose graphs may have corners. Limits at infinity are briefly described,
while Problem 7.9.4 introduces the general definition of asymptotes. In this section we also point out that
differentiable functions are continuous.
A rigorous definition of limits and continuity is discussed in small print. We feel that at least some
students can already benefit at this stage from seeing the -definition of limits and continuity, although a real
understanding of these concepts is beyond this text.
Section 7.10 has a discussion of the intermediate value theorem. Newtons method for finding the roots
of equations also receives brief treatment.
Section 7.11 deals with infinite sequences and has a short description of irrational numbers as limits of
sequences. The chapter ends with Section 7.12 discussing lHpitals rule for evaluating limits, which is quite
often used in economics.

7.1
dv dv dv 2u + v dv
2. 2u + v + u 3v 2 = 0, so = 2 . Hence = 0 when v = 2u (provided 3v 2 u 6 = 0).
du du du 3v u du
Substituting for v in the original equation yields 8u3 u2 = 0. So the only point on the curve where
dv/du = 0 and u 6= 0 is (u, v) = (1/8, 1/4).

4. (a) y = x/y (b) y = y/x (c) y = 2x(2x 2 y 3 )/y 2 (3x 2 + 4y)
1
6. y = . (Differentiation w.r.t. x yields 3x 2 F (xy) + x 3 F (xy)(y + xy ) + exy (y + xy ) = 1.
F (0)
+1
Then put x = 1, y = 0. Note that F is a function of only one variable, with argument xy.)

CHAPTER 7 DERIVATIVES IN USE 25
7.2
2. (a) 1 = C (Q )(dQ /dP ), so dQ /dP = 1/C (Q ) (b) dQ /dP > 0, which is reasonable because
if the price received by the producer increases, the optimal production should increase.
4. (a) Y = f (Y )+I +Xg(Y
). Differentiating
w.r.t. I yields dY /dI = f (Y )(dY /dI )+1g (Y )(dY /dI ).
Thus, dY /dI = 1/ 1 f (Y ) + g (Y ) . Imports should increase when income increases, so g (Y ) > 0.
We find that dY /dI > 0. (b) d 2 Y /dI 2 = (f g )(dY /dI )/(1f +g )2 = (f g )/(1f +g )3 .
7.3
4x 2 (3 x 2 )
2. (a) f (x) = . So f increases in [ 3, 3 ], and decreases in [2, 3 ] and in [ 3, 2].
3 4 x2
See Figure
M7.3.2. (b) f has an inverse in the interval [0, 3] because f is strictly increasing there.
g ( 13 3 ) = 1/f (1) = 3 3/8.
y
2
1 r + dr
x r
2 1 1 2
1
7.4
1 1 10
2. 2
x. (f (0) = 1/9, f (x) = 10(5x + 3)3 , so f (0) = 10/27.)
(5x + 3) 9 27

4. AK A 1 + (K 1)
6. (a) (px p1 + qx q1 ) dx (b) (p + q)x p+q1 dx (c) rp(px + q)r1 dx (d) (pepx + qeqx )dx
8. (a) (i) 1y = 0.61, dy = 0.6 (ii) 1y = 0.0601, dy = 0.06
(b) (i) 1y = 0.011494, dy = 0.011111 (ii) 1y = 0.001115, dy = 0.001111
(c) (i) 1y = 0.012461, dy = 0.0125 (ii) 1y = 0.002498, dy = 0.0025
10. (a) A(r + dr) A(r) is the shaded area in Fig. M7.4.10. It is approximately equal to the length of the
inner circle, 2 r, times dr.
(b) V (r + dr) V (r) is the volume of the shell between the ball with radius r + dr and the ball with
radius r. It is approximately equal to the area of the inner ball 4 r 2 times dr.
7.5
2. p(x) = x 21 x 2 + 31 x 3 41 x 4 + 51 x 5
4. Follows from formula (1) with f = U , a = y, x = y + M s.

6. We find x(0) = 2[x(0)]2 = 2. Differentiating the expression for x(t) yields x(t) = x(t) + t x(t) +
4[x(t)]x(t), and so x(0) = x(0)+4[x(0)]x(0) = 1+412 = 9. Hence, x(t) x(0)+ x(0)t + 21 x(0)t 2 =
1 + 2t + 29 t 2 .
8. Use (2) with f (x) = (1+x)n and x = p/100. Then f (x) = n(1+x)n1 and f (x) = n(n1)(1+x)n2 .
The approximation follows.
7.6

2. (a) 25 = 3(1 2/27)1/3 3(1 31 27 2
91 2742 ) 2.924
3
1 2 1
(b) 33 = 2(1 + 1/32)1/5 2(1 + 532
5
25 322
) 2.0125

4. (a) p(x) = 1 + 1
x 1 2
x (b) |R 3 (x)| = 1 g (c)x 3 = 1 10 (1 + c)8/3 x 3 5 x 3
3 9 3! 6 27 81
(c) 3 1003 = 10(1+3103 )1/3 10.0099900, using part (a) to approximate (1+3103 )1/3 . The error
5
in (b) is |R3 (x)| 81 (3 103 )3 = 53 109 . So the error in 3 1003 is 10|R3 (x)| = 50 3 10
9 < 2 108 ,
and the answer is correct to 7 decimal places.
7.7
2. ElK T = 1.06. A 1% increase in expenditure on road building leads to an increase in the traffic volume
of approx. 1.06 %.
4. (a) Elx eax = (x/eax )aeax = ax (b) 1/ ln x (c) p + ax (d) p + 1/ ln x
6. Elr D = 1.23. This means that a 1% increase in income leads to an increase in the demand for apples of
approx. 1.23 %
8. (a) See Fig. M7.7.8. (The straight line has been drawn through the lowest and highest points.)
T 36.3 35.0 33.9 32.4 24.7 24.2
ln n 5.04 4.89 4.70 4.54 3.64 3.58
(b) f (T ) = 1.99e0.12T (c) The fall in temperature that halves the pulse rate is (ln 2)/0.12 5.8 degrees.
ln n
T
20 25 30 35 40
Figure M7.7.8
10. (a) Elx A = (x/A)(dA/dx) = 0

x x xf xg
(b) Elx (fg) = (fg) = (f g + fg ) = + = Elx f + Elx g
fg fg f g


f xg f xg gf fg xf xg
(c) Elx = = = = Elx f Elx g
g f g f g2 f g
xf xg
f + g
x(f + g ) f g f Elx f + gElx g
(d) Elx (f + g) = = =
f +g f +g f +g
(e) Is like (d), but with +g replaced by g, and +g by g .
x dz x u dz du
(f) z = f (g(u)), u = g(x) Elx z = = = Elu f (u) Elx u
z dx u z du dx
7.8
2. f is discontinuous at x = 0. g is continuous at x = 2. The graphs of f and g are shown in Figs. M7.8.2a
and M7.8.2b.
y y
4 4
3 3 g
2 2
1 1
2 1 2 3 x 1 2 3 4 5 x
1
f
2 2
3
Figure M7.8.2a Figure M7.8.2b
4. See Fig. M7.8.4; y is discontinuous at x = a.

6. See Fig. M7.8.6. (This example shows that the commonly seen statement: if the inverse function ex-
ists, the original and the inverse function must both be monotonic is wrong. This claim is correct for
continuous functions, however.)
y
y
y f (x)
x
1
a x
x
Figure M7.8.4 Figure M7.8.6 Figure M7.9.6
7.9
1 3 r s
x3 2 2 + 3x 3 + 2/x
2. (a) 2 = x x 0 as x . (b) = 3 as x . (c) a 2
x +1 1 x1 1 1/x
1+ 2
x

4. (a) y = x 1 (x = 1 is a vertical asymptote). (b) y = 2x 3 (c) y = 3x + 5 (x = 1 is a vertical

asymptote). (d) y = 5x (x = 1 is a vertical asymptote).
6. f (0+ ) = 1 and f (0 ) = 0. See Fig. M7.9.6.
7.10
2. A persons height is a continuous function of time (even if growth occurs in intermittent spurts, often
overnight). The intermediate value theorem (and common sense) give the conclusion.
4. Integer root: x = 3. Newtons method gives 1.880, 0.346, and 1.533 for the three other roots.
6. If f (x0 ) and f (x0 ) have the same sign (as in Fig. 2), then (1) implies that x1 < x0 . But if they have
opposite signs, then x1 > x0 .
7.11
3n 3
2. (a) Converges to 5. (b) Diverges to . (c) Converges to 3 2/2. (sn = p = p
n 2 1/n2 2 1/n2
3/ 2 = 3 2/2 as n )
7.12
x 2 a2 0 2x x 2 a2 (x a)(x + a)
2. (a) lim = = lim = 2a (or = = x + a 2a as x a.)
xa x a 0 xa 1 xa xa
2(1 + x)1/2 2 x 0 (1 + x)1/2 1 0
(b) lim 2 1/2
= = lim 2 1/2
= =
x0 2(1 + x + x ) 2x 0 x0 (1 + 2x)(1 + x + x ) 1 0
(1/2)(1 + x)3/2 1/2 1
lim 2 1/2 2 2 3/2
= =
x0 2(1 + x + x ) + (1/2)(1 + 2x) (1 + x + x ) 2 1/2 3
4. The second fraction is not 0/0. The correct limit is 5/2.
f (x) f (1/t) 0 f (1/t)(1/t 2 ) f (1/t) f (x)
6. lim = lim+ = = lim+ = lim = lim
x g(x) t0 g(1/t) 0 t0 g (1/t)(1/t 2 ) t0+ g (1/t) x g (x)

y2 y
2. 5y 4 y y 2 2xyy = 0, so y = = for y 6 = 0 and 2x 6= 5y 3 . Because no point
5y 4 2xy 5y 3 2x
with y = 0 can satisfy the given equation, y is never 0.
4. (a) y = 4/13 (Implicit differentiation yields () 2xy + x 2 y + 9y 2 y = 0.)
(b) Differentiating () w.r.t. x yields: 2y + 2xy + 2xy + x 2 y + 18yy y + 9y 2 y = 0. Inserting x = 2,
y = 1, and y = 4/13 gives the answer.
(2/x)(1 + 15 ln x)
6. y = = 0 for x = e5 .
1 + 1/y
8. (a) f (e2 ) = 2 and f (x) = ln x(ln x 1)2 = 0 for ln x = 0 and ln x = 1, so x = 1 or x = e.
(b) f (x) = (3/x)(ln x 1)(ln x 1/3) > 0 for x > e, and so f is strictly increasing in [e, ). It
therefore has an inverse h. According to (7.3.3), because f (e2 ) = 2, we have h (2) = 1/f (e2 ) = e2 /5.

10. (a) x dx/ 1 + x 2 (b) 8 r dr (c) 400K 3 dK (d) 3x 2 dx/(1 x 3 )

1
12. Letting x = 2 and n = 5, formula (7.6.6) yields,
1
1
2 ( 21 )2 ( 1 )3 ( 1 )4 ( 1 )5 ( 1 )6
e2 = 1 + + + 2 + 2 + 2 + 2 ec
1! 2! 3! 4! 5! 6!
( 1 )6 ( 21 )6
where c is some number between 0 and 21 . Now, R6 ( 21 ) = 6! 2
ec < 6! 2 = 1
23040 0.0004340, where
1
we used the fact that since c < 21 , ec < e 2 < 2. Thus it follows that
1
1
2 ( 21 )2 ( 1 )3 ( 1 )4 ( 1 )5 1 1 1 1 1
e2 1 + + + 2 + 2 + 2 =1+ + + + + 1.6486979
1! 2! 3! 4! 5! 2 8 48 384 3840
1
The error is less than 0.000043, and e 2 1.649 correct to 3 decimals.
14. We find y = 1/2 and y = 1/8, so y(x) 1 + 21 x + 1 2
16 x .
16. Elr (Dmarg ) = 0.165 and Elr (Dmah ) = 2.39. When income increased by 1%, the demand for mar-
garine decreased by approximately 0.165 %, while the demand for meals away from home increased by
approximately 2.39 %.
18. Put f (x) = x 3 x 5. Then f (x) = 3x 2 1. Taking x0 = 2, formula (7.10.1) with n = 1 gives
x1 = 2 f (2)/f (2) = 1 1/11 1.909.

3 x + 17
20. (a) 2 (b) Approaches +. (c) approaches + as x 1 , but as x 1+ ,
x+1
so there is no limit as x 1.
2ex+1 x 2 4x 5 0 2ex+1 2x 4 0
22. (a) 1 (b) 1/16 (c) lim = = lim = =
x1 (x + 1)3 0 x1 3(x + 1)2 0
2ex+1 2 0 2ex+1 1
lim = = lim = .
x1 6(x + 1) 0 x1 6 3
24. Does not exist if b 6= d. If b = d, the limit is 0/0 andby lHpitals
rule, it is
1 1/2 1 1/2
limx0 2 a(ax + b) 2 c(cx + d) /1 = a/2 b c/2 d = (a c)/2 b.
26. x1 = 0.9f (0.9)/f (0.9) 0.9247924, x2 = x1 f (x1 )/f (x1 ) 0.9279565, x3 = x2 f (x2 )/f (x2 )
0.9280338, and x4 = x3 f (x3 )/f (x3 ) 0.9280339. So it seems that the answer correct to 3 decimals
is 0.928.

30 CHAPTER 8 SINGLE-VARIABLE OPTIMIZATION
Chapter 8 Single-Variable Optimization

The previous chapters have presented most of the mathematical functions used in economic analysis, as well
as rules for differentiating them, etc. After these essential preliminaries, Chapter 8 turns to the systematic
study of single variable optimization problems.
In Section 8.1 we begin by giving algebraic (non-calculus) definitions of maxima and minima. Then we
define a stationary point, and discuss the role played by stationary points in optimization theory. Students
should be encouraged even more strongly than usual to do at least some of the problems in this section in
order to understand that sometimes extreme points can be found by merely examining the formula for the
function, without any need for differentiation.
Section 8.2 gives an extremely useful and important test for (global) maxima and minima, based only
on the variation in the sign of the first derivative (Theorem 8.2.1). When this test is passed, there is no
need whatsoever to consider the second derivative. (We emphasize this point because it is often neglected by
economists who, in our experience, are routinely taught to consider second-derivative tests even when there
is no need to do so.)
Theorem 8.2.2 claims that, at an interior stationary point of an interval in which a function is concave (or
convex), that function has a maximum (or minimum) over that interval. This is a result of great significance
in many economic applications. It is also a result that extends to functions of n variables which are concave
or convex on some convex set. The case n = 2 is covered in Theorem 13.2.1.
Section 8.3 presents some interesting economic examples. The economic interpretation of condition ()
in Example 1 should be noted. Example 3 illustrates an important type of sensitivity analysis of an optimization
problem, and illustrates the envelope theorem in a special case.
Next, in Section 8.4, the extreme-value theorem is formulated. It is then pointed out that extreme points,
if they are not stationary points, must be either end points of an interval on which the function is defined, or
else points where the function is not even differentiable. This is the justification for the usual procedure of
searching for extrema by considering stationary points. But it is important to realize when the procedure fails
because the relevant extreme point is not an interior point of the domain at which the function is differentiable.
Otherwise, one would be restricted to interior and smooth maxima for the rest of ones work in economic
analysisand, of course, corner solutions are often important. Finally, the mean-value theorem is formulated
and used to confirm the tests for increasing/decreasing functions discussed in Section 6.3.
In Section 8.5 further economic examples are studied. Again, the economic interpretations of the first-
order conditions should be stressed.
Section 8.6 moves on to local rather than global extreme points. First, these are defined algebraically
(using the concept of a neighbouring interval about a local extreme point). Theorem 8.6.1 gives conditions
for a stationary point to be a local maximum or a local minimum, or neither, based on the variation in the sign
of the derivative near the stationary point. (These tests are often erroneously claimed to be not only sufficient
but also necessary for local optimality.)
Then Theorem 8.6.2 proceeds to the rather standard second-derivative test. This gives a sufficient condition
for a (strict) local maximum (or minimum), based on the function being twice differentiable around the
stationary point, with its second derivative being negative (or positive) at that point. (If f (c) = 0, the test
fails. Then Theorem 8.6.1 can usually be applied. Tests based on evaluating higher order derivatives at c, we
believe, are only of marginal interest to economists. Again, these nth order tests are often erroneously claimed
to be necessary and sufficient.) As a simple exercise in comparative statics, Example 5 shows how the second
derivative at an optimum can be used to infer the sign of a response to an exogenous shock.

CHAPTER 8 SINGLE-VARIABLE OPTIMIZATION 31
Section 8.7 considers inflection points, for which Theorem 8.7.1 gives a simple test. This section also
provides an alternative definition of concavity/convexity based on line segments joining pairs of points on
the graph. This leads next to standard definitions of strict concavity or convexity, for which there are obvious
second-derivative sufficient conditions. These ideas will be generalized and extensively discussed in FMEA.

8.1
2
2. (a) Minimum 1 at x = 0. (F (0) = 1 and 1 for all x because 2 + x 2 2 and so
2 + x2
2
1.) No maximum. (b) Maximum 2 at x = 1. No minimum.
2 + x2
(c) Minimum 99 at x = 0. No maximum. (When x , H (x) 100.)
8.2

8(2
3x)(2 + 3x)
2. h (x)
= . The function has a maximum at x = 2 3/3 and a minimum at
(3x 2 +
4)2
x = 2 3/3, with h(2 3/3) = 2 3/3 and h(2 3/3) = 2 3/3.
4. f (x) = [4x(x 4 + 1) 2x 2 4x 3 ]/(x 4 + 1)2 , then simplify and factor. f on [0, ) has maximum 1 at
x = 1, because f (x) increases in [0, 1] and decreases in [1, ).
6. f (x) = 3e3x 6ex = 3ex (e2x 2). So f (x) = 0 when e2x = 2 or 2x = ln 2, so x = 21 ln 2. If
x < 21 ln 2 then f (x) < 0, and if x > 21 ln 2 then f (x) > 0, so x = 21 ln 2 is a minimum point.
f (x) = ex (e2x 6) tends to + as x , so f has no maximum.
1
8. (a) x = 3 ln 2 is a minimum point (y is convex). (b) x = 31 (a + 2b) is a maximum point (y is concave).
1
(c) x = 5 is a maximum point (y is concave).
10. (a) f (x) = k Aex = 0 when x0 = (1/) ln(A/k). Note that x0 > 0 iff A > k. Moreover,
f (x) < 0 if x < x0 and f (x) > 0 if x > x0 , so x0 solves the minimization problem.
(b) Substituting for A in the answer to (a) gives the expression for the optimal height x0 . Its value increases
as p0 (probability of flooding) or V (cost of flooding) increases, but decreases as (interest rate) or k
(marginal construction costs) increases. The signs of these responses are obviously what an economist
would expect.
8.3
2. (a) (Q) = (102 2Q)Q (2Q + 21 Q2 ) = 100Q 25 Q2 , which is maximized at Q = 20.
(Q) = 100Q 25 Q2 tQ, which is maximized at Q = 20 t/5.
(b) b
4. p (x) = kcecx , p (x) = kc2 ecx . No maximum exists, and p(x) a + k as x . See
Fig. M8.3.2.
8.4
2. (a) Maximum 1 at x = 0. Minimum 7 at x = 3.
(b) Maximum 10 at x = 1 and x = 2. Minimum 6 at x = 1.
(c) Maximum 5/2 at x = 1/2 and x = 2. Minimum
2at x = 1.
(d) Maximum 4 at x = 1. Minimum 6 3 at x = 3.
(e) Maximum 4.5 109 at x = 3000. Minimum 0 at x = 0.

32 CHAPTER 8 SINGLE-VARIABLE OPTIMIZATION
R(Q) = 80Q
p 4000
a+k 3000
p(x) = a + k(1 ecx ) C(Q) = Q2 + 10Q + 900
2000
a 1000
x Q
10 Q0 30 Q 40 50
2 2
4. g (x) = 25 xex (1 e22x ). Stationary points: x = 0 and x = 1. Here x = 2 is a maximum point,
x = 1 and x = 1 are minimum points. (Note that g(2) = 15 (e4 + e2 ) > g(0) = 51 (1 + e2 ).)

6. (a) x = 3/2 (b) x = 2/2 (c) x = 12 (d) x = 3
8. f is not continuous at x = 1 and x = 1. It has no maximum because f (x) is arbitrarily close to 1 for
x sufficiently close to 1. But there is no value of x for which f (x) = 1. Similarly, there is no minimum.
8.5
Q [0, 50], that is Q2 +70Q900 0
2. (a) See Fig. M8.5.2. (b) (i) The requirement is (Q) 0 and
and Q [0, 50]. The firm must produce at least Q0 = 35 5 13 17 units. (ii) Profits are maximized
at Q = 35.
4. (i) Q = 450 (ii) Q = 550 (iii) Q = 0
6. (Q) is stationary at Q = (P /ab)1/(b1) . Moreover, (Q) = ab(b 1)Qb2 < 0 for all Q > 0, so
this is a maximum point.
8.6
2. (a) No local extreme points. (b) Local maximum 10 at x = 1. Local minimum 6 at x = 1.
(c) Local maximum 2 at x = 1.Local minimum 2 at x =
1.
(d) Local maximum 6 3 at x = 3. Local minimum 6 3 at x = 3.
(e) No local maximum point. Local minimum 1/2 at x = 3.
(f) Local maximum 2 at x = 2. Local minimum 2 at x = 0.
4. a and d are local minimum points, whereas c is a local maximum point for f . Points b and e are neither.
6. (a) x = 3 is a local (and global) minimum point. No local maximum points. (b) x = 0 is a local
minimum point and x = 2/ ln 2 is a local maximum point. (Note that f (x) = x2x (2 + x ln 2).)
8.7

2. (a) f (x)
= 2x(x 2 3)(1 + x 2 )3 , so f is convex in [ 3, 0] and in [ 3, ). Inflection points are at

x = 3, 0, and 3.
(b) g (x) = 4(1 + x)3 > 0 when x > 1, so g is (strictly) convex in (1, ). No inflection points.
(c) h (x) = (2 + x)ex , so h is convex in [2, ) and x = 2 is an inflection point.

CHAPTER 8 SINGLE-VARIABLE OPTIMIZATION 33

4. (a) For x > 0 one has R = p x, C = wx + F , and (x) = p x wx F .

(b) (x) = p(1/2 x) w = 0, or p/2 x = w. (Marginal cost = price times marginal product.) Then
x = p2 /4w 2 . Moreover, (x) = 41 px 3/2 < 0 for all x > 0, so profit is maximized over the interval
(0, ). When x = p2 /4w2 , then = p2 /2w p 2 /4w F = p 2 /4w F . So this is a profit maximum
if F p2 /4w; otherwise, the firm does better not to start up and to choose x = 0.
6. a = 2/5, b = 3/5 (f (1) = 1 gives a + b = 1. Moreover, f (x) = 3ax 2 + 2bx and f (x) =
6ax + 2b, so f (1/2) = 0 yields 3a + 2b = 0.)
y
y 2
f g
1
x
1 1 2 3 4 5
1
x 2
Figure M8.7.8 Figure M8.R.4
8. See Fig. M8.7.8. Use definition (2).

2. (a) L = 160, L = 120 (b) Q (120) = Q(120)/120 = 720. In general (see Example 6.7.6),
(d/dL)(Q(L)/L) = (1/L)(Q (L) Q(L)/L). If L > 0 maximizes output per worker, one must have
Q (L) = Q(L)/L.
3
27e at x = 3 and no minimum. f has inflection points at x = 0,
4. (a) f hasa (global) maximum
x = 3 3, and x = 3 + 3. (b) lim f (x) = 0, lim f (x) = . See Fig. M8.R.4.
x x
6. (a) f (x) = 2x + 1 ex , f (x) = 2 + ex . f (x) is strictly increasing in [3, ln 2] and strictly
decreasing in [ ln 2, 3].
(b) Note that f (3) = 7 e3 < 0, f ( ln 2) = 2 ln 2 1 > 0, and f (3) = 5 e3 < 0. By the
intermediate-value theorem and the strict monotonicity of f (x) on both sides of x = ln 2, it follows
that f (x) = 0 has just one solution (say, x0 ) in (3, ln 2), and another (actually x = 0) in ( ln 2, 3).
(c) Because f decreases in [3, x0 ], increases in [x0 , 0], then decreases again in [0, 3], the candidate
maxima are x = 3 and x = 0. Because f (3) = e3 12 > f (0) = 1, the maximum is at x = 3.
8. (a) f (x) = 4e4x +8ex 32e2x , f (x) = 16e4x +8ex +64e2x (b) f (x) = 4e2x (e3x +4)(e3x 2), so
f (x) is increasing in [ 31 ln 2, ), decreasing in (, 31 ln 2]. f (x) > 0 for all x so f is strictly convex.
(c) 31 ln 2 is a (global) minimum. No maximum exists because f (x) as x .
10. (a) g (x) = (a 1)(1 ca x a ) and g (x) = a(a 1)ca x a1 . Moreover, limx0+ g(x) = and
limx g(x) = . (b) gmin = g(c) = a(c 1). (Look at the sign of the derivative.)
(c) Because g(c) = a(c 1) < 0, the intermediate-value theorem and strict monotonicity of g(x) on each
side of x = c imply that there is one root in each of the intervals (0, c) and (c, ). But g(1) = ca 1 < 0
and g(a/(a 1)) = ca (a/(a 1))1a > 0, so the root that is greater than c must lie in (1, a/(a 1)).

34 CHAPTER 9 INTEGRATION
Chapter 9 Integration
Integration is an important mathematical technique which features occasionally in economic analysis. Indeed,
after this chapter, integrals appear only rarely in the rest of the book. However, when we study differential
equations and control theory in FMEA, integrals become important. Despite this, undergraduate economics
students do need a thorough understanding of elementary ideas in the theory of integration, not so much for
the occasional economic application they may see, but because integrals play such an enormous role in the
statistical foundations of econometrics.
Section 9.1 introduces an indefinite integral as, in effect, an antiderivativei.e., a function whose de-
rivative is the original function. Indefinite integrals are defined only up to an arbitrary additive constant of
integration. General rules for integration are presented.
Section 9.2 shows how integrals are used to find the area of certain plane regions. In fact, it is shown that
if A(x) denotes the area under the graph of the non-negative valued function f (x) over the interval [a, x], then
A(x) is an indefinite integral of f (x). Students should be made aware that this discovery is not another boring
fact, but historically a fundamental breakthrough in the history of science. On this basis the area under the
graphs of complicated functions can often be calculated with ease. We next introduce definite integrals. They
can be found, of course, as differences in the value of any indefinite integral (or antiderivative), provided that
such an indefinite integral is known. The section concludes with a brief consideration of the areas associated
with graphs of functions that may have negative values.
In Section 9.3 standard properties of definite integrals are explainede.g., the integral of a sum is the
sum of the integrals, etc. The rules for differentiating a definite integral w.r.t. both its upper and lower limits
of integration are given. It is pointed out that any continuous function, at least, can be integrated, and a list of
insoluble integrals is given. Also, it is briefly explained how the NewtonLeibniz integral presented in
the chapter can be extended to Riemann integrals (at least for bounded functions).
After three sections of almost pure mathematics comes Section 9.4, with economic applications. The first
concerns extraction from an oil well. It brings out the elementary but important idea that changes in stocks
are integrals of flows. There follows a rather extended discussion of the cumulative distribution function of
income, defined as the integral of a density function. It is shown how one can find total and mean income
for all individuals with income in a certain interval. There is also a discussion of how, when all individuals
demands for a commodity are a known function of their incomes, it is possible to find total demand for that
commodity once the income distribution is also known. The section concludes with a brief discussion of
consumer and producer surplus.
Next, Section 9.5 presents the method of integration by parts. This is the counterpart for integrals of the
product rule for differentiation. It is a tricky technique to master, however, because it is not very obvious
which part of the integral needs to be integrated first. So students should be encouraged to attempt as many
problems as possible.
The second important technique for economics students to learn is that of integration by substitution, as
discussed in Section 9.6. This is the counterpart for integration of the chain rule for differentiating a function
of a function.
In Section 9.7 we study integrals over infinite intervals, partly because some important economic models
involve an infinite planning horizon, partly because this generalization is useful in statistics.
The chapter ends with a short introduction to differential equations. The first-order equations for propor-
tional growth, growth towards an upper limit, and logistic growth are presented and solved. Of course, FMEA
provides a much more extensive discussion of differential equations.

CHAPTER 9 INTEGRATION 35

9.1
1
2. (a) ex + C (b) 4e 4 x + C (c) 23 e2x + C (d) (1/ ln 2)2x + C
4. (a) 41 t 4 + t 2 3t + C (b) 13 (x 1)3 + C (c) 13 x 3 + 21 x 2 2x + C
(d) 41 (x + 2)4 + C (e) 31 e3x 21 e2x + ex + C (f) 13 x 3 3x + 4 ln |x| + C
6. (a) and (b): Differentiate the right-hand side and check that you get the integrand. (For (a) see also
Problem 9.5.4.)
8. The graph of f (x) in Fig. 2 can be that of a cubic function, with roots at 3, 1, and 1, and with
f (0) = 1. So f (x) = 31 (x + 3)(x + 1)(x 1) = 13 x 3 + x 2 13 x 1. If f (0) = 0, integrating gives
1 4
f (x) = 12 x + 31 x 3 61 x 2 x. Figure M9.1.8 is the graph of this f .
10. (a) Differentiate the right-hand side. (Once we have learned integration by substitution in Section 9.6,
1

this is an easy problem.) (b) (i) 10 (2x + 1)5 + C (ii) 23 (x + 2)3/2 + C (iii) 2 4 x + C
12. The general form for f is f (x) = 31 x 3 + A, so that for f is f (x) = 12
1 4
x + Ax + B. If we require that
1 4
f (0) = 1 and f (0) = 1, then B = 1 and A = 1, so f (x) = 12 x x + 1.
y y y
f
2 2
1 1
f (x) = ex
4 3 2 1 1 2 x 1 1 2 x
1 1
1 1 x
9.2
R2 2
2. (a) 0 3x 2 dx = 0 x 3 = 8 (b) 1/7 (c) e 1/e. (See the shaded area in Fig. M9.2.2.) (d) 9/10
R1 1
4. A = 21 1 (ex + ex ) dx = 21 1 (ex ex ) = e e1

6. (a) f (x) = 3x 2 6x + 2. With x0 = 1 13 3 and x1 = 1 + 13 3, f (x) increases in (, x0 ) and in
R1
(x1 , ). (b) See Fig. M9.2.6. The shaded area is 0 f (x) dx = 10 ( 41 x 4 x 3 + x 2 ) = 41 .
8. (a) 6/5 (b) 26/3 (c) (e 1)/ (d) ln 2
9.3
1 1
2. (a) + (b) f (x) = 4x 3 3x 2 + 5. (f (1) = 6 gives a + b = 6, f (1) = 18 gives
p+q +1 p+r +1
2a + b = 18.
R 2 Then a = 12 and b = 6, so f (x) = 12x 2 6x. By integration, f (x) = 4x 3 3x 2 + C.
Requiring 0 (4x 3x 2 + C) dx = 18 gives C = 5.)
3

4. (a) Use formula (6) to obtain F (x) = x 2 + 2. To find G (x), put u = x 2 and use formula (8). Then
G (x) = [(x 2 )2 + 2]2x = 2x 5 + 4x. (One can also evaluate the integrals and then differentiate: G(x) =
x 2 1 3
( t + 2t) = 1 x 6 + 2x 2 , so G (x) = 2x 5 + 4x.) (b) H (t) = 2tK(t 2 )et 2 (use formula (8)).
0 3 3
R3 2
6. The parabolas intersect at (0, 0) and (3, 3), so A = 0 ( 3x x + 2x) dx = 6. See Fig. M9.3.6.
2
8. (a) f (x) = > 0 for x > 0 and f has range (, ), so f has an inverse g
x + 4 ( x + 4 2)
defined on (, ). We find that the inverse is g(x) = ex/2 + 4ex/4 .
(b) See Fig. M9.3.8. (c) InR Fig. M9.3.8 the graphs of f and g are symmetric about the line y = x, so
a
A = area B = 10a 0 (ex/2 + 4ex/4 ) dx= 10a + 18 2ea/2 16ea/4 . Because a = f (10) =
area
4 ln( 14 2), this simplifies to 10a + 14 8 14 6.26.
y
a
10
B y=x
y
y + 1 = (x 1)2 5 150 g f
4 g
y2 = 3x
3 A a 100
2 5 5 10 x
1 50
f
x
1 1 3 4 5 6 5 t
1 2 4 6 8 10
9.4
R 2b
2. (a) m = 2b ln 2. (The relevant number of individuals is n b Br 2 dr = nB/2b, whose total income is
R 2b R 2b
M = n b Br 1 dr = nB ln 2.) (b) x(p) = b nAp r Br 2 dr = nABp b1 (21 1)/( 1)
4. (a) See Fig. M9.4.4.Z (f and g both have maximum value 4000/27 148 at t = 20/3 and at t = 10/3,
t 1
respectively.) (b) g( ) f ( ) d = t 2 (t 10)2 0 for all t.
Z 10 0 Z 2
10
(c) p(t)f (t) dt = t 3 + 9t 2 + 11t 11 + 11/(t + 1) dt = 940 + 11 ln 11 966.38,
Z 10 0 Z 10 0
3
p(t)g(t)dt = t 19t 2 + 79t + 121 121/(t + 1) dt = 3980/3 121 ln 11 1036.52.
0 0
Profile g should be chosen.
6000
6. Equilibrium when = Q + 10. The only positive solution is Q = 50, and then P = 60.
Z 50 Q + 50 50
6000
CS = 60 dQ = [6000 ln(Q + 50) 60Q] = 6000 ln 2 3000,
Q + 50
Z 050 0
PS = (50 Q) dQ = 1250
0

9.5
Z 1 1 Z 1 Z 1
1 2 1 2 1 1 1 x2
2. (a) x ln(x + 2) dx = ln(x + 2)
2x 2x x dx = 2 ln 3 2 dx =
1 1 1 +2 1 x+2
Z 1
1 1 4
2 ln 3 2 x 2 + dx = 2 23 ln 3 (b) 8/(ln 2) 3/(ln 2)2 (c) e 2
1 x + 2
4. Use (1) with f (x) = ln x and g (x) = x . (Alternatively, simply differentiate the right-hand side.)
9.6
1
R
2. (a) 2 6 2
24 (2x + 3) + C. (Substitute u = 2x + 3, so du = 4x dx. The integral becomes 41 u5 du.)
1 x 3 +2 3 2
(b) 3e + C. (Substitute u = ex +2 .) (c) 41 ln(x + 2) + C. (Substitute u = ln(x + 2).)
2 5/2 2 (1 + x)3/2 + C. (Substitute u =

(d) 5 (1 + x) 3 1 + x.)
1 1
(e) + + C. (Substitute u = 1 + x 2 , or u = (1 + x 2 )1 .)
2(1 + x ) 4(1 + x 2 )2
2

2
(f) 15 (4 x 3 )5/2 98 (4 x 3 )3/2 + C. (Substitute u = 4 x 3 .)
Z x x
2t 2
4. dt = ln(t 2 2t) = ln(x 2 2x) ln 3 = ln 1 (x 2 2x), so the given equation reduces to
t 2 2t 3
3 3
1 2 2 2
3 (x 2x) = 3 x 1. Hence, x 4x + 3 = 0, with solutions x = 1 and x = 3. But only x = 3 is in
the specified domain. So the solution is x = 3.
6. (a) 1/70 (The integrand can be written as x 4 (x 5 1)13 . Then put u = x 5 1.)

(b) 2 x ln x 4 x + C. (Let
p u = x. Then u2 = x, 2udu = dx, etc. Alternatively, use integration by

parts.) (c) 8/3. (Let u = 1 + x.)
9.7
Z Z
+ b
1 1 b 1
2. (a) f (x) dx = dx = x= (b a) = 1
a ba ba a ba
Z + Z b b
x 1 2 1 1
(b) xf (x) dx = dx = x = (b2 a 2 ) = (a + b)
a ba 2(b a) a 2(b a) 2
1 b
1 b 3 a3 1
(c) x3 = = (a 2 + ab + b2 )
3(b a) a 3 ba 3
Z b b
4. The first integral diverges because [x/(1+x 2 )] dx = 21 ln(1+x 2 ) = 21 ln(1+b2 ) as b .
Z b 0 0
b
1
On the other hand, [x/(1 + x 2 )] dx = ln(1 + x 2 ) = 0 for all b, so the limit as b is 0.
b b 2
Z b b
1 1 1
6. 2
2
for x 1, and 2
dx = (1/x) = 1 1/b 1 as b , so by Theorem 9.7.1
1+x x 1 x 1
the given integral converges.
Z Z
1 rs 1 r 2( s) rs 2 1 r
8. (a) z = e ds = (1 e ) (b) z = e ds = 1 (1 e )
0 r 0 2 r r
Z 1
ln x 1
10. Using the answer to Problem 9.6.6(b), dx =h (2 x ln x 4 x) = 4 (2 h ln h 4 h)
h x
4 as h 0+ , so the given integral converges to 4. ( h ln h = ln h/ h1/2 0, by lHpitals rule.)

Z + Z +
1 2
12. (a) The suggested substitution gives f (x) dx = eu du = 1, by (8).
Z + Z +
1 2
(b) Here xf (x) dx = ( + 2 u)eu du = , using part (a) and Example 3.

Z + Z +
1 2
(c) Here I = 2
x f (x) dx = (2 2 u2 + 2 2 u + 2 )eu du

Z Z Z +
2 2 + 2 u2 2 2 + u2 2 2
= u e du + ue du + eu du = 2 + 0 + 2 . (Note how
Z Z Z +
2 u2 1 u2 1 u2 2
integration by parts gives u e du = ue + e du, so u2 eu du = 21 .)
2 2
9.8
2. (a) K(t) = (K0 I /)et + I / (b) (i) K(t) = 200 50e0.05t and K(t) tends to 200 from below
as t . (ii) K(t) = 200 + 50e0.05t , and K(t) tends to 200 from above as t .
4. N (t) = 0.02N (t) + 4 104 . The solution with N (0) = 2 106 is N (t) = 2 106 (2e0.02t 1).
6. The percentage surviving after t seconds satisfies p(t) = 100et , where p(7) = 70.5 and so =
ln 0.705/7 0.05. Thus p(30) = 100e30 22.3% are still alive after 30 seconds. Because
100et = 5 when t ln 20/0.05 60, it takes about 60 seconds to kill 95%.
8. (a) In 1950 there were about 276 thousand. In the next 10 years the number increased by 155 thousand.
(b) y 479.36 as t . See Fig. M9.8.8 for the graph.
Tractors (in 1000)

500
400
300
200
100
t
5 10 15 20
(1950) (1960) (1970)
Figure M9.8.8
10. At about 11:26. (Measuring time in hours, with t = 0 being 12 noon, one has T = k(20T ) with T (0) =
35 and T (1) = 32. So the body temperature at time t is T (t) = 20 + 15ekt with k = ln(5/4). Assuming
that the temperature was the normal 37 degrees at the time of death t , then t = ln(17/15)/ ln(5/4)
0.56 hours, or about 34 minutes before 12:00.)

25 25
2. (a) e2x + C (b) 21 x 2 2 e +C (c) 31 e3x + 31 e3x + C (d) 2 ln |x + 5| + C

1
1
4. (a) 5/4, according to Example 9.7.2. (b) 20 (1 + x 4 )5 = 31/20
0
R

(c) 5tet 0 5et dt = 5 et = 5
0 0
Re e Re e

(d) 1 (ln x)2 dx = x(ln x)2 2 1 ln x dx = e 2 (x ln x x) = e 2
1 1
2 0
2 3
(e) 9 (x + 1)3/2 = 9 (93/2 1) = 52/9 (f) 3 ln(e3z + 5) = 13 (ln 6 ln 5) =
2 1 1
3 ln(6/5)
0
Z x x

6. F (x) = 4( x 1). (Hint: (u1/2 + xu1/2 ) du = 2 3/2
3u + 2xu1/2 = 83 x 3/2 16
3 4x.)
4 4
x
8. C(x) = (e 1) + x + C0

10. P = Q = 5, CS = 50 ln 2 25, PS = 1.25
3 1
12. (a) x = Ae3t (b) x = Ae4t + 3 (c) x = (d) x = Ae 5 t (e) x = Ae2t + 5/3
12 Ae3t
21 1
(f) x = = (B = 2A)
21 t 21 t
1 Ae Be 2

b + I (1a)t b + I
14. (a) Y = (a 1)Y + (b + I) (b) Y = Y0 e +
1a 1a

40 CHAPTER 10 INTEREST RATES AND PRESENT VALUES
Chapter 10 Interest Rates and Present Values

This chapter essentially deals with the calculation of interest and present values. Section 10.1 considers
different interest periods, and defines the effective yearly interest rate. Section 10.2 considers the case when
interest is compounded so frequently that one may as well treat time as continuous and have continuous
compounding of interest.
The idea of present values is important for economists and it is discussed in Section 10.3.
Section 10.4 is concerned with geometric series and their sums. The condition for an infinite geometric
series to have a convergent sum is presented. Of course, this topic is not specific to finance, but we include
it here because it can be motivated by the need to discount payment streams extending into the indefinite
future. When the rate of discount is fixed, this is treated in Section 10.5, where the present and future values
of annuities are also derived.
Mortgage repayments are the topic in Section 10.6, which also includes a short subsection on deposits
within an interest period. The final Section 10.7 gives a brief discussion of internal rates of return.
10.1
2. (a) 5000(1+0.03)10 6719.58 (b) 37.17 years. (5000(1.03)t = 35000, so t = ln 3/ ln 1.03 37.17.)
4. (a) (i) After 2 years: 2000(1.07)2 = 2289.80 (ii) After 10 years: 2000(1.07)10 3934.30
(b) t = ln 3/ ln 1.07 16.2 years.
6. The effective yearly rate for alternative (ii) is (1 + 0.2/4)4 1 = 1.054 1 0.2155 > 0.215, so
alternative (i) is (slightly) cheaper.

8. Let the nominal yearly rate be r. By (2), 0.28 = (1 + r/4)4 1, so r = 4( 4 1.28 1) 0.25, or 25%.
10.2
2. (a) (i) 1000(1 + 0.05)10 1629 (ii) 1000(1 + 0.05/12)120 1647 (iii) 1000e0.0510 1649
(b) (i) 1000(1 + 0.05)50 11467 (ii) 1000(1 + 0.05/12)600 12119 (iii) 1000e0.0550 12182

4. e0.1t = 1/10, so 0.1t = ln 10, hence t = ln 10/0.1 23.
6. h (u) = u/(1 + u)2 > 0 for u > 0, so h(u) > 0 for u > 0, implying that g (x)/g(x) = h(r/x) > 0
for all x > 0. So g(x) is strictly increasing for x > 0. Because g(x) er as x , it follows that
g(x) < er for all x > 0. Continuous compounding of interest is best for the lender.
10.3
2. (i) The present value is 50 000 1.05755 37 806.64. (ii) 50 000 e0.05755 37 506.83
10.4
2. (a) (1/5)/(1 1/5) = 1/4 (b) 0.1/(1 0.1) = 1/9 (c) 517/[1 (1.1)1 ] = 5687
(d) a/[1 (1 + a)1 ] = 1 + a (e) 5/(1 3/7) = 35/4

CHAPTER 10 INTEREST RATES AND PRESENT VALUES 41
4. (a) Quotient k = 1/p. Converges to 1/(p 1) for |p| > 1.

(b) Quotient k = 1/ x. Converges to x x/( x 1) for x > 1, that is, for x > 1.
(c) Quotient k = x 2 . Converges to x 2 /(1 x 2 ) for |x| < 1.
6. Let x denote the number of years beyond 1971 that the Earths extractable resources of iron will last. Then
we get 794+7941.05+ +794(1.05)x = 249103 . Using (3), 794[1(1.05)x+1 ]/(11.05) = 249103
or (1.05)x+1 = 249 103 0.05/794 16.68. Using a calculator, we find x (ln 16.68/ ln 1.05) 1
56.68, so the resources will be exhausted part way through the year 2028.
P (t)ert P (t) rt rt
(t) = P (t)(e 1) P (t)re , and t > 0 can only
8. (a) f (t) = = (b) Here f
1 ert ert 1 (ert 1)2

maximize f (t) if f (t ) = 0, that is, if P (t )(e 1) = rP (t )ert , which implies that P (t ) =
rt

r P (t )/(1 ert ). (c) Using lHpitals rule, P (t )/P (t ) 1/t as r 0.
10.5
2. (a) 10 years ago the amount was: 100 000(1.04)10 67556.42
1.064 1
(b) 10000(1.063 + 1.062 + 1.06 + 1) = 10 000 43746.16
1.06 1
1 (1.06)5
4. Offer (a) is better. The second offer has present value 4600 20 540.
1 (1.06)1
6. If the largest amount is a, then according to formula (4), a/r = K, so that a = rK.
R 15 15 1 0.06t
8. PDV = 0 500e0.06t dt = 500 0 0.06 e 500
= 0.06 1 e0.9 4945.25.
FDV = e0.0615 PDV = e0.9 PDV 2.4596 4945.25 12163.3.
10.6
(0.07/12) 80000
2. Using (2) we get a = 928.87.
1 (1 + 0.07/12)120
12 000 1.115
4. Schedule (b) has present value [1 (1.115)8 ] 67 644.42.
0.115
7000
Schedule (c) has present value 22 000 + [1 (1.115)12 ] 66 384.08.
0.115
Thus schedule (c) is cheapest. When the interest rate becomes 12.5 %, schedules (b) and (c) have present
values equal to 65907.61 and 64374.33, respectively.
10.7
a a
2. Equation (1) is here + + = a0 , which yields a/r = a0 , so r = a/a0 .
1+r (1 + r)2

4. $ 1.55 million. (400 000 1/1.175 + (1/1.175)2 + + (1/1.175)7 1 546 522.94.)

6. Applying (10.5.2) with a = 1000 and n = 5 gives the equation P5 = (1000/r) 1 1/(1 + r)5 = 4340
to be solved for r. For r = 0.05%, the present value is $4329.48; for r = 0.045%, the present value is
$4389.98. Because dP5 /dr < 0, it follows that p is a little less than 5%.

42 CHAPTER 10 INTEREST RATES AND PRESENT VALUES

2. (a) 8000 1.053 = 9261 (b) 8000 1.0513 15085.19 (c) (1.05)t = 4, so t = ln 4/1.05 28.5
4. 15 000e0.0712 34745.50
6. (a) 44/(1 0.56) = 100 (b) 20/[1 (1.2)1 ] = 120 (c) 3/(1 2/5) = 5
(d) 400/(1 201 ) = 8000/19
5000
8. (a) 5000(1.04)4 = 5849.29. (b) [(1.04)4 1] = 21 232.32
0.04
(c) The last payment will be on 1st January 2006, when the initial balance of 10 000 will have earned
interest for 10 years. So K must solve 10 000 (1.04)10 + K[(1.04)8 1]/0.04 = 70 000. We find that
K 5990.49.
3200
10. (a) Present value: [1 (1.08)10 ] = 21 472.26.
0.08
3000
(b) Present value: 7000 + [1 1.085 ] = 18 978.13.
0.08
4000
(c) Four years ahead the present value is [1 (1.08)10 ] = 26 840.33. The present value when
0.08
Lucy makes her choice is 26 840.33 1.084 = 19 728.44. She should choose option (a) .

CHAPTER 11 FUNCTIONS OF MANY VARIABLES 43
Chapter 11 Functions of Many Variables

In principle, this chapter represents the first occasion on which the student will ever have seen functions of
many variables. For this reason, the rather slow start provided by Section 11.1 does seem required. In fact,
the first three sections deal exclusively with functions of two variables. Though definition (1) is abstract, it is
immediately followed by concrete examples. As usual, first comes the mechanical Example 1, which prepares
the student for problems 14. Then come, in rapid succession, a demand function with both price and income
as variables, a CobbDouglas production function, and an example that prepares for the later discussion of
homogeneous functions. Problem 5 is especially relevant at this stage. The final discussion of domains may
seem to deserve lower priority, but difficulties may arise later if this topic is neglected.
Section 11.2 introduces the key idea of partial differentiation. We try to make it as easy as possible
by borrowing from earlier work for functions of one variable. In addition to mastering the technique for
partially differentiating simple functions, the student should be told how important it is to understand the
basic definitions of the partial derivatives given in (2) and (3), as well as the approximations in (4) to (6). In
particular, economists often interpret the partial derivative f1 (x, y) as approximately equal to the change in
f (x, y) that results from increasing x by one unit while holding y constant. (Of course, it is better in principle
to interpret f1 (x, y)h as approximately equal to f (x + h) f (x, y), for h small.)
After the mechanics of partial derivatives, Section 11.3 turns to geometric representations of functions.
Graphs are useful for functions of one variable. They are less useful for functions of two variables (or more)
because of the difficulty of drawing in three (or more) dimensions. Nevertheless, for many functions of two
variables, level curves are a useful device that often arises in economic contextsespecially as indifference
curves, isoquants, etc. The relationship between partial derivatives, graphs, and level curves is important.
Section 11.4 explains how an equation in three variables can be represented, in principle, by a surface in
three dimensional space. An easy example is the budget plane if there are three commodities. The distance
formula in three dimensional space is derived and the equation for a sphere in three dimensions follows easily.
Following all the work in Sections 11.111.3 with functions of two variables, we hope that the extension
to n variables in Sections 11.5 and 11.6 will seem routine. The general CobbDouglas function and its log-
linear transformation are important in economics. Example 11.5.2 concerns the arithmetic, geometric, and
harmonic means. Our discussion of continuity and the Euclidean n-dimensional space Rn (for n > 3) is brief.
These concepts will tax the powers of most undergraduate students if taken beyond this very intuitive level.
The only really new concept in Section 11.6 is the Hessian matrix. Youngs Theorem 11.6.1 is stated
carefully, as is the formal definition of the partial derivative, and the concept of a C k function. The latter is
used occasionally in later chapters.
Section 11.7 brings in economic applications, starting with marginal products. The notation FK , FL , FT
is preferred for these (when K is capital, L is labour, and T is land) because it makes clear what factor is being
varied in each case. Brief mention is made of complementary factors and, more importantly, of diminishing
marginal products. These are applications of second partial derivatives, of course.
This chapter ends with a brief discussion of partial elasticities in Section 11.8. These, of course, arise in
many economic contexts.

11.1
2. f (0, 1) = 0, f (1, 2) = 4, f (104 , 102 ) = 1, f (a, a) = a 3 , f (a + h, b) = (a + h)b2 = ab2 + hb2 ,
and f (a, b + k) f (a, b) = 2abk + ak 2 .

44 CHAPTER 11 FUNCTIONS OF MANY VARIABLES
4. (a) f (1, 2) = 1, f (a, a) = 4a 2 , f (a + h, b) f (a, b) = 2(a + b)h + h2

(b) f (tx, ty) = (tx)2 + 2(tx)(ty) + (ty)2 = t 2 (x 2 + 2xy + y 2 ) = t 2 f (x, y) for all t, including t = 2.
6. (a) y 6 = x 2 (b) x 2 + y 2 2 (c) 1 x 2 + y 2 4. The domains in (b) and (c) are the shaded sets
shown in Figs. M11.1.6b and c.
y y
2
x2 + y2 2
y
x
x 1 2 x
Figure M11.1.6b Figure M11.1.6c Figure M11.3.6
11.2
2. (a) z/x = 2x, z/y = 6y (b) z/x = y, z/y = x
(c) z/x = 20x 3 y 2 2y 5 , z/y = 10x 4 y 10xy 4 (d) z/x = z/y = ex+y
(e) z/x = yexy , z/y = xexy (f) z/x = ex /y, z/y = ex /y 2
(g) z/x = z/y = 1/(x + y) (h) z/x = 1/x, z/y = 1/y
4. (a) zx = 3, zy = 4, and zxx = z = z = z = 0
xy yx yy
(b) zx = 3x 2 y 2 , zy = 2x 3 y, zxx
= 6xy 2 , z = 2x 3 , and z = z = 6x 2 y
yy xy yx
(c) zx = 5x 4 6xy, zy = 3x 2 + 6y 5 , zxx = 20x 3 6y, z = 30y 4 , and z = z = 6x
yy xy yx
(d) zx = 1/y, zy = x/y 2 , zxx = 0, z = 2x/y 3 , and z = z = 1/y 2
yy xy yx
(e) zx = 2y(x + y)2 , zy = 2x(x + y)2 , zxx = 4y(x + y)3 , z = 4x(x + y)3 , and z =
yy xy
= 2(x y)(x + y)3 (f) z = x(x 2 + y 2 )1/2 , z = y(x 2 + y 2 )1/2 , z = y 2 (x 2 + y 2 )3/2 ,
zyx x y xx
= x 2 (x 2 + y 2 )3/2 , and z = z = xy(x 2 + y 2 )3/2
zyy xy yx
6. (a) FS = 2.260.44S 0.56 E 0.48 = 0.9944S 0.56 E 0.48 , FE = 2.260.48S 0.44 E 0.52 = 1.0848S 0.44 E 0.52
(b) SFS + EFE = S 2.26 0.44S 0.56 E 0.48 + E 2.26 0.48S 0.44 E 0.52 = 0.44 F + 0.48 F = 0.92 F ,
so k = 0.92.
8. Here z/x = x/(x 2 + y 2 ), z/y = y/(x 2 + y 2 ), 2 z/x 2 = (y 2 x 2 )/(x 2 + y 2 )2 , and 2 z/y 2 =
(x 2 y 2 )/(x 2 + y 2 )2 . Thus, 2 z/x 2 + 2 z/y 2 = 0.
11.3
2. (a) A straight line through (0, 2, 3) parallel to the x-axis.
(b) A plane parallel to the z-axis whose intersection with the xy-plane is the line y = x.
2 2
4. f (x, y) = ex y + (x 2 y 2 )2 = ec + c2 when x 2 y 2 = c, so the last equation represents a level curve
of f at height ec + c2 .

CHAPTER 11 FUNCTIONS OF MANY VARIABLES 45
6. Generally, the graph of g(x, y) = f (x) in 3-space consists of a surface traced out by moving the graph
of z = f (x) parallel to the y-axis in both directions. The graph of g(x, y) = x is the plane through the
y-axis at a 45 angle with the xy-plane. The graph of g(x, y) = x 3 is shown in Fig. M11.3.6. (Only a
portion of the the unbounded graph is indicated, of course.)
8. (a) f (2, 3) = 8. f (x, 3) = 8 has the solutions x = 2 and x = 5 (b) As y varies with x = 2 fixed,
the minimum of f (2, y) is 8 when y = 3. (c) A: f1 (x, y) > 0, f2 (x, y) > 0. B: f1 (x, y) < 0,
f2 (x, y) < 0. C: f1 (x, y) = 0, f2 (x, y) = 0. At A, f1 2/1 = 2 and f2 2/0.6 = 10/3.
R1 R2
10. F (1, 0) = F (0, 0) + 0 F1 (x, 0) dx 2, F (2, 0) = F (1, 0) + 1 F1 (x, 0) dx F (1, 0) + 2,
R1 R1
F (0, 1) = F (0, 0) + 0 F2 (0, y) dy 1, F (1, 1) = F (0, 1) + 0 F1 (x, 1) dx F (0, 1) + 2,
R1
F (1, 1) = F (1, 0) + 0 F2 (1, y) dy F (1, 0) + 1.
11.4
p
2. d = (4 (1))2 + (2 2)2 + (0 3)2 = 25 + 16 + 9 = 50 = 5 2.
4. (x 2)2 + (y 1)2 + (z 1)2 = 25
6. (x 4)2 + (y 4)2 + (z 21 )2 measures the square of the distance from the point (4, 4, 21 ) to the point
(x, y, z) on the paraboloid.
11.5
2. (a) F (K + 1, L, T ) F (K, L, T ) is the increase in output from increasing capital input by one unit.
(b) F (K, L, T ) = AK a Lb T c , where A, a, b, and c are constants, A > 0.
(c) F (tK, tL, tT ) = t a+b+c F (K, L, T )
4. You drive (5/60) 0 + (10/60) 30 + (20/60) 60 + (15/60) 80 = 45 kilometres in 5 + 10 + 20 + 15 = 50
minutes, so the average speed is 45 60/50 = 54 kph.
6. (a) Each machine would produce 60 units per day, so each unit produced would require 480/60 = 8
P P
minutes. (b) Total output is ni=1 (T /ti ) = T ni=1 (1/ti ). If all
P n machines were equally efficient,
Pn n
the time needed for each unit would be nT T i=1 (1/ti ) = n i=1 (1/ti ), the harmonic mean of
t1 , . . . , tn .
11.6
2. (a) f1 = 2x, f2 = 3y 2 , and f3 = 4z3 (b) f1 = 10x, f2 = 9y 2 , and f3 = 12z3
(c) f1 = yz, f2 = xz, and f3 = xy (d) f1 = 4x 3 /yz, f2 = x 4 /y 2 z, and f3 = x 4 /yz2
(e) f1 = 12x(x 2 + y 3 + z4 )5 , f2 = 18y 2 (x 2 + y 3 + z4 )5 , and f3 = 24z3 (x 2 + y 3 + z4 )5
(f) f1 = yzexyz , f2 = xzexyz , and f3 = xyexyz
4. First-order partials: w1 = 3yz + 2xy z3 , w2 = 3xz + x 2 , w3 = 3xy 3xz2 . Second-order partials:

w11 = 2y, w12 = w21 = 3z + 2x, w13 = w31 = 3y 3z2 , w22 = 0, w23 = w32 = 3x, w33 = 6xz.

2a 0 0 a(a 1)g/x 2 abg/xy acg/xz

6. (a) 0 2b 0 (b) abg/xy b(b 1)g/y 2 bcg/yz , in concise form.
0 0 2c acg/xz bcg/yz c(c 1)g/z2
z 1 z z
8. fx = y z x y , fy = zy z1 (ln x)x y , fz = y z (ln x)(ln y)x y

46 CHAPTER 11 FUNCTIONS OF MANY VARIABLES
11.7
2. Simplified answers: (a) KYK + LYL = aY (b) KYK + LYL = (a + b)Y (c) KYK + LYL = Y
4. D/p and E/q are normally negative, because the demand for a commodity goes down when the
price of that commodity increases. If the commodities are substitutes, this means that demand increases
when the price of the other good increases. So the usual signs are D/q > 0 and E/p > 0.
6. KYK + LYL = mY . (YK = (m/)a()Aet K 1 [aK + (1 a)L ](m/)1 and
YL = (m/)(1 a)()Aet L1 [aK + (1 a)L ](m/)1 , so
KYK + LYL = (m/)()[aK + (1 a)L ]m/ = mY.
11.8
2. Let z = ug with u = ax1d + bx2d + cx3d . Then El1 z = Elu ug El1 u = g(x1 /u)adx1d1 = adgx1d /u.
Similarly, El2 z = bdgx2d /u and El3 z = cdgx3d /u, so El1 z+El2 z+El3 z = dg(ax1d +bx2d +cx3d )/u = dg.
(This result follows easily from the fact that the function is homogeneous of degree dg (see Problem
12.7.2(b) and the elasticity form (12.7.3) of the Euler equation.)
pD(p, m) mDm (p, m) D(p, m) pD(p, m)
4. =p 2
= [Elm D(p, m)1] > 0 iff Elm D(p, m) > 1,
m m m m2
so pD/m increases with m if Elm D > 1. (Using the formulas in Problem 7.7.10, the result also follows
from the fact that Elm (pD(p, m)/m) = Elm D(p, m) Elm m = Elm D(p, m) 1.)

2. f (1, 2) = 10, f (2a, 2a) = 4a 2 , f (a, b + k) f (a, b) = 6bk 3k 2 , f (tx, ty) t 2 f (x, y) = 0.
4. (a) Y /K 0.083K 0.356 S 0.562 and Y /S 0.035K 1.356 S 0.438 .
(b) The catch becomes 21.356+0.562 = 21.918 3.779 times higher.

6. FK = aF /K, FL = bF /L, and FM
= cF /M, so KFK + LFL + MFM = (a + b + c)F .
8. (a) g(2, 1, 1) = 2, g(3, 4, 2) = 352, and g(1, 1, a + h) g(1, 1, a) = 2ah + h2 h.
(b) g/x = 4x 4y 4, g/y = 4x + 20y 28, g/z = 2z 1. The Hessian matrix of the
4 4 0
second-order partials is: 4 20 0 .
0 0 2

10. (a) z/x = 10xy 4 (x 2 y 4 + 2)4 (b) K(F /K) = 2 K( K + L)(1/2 K) = K + L
(c) KFK + LFL = K(1/a)aK a1 (K a + La )1/a1 + L(1/a)aLa1 (K a + La )1/a1 =
(K a + La )(K a + La )1/a1 = F (d) g/t = 3/w + 2wt, so 2 g/wt = 3/w 2 + 2t
(e) g3 = t3 (t12 + t22 + t32 )1/2 (f) f1 = 4xyz + 2xz2 , f13

= 4xy + 4xz
12. If x y = c, then F (x, y) = ln(x y)2 + e2(xy) = ln c2 + e2c , a constant.
2x 2 2y 2
14. (a) Elx z = 3, Ely z = 4 (b) Elx z = , El y z =
(x 2 + y 2 ) ln(x 2 + y 2 ) (x 2 + y 2 ) ln(x 2 + y 2 )
(c) Elx z = Elx (e e ) = Elx e = x, Ely z = y (d) Elx z = x /(x + y ), Ely z = y 2 /(x 2 + y 2 )
x y x 2 2 2

CHAPTER 12 TOOLS FOR COMPARATIVE STATICS 47
Chapter 12 Tools for Comparative Statics
Chapter 12 gathers together a number of tools and techniques from the calculus of many variables that are much
used by economists. It also introduces important concepts such as elasticity of substitution, homogeneous
and homothetic functions, linear approximations, and differentials.
First, however, Section 12.1 deals with the chain rule for differentiating functions of many variables,
when these variables all depend on a single variable such as time. Of course, it begins with the two variable
case, and Examples 1 and 2 where the validity of the rule is checked by direct differentiation. Examples 3
and 4 are just two of the many obvious economic applications that spring to mind immediately. Finally, the
section closes with an attempt (in small print) to justify the chain rule.
Next, Section 12.2 considers chain rules for functions of many variables that are themselves functions of
many variables. It is necessary to learn how to keep notation straight, and under control!
Section 12.3 deals with implicit differentiation, generalizing the results from Section 7.1. It introduces
formula (1), giving the slope of a level curve for the two-variable case. This formula is then applied mech-
anically in Examples 1 to 3, and to the economic Example 4 on the incidence of taxation in a simple model
of supply and demand. Next comes a formula for the second derivative. For those students who have already
learned about determinants, the second derivative can also be expressed as a bordered Hessian (though we do
not use this term).
Section 12.4 develops more general results on implicit differentiation. Two interesting examples are a
standard profit maximization model, and then a simple search model.
Section 12.5 introduces the elasticity of substitution, with standard applications to the CobbDouglas
and CES (constant elasticity of substitution) functions.
Homogeneous functions receive much attention in economics. Section 12.6 introduces them for the case
of two variables, along with Eulers theorem and other properties. Some geometric properties that may be
helpful are discussed.
Section 12.7 considers homogeneous functions of many variables. After proving that a function is homo-
geneous if and only if it satisfies the Euler equation, we show that the other relevant properties of functions of
two variables generalize appropriately. Economic applications are then considered. In particular, the relation-
ship between homogeneity and returns to scale is considered, as well as homogeneity of demand functions.
The section concludes with an introduction to homothetic functions, and how they generalize homogeneous
functions.
Next, Section 12.8 considers linear approximations and tangent planes.
Section 12.9 turns to differentials of functions of several variables. Rules for differentials are presented,
as well as an explanation of how they are natural counterparts of rules for derivatives. The invariance of the
differential, however, deserves more careful consideration, though most economists have neglected it.
Section 12.10, on systems of equations, begins with an elementary discussion of degrees of freedom,
and the associated counting rule. Some examples illustrate difficulties with the counting rule for nonlinear
functions.1
1
We do not discuss the concept of functional dependence. Most texts in mathematics for economists state erro-
neously that functional dependence of n functions is equivalent to the vanishing of the associated Jacobian determinant.
For a correct account, see J. E. Marsden and M. J. Hoffman: Elementary Classical Analysis, 2. ed. W. H. Freeman and
Co. (1993).

48 CHAPTER 12 TOOLS FOR COMPARATIVE STATICS
The main part of Section 12.11 consists of Examples 13 showing how to find partial derivatives from
differentials. Whereas Examples 1 and 2 are purely mechanical, with no economic distractions, Example 3 is
based on a rather standard macroeconomic model.
The general case of a system of structural equations is briefly discussed, as is the distinction between
endogenous and exogenous variables. Finally, the concept of reduced form is defined in this setting.

12.1

dz y x 1 ln t t + 1 ln(t + 1)
2. (a) = ln y + 1+ + ln x = ln(ln t) + + +
dt x y t t +1 t ln t t
dz 1 1
(b) = Aaeat + Bbebt = a + b
dt x y

4. dY /dt = 10L 21 K 1/2 0.2 + 10K 21 L1/2 0.5e0.1t = 35 7 5/100 when t = 0 and so
K = L = 5.
6. Let U (x) = u(x, h(x)) = ln[x + (ax 4 + b)/3 ] 3 ln(ax 4 + b).
x 1 (3b ax 4 )
Then U (x) = 4 /3 4
. So U (x ) = 0 at x = 4 3b/a, whereas U (x) > 0 for
3[x + (ax + b) ](ax + b)
x < x and U (x) < 0 for x > x . Hence x maximizes U .
12.2
2. (a) z/t = y 2 + 2xy2ts = 5t 4 s 2 + 4t 3 s 4 , z/s = y 2 2s + 2xyt 2 = 2t 5 s + 4t 4 s 3
z 2(1 s)ets+t+s z 2(1 t)ets+t+s
(b) = and =
t (et+s + ets )2 s (et+s + ets )2
4. z/t1 = F (x)f1 (t1 , t2 ), z/t2 = F (x)f2 (t1 , t2 )
C Q1 Q2 Q1
= 1 A(a + 2cAp11 p2 1 )p11 1 p2 1 + 2 bBp12 1 p2 2

6. =a +b + 2cQ1
p1 p1 p1 p1
C 1 1 1 1 1 1
= 1 A(a + 2cAp1 p2 )p1 p2 2 bBp12 p2 2
p2
u f x f y f z f w u
8. (a) = + + + (b) = yzw + xzw + xyws + xyz(1/s) = 28
r x r y r z r w r r
12.3
2. See the answers to Problem 7.1.1. (For (a): Put F (x, y) = x 2 y. Then F1 = 2xy, F2 = x 2 , F11
=
2
2y, F12 = 2x, F 22 = 0, so y = F1 /F2 = 2xy/x = 2y/x. Moreover, according to (3),
y = (1/(F2 )3 ) F11
(F2 )2 2F12
F1 F2 + F22 (F1 )2 = (1/x 6 )[2yx 4 2(2x)(2xy)x 2 ] = 6y/x 2 .)
6x 3y 2
4. h (x) = = 1 at (x, y) = (1, 1).
6xy + 3y 2 + 6y
6. Differentiating the equation w.r.t. x gives (i) 1 azx = f (y bz)(bzx ). Differentiating w.r.t. y
gives (ii) azy = f (y bz)(1 bzy ). If bzx 6= 0, solving (i) for f and inserting it into (ii) yields
azx + bzy = 1. If bzx = 0, then (i) implies azx = 1. But then zx 6 = 0, so b = 0 and then again
azx + bzy = 1.

12.4
2. Differentiating partially w.r.t. x yields () 3x 2 + 3z2 zx 3zx = 0, so zx = x 2 /(1 z2 ). By symmetry,
zy = y 2 /(1 z2 ). To find zxy , differentiate () w.r.t. y to obtain 6zz z + 3z2 z 3z = 0, so
y x xy xy
= 2zx 2 y 2 /(1 z2 )3 . (Alternatively, differentiate z = x 2 /(1 z2 ) w.r.t. y.)
zxy x
yx y1 + zx ln z x y ln x + zy z1
4. zx = and zy

=
y z ln y + xzx1 y z ln y + xzx1
Fx (1, 3)
6. (a) F (1, 3) = 4. The equation for the tangent is y 3 = (x 1) with Fx (1, 3) = 10 and
Fy (1, 3)
y y y y
Fy (1, 3) = 5, so y = 2x + 5. (b) = , =
K K(1 + 2c ln y) L L(1 + 2c ln y)
12.5
2. (a) Ryx = (x/y)a1 = (y/x)1a (b) yx = ElRyx (y/x) = ElRyx (Ryx )1/(1a) = 1/(1 a)
12.6
2. x(tp, tr) = A(tp)1.5 (tr)2.08 = At 1.5 p1.5 t 2.08 r 2.08 = t 1.5 t 2.08 Ap1.5 r 2.08 = t 0.58 x(p, r), so the
function is homogeneous of degree 0.58 . (Alternatively, use the result in Example 11.1.4.)
4. Checking definition (1) shows that f is homogeneous of degree 0. Then using the partial derivatives
found in Example 11.2.1(b), we confirm that xf1 (x, y) + yf2 (x, y) = 0.
6. Definition (1) requires that for some number k one has, t 3 x 3 + t 2 xy = t k (x 3 + xy) for all t > 0 and
all (x, y). In particular, for x = y = 1, we must have t 3 + t 2 = 2t k . For t = 2, we get 12 = 2 2k , or
2k = 6. For t = 4, we get 80 = 2 4k , or 4k = 40. But 2k = 6 implies 4k = 36. So the two values of k
must actually be different, implying that f is not homogeneous of any degree.
8. Let C and D denote the the numerator and the denominator in the expression for yx in Problem 12.5.3.
Because F is homogeneous of degree one, Eulers theorem implies that C = F1 F2 F , and () implies

that xF 11 = yF12 and yF22 = xF21 = xF12 . Hence, D = xy (F2 )2 F11

2F1 F2 F12

+(F1 )2 F22

=

F12 y 2 (F2 )2 + 2xyF1 F2 + x 2 (F1 )2 = F12
(xF1 + yF2 )2 = F12
2
F , using Eulers theorem again.
Hence xy = C/D = (F1 F2 F )/(F12 2
F ) = F1 F2 /F F12

.
12.7
2. (a) The function is homogeneous of degree 1 because

(tx1 tx2 tx3 )2 1 1 1
F (tx1 , tx2 , tx3 ) = + +
(tx1 )4 + (tx2 )4 + (tx3 )4 tx1 tx2 tx3
6 2
t (x1 x2 x3 ) 1 1 1 1
= 4 4 4 4 t
+ + = tF (x1 , x2 , x3 )
t (x1 + x2 + x3 ) x1 x2 x3
g g
(b) G(tx1 , tx2 , tx3 ) = a(tx1 )d + b(tx2 )d + c(tx3 )d = t d (ax1d + bx2d + cx3d )
= t dg G(x1 , x2 , x3 ), so G is homogeneous of degree dg.

P P P
4. vi = ui a/(x1 + + xn ), so ni=1 xi vi = ni=1 xi ui ni=1 axi /(x1 + + xn ) =
Pn
a [a/(x1 + + xn )] i=1 xi = a a = 0. By Eulers theorem, v is homogeneous of degree 0.
6. Let 1 = ln C(tw, y) ln C(w, t). It suffices to prove that 1 = ln t, because then C(tw, y)/C(w, y) =
ln et = t. We find that
n
X n n
1 X X
1= ai [ln(twi ) ln wi ] + aij [ln(twi ) ln(twj ) ln wi ln wj ] + ln y bi [ln(twi ) ln wi ]
2
i=1 i,j =1 i=1
Because ln(twi ) ln wi = ln t + ln wi ln wi = ln t and also ln(twi ) ln(twj ) ln wi ln wj = (ln t)2 +

ln t ln wi + ln t ln wj , this can be reduced to
n
X n
X n
X n
X n
X n
X n
X
1 2 1 1
1 = ln t ai + (ln t) aij + (ln t) ln wi aij + (ln t) ln wj aij + ln y ln t bi
2 2 2
i=1 i,j =1 j =1 i=1 i=1 j =1 i=1
Finally, this simplifies to 1 = ln t + 0 + 0 + 0 + 0 = ln t because of the specified restrictions on the

parameters ai , aij , and bi .
12.8
x x0 y y0
2. f (x, y) Ax0a y0b + aAx0a1 y0b (x x0 ) + bAx0a y0b1 (y y0 ) = Ax0a y0b 1 + a +b
x0 y0
4. f (0.98, 1.01) 5 6(0.98 1) + 9(1.01 + 1) = 4.97. The exact value is 4.970614, so the
error is 0.000614.
6. v(1.01, 0.02) v(1, 0) + v1 (1, 0) 0.01 + v2 (1, 0) 0.02 = 1 1/150
8. g(0) = f (x0 ), g(1) = f (x). Using formula (12.2.3), it follows that
g (t) = f1 (x0 + t (x x0 ))(x1 x10 ) + + fn (x0 + t (x x0 ))(xn xn0 ). Putting t = 0 gives
g (0) = f1 (x0 )(x1 x10 ) + + fn (x0 )(xn xn0 ), and the conclusion follows.
12.9
2 2
2. (a) dz = 3x 2 dx + 3y 2 dy (b) dz = ey (dx + 2xy dy) (c) dz = (x dx y dy)
x2 y2
4. Let T (x, y, z) = [x 2 + y 2 + z2 ]1/2 = u1/2 , where u = x 2 + y 2 + z2 . Then dT = 21 u1/2 du =
1 1/2
2u (2xdx + 2ydy + 2zdz). For x = 2, y = 3, and z = 6, we have u = 49, T = 7 and dT =
1
7 (xdx + ydy + zdz) = 17 (2dx + 3dy + 6dz). Thus, T (2 + 0.01, 3 0.01, 6 + 0.02) T (2, 3, 6) +
1
7 [2 0.01 + 3(0.01)
+ 6 0.02] = 7 + 17 0.11 7.015714. (A calculator or computer gives a better
approximation: 49.2206 7.015739.)
6. (a) dX = AN 1 et dN + AN et dt
(b) dX1 = BEXE1 N 1E dX + B(1 E)X E N E dN
8. d(ln z) = a1 d(ln x1 ) + + an d(ln xn ), so dz/z = a1 dx1 /x1 + a2 dx2 /x2 + + an dxn /xn .
12.10
2. There are 6 variables Y , C, I , G, T , and r, and 3 equations. So there are 6 3 = 3 degrees of freedom.

12.11
2. (a) Differentiating yields: u3 dx + x3u2 du + dv = 2y dy and 3v du + 3u dv dx = 0.
Solving for du and dv yields, with D = 9xu3 3v,
du = D 1 (3u4 1) dx + D 1 6yu dy and dv = D 1 (3xu2 + 3u3 v) dx + D 1 (6yv) dy
(b) ux = D 1 (3u4 1), vx = D 1 (3xu2 + 3u3 v) (c) ux = 283/81 and vx = 64/27
4. With a fixed, I (r) dr = S (Y ) dY and a dY + L (r) dr = dM. Solving for dY and dr in terms of dM
gives
Y I (r) r S (Y )
= and =
M aI (r) + L (r)S (Y ) M aI (r) + L (r)S (Y )

dY = dC + dI + dG,
6. (a) Differentiation yields the equations:

= FY dY + FT dT + Fr dr , and
dC
dI = fY dY + fr dr. Hence, dY = FT dT + dG + (Fr + fr ) dr /(1 FY fY ).
(b) Y /T = FT /(1 FY fY ) < 0, so Y decreases as T increases. But if dT = dG with dr = 0,
then dY = 1 + FT dT /(1 FY fY ), which is positive provided that FT > 1.
8. (a) There are 3 variables and 2 equations, so there is (in general) one degree of freedom.
(b) Differentiation gives 0 = lP dy + L (r)dr and Sy dy + Sr dr + Sg dg = Iy dy + Ir dr. We find
dy L (r)Sg dr lP Sg
= , =
dg L (r)(Sy Iy ) lP (Sr Ir ) dg L (r)(Sy Iy ) lP (Sr Ir )

10. Differentiating while putting dp2 = dm = 0, gives: (i) U11 dx1 + U12 dx2 = p1 d + dp1 ;

(ii) U21 dx1 + U22 dx2 = p2 d; (iii) p1 dx1 + dp1 x1 + p2 dx2 = 0. Solving for dx1 , we obtain

x1 p 2 + x1 (p2 U12 p1 U22 )
= 2 2 2
p1 p1 U22 2p1 p2 U12 + p2 U11
2. z/t = G1 (u, v)1 (t, s) and z/s = G1 (u, v)2 (t, s) + G2 (u, v) (s)

4. dX/dN = g(u) + g (u) (N ) u , where u = (N )/N,
2
d 2 X/dN 2 = (1/N )g (u) (N ) u + g (u) (N )
6. Differentiating each side w.r.t. x while holding y constant gives 3x 2 ln x + x 2 = (6z2 ln z + 2z2 )z1 .
When x = y = z = e, this gives z1 = 1/2. Differentiating a second time gives 6x ln x + 5x =
(12z ln z + 10z)(z1 )2 + (6z2 ln z + 2z2 )z11

. When x = y = z = e and z1 = 1/2, this gives z11

= 11/16e.
8. (a) MRS = Ryx = 2y/3x (b) MRS = Ryx = y/(x + 1) (c) MRS = Ryx = (y/x)3
10. Since y/x = (Ryx )1/3 , yx = ElRyx (y/x) = 1/3.
12. (a) 1 (b) k (c) 0

14. Differentiate f (tx1 , . . . , txn ) = g(t)f (x1 , . . . , xn ) w.r.t. t and put t = 1, as in the proof of Eulers
P
theorem (Theorem 12.7.1). This yields ni=1 xi fi (x1 , . . . , xn ) = g (1)f (x1 , . . . , xn ). Thus, by Eulers
theorem, f must be homogeneous of degree g (1). In fact, g(t) = t k where k = g (1).
16. (a) Differentiating, then gathering all terms in dp and dL on the left-hand side, yields
(i) F (L) dp + pF (L) dL = dw (ii) F (L) dp + (pF (L) w) dL = L dw + dB
Since we know that pF (L) = w, (ii) implies that dp = (Ldw + dB)/F (L). Substituting this into (i)
and solving for dL, we obtain dL = [(F (L) LF (L))dw F (L)dB]/pF (L)F (L). It follows that
p L p 1 L F (L) LF (L) L F (L)

= , = , = , =
w F (L) B F (L) w pF (L)F (L) B pF (L)F (L)
(b) We know that p > 0, F (L) > 0, and F (L) < 0. Also, F (L) = (wL + B)/p > 0. Hence, it is
clear that p/w > 0, p/B > 0, and L/B > 0.
To find the sign of L/w, we need the sign of F (L) LF (L). From the equations in the model, we
get F (L) = w/p and F (L) = (wL + B)/p, so F (L) LF (L) = B/p > 0. Therefore L/w < 0.

CHAPTER 13 MULTIVARIABLE OPTIMIZATION 53
Chapter 13 Multivariable Optimization

Chapter 13 is devoted to multivariable optimization. The first five sections treat the case of only two variables.
Section 13.1 begins with a graphical explanation of the first-order conditions stated in Theorem 13.1.1.
These are first applied to a simple problem without any particular economic interpretation, and then to simple
profit maximization in Examples 24. We believe the economic interpretations of conditions () in Example
3 are worth the students serious attention. Example 5 looks at a monopolist who sells his product in two
separate markets. Again, the economic interpretations of the first-order conditions are important.
In identifying global maxima or minima, the theory of concave or convex functions provides sufficient
conditions that are enormously important in economic analysis. Theorem 13.2.1 gives the precise result for
functions of two variables. It is applied in Example 5 to a cost minimization problem where one of three
variables has to be eliminated.
Stationary points can be local or global maxima, local or global minima, or saddle points. Section 13.3
discusses ways of distinguishing these different possibilities using second-order conditions that are formally
stated in Theorem 13.3.1. (The proof is indicated at the end of the section.) Several examples and many
problems are supplied.
Probably the most important application of multivariable calculus to economic theory is as a tool for
solving optimization problems. So interesting economic examples abound. Of these, Section 13.4 presents a
small selection, including some simple models of discriminating monopolists and monopsonists.
Section 13.5 begins with a brief discussion of some essential topological concepts in the planenamely,
interior points, open sets, boundary points, closed sets, bounded sets, and compact sets. It is pointed out that
closed sets are typically determined by weak inequalities applied to continuous functions, and open sets by
strict inequalities applied to such functions. The extreme-value theorem claims that a continuous function
defined on a closed and bounded set in the plane has both a maximum and minimum. This result was stated
for functions of one variable in Theorem 8.4.1. The corresponding statement for functions of two variables is
given in Theorem 13.5.1. Frame (1) outlines a typical procedure for finding maximum and minimum values.
It pays careful attention to the possibility that these may occur at boundary points, by insisting on finding
maxima and minima on the boundary as well as in the interior. Examples 1 and 2 illustrate the procedure.
Section 13.6 discusses the extension of results in earlier sections to functions of n variables. It begins
with formal definitions of maxima and minima that must be clearly understood. Topology in Rn is briefly
mentioned; the main difference from R2 is that the word circle (or disk) must be replaced by ball
in any definition. Only the most essential concepts receive attention, however. Theorem 13.6.3 states the
invariance of any maximum point to (strictly) increasing transformations of the maximand. This result often
gets overlooked despite its enormous importance and usefulness in economics. Indeed, the different role
of strictly increasing and merely increasing transformations, illustrated in Problem 2, should also be noted
carefully. Finally, Section 13.7 discusses the envelope theorem which appears so often in economics. Problem
4 is typical of the way in which the envelope theorem can give interesting economic conclusions which are
far from evident to the uninitiated, and impossible even to understand without some facility in multivariable
calculus.

54 CHAPTER 13 MULTIVARIABLE OPTIMIZATION

13.1
2. (a) (x, y) = (3, 4). (f1 (x, y) = 2x 6 and f2 (x, y) = 2y + 8, so the only stationary point is
(x, y) = (3, 4).) (b) f (x, y) = x 2 6x+32 +y 2 +8y+42 +3532 42 = (x3)2 +(y+4)2 +10 10
for all (x, y), whereas f (3, 4) = 10, so (3, 4) minimizes f .
4. (a) P (10, 8) = P (12, 10) = 98 (b) (x, y) = (11, 9) maximizes P , and P (11, 9) = 100.
13.2
2. (a) Profit: (x, y) = 24x + 12y C(x, y) = 2x 2 4y 2 + 4xy + 64x + 32y 514. Maximum at

x = 40, y = 24. (Then (40, 24) = 1150.) Since 11 = 4 0, 22 = 8 0, and 11 22
2
(12 ) = 16 0, we have found the maximum. (b) x = 34, y = 20. (With y = 54 x, profits are
= 2x 2 4(54 x)2 + 4x(54 x) + 64x + 32(54 x) 514 = 10x 2 + 680x 10450, which
has a maximum at x = 34. Then y = 54 34 = 20. The maximum value is 1110.)
4. (a) (x, y) = px + qy C(x, y) = (25 x)x + (24 2y)y (3x 2 + 3xy + y 2 ) = 4x 2 3xy
3y 2 + 25x + 24y. (b) 1 = 8x 3y + 25 = 0 and 2 = 3x 6y + 24 = 0 when (x, y) = (2, 3).
2
Moreover, then 11 = 8 0, 22 = 6 0, and 11 22 (12 ) = (8)(6) (3)2 = 39 0.
So (x, y) = (2, 3) maximizes profits.
6. (a) The first-order conditions are: 1 = p2x = 0, 2 = q 2y = 0. Thus the only stationary point is
2
x = p/2, y = q/2. Moreover, 11 = 2 0, 22 = 2 0, and 11 22 (12 ) = 4 0,

so (x , y ) = (p/2, q/2) maximizes profits.
(b) (p, q) = px + qy (x )2 (y )2 = p 2 /4 + q 2 /2. Hence (p, q)/p = p/2 = x :
Increasing the price p by one unit increases the optimal profit by x , the optimal production level of the
first good. (p, q)/q = y has a similar interpretation.

8. The second-order partial derivatives are f11 = a(a 1)Ax a2 y b, f12 = f21
= abAx a1 y b1 , and
2
f22 = b(b1)Ax a y b2 . Thus, f11 f22 (f12 ) = abA2 x 2a2 y 2b2 1(a +b) . Suppose that a +b 1.
2
Then a 1 and b 1 as well. If x > 0 and y > 0, then f11 0 and f22 0, and f11 f22 (f12 ) 0.
We conclude from Note 13.2.2 that f is concave for x > 0, y > 0 when a + b 1.
13.3
2. (a) f1 = 2x + 2y 2 , f2 = 4xy + 4y, f11

= 2, f12
= 4y, f22 = 4x + 4
(b) f2 = 0 4y(x + 1) = 0 x = 1 or y = 0. If x = 1, then f1 = 0 for y = 1. If

y = 0, then f1 = 0 for x = 0. Thus we get the three stationary points classified in the table:
(x, y) A B C AC B 2 Type of stationary point:
(0, 0) 2 0 4 8 Local minimum point
(1, 1) 2 4 0 16 Saddle point
(1, 1) 2 4 0 16 Saddle point

4. (a) f1 (x, y) = 24x 2 12y = 12(2x 2 y), f2 (x, y) = 12x + 3y 2 = 3(y 2 4x). The stationary
points are thus the solutions of the system (i) 2x 2 y = 0, (ii) 4x + y 2 = 0. From (i) we get y = 2x 2
which inserted into (ii) yields 4x(x 3 1) = 0. The only solutions of this equation are x = 0 and x = 1.

The stationary points are thus (0, 0) and (1, 2). Moreover, f11 (x, y) = 48x, f12 (x, y) = 12, and

f22 (x, y) = 6y. The stationary points are classified in the table:
(0, 0) 0 12 0 144 Saddle point
(b) Since g(x, y) is a square, g(x, y) 0 for all x and y. On the other hand, g(0, 0) = 0, so (0, 0) is
a global minimum point for g. (There are many global minimum points for g, namely all points (x, y)
satisfying 8x 3 12xy + y 3 = 0.)
6. In all three cases we easily find that (0, 0) is a stationary point where z = 0 and A = B = C = 0, so
AC B 2 = 0. In case (a), z 0 for all (x, y), so the origin is a maximum point. In case (b), z 0 for all
(x, y), so the origin is a minimum point. In (c), z takes positive and negative values at points arbitrarily
close to the origin, so it is a saddle point.
2xy x2
8. (a) f is defined for x = 0 and for y > 1/x 2 . (b) f1 (x, y) = and f
2 (x, y) = . Here
1 + x2y 1 + x2y
f1 = f2 = 0 at all points (0, b) with b R. (c) Because AC B 2 = 0 when (x, y) = (0, b), the
second-derivative test fails. But by studying the function directly, we see that (0, b) is a local maximum
point if b < 0; a saddle point if b = 0; and a local minimum point if b > 0. See Fig. M13.3.8.
z
z = ln(1 + x 2 y)
y
Figure M13.3.8
13.4
a + b c + d
2. (a) = bp 2 dq 2 + (a + b)p + (c + d)q (a + c), p =,q = .
2b 2d

The second-order conditions are obviously satisfied because 11 = 2b, 12 = 0, and 22 = 2d.
2 2
(b) The new profit function is = bp dp + (a + b)p + (c + d)p (a + c) and the price
a + c + (b + d)
which maximizes profits is p = .
2(b + d)

a c a2 c2 a+c
(c) If = 0, then p = ,q = , and (p , q ) = + . Moreover, p = with
2b 2d 4b 4d 2(b + d)
(a + c)2 (ad bc)2
(p) = , and (p , q ) (p) = 0. Note that the difference is 0 when
4(b + d) 4bd(b + d)
ad = bc, in which case p = q , so the firm wants to charge the same price in each market anyway.
4. (a) Let (x0 , y0 ) = (0, 11.29), (x1 , y1 ) = (1, 11.40), (x2 , y2 ) = (2, 11.49), and (x3 , y3 ) = (3, 11.61),
so that x0 corresponds to 1970, etc. (The numbers yt are approximate, as are most subsequent results.)
We find that x = 41 (0 + 1 + 2 + 3) = 1.5, y = 41 (11.29 + 11.40 + 11.49 + 11.61) = 11.45,
and xx = 41 [(0 1.5)2 + (1 1.5)2 + (2 1.5)2 + (3 1.5)2 ] = 1.25. Moreover, we find xy =
1
4 [(1.5)(11.2911.45)+(0.5)(11.4011.45)+(0.5)(11.4911.45)+(1.5)(11.6111.45)], which
is equal to 0.13125, and so a = xy /xx = 0.105 and b = y ax 11.45 0.105 1.5 = 11.29.
(b) With z0 = ln 274, z1 = ln 307, z2 = ln 436, and z3 = ln 524, we have (x0 , z0 ) = (0, 5.61),
(x1 , z1 ) = (1, 5.73), (x2 , z2 ) = (2, 6.08), and (x3 , z3 ) = (3, 6.26). As before, x = 1.5 and xx = 1.25.
Moreover, z = 41 (5.61 + 5.73 + 6.08 + 6.26) = 5.92 and
xz 41 [(1.5)(5.615.92)+(0.5)(5.735.92)+(0.5)(6.085.92)+(1.5)(6.265.92)] = 0.2875
Hence c = xz /xx = 0.23, d = z cx = 5.92 0.23 1.5 = 5.575.

(c) With ln (GNP) = 0.105x + 11.25, GNP = e11.25 e0.105x = 80017e0.105x . Likewise, FA = 256e0.23x .
The requirement that FA = 0.01 GNP implies that e0.23x0.105x = 80017/25600, and so 0.125x =
ln(80017/25600). Thus x = ln(80017/25600)/0.125 = 9.12. Since x = 0 corresponds to 1970, the
goal would have been reached in 1979.
13.5
2. (a) Stationary points occur where both (i) f1 (x, y) = 3x 2 9y = 0 and (ii) f2 (x, y) = 3y 2 9x = 0.
From (ii) we get x = 13 y 2 , which inserted into (i) yields y 4 = 27y, whose only solutions are y = 0 and
y = 3. So the only interior stationary point is (3, 3). Along the edges x = 0 and y = 0 of the boundary,
the function has no extreme point except at a corner. Along the edges x = 4 and y = 4, we find two
candidates for extreme points at (4, 2 3) and (2 3, 4). There are also the four corners (0, 0), (4, 0),
(0, 4), and (4, 4). Comparing the values of f at all these points, we find that f has maximum 91 at (0, 4)
and at (4, 0), and minimum 0 at (3, 3). (Extreme points exist by the extreme-value theorem.)
(b) At any stationary point, f1 = 2x 1 = 0 and f2 = 4y = 0, so f is stationary at the interior point
( 21 , 0) of the domain. Along the boundary x 2 + y 2 = 1, the behaviour of f is described by the function
g(x) = x 2 + 2(1 x 2 ) x = 2 x x 2 , with x in [1, 1]. We see that g (x) = 1 2x = 0 at
x = 21 , so the points ( 21 , 21 3) are optimality candidates. Moreover, g(1) = 2 and g(1) = 0,
so (1, 0) and (1, 0) are also optimality candidates.
Comparing the values
of f at all these points, we
conclude that f has maximum 9/4 at (1/2, 3/2) and at (1/2, 3/2). The minimum is 1/4 at
(1/2, 0).
(c) At any stationary point, f1 = 3x 2 2x = 0 and f2 = 2y = 0, so f is stationary at ( 23 , 0) (and at
the boundary point (0, 0)). Along the edge x = 0, y [1, 1], f (0, y) = 3 y 2 , which is largest at
y = 0 and smallest at y = 1. Along the edge x 2 + y 2 = 1, x [0, 1], one has f (x, y) = 2 + x 3 , which
is strictly increasing, smallest at x = 0, largest at x = 1. Comparing the values of f at all these points,
we conclude that the maximum is 3 at (0, 0) and at (1, 0). The minimum is 2 at (0, 1) and at (0, 1).

2 2 2 2
(d) f1 (x, y) = (2x 2 5x + 3)ex x (2y 1)e(y2) and f2 (x, y) = 2(2y 2 5y + 3)(x 2)ex x e(y2) ,
so the function f has five stationary points, with (x, y) = (1, 1), ( 23 , 1), (1, 23 ), ( 23 , 23 ), or (2, 21 ). None
of these are in the interior of the domain, however. So any maximum or minimum must be on one of the
four edges, or at the four corners. Obviously f = 0 along the edges x = 2 and y = 21 , as well as at the
three corners which make up the ends of these edges. At the other corner, f (0, 0) = 2e4 . Along the edge
x = 0 one has f2 < 0 for all y in [0, 21 ], so only the corners are relevant. It remains to consider the edge
y = 0, along which f1 < 0 for 0 x < 1 and for 23 < x 2, but f1 > 0 for 1 < x < 23 . Note that
f (1, 0) = e4 > 0 and f ( 23 , 0) = 21 e19/4 < 2e4 . We conclude that the maximum is 2e4 at (0, 0) and the
minimum is 0 at (x, 1/2) for all x [0, 2], and at (2, y) for all y [0, 1/2].
4. (a) The first-order conditions 2axy + by + 2y 2 = 0 and ax 2 + bx + 4xy = 0 must have (x, y) =
(2/3, 1/3) as a solution. So a = 1 and b = 2. Also c = 1/27, so that f (2/3, 1/3) = 1/9. Because

A = f11 (2/3, 1/3) = 2/3, B = f12 (2/3, 1/3) = 2/3, and C = f22 (2/3, 1/3) = 8/3, Theorem 13.3.1
shows that this is a local minimum.
(b) Maximum 193/27 at (2/3, 8/3). Minimum 1/9 at (2/3, 1/3). (The extreme points must be stationary
points in the interior of the admissible set or else points on the boundary. Note that we must check if
there are stationary points in the interior other than (2/3, 1/3). The first-order conditions reduce to
2y(x 1 + y) = 0 and x(x 2 + 4y) = 0. The stationary points satisfying these equations are (0, 0),
(0, 1), (2, 0), and (2/3, 1/3). Three of these four are on the boundary. The maximum is a point on the
boundary line 2x + y = 4.)
6. (a) Closed and bounded, so compact. (b) Open and unbounded. (c) Closed and bounded, so compact.
(d) Closed and unbounded. (e) Closed and unbounded. (f) Open and unbounded.
8. (a) The domain D is shown in Fig. M13.5.8.
The first order partials
are
41 (x+y 2 ) 1 41 (x+y 2 ) 1
f1 (x, y) = e 1 4 (x + y) , f2 (x, y) = e 1 2 y(x + y) .
(b) The unique stationary point (7/2, 1/2) gives the maximum value 4e15/16 .
D
y =1x
x
Figure M13.5.8
13.6
2 2
2. (a) f (x) = ex and g(x) = F (f (x)) = ln(ex ) = x 2 both have a unique maximum at x = 0.
(b) Only x = 0 maximizes f (x). But g(x) = 5 is maximized at every point x because it is a constant.
2 2
4. fx = 6x 2 + 30x 36, fy = 2 ey , fz = 3 + ez . The 8 stationary points are (x, y, z) =

(3, ln 2, ln 3 ), and (x, y, z) = (2, ln 2, ln 3 ), where all combinations of signs are allowed.
13.7
8 3 3 1 2 2 1
2. (a) K =
27 p r , L = 4 p w , T = 3 3 p
3/2 q 3/2
4 2 2
(b) Q = 9p r + 21 pw1 + 13 p 1/2 q 1/2 , so Q/r = 98 p 2 r 3 = K /p

4. (a) First-order conditions: (i) R1 C1 + s = 0, (ii) R2 C2 t = 0.

2 2
(b) 11 = R11 C11 < 0 and D = 11 22 (12 ) = (R11 C11 )(R22 C22 ) (R12 C12 ) > 0.
(c) Taking the total differentials of (i) and (ii) yields

(R11 C11
)dx1 + (R12
C12 )dx2 = ds,
(R21
C21
)dx1 + (R22
C22 )dx2 = dt
Solving for dx1 and dx2 yields, after rearranging,

(R22 C22 )ds (R12 C12 )dt (R C21

)ds + (R11
C11 )dt
dx1 = , dx2 = 21
D D
From this we find that the partial derivatives are
x1
R22
+ C22 x1
R12
+ C12 x2 R C21

x2 R C11

= > 0, = > 0, = 21 < 0, = 11 <0
s D t D s D t D
where the signs follow from the assumptions in the problem and the fact that D > 0 from (b). Note
that these signs accord with economic intuition. For example, if the tax on good 2 increases, then the
production of good 1 increases, while the production of good 2 decreases.

(d) Follows from the expressions in (c) because R12 = R21 and C12 = C21 .
2. (a) The profit is (Q1 , Q2 ) = 120Q1 + 90Q2 0.1Q21 0.1Q1 Q2 0.1Q22 . The first-order conditions
give Q1 = 500, Q2 = 200. We easily see that the second-order conditions are satisfied.
(b) The profit is now (Q1 , Q2 ) = P1 Q1 + 90Q2 0.1Q21 0.1Q1 Q2 0.1Q22 . The first-order
conditions give 1 = P1 0.2Q1 0.1Q2 = 0 and 2 = 90 0.1Q1 0.2Q2 = 0. If we require
Q1 = 400, then P1 80 0.1Q2 = 0 and 90 40 0.2Q2 = 0. It follows that Q2 = 250 and so
P1 = 105. Second-order conditions are easily checked.
4. (a) f1 = 2(x + y 2) + 4x(x 2 + y 2), f2 = 2(x + y 2) + 2(x 2 + y 2), f11

= 12x 2 + 4y 6,

f12 = f21 = 4x+2, f22 = 4. (b) f1 = f2 = 0 implies that x+y2 = 2x(x 2 +y2) = (x 2 +y2).
The last equality gives x = 21 or y = 2 x 2 . Inserting each of these into f2 = 0 yields three stationary
points, which are classified in the following table.
(1/2, 13/8) 7/2 4 4 2 Saddle point
(c) f (0, 2) = f (1, 1) = 0, and f (x, y) 0 for all (x, y), so these are (global) minimum points.

6. (a) f is stationary when 3x 2 2xy = x 2 + 2y = 0. Substituting 2y = x 2 in the first expression yields

3x 2 = x 3 , implying
that x 1=0 or x= 3. There are two stationary points at (0, 0 2and (3, 9/2).
1
(b) (0, 0), ( 2 2, 2), ( 2 2, 2). (The stationary points must satisfy y(8x 5xy + 1) = 0 and
x(2y 2 5xy + 1) = 0. Notice that at any stationary point, x = 0 implies y = 0, and conversely. So
apart from (0, 0), any other stationary point must satisfy 5xy 1 = 8x 2 = 2y 2 . Since no solution of
these last two equations can satisfy y = 2x, the only possibility is that y = 2x and so 2x 2 = 1.)
1 xy
f11 f12 2e ex 21 exy
8.
= 1 xy 1 xy
.
f21 f22 2e 2e ey
It follows that f11
= ex ( 21 ey 1) < 0 because ey < 1 < 2. In the same way, f22
< 0. Moreover,
2 1 xy x y
f11 f22 (f12 ) = 2 e (2 e e ) > 0, so f is concave.
10. (a) f1 = f2 = 0 implies that 2x y 3x 2 = 2y x = 0. So y = 21 x and 25 x 3x 2 = 0, implying
that x = 0 or 65 . There are two stationary points, which are classified in the following table.
(0, 0) 2 1 2 5 Saddle point
(5/6, 5/12) 3 1 2 5 Local maximum point

2
(b) f11 = 2 6x 0 x 1/3, while f22 = 2 0, and f11 f22 (f12 ) = 12x 5 0
x 5/12. We conclude that f is concave if and only if x 5/12. The largest value of f in S is 125/432,
obtained at (5/6, 5/12).

60 CHAPTER 14 CONSTRAINED OPTIMIZATION
Chapter 14 Constrained Optimization
Constrained optimization is the central tool of mathematical economics. Economics is about making the best
use of scarce resources, and scarcity is naturally represented by constraints. The usual technique for solving
constrained optimization problems is the Lagrange multiplier method. Most of this chapter is concerned with
equality constraints; the last two sections offer a brief introduction to inequality constraints.
For the case of two variables, the Lagrange multiplier method is the subject of Section 14.1. The method
is set out as a procedure in a large frame that should be memorized by the students. Partly because it easily
leads to errors when there are inequality constraints, we prefer not to set to zero the partial derivative of
the Lagrangian w.r.t. the multiplier (as is done in many other textbooks).2 Example 2 is important because
it explains what needs to be done when the equation system has multiple solutions. The consumer demand
problem with a CobbDouglas utility function is considered in Example 3, and the appropriate demand
functions are derived.
Section 14.2 presents the standard economic interpretation of Lagrange multipliers as shadow prices.
For the case when the (optimal) value function is differentiable, we present an easy proof that its derivative
is equal to the Lagrange multiplier. Example 1 illustrates its use in economics. The section ends with a rich
selection of problems.
Section 14.3 is devoted to explaining first with a geometric argument why the Lagrangian method works
when it does. Then an analytic argument for the first-order conditions is given, and a precise result is formulated
in Theorem 14.3.1. Example 1 considers a utility maximization problem in which a mechanical application
of the Lagrangian method fails to give the right answer. Problem 1 shows that the solution to a constrained
maximization problem need not maximize the Lagrangian, even though (under usual conditions) it must make
the Lagrangian stationary. This is another point that is rather commonly misunderstood. Let the teacher beware
of misleading attempts to motivate the Lagrangian technique! Fortunately, the error found in the quotation in
Problem 2 is far less common.
In Section 14.4 we turn from necessary to sufficient conditions for optimality. Of these, the simplest to
state is that a global maximum (or minimum) of the Lagrangian which satisfies the constraint is optimal. This
could have been stated as a theorem in its own right. In practice, however, there are difficulties in verifying that
a global maximum (or minimum) has been found, except in the important case when the Lagrangian happens
to be concave (or convex). This is the case mentioned in Theorem 14.4.1. (The appropriate generalization to
quasi-concave functions is not, in our view, an essential topic for an undergraduate course.)
The concept of a local constrained extremum is introduced. Theorem 14.4.2 then gives sufficient condi-
tions for the Lagrangian method to find a local constrained extremum. In smaller print the sufficient conditions
in Theorem 14.4.2 are expressed in terms of the sign of the bordered Hessian determinant.
Sections 14.114.4 consider only functions of two variables. The extension to n variables in Section
14.5 is mostly routine, though actually solving problems with many variables is often very complicated.
Example 3 discusses how to derive individual demand functions in a general n good setting. Then the case
of m equality constraints is considered. The Lagrangian needs as many Lagrange multipliers as there are
constraints, of course. This gives the right number of equations when the first-order conditions are combined
with the constraints themselves. In Example 5 we study a welfare maximization problem in a simple exchange
economy. Problem 3 is an interesting economic example, involving the optimal choice of leisure as well as
2
The equality constraint is an obvious necessary condition for optimality. There is no reason to differentiate
anything to demonstrate this.

CHAPTER 14 CONSTRAINED OPTIMIZATION 61
consumption of two goods. It also shows the possibility of a corner solution with zero labour supply in case
unearned income is sufficiently high.
Section 14.6 deals with sensitivity analysis. First, the standard interpretation of Lagrange multipliers
as shadow prices is mentioned. With m constraints, the value function depends on m variables. Where it is
differentiable, the m Lagrange multipliers are partial derivatives of the value function. Example 1 verifies the
properties of Lagrange multipliers numerically for a case with two constraints.
Adjusting the right-hand side of an equality constraint is just one of many perturbations to which a
constrained optimization problem can be subjected. Other perturbations are considered in this section; in
economics they feature constantly in comparative static propositions, of course. For such perturbations, frame
(6) states one version of the envelope theorem, which often arises in economic analysis. Examples 2, 3, and
4 involve interesting economic applications.
The final Sections 14.7 and 14.8 give a brief introduction to nonlinear programmingi.e., optimization
problems where the contraints are expressed as inequalities, rather than equalities. For the case of two variables
and one constraint, a frame sets out a recipe for solving the problem based on the Lagrange multiplier method.
Part of the recipe involves a complementary slackness condition. Theorem 14.7.1 emphasizes how this method
leads to a sufficient condition when the Lagrangian is concave. Section 14.7 concludes with a graphical
explanation of the complementary slackness condition.
In Section 14.8, the recipe of Section 14.7 is extended to general problems with n variables and m
constraints. Particular attention is paid to nonnegativity restrictions on the variables. The Lagrange multipliers
are related to partial derivatives of the (optimal) value function.

14.1
2. (a) (x, y) = (4/5, 8/5) with = 8/5. (With L(x, y) = x 2 +y 2 (x +2y 4), the first-order conditions
are L1 = 2x = 0 and L2 = 2y 2 = 0. From these equations we get 2x = y, which inserted into
the constraint gives x + 4x = 4. So x = 4/5 and y = 2x = 8/5, with = 2x = 8/5.)
(b) The same method as in (a) gives 2x = 0 and 4y = 0, so x = 2y. From the constraint we get
x = 8 and y = 4, with = 16.
4. (a) With L = x 2 + y 2 2x + 1 (x 2 + 4y 2 16), the first-order conditions are
2x 2 2x = 0 and
2y 8y = 0. One possibility is = 1/4, in which case x = 4/3 and so y = 4 2/3. Alternatively, y =
0, in which case x = 4. So there are four
solution candidates: (i) (x, y, ) = (4, 0,3/4), (ii) (x, y, ) =
(4, 0, 5/4), (iii) (x, y, ) = (4/3, 4 2/3, 1/4), and (iv) (x, y, ) = (4/3, 4 2/3, 1/4). Of these,
the second is the maximum point (while (iii) and (iv) are the minimum points).
(b) The first-order conditions imply that 2x + 3y = = 3x + 2y. So (x, y) = (50, 50) with = 250.
6. (a) With L(x, y) = 100 ex ey (px + qy m), Lx = Ly = 0 when ex = p and ey = q.
Hence, x = ln(p) = ln ln p, y = ln ln q. Inserting these expressions for x and y
into the constraint, then solving for ln , yields ln = (m + p ln p + q ln q)/(p + q). Therefore
x(p, q, m) = [m + q ln(q/p)]/(p + q), y(p, q, m) = [m + p ln(p/q)]/(p + q).
(b) x(tp, tq, tm) = [tm + tq ln(tq/tp)]/(tp + tq) = x(p, q, m), so x is homogeneous of degree 0. In
the same way we see that y(p, q, m) is homogeneous of degree 0.
14.2
2. (a) L(x, y) = x + y (x 2 + y 1). The equations L1 = 1 2x = 0, L2 = 1 = 0, and x 2 + y = 1

have the solution x = 1/2, y = 3/4, and = 1. (b) The solution is illustrated in Fig. M14.2.2. The
minimization problem has no solution because f (x, 1 x 2 ) = x + 1 x 2 as x .
(c) The new solution is x = 0.5 and y = 0.85. The change in the value function is f (1.1) f (1) =
(0.5 + 0.85) (0.5 + 0.75) = 0.1. Because = 1, one has dc = 1 0.1 = 0.1. So, in this case, (3)
is satisfied with equality. (This is because of the special form of the functions f and g.)
y
6
5
4 y 2 = x(x + 1)2
y
3
2 (x, y)
d 1
( 21 , 43 ) x
y = 1 x2 321
1
1 2 3 4
2
3
x 4
1 1 5
x+y = 4 5
1
x+y = 2 6
4. (a) x = 4, y = 24, = 1/4. (b) y = 97/4, x = 4. 1U = 105/4 104/4 = 1/4, the value of the
Lagrange multiplier from (a). (There is exact equality here because U is linear in one of the variables.)
(c) x = q 2 /4p, y = m/q q/4p. (Note that y > 0 iff m > q 2 /4p.)
14.3
2. The problem with systems of three equations and two unknowns is not that they are merely difficult to
solve but that they are usually inconsistenti.e., it is impossible to solve them. The equations fx (x, y) =
fy (x, y) = 0 are NOT valid at the optimal point.
4. The minimum is 1 at (x, y) = (1, 0). Actually, this problem is quite tricky. The Lagrangian can be
written in the form L = (x + 2)2 + (1 )y 2 + x(x + 1)2 . The only stationary point which satisfies
the constraint is (0, 0), with = 4, and with f (0, 0) = 4. (In fact, L2 = 0 only if = 1 or y = 0. For
= 1, L1 = 3(x + 1)2 + 2 > 0 for all x. For y = 0, the constraint gives x = 0 or x = 1. But x = 1
gives L1 = 2, so x = 0 is necessary for a stationary point.) Yet at (1, 0) both g1 (1, 0) and g2 (1, 0)
are 0, and the Lagrange multiplier method fails. The given problem is to minimize (the square of) the
distance from (2, 0) to a point on the graph of g(x, y) = 0. But the graph consists of the isolated point
(1, 0) and a smooth curve, as illustrated in Fig. M14.3.4.
14.4
2. With L(x, y) = x 2 + y 2 (x + 2y a) the first-order conditions are 2x = 0 = 2y 2.
So (x, y) = (a/5, 2a/5) with = 2a/5. Applying Theorem 14.4.2 with f (x, y) = x 2 + y 2 and
g(x, y) = x + 2y, we find that D(x, y, ) = 10, so the only stationary point is a local minimum. (By
reducing the problem to a one-variable problem, it is easy to see that it is actually a global minimum.)
Geometrically, the problem is to find that point on the straight line x + 2y = a that is closest to the origin.

14.5
2. Here L = x + 4y + 3z (x 2 + 2y 2 + 31 z2 b). So necessary conditions are:
(i) L1 = 1 2x = 0; (ii) L2 = 4 4y = 0; (iii) L3 = 3 23 z = 0. It follows that 6 = 0,
= 1/2, y = 1/, z = 9/2. Inserting these values into the constraint yields 2
and so x = 9/b, so
= 3/ b. The value of the objective function is x + 4y + 3z
= 18/, so = 3/ b determines the
minimum point. This is (x, y, z) = (a, 2a, 9a), where a = b/6.
4. The constraints reduce to h + 2k + = 0 and 2h k 3 = 0, so k = h and = h. But then

f (x, y, z) = x 2 + y 2 + z2 = 200 + 3h2 200 for all h, so f is maximized for h = 0. Then k = = 0
also, and we conclude that (x, y, z) = (10, 10, 0) solves the minimization problem.
6. (a) c1i = 1 r i /p i i i i i
1 , c2 = 2 r /p2 . (Write the utility function in the equivalent form U (c1 , c2 ) =
i 1 i 2
ln (c1 ) (c2 ) . Using Theorem 13.6.3, the result follows immediately from () in Example 14.1.3,
recalling that 1 + 2 = 1.) (b) In Example 5, 1 = 1 /e1 , 2 = 2 /e2 . So p1 = 1 and p2 = 2
implies c1i = 1 r i /p1 = r i e1 = i e1 provided that r i = i , which also gives c2i = i e2 . Because
p1 e1 + p2 e2 = 1 e1 + 2 e2 = 1 + 2 = 1, it follows that r i = i (p1 e1 + p2 e2 ) for i = 1, 2.
8. (a) With a CobbDouglas utility function, Uk (x) = k U (x)/xk , so from (6) (with j = 1), we have
pk /p1 = Uk (x)/U1 (x) = k x1 /1 xk . Thus pk xk = (ak /a1 )p1 x1 . Inserted into the budget constraint, we
have p1 x1 + (a2 /a1 )p1 x1 + + (an /a1 )p1 x1 = m, which implies that p1 x1 = a1 m/(a1 + + an ).
Similarly, pk xk = ak m/(a1 + + an ) for k = 1, . . . , n.
(b) From (6) (with j = 1), we get xka1 /x1a1 = pk /p1 and so xk /x1 = (pk /p1 )1/(1a) or pk xk /p1 x1 =
(pk /p1 )11/(1a) = (pk /p1 )a/(1a) . The budget constraint gives
n
X
a/(1a) a/(1a)
p1 x1 1 + (p2 /p1 )a/(1a) + + (pn /p1 )a/(1a) = m. So p1 x1 = mp1 pi .
i=1
.X
n
1/(1a) a/(1a)
Arguing similarly for each k, one has xk = mpk pi for k = 1, . . . , n.
i=1
10. Differentiating the constraint w.r.t. x1 (keeping x2 fixed) yields g1 + g3 (x3 /x1 ) = 0, which implies
(i) x3 /x1 = g1 /g3 . Similarly, (ii) x3 /x2 = g2 /g3 . The first-order conditions for the maximiza-
tion of z = f (x1 , x2 , x3 ), where x3 is a function of (x1 , x2 ), are (iii) z/x1 = f1 + f3 (x3 /x1 ) = 0 and
(iv) z/x2 = f2 + f3 (x3 /x2 ) = 0. Substitute from (i) and (ii) into (iii) and (iv), letting = f3 /g3 .
This yields the equations f1 g1 = 0 and f2 g2 = 0. Also, by definition of , f3 g3 = 0. These
are the conditions (3) for n = 3.
14.6
p
2. (a) x = aM/, y = bM/, z = cM/ , = 1/(2M), where M = L/ a 2 / + b2 / + c2 / . (The
first-order conditions give x = a/2, y = b/2, z = c/2 . Substituting in the constraint and solving
for gives the solution.) (b) We find that M = L/5, and the given values of x, y, and z follow. For
L = 100 one has M = 2 and = 1/4. The increase in the maximalvalue as L increases from 100 to
101, is approximately 1 = 0.25. The actual increase is 5( 101 100 ) 0.249378.
4. K = 21/3 r 1/3 w1/3 Q4/3 , L = 22/3 r 2/3 w 2/3 Q4/3 , C = 3 22/3 r 2/3 w 1/3 Q4/3 ,
= 24/3 r 2/3 w1/3 Q1/3 . The equalities () are easily verified.

6. (a) Given L = ex + y + z (x + y + z 1) (x 2 + y 2 + z2 1), the first-order conditions are

(i) ex 2x = 0; (ii) 1 2y = 0; (iii) 1 2z = 0. From (ii) and (iii) it follows that
(y z) = 0. If = 0, then = 1, x = 0, and y + z = 1, y 2 + z2 = 1. The stationary points occur
where (x, y, z) = (0, 1, 0) or (0, 0, 1), in which case ex + y + z = 2. If 6 = 0, then y = z = 21 (1 x),
implying that x 2 + 21 (1 x)2 = 1, so x = 1 or 31 . The maximum is at (1, 0, 0) where ex + y + z = e.
Another stationary point is at ( 31 , 23 , 23 ) where ex + y + z = e1/3 + 43 < 73 < e. The relevant Lagrange
multipliers are = 1, = 21 (e 1). (b) 1f (0.02) + (0.02) = 0.01(3 e) 0.0028.
14.7
2. (a) The Lagrangian is L(x, y) = x 2 + 2y 2 x (x 2 + y 2 1). Conditions (2)(4) take the form:
2x 1 2x = 0; 4y 2y = 0; 0 ( = 0 if x 2 + y 2 <1). (b) Candidates: (1/2, 0) with
= 0; (1,
0) with = 1/2; (1,
0) with = 3/2; and (1/2, 3/2) with = 2. Maximum 9/4 at
(1/2, 3/2) and at (1/2, 3/2). The extreme-value theorem confirms that this is the solution.
4. (a) f1 (x, y) 0 (f1 (x, y) = 0 if x < b) and f2 (x, y) = 0 (b) f1 (x, y) = 0 and f2 (x, y) 0
(f2 (x, y) = 0 if y > a)
14.8
2. (a) The admissible set is the shaded region in Fig. M14.8.2.
(b) With the constraints g1 (x, y) = x y 4, g2 (x, y) = x 1, g3 (x, y) = y 1, the
Lagrangian is L = x + y ex ex+y 1 (x y + 4) 2 (x 1) 3 (y + 1). The first-order
conditions are that there exist nonnegative numbers 1 , 2 , and 3 such that:
(i) Lx = 1 ex ex+y + 1 + 2 = 0; (ii) Ly = 1 ex+y + 1 + 3 = 0; (iii) 1 (x y + 4) = 0;
(iv) 2 (x 1) = 0; (v) 3 (y + 1) = 0. (We formulate the complementary slackness conditions as
in (14.7.5).) From (ii), ex+y = 1 + 1 + 3 . Inserting this into (i) yields 2 = ex + 3 ex > 0.
Because 2 > 0, (iv) implies that x = 1. So any solution must lie on the line (II) in the figure, which
shows that the third constraint must be slack. (Algebraically, because x + y 4 and x = 1, we have
y 4 x = 5 > 1.) So from (v) we get 3 = 0, and then (ii) gives 1 = ex+y 1 e4 1 > 0.
Thus from (iii), the first constraint is active, so y = 4 x = 5. Hence the only possible solution is
(x , y ) = (1, 5). Because L(x, y) is concave, we have found the optimal point.
y
7
II
6
(1, 5) 5
4
3
2 I
(3, 1)
1
III
x
2 1 1 2 3 4 5 6 7
Figure M14.8.2
4. (a) The Lagrangian is L = y x 2 + y + (y x + 2) (y 2 x), which is stationary when

(i) 2x + = 0; (ii) 1 + + 2y = 0. Complementary slackness requires in addition,

(iii) 0 ( = 0 if y > 0), (iv) 0 ( = 0 if y x + 2 > 0); (v) 0 ( = 0 if y 2 < x).

From (ii), 2y = 1 + + > 0, so y > 0. Then (iii) implies = 0, and 2y = 1 + . From (i),
x = 21 ( ). But x y 2 > 0, so > 0, and from (v), y 2 = x.
Suppose > 0. Then y x + 2 = y y 2 + 2 = 0 with roots y = 1 and y = 2. Only y = 2 is
feasible. Then x = y 2 = 4. Because = 0, conditions (i) and (ii) become + = 8 and 4 = 1,
so = 7/3, which contradicts 0, so (x, y) = (4, 2) is not a candidate. Therefore = 0 after
all. Thus x = 21 = y 2 and, by (ii), 1 = 2y = 4y 3 . Hence y = 41/3 , x = 42/3 . This is the only
remaining candidate. It is the solution with = 0, = 0, and = 1/2y = 41/6 .
(b) We write the problem as max xeyx 2ey subject to y 1 + 21 x, x 0, y 0. The Lagrangian is
L = xeyx 2ey (y 1 x/2), so the first-order conditions (4) and (5) are:
(i) Lx = eyx xeyx + 21 0 (= 0 if x > 0); (ii) Ly = xeyx 2e 0 (= 0 if y > 0);
(iii) 0 ( = 0 if y < 1 + 21 x).
From (i) we have x 1 + 21 exy 1, so (iv) (x 1)eyx = 21 . Suppose > 0. Then (iii)
implies (v) y = 1 + 21 x > 0. From (ii) and (iv) we then have xeyx = 2e + = eyx + 21 and so
1 1
= 2eyx 4e = 2e(e 2 x 2), by (v). But then > 0 implies that e 2 x > 2, which contradicts x 0.
When = 0, (iv) gives x = 1. If y > 0, then (ii) yields ey1 = 2e, and so y 1 = ln(2e) > 21 x when
x = 1. Thus feasibility requires that y = 0, so we see that (x, y) = (1, 0) is the only point satisfying all
the conditions, with = 0.

2m 3m m 2m 3m 2m
2. (a) x = ,y = (b) x = ,y = (c) x = ,y =
5p 5q 3p 3q 5p 5q
4. (a) Interpretation: Maximize the volume of a rectangular box with a given surface area A.
(b) With the Lagrangian L = xyz (2xy + 2xz + 2yz A), the first-order conditions yield
yz (2y + 2z) = 0, xz (2x + 2z) = 0, xy (2x + 2y) = 0. (The optimal solution obviously
yz xz xy
requires all three variables to be positive.) It follows that = = = . The other
y+z x+z x+y
required equality is just the constraint divided by 2.
(c) From yz/(y + z) = xz/(x + z) we get y(x + z) = x(y + z), or yz = xz, so x = y because z > 0.
Similarly, from xz/(x + z) = xy/(x + y) we get y = z. Inserting y = z = x into the constraint and

solving for x yields x = A/6, and so the optimal solution is x = y = z = A/6. The optimal box is
a cube whose six faces all have area A/6.

6. With the Lagrangian L = (1/x) + 4y + 4z (x + y + z 2) x 2 + y 2 + z2 3/2 , the first-order
conditions yield (i) 1/x 2 2x = 0; (ii) 4 2y = 0; (iii) 4 2z = 0. Equations (ii)
and (iii) give y = z, i.e. (iv) (y z) = 0.
A. Suppose 6= 0. Then (iv) implies y = z, which inserted into the constraints gives x = 2 2y and
x 2 + 2y 2 = 3/2, so (2 2y)2 + 2y 2 = 3/2, or 6y 2 8y + 5/2 = 0. The solutions of this equation
are y = 5/6 and y = 1/2. This gives the following solution candidates: P1 = (1/3, 5/6, 5/6) and
P2 = (1, 1/2, 1/2). The function values are are 11/3 and 3 respectively.
B. Suppose = 0. From (ii) we have = 4, and then (i) gives 1/x 2 = 4, so x = 1/2. Inserting x = 1/2
into the first constraint gives z = 3/2 y, and then the second constraint gives 2y 2 3y + 1 = 0, with the
roots y = 1 and y = 1/2. The new solution candidates are P3 = (1/2, 1, 1/2) and P4 = (1/2, 1/2, 1).
The function values are both 4.

It remains to consider the case x = 1/2. But then the first constraint gives z = 5/2 y, and the
second constraint gives the equation 2y 2 5y + 5 = 0, which has no real solutions.
By evaluating the objective function f at all the four candidate points, we find that f has its largest
value at P3 and P4 , with fmax = f (1/2, 1, 1/2) = f (1/2, 1/2, 1) = 4.
8. (a) The maximum value is 20 attained at (x, y, z) = (4, 0, 0). The minimum value is 7.5 attained at all
points (x, y, z) satisfying x = 1 and y 2 + z2 = 7.5. (With as the Lagrange multiplier, the first-order
conditions imply (i) 2x + 1 = 2x; (ii) 2y = 4y; (iii) 2z = 4z. From (ii), y = 0 or = 1/2. From
(iii), z = 0 or = 1/2. Suppose first that = 1/2. Then (i) implies x = 1 and 2(y 2 + z2 ) = 15,
which gives the minimum value. On the other hand, if 6 = 1/2, then (ii) and (iii) give y = z = 0. Then
x = 4, and x = 4 gives the maximum. By the extreme value theorem, maximum and minimum values
exist in (a) as well as in (b).)
(b) The function f has a unique stationary point in the interior of S at (x, y, z) = (1/2, 0, 0), with
f (1/2, 0, 0) = 1/4. We conclude that the maximum point of f in S is at (4, 0, 0) and the minimum
is at (1/2, 0, 0).
10. (a) The Lagrangian is L(x, y) = U (x, y) [py w(24 x)]. The first-order conditions imply
pU1 = wU2 = wp, which immediately yields (). (b) Differentiating () and () w.r.t. w gives

pyw = 24 x wxw and p(U11 xw + U12
yw ) = U2 + w(U21
xw + U22 yw ). Solving for xw = x/w
yields the given formula. (c) x/w = (4 ln 18 5)/(ln 18 + 1).
12. (a) See Fig. M14.R.12. Note that x ln(3/2). (b) With and as the Lagrange multipliers, the first-
order conditions are: (i) (2x + 1) + ex = 0; (ii) y + = 0; (iii) 0 ( = 0 if ex < y);
(iv) 0 ( = 0 if y < 2/3). From (i), = (2x + 1)ex 3/2, because of (a). From (ii), =
y 3/2 2/3 > 0, so y = 2/3 because of (iii). Solution: (x , y ) = (ln(3/2), 2/3), with
= 3[ln(3/2) + 1/2], = 3 ln(3/2) + 5/6. (The Lagrangian is concave so this is the solution.)
y
2
y = ex
1
(x , y )
y = 2/3
1 2 x
Figure M14.R.12

CHAPTER 15 MATRIX AND VECTOR ALGEBRA 67
Chapter 15 Matrix and Vector Algebra

Chapter 15 begins our treatment of matrix algebra by presenting matrices, vectors, and simple algebraic rules
for working with them. Section 15.1 introduces a general notation for linear simultaneous equations, and
motivates them through a simple case of the Leontief inputoutput system.
Matrices are defined in Section 15.2 as rectangular arrays of numbers. (The idea that matrices represent
linear transformations is, in our view, better postponed to a more advanced and abstract book on linear
algebra.) Particular examples are the coefficient matrix for a system of linear simultaneous equations, and
a spreadsheet recording the value of retail sales for different commodities at different sales outlets. Then
equality, addition, and multiplication by a scalar are defined in obvious ways.
The rules for matrix multiplication are far from intuitively obvious to a student seeing them for the first
time. Section 15.3 therefore takes some pains to explain how C = AB is defined for a 2 3 matrix A and a
3 2 matrix B in order that the 2 2 matrix C should satisfy Cx = A(Bx) for all 2-vectors x. This, of course,
builds on the rules for multiplying a coefficient matrix by a vector, which are also explained in Section 15.3.
Thereafter, the rule for multiplying general m n and n p matrices is set out and illustrated. Attention is
drawn to the important fact that matrix multiplication is generally noncommutativei.e., generally AB 6 = BA.
The section concludes by reviewing how a general system of equations can be expressed concisely in matrix
form.
Next, Section 15.4 turns to the associative law and (since multiplication is non-commutative) the two
versions of the distributive law for matrix multiplication. Then powers of square matrices are defined, as is
the identity matrix of any finite dimension. The section concludes by warning against algebraic errors such
as replacing AB by BA, or asserting that if AB = 0, then either A = 0 or B = 0 (or both).
Section 15.5 gives a brief discussion of transpose matrices, including general rules for transposition, and
of symmetric matrices.
A powerful general method for solving linear systems of equations is Gaussian elimination, which is
the subject of Section 15.6. The staircase method is explained and a convenient notation for elementary
operations is introduced.
Next, Section 15.7 introduces n-dimensional vectors as lists of n numbers. The rules of addition, sub-
traction, and multiplication by a scalar are described. Linear combinations are defined.
Particularly for two-dimensional vectors, the fact that there is a simple geometric interpretation can help
students enormously. This is the topic of Section 15.8, which also discusses the parallelogram law of vector
addition, and how to represent multiplication by a scalar. The section concludes with some remarks about
n-dimensional geometry.
Calculating the inner (or scalar) product of two n-vectors is an important step on the way to finding
the product of two matrices. In economics, it is common to consider corresponding vectors of prices and of
quantities. Then the inner product of a price and quantity vector represents the value of a commodity bundle.
This is used in Section 15.8 to motivate the concept. Rules for manipulation are presented. The length of a
vector is defined. The CauchySchwarz inequality is stated. The section concludes with a brief discussion of
orthogonality, a concept that is useful in econometrics.
Geometric ideas receive more attention in Section 15.9. A diagram is used to motivate the algebraic
definition of a line in a space of three dimensions. This algebraic definition is then extended to n dimensions.
Similarly, a diagram of a plane in three dimensions is used to motivate the algebraic definition, which is then
extended to n dimensions.

68 CHAPTER 15 MATRIX AND VECTOR ALGEBRA
15.1
2. Yes, the system is linear in a, b, c, and d.
4. 2x1 + 3x2 + 4x3 = 1

3x1 + 4x2 + 5x3 = 2
4x1 + 5x2 + 6x3 = 3
6. (a) The commodity bundle owned by individual j . (b) ai1 +ai2 + +ain is the total stock of commodity i.
The first case is when i = 1. (c) p1 a1j + p2 a2j + + pm amj
15.2
2. u = 3 and v = 2. (Equating the elements in row 1 and column 3 gives u = 3. Then, equating those in
row 2 and column 3 gives u v = 5 and so v = 2. The other elements then need to be checked, but
this is obvious.)

1 0 4 1 2 6 3 8 20
4. A + B = ,AB= , and 5A 3B =
2 4 16 2 2 2 10 12 8
15.3

1 15 0 0 0 0
2. (i) (ii) and (iii): AB = C(AB) = , since AB = .
6 13 0 0 0 0

1 2 1 x1 4
1 1 x1 3
4. (a) = (b) 1 1 1 x2 = 5

3 5 x2 5
2 3 1 x3 1
x1
2 3 1 0
(c) x2 =
1 1 1 0
x3

0.2875
6. T(Ts) = 0.2250
0.4875
15.4
2. (ax 2 + by 2 + cz2 + 2dxy + 2exz + 2fyz) (a 1 1 matrix)
4. Equality in (a) as well as in (b) if and only if AB = BA. ((A+B)(AB) = A2 AB+BAB2 6 = A2 B2

unless AB = BA. The other case is similar.)
6. (a) Direct verification by matrix multiplication. (b) AA = (AB)A = A(BA) = AB = A, so A is

idempotent. Then just interchange A and B to show that B is idempotent.
(c) As the induction hypothesis, suppose that Ak = A, which is true for k = 1. Then Ak+1 = Ak A =
AA = A, which completes the proof by induction.

15.5

3 1 0 2 3 1 6 2
2. A = ,B =
, (A + B) =
, (A) = ,
2 5 2 2 4 7 4 10

4 10 4 10 2 4
AB = , (AB) = = B A , and A B =
10 8 10 8 10 14
Verifying the rules in (2) is now very easy.
4. Symmetry requires a 2 1 = a + 1 and a 2 + 4 = 4a. The second equation has the unique root a = 2,
which also satisfies the first equation.
6. By the associative law and rule (15.5.2)(d) for transposition, (A1 A2 A3 ) = (A1 (A2 A3 )) = (A2 A3 ) A1 =
(A3 A2 )A1 = A3 A2 A1 . In general, for any natural number n > 3, one has
(A1 A2 An ) ((A1 A2 An1 )An ) = An (A1 A2 An1 ) . As the induction hypothesis, suppose the
result is true for n 1. Then the last expression becomes An An1 , A2 A1 , so the result is true for n.
15.6
2. Gaussian elimination yields:

1 1 1 1 1 1 1 1 1 1
1 1 2 2 0 2 3 1 21
1 2 a b 0 1 a+1 b1

1 1 1 1 1 1 1 1
0 1 3/2 1/2 1 0 1 3/2 1/2
0 1 a+1 b1 0 0 a + 5/2 b 1/2
For any z, the first two equations imply that y = 21 + 23 z and x = 1 y + z =
3 1
2 2 z. From the last
equation we see that for a 6 = 25 , there is a unique solution with z = (b 21 ) (a + 25 ). For a = 25 ,
there are no solutions if b 6= 21 , but z is arbitrary (one degree of freedom) if b = 21 .
4. (a) After moving the first row down to row number three, Gaussian elimination yields the matrix

1 2 1 b2
0 1 2 3
b2 1 b3
2 2
0 0 3 4a b1 + (2a 23 )b2 + ( 21 a)b3
Obviously, there is a unique solution iff a 6 = 3/4.
(b) Put a = 3/4 in part (a). Then the last row in the matrix in (a) becomes (0, 0, 0, b1 41 b3 ). It follows
that if b1 6= 41 b3 there is no solution. If b1 = 41 b3 there are an infinite number of solutions. In fact,
x = 2b2 + b3 5t, y = 23 b2 21 b3 + 2t, z = t, with t R.
15.7
2. a + b + c = (1, 6, 4), a 2b + 2c = (3, 10, 2), 3a + 2b 3c = (9, 6, 9),
a b c = (a + b + c) = (1, 6, 4).
4. (a) xi = 0 for all i. (b) Nothing, because 0 x = 0 for all x.
6. 4x 2x = 7a + 8b a, so 2x = 6a + 8b, and x = 3a + 4b.
8. The scalar product of the two vectors is x 2 +(x 1)x +33x = x 2 +x 2 x +9x = 2x 2 +8x = 2x(x +4),
which is 0 for x = 0 and x = 4.

70 CHAPTER 15 MATRIX AND VECTOR ALGEBRA
10. (a) The firms revenue is pz. Its costs are px. (b) Profit = revenue costs = pzpx = p(zx) = py.
If p y < 0, the firm makes a loss equal to p y.
15.8
2. (a) = 0 gives x = (3, 1), = 1/4 gives x = (2, 5/4), = 1/2 gives x = (1, 3/2),
= 3/4 gives x = (0, 7/4), and = 1 gives x = (1, 2). See Fig. M15.8.2.
(b) As runs through [0, 1], the vector x will run through all points on the line segment S between a and
b. In fact, according to the pointpoint formula, the line L through (3, 1) and (1, 2) has the equation
x2 = 41 x1 + 47 or x1 + 4x2 = 7. The line segment S is traced out by having x1 run through [3, 1] as x2
runs through [1, 2]. Now, (1 )a + b = (3 4, 1 + ). Any point (x1 , x2 ) on L satisfies x1 + 4x2 = 7
and equals (3 4, 1 + ) for = 41 (3 x1 ) = x2 1. Any point on the segment of this line between
a = (3, 1) and b = (1, 2) equals (3 4, 1 + ) for some [0, 1].
y
=1
= 3/4
= 1/2
= 1/4
1 =0
b
a
x
1 1 2 3
Figure M15.8.2
4. (a) x1 (1, 2, 1) + x2 (3, 0, 2) = (x1 3x2 , 2x1 , x1 2x2 ) = (5, 4, 4) when x1 = 2 and x2 = 1.
(b) x1 and x2 must satisfy x1 (1, 2, 1) + x2 (3, 0, 2) = (3, 6, 1). Then x1 3x2 = 3, 2x1 = 6, and
x1 2x2 = 1. The first two equations yield x1 = 3 and x2 = 2; then the last equation is not satisfied.
6. The vectors are orthogonal iff their scalar product is 0, that is iff x 2 x 8 2x + x = x 2 2x 8 = 0,
which is the case for x = 2 and x = 4.
8. (||a|| + ||b||)2 ||a + b||2 = ||a||2 + 2||a|| ||b|| + ||b||2 (a + b)(a + b) = 2(||a|| ||b|| a b)
2(||a|| ||b|| |a b|) 0 by the CauchySchwarz inequality (2).
15.9
2. (a) Direct verification. (To show that a lies on L, put t = 0.) (b) The direction of L is given by the vector
(1, 2, 1), and the equation of the orthogonal plane is (1)(x1 2) + 2(x2 (1)) + 1 (x3 3) = 0,
or x1 + 2x2 + x3 = 1. (c) We must have 3(t + 2) + 5(2t 1) (t + 3) = 6, and so t = 4/3. The
point of intersection is then (2/3, 5/3, 13/3).
4. (a) Direct verification. (b) (x1 , x2 , x3 ) = (2, 1, 1) + t (1, 2, 3) = (2 t, 1 + 2t, 1 + 3t).

3 2 3 4 2 4
2. (a) A B = (b) A + B 2C = (c) AB =
2 2 2 8 2 3
2 1 2 2 2
(d) C(AB) = (e) AD = (f) DC is not defined.
6 8 0 2 3


1 1 1 1 x a

2 5 x1 3 1 3 2 4y b

4. (a) = (b)
1 4 8 =
5 8 x2 5 0 z c
2 0 1 1 t d
a 1 a+1 x b1
(c) 1 2 1 y = b2
3 4 7 z b3

p3 + p2 q 2p 2 q + 2pq 2 pq 2 + q 3
6. (a) TS = 21 p 3 + 21 p 2 + 21 p 2 q p 2 q + pq + pq 2 21 pq 2 + 21 q 2 + 21 q 3 = S because p + q = 1.
p3 + p2 q 2p 2 q + 2pq 2 pq 2 + q 3
A similar argument shows that T2 = 21 T + 21 S. To prove the last equality, we do not have to consider the
individual elements: T3 = TT2 = T( 21 T + 21 S) = 21 T2 + 21 TS = 21 ( 21 T + 21 S) + 21 S = 41 T + 43 S.
(b) The appropriate formula is () Tn = 21n T + (1 21n )S. This formula is correct for n = 1
(and for n = 2, 3). Suppose () is true for n = k. Then using the two first equalities in (a), Tk+1 =
TTk = T(21k T + (1 21k )S) = 21k T2 + (1 21k )TS = 21k ( 21 T + 21 S) + (1 21k )S =
2k T + 2k S + S 2 2k S = 2k T + (1 2k )S, which is formula () for n = k + 1.

1 4 1 2 1 4 1 1 4 1 1 0 5
8. (a)
2 2 8 0 6 6 1/6 0 1 1 4 0 1 1
The solution is x1 = 5, x2 = 1.

2 2 1 2 1 3 1 0 2 3
(b) 1 3 1 0 2 2 1 2
3 4 1 1 3 4 1 1
1 3 1 0 1 3 1 0
0 8 3 2 1/8 0 1 3/8 1/4 13
0 13 4 1 0 13
4 1
1 3 1 0 1 3 1 0
0 1 3/8 1/4
0 1 3/8 1/4 3
0 0 7/8 9/4 8/7 0 0
1 18/7
1 0 1/8 3/4 1 0 0 3/7
0 1 3/8 1/4 0 1 0 5/7
0 0 1 18/7 3/8 1/8 0 0 1 18/7
The solution is x1 = 3/7, x2 = 5/7, x3 = 18/7.

1 3 4 0 5 1 3 4 0
(c)
5 1 1 0 0 14 19 0 1/14

1 3 4 0 1 0 1/14 0

0 1 19/14 0 3 0 1 19/14 0
The solution is x1 = (1/14)x3 , x2 = (19/14)x3 , where x3 is arbitrary. (One degree of freedom.)

10. kak
= 35, kbk = 11, and kck = 69. Also, |a b| = |(1) 1 + 5 1 + 3 (3)| = | 5| = 5, and
35 11 = 385 is obviously greater than 5, so the CauchySchwarz inequality is satisfied.
12. If PQ QP = P, then PQ = QP + P, and so P2 Q = P(PQ) = P(QP + P) = (PQ)P + P2 =

(QP+P)P+P2 = QP2 +2P2 . Thus, P2 QQP2 = 2P2 . Moreover, P3 Q = P(P2 Q) = P(QP2 +2P2 ) =
(PQ)P2 + 2P3 = (QP + P)P2 + 2P3 = QP3 + 3P3 Hence, P3 Q QP3 = 3P3 .

72 CHAPTER 16 DETERMINANTS AND INVERSE MATRICES
Chapter 16 Determinants and Inverse Matrices

In this chapter, Section 16.1 begins with determinants of order 2, whose definition is motivated by the form of
the typical solution to two linear simultaneous equations in two unknowns. In fact, determinants are defined
so that Cramers rule for solving such equations is valid. Then it is shown that the determinant of a 22 matrix
is equal to the area of the parallelogram generated by its two columns, regarded as a pair of 2-vectors.
Next, Section 16.2 moves on to three equations in three unknowns. This gives rise to a complicated
formula, which is then broken down so that it appears as the expansion of the determinant by the cofactors of
its first row. It is claimed that the determinant of a 3 3 matrix equals the volume of the (3-dimensional)
parallelepiped generated by its columns, regarded as three 3-vectors. Sarruss rule for evaluating 3 3
determinants is then introduced.
Section 16.3 proposes a formula for n n determinants that is essentially the sum of n! terms, each term
being the product of n elements, of which one is chosen from each row and one from each column of the
matrix. Then a sign rule is given to determine whether this product should have a + or attached in forming
the overall sum. Of course, this formula is virtually useless except in a few special cases of sparse matrices
where virtually all the n! terms collapse to 0.
In fact, Section 16.3 really helps clear the ground for the fundamental Rules for Determinants that are
enunciated in Theorem 16.4.1. Particularly important is the rule saying that determinants are unchanged by
the elementary row (or column) operations of adding a multiple of one row (or column) to any other row (or
column). These rules are illustrated by showing how they apply to particular 2 2 matrices. There are also
two pages of discussion concerning how to prove most parts of Theorem 16.4.1. We envisage, however, that
most students will prefer to take on trust the key results set out in the theorem.
The topic of Section 16.5 is expansion by cofactors in generalnot just the expansion of 33 determinants
by the first row that was considered in Section 16.2. Also important is the fact that expansion by alien
cofactors always gives zero, because it is like finding the determinant of a matrix in which one row or column
has been repeated. The section concludes with a proof that the rule for expanding cofactors is valid, though
once again, this is a result that we expect most students will prefer to take on trust. By the end of this section
students should have learned how to find the determinant of a matrix in any one of several equivalent ways
expansion by cofactors, the full expansion of n! terms, or by elementary row or column operations designed
to produce a sparse matrixeven one that is upper or lower triangular, and so with determinant equal to
the product of its diagonal terms.
Having completed the discussion of determinants per se, Section 16.6 moves on to the inverse of a square
matrix. After the definition, it is claimed that an inverse exists iff the matrix is non-singular in the sense of
having a non-zero determinant. It is shown that, if an inverse exists, it must be unique. The general formula for
the inverse of a 2 2 matrix is given. Then Theorem 16.6.1 sets out some key properties of inverse matrices.
These include the results that the inverse of the inverse is the original matrix, that the inverse of a matrix
product is the reverse product of the inverses, and that the inverse of the transpose is equal to the transpose
of the inverse. Finally, Theorem 16.6.2 states how to solve linear equations by matrix inversion, in case the
relevant matrix is non-singular.
An explicit formula for the inverse of a matrix is the main topic of Section 16.7. Indeed, based on the
rules for expansion by cofactors (including alien cofactors), it is shown that one can find the adjoint matrix,
defined as the transpose of the matrix of cofactors, then divide the adjoint by the (non-zero) determinant of
the (non-singular) matrix. This is the general formula for the inverse set out in Theorem 16.7.1. The section
concludes by outlining how to find the inverse matrix by the generally more practical procedure of elementary
row operations.

CHAPTER 16 DETERMINANTS AND INVERSE MATRICES 73
Section 16.8 explains how a general linear system of n equations in n unknowns with a non-singular
coefficient matrix can be solved by means of Cramers rule (Theorem 16.8.1). This result is of theoretical
interest, but for systems with more than 4 unknowns, the calculation of the appropriate determinants usually
becomes computationally impractical. Finally, whereas a linear system of n equations in n unknowns is
usually soluble only when its coefficient matrix is non-singular, a system of homogeneous equations (with
zero right-hand sides) has a non-trivial solution iff the coefficient matrix is singular. This is Theorem 16.8.2.
To conclude the chapter, Section 16.9 gives a brief introduction to the Leontief model.

16.1
3 0
2. See Fig. M16.1.2. The shaded area is 18 = .
2 6
(2, 6)
(3, 0)
Figure M16.1.2

a11 b11 + a12 b21 a11 b12 + a12 b22
4. The matrix product is AB = , implying that
a21 b11 + a22 b21 a21 b12 + a22 b22
|AB| = (a11 b11 + a12 b21 )(a21 b12 + a22 b22 ) (a11 b12 + a12 b22 )(a21 b11 + a22 b21 ). On the other hand,
|A||B| = (a11 a22 a12 a21 )(b11 b22 b12 b21 ). A tedious process of expanding each expression, then
cancelling four terms in the expression of |A||B|, reveals that the two expressions are equal.
6. We write the system as
Y C = I 0 + G0
bY + C = a
Then Cramers rule yields

I0 + G0 1 1 I 0 + G0

a 1 a + I 0 + G0 b a a + b(I0 + G0 )
Y = = , C= =
1 1 1b 1 1 1b

b 1 b 1
The expression for Y is most easily found by substituting the second equation into the first, and then
solving for Y . Then use C = a + bY to find C.
16.2

1 1 1
2. AB = 7 13 13 , |A| = 2, |B| = 3, |AB| = |A| |B| = 6
5 9 10

4. By Sarrus rule the determinant is (1 + a)(1 + b)(1 + c) + 1 + 1 (1 + b) (1 + a) (1 + c), which

reduces to the given expression.
6. (a) Substituting T = d + tY into the expression for C gives C = a bd + b(1 t)Y . Substituting for
C in the expression for Y then yields Y = a + b(Y d tY ) + A0 . Then solve for Y , T , and C in turn
to derive the answers givenin (b) below.
1 1 0 Y A0

(b) We write the system as b 1 b
C = a . Then Cramers rule yields
t 0 1 T d

A0 1 0 1 A0 0

a 1 b b a b

d 0 1 a bd + A0 t d 1 a bd + A0 b(1 t)
Y = = , C= =
1 1 0 1 b(1 t) 1 1 0 1 b(1 t)

b
1 b b 1 b
t 0 1 t 0 1

1 1 A0

b 1 a

t 0 d t (a + A0 ) + (1 b)d
T = =
1 1 0 1 b(1 t)

b 1 b

t 0 1
(This problem is meant to train you in using Cramers rule. Note how systematic elimination is much
more efficient.)
16.3
2. +a12 a23 a35 a41 a54 (Four lines between pairs of boxed elements rise as one goes to the right.)
a b
11 a12 . . . a1n 11 b12 . . . b1n
0 a22 . . . a2n 0 b22 . . . b2n
4. With A = ... .. .. ..
and B = ... .. .. .. ,
. . . . . .
0 0 . . . ann 0 0 . . . bnn
the product AB is easily seen to be upper triangular, with the elements a11 b11 , a22 b22 , . . . , ann bnn on the
main diagonal. The determinant |AB| is, according to (4), the product of these n numbers. On the other
hand, |A| = a11 a22 ann , and |B| = b11 b22 bnn , so the required equality follows immediately.
16.4

2 1 1

2. A = 1 0 2 , |A| = |A | = 2
3 1 5
4. |AB| = |A||B| = 12, 3|A| = 9, | 2B| = (2)3 (4) = 32, |4A| = 43 |A| = 4 3 3 = 192,

1 0 0
|A| + |B| = 1, whereas |A + B| is not determined. (As an example showing this, let A1 = 0 1 0
0 0 3


1 0 0
and B1 = 0 1 0 . Then |A1 | = 3, |B1 | = 4, and |A1 + B1 | = 4. On the other hand, if
0 0 4
3 0 0 2 0 0

A2 = 0
1 0 and B2 = 0 2 0 , then |A2 | = 3, |B2 | = 4, and |A2 + B2 | = 10.)
0 0 1 0 0 1
6. (a) The first and the second columns are proportional, so the determinant is 0 by part E of Theorem 16.4.1.
(b) Add the second column to the third. This makes the first and third columns proportional.
(c) The term x y is a common factor for each entry in the first row. If x 6= y, the first two rows are
proportional. If x = y, the first row has all elements 0. In either case the determinant is 0.
8. By Sarruss rule, for example, |Aa | = a(a 2 + 1) + 4 + 4 4(a 2 + 1) a 4 = a 2 (a 4), so |A1 | = 3
and |A61 | = |A1 |6 = (3)6 = 729.
10. (a) Because A2 = In it follows from part G of Theorem 16.4.1 that |A|2 = |In | = 1, and so |A| = 1.
(b) Direct verification by matrix multiplication.
(c) (In A)(In + A) = In2 + A A A2 = In A2 = 0 A2 = In .
12. Add each of the last n 1 rows to the first row. Each element in the first row then becomes na + b. So

1 1 ... 1

a a+b ... a
Dn = (na + b) .. .. .. ..

. . . .

a a ... a + b
Next, add the first row multiplied by a to all the other n 1 rows. The result is an upper triangular
matrix whose diagonal elements are 1, b, b, ..., b, with product equal to bn1 . The conclusion follows
easily.
16.5
2. (a) abc (b) abcd (c) 6 4 3 5 1 = 360
16.6
2. Multiply the two matrices to get I3 .
1
x 2 3 3 4 3 3 3
4. (a) = = =
y 3 4 5 3 2 5 1

x 4 3 8 1 x 0
(b) = = (c) =
y 3 2 11 2 y 0

0 1 1 1 1 2
2
6. (a) |A| = 1, A = 1 1 2 , A = 2 2 3 . Direct verification yields A3 2A2 + A I = 0.
3
1 1 1 1 2 2
(b) The equality shown in (a) is equivalent to A(A I)2 = I, so A1 = (A I)2 .

0 0 1
(c) Choose P = (A I)1 = 1 0 1 , so that A = [(A I)2 ]1 = P2 . The matrix P also works.
0 1 0

8. (a) A2 = (PDP1 )(PDP1 ) = PD(P1 P)DP1 = PDIDP1 = P(D)2 P1 .

(b) Suppose the formula is valid for m = k. Then Ak+1 = AAk = PDP1 (PDk P1 ) = PD(P1 P)Dk P1
= PDIDk P1 = PDDk P1 = PDk+1 P1 .
10. (a) If C2 + C = I, then C(C + I) = I, and so C1 = C + I = I + C.

(b) Because C2 = I C, it follows that C3 = C2 C = (I C)C = C C2 = C (I C) = I + 2C.
Moreover, C4 = C3 C = (I + 2C)C = C + 2C2 = C + 2(I C) = 2I 3C.

s + 17 4t 16 0
1
12. (a) AT = 12 2s + 10 5t 8 0 = I3 , and so T = A1 , for s = 5, t = 4.
3s + 15 4t 16 12
(b) BX = 2X + C BX 2X = C (B 2I)X = C. But it is easy to see that B 2I = A.
So BX = 2X + C AX = C X = A1 C = TC, when s = 5 and t = 4. Hence,

5 4 3 2 3 0 1 1/2 0 0 1/6
1
X = 12 7 8 31 0 3 1 = 1/2 3 3 1/6
1 4 3 0 5 4 1 1/2 1 2 1/6
16.7
2. The determinant is 72. The cofactors are

0 3 6 3 6 0
C11 = = 3, C12 = = 18, C13 = =6
1 1 4 1 4 1

3 2 2 2 2 3
C21 = = 5, C22 = = 6, C23 = = 14
1 1 4 1 4 1

3 2 2 2 2 3
C31 = = 9, C32 = = 18, C33 = = 18
0 3 6 3 6 0

C C21 C31 3 5 9
1 11 1
so the inverse is C12 C22 C32 = 18 6 18 .
|A| 72
C13 C23 C33 6 14 18
4. Let B denote the n p matrix whose ith column has the elements b1i , b2i , . . . , bni . The p systems of
n equations in n unknowns can be expressed as AX = B, where A is n n and X is n p. Following
the method illustrated in Example 2, exactly the same row operations that transform the n 2n matrix
(A : I) into (I : A1 ) will also transform the n (n + p) matrix (A : B) into (I : B ), where
B is the matrix with elements bij . (In fact, because these row operations are together equivalent to
premultiplication by A1 , it must be true that B = A1 B.) When k = r, the solution to the system is
.
x1 = b1r , x2 = b2r , . . . , xn = bnr
16.8
2. The determinant of the system is equal to 10, so the solution is unique. The determinants in (2) are here

b1 1 0 3 b1 0 3 1 b1

D1 = b2 1 2 ,
D2 = 1 b2 2 ,
D3 = 1 1 b2
b 3 1 2 b3 1 2 3 b3
3

Expanding each of these determinants by the column (b1 , b2 , b3 ), we find that D1 = 5b1 + b2 + 2b3 ,
D2 = 5b1 3b2 6b3 , D3 = 5b1 7b2 4b3 . Hence, x1 = 21 b1 101
b2 51 b3 , x2 = 21 b1 + 10
3
b2 + 53 b3 ,
7
x3 = 21 b1 + 10 b2 + 25 b3 .
16.9
2. (a) No sector delivers to itself. (b) The total amount of good i needed to produce one unit of each good.
(c) This column vector gives the number of units of each good which are needed to produce one unit of
good j . (d) No meaningful economic interpretation. (The goods are usually measured in different units,
so it is meaningless to add them together.)
4. The Leontief system for this three-sector model is as follows:
0.9x1 0.2x2 0.1x3 = 85
0.3x1 + 0.8x2 0.2x3 = 95
0.2x1 0.2x2 + 0.9x3 = 20
which does have the claimed solution.
6. The quantity vector x0 must satisfy () (In A)x0 = b and the price vector p0 must satisfy ()
p0 (In A) = v . Multiplying () from the right by x0 yields v x0 = (p0 (In A))x0 = p0 ((In A)x0 ) =
p0 b.

2. (a) 4 (b) 1. (Suggestion: Subtract row 1 from rows 2 and 3. Then subtract twice row 2 from row 3.
The resulting determinant has only one non-zero term in its third row.) (c) 1. (Suggestion: Use exactly
the same row operations as in (b).)
4. (a) |At | = t + 1, so At has an inverse iff t 6= 1. (b) Multiplying the given equation from the right by
0 0 1
A1 yields BA1 + X = I3 . Hence X = I3 BA1 = 0 0 1 .
2 1 0

2 4 t

6. The determinant of the coefficient matrix is 3 1 t = 5t 2 45t + 40 = 5(t 1)(t 8).
t 2 7 4
So by Cramers rule, there is a unique solution iff t 6= 1 and t 6= 8.
8. (a) (In + aU)(In + bU) = In2 + bU + aU+ abU2 = In + (a + b + nab)U, because U2 = nU, as is easily
4 3 3

verified. (b) Note first that 3 4 3 = I3 + 3U. Moreover, (a) implies that
3 3 4
4 3 3
(I3 + 3U)(I3 + bU) = I3 + (3 + 10b)U = I3 for b = 3/10. Thus the inverse of 3 4 3 is
3 3 4
7 3 3
1
I3 (3/10)U = 3 7 3 .
10
3 3 7

2 11 6
10. (a) |A| = 2. A 2I2 = = A, so A2 + cA = 2I2 if c = 1.
18 10
(b) If B2 = A, then |B|2 = |A| = 2, which is impossible.

12. (a) For a 6 = 1 and a 6= 2, there is a unique solution. If a = 1, there is no solution. If a = 2, there are
infinitely many solutions. (These are a special case of the results in (b).)
(b) Using Gaussian elimination (where we start by interchanging rows 1 and 3), we obtain the matrix

1 0 3 b3
0 1 a2 + 6 b2 2b3
0 0 a 2 + 3a 2 b1 b2 + (2 a)b3
The last row corresponds to the equation (a 2 + 3a 2)z = b1 b2 + (2 a)b3 . If a 2 + 3a 2 =

(a 1)(a 2) 6= 0, i.e. if a 6= 1 and a 6 = 2, then the system has a unique solution. When a = 1 and
b1 b2 + b3 = 0, or when a = 2 and b1 = b2 , there are infinitely many solutions. Otherwise there is no
solution.
14. Using Gaussian elimination, one can show that there is a unique solution when a 6= 0 and b 6 = 2. The
solution is:
2b 2a + 3 2a + b 9 2ab 2a 2b 3 7 2a
x= , y= , z= , u=
a(b 2) b2 a(b 2) b2

CHAPTER 17 LINEAR PROGRAMMING 79
Chapter 17 Linear Programming

Chapter 17 presents some of the main ideas behind linear programming (LP). But it does not cover the simplex
method. This we regard as an important computational technique that is suitable for a specialist course, but
not really essential knowledge for most economists.
Section 17.1 begins by considering problems with only two choice variables, which can therefore be
solved graphically. Example 1 is a detailed discussion of a simple introductory bakers problem; Example 2
moves rather faster. Then, after a brief discussion of when graphical techniques will work, the latter part of
the section presents the general LP problem. It also gives an n-dimensional geometric interpretation of such
problems, enabling the basic idea of the simplex method to be visualized.
Duality theory is introduced in Section 17.2. Example 1 revisits the bakers problem and explicitly
calculates the marginal profit from relaxing each of the three main constraints. The resulting numbers (which
will be interpreted as shadow prices later) are then used to verify algebraically the solution found by geometric
techniques. Then it is shown that, in this same example, these shadow prices solve the dual problem. This is
set up as a problem facing somebody who wants to pay as little as possible to take over the resources available
to the baker, while offering just enough to ensure that the baker is compensated sufficiently for these resources.
The resulting prices then enable the baker to earn just as much profit as by continuing in business. After the
example, it is shown how to set up the dual of a general LP problem. Finally, the concise matrix formulations
of the primal and dual problems are presented.
Section 17.3 begins with a further discussion of the bakers problem. This time, the point is made that
any price vector that is feasible for the dual problem must give a value for the resources that is not less than the
profit earned from any activity vector that is feasible for the primal. Indeed, this is stated as Theorem 17.3.1,
and proved in general linear programs by a very easy argument (which does, however, involve recognizing
how the same double sum can be written in two different ways). Moreover, Theorem 17.3.2 shows that if
the value of primal and dual are made equal, then both problems have been solved. This has an even easier
proof. The section concludes with the Duality Theorem 17.3.3, claiming: (i) that if the primal has an optimal
solution, then so does the dual; (ii) the values of both primal and dual must be equal when both are solved;
and (iii) unless there is an upper bound in the value of the primal, the dual is infeasible. Proving (i) and (ii)
involves relatively advanced arguments (or at least an understanding of the simplex method) so the proofs are
omitted. An easy proof of (iii) is provided, however.
Section 17.4 provides a standard economic interpretation of any LP problem in terms of activity vectors
and scarce resources. It also interprets shadow prices as those ensuring that the shadow profit on any activity
is non-positive, and interprets the dual problem in rather the same way as was done in Section 17.2. Finally,
it is shown how shadow prices can be used to estimate changes in the optimized value of the primal from
changes in the resources.
The last topic of the chapter is complementary slackness, the subject of Section 17.5. The bakers problem
is used to illustrate how positive activity levels are associated with binding constraints in the dual, and positive
shadow prices with binding constraints in the primal. Then, for the case of a general primal with two variables
and three constraints, the part of the duality theorem claiming equality of the values of primal and dual is
invoked. It shows that the pattern observed in the bakers problem always holds. This complementary slackness
property is then stated and proved as Theorem 17.5.1. (The fairly long proof is in small print, especially as
it uses vector and matrix notation at the beginning. But it is not hard.) The standard interpretations of the
complementary slackness conditions are then presented. Example 1 is provided to show how, given the
solution to the dual problem with two variables (which can be found graphically, of course), complementary
slackness can be used to identify the set of binding constraints in the primal, and also the activity levels which

80 CHAPTER 17 LINEAR PROGRAMMING
must be zero. Then, of course, the primal can be solved from simultaneous equations. This is an important
technique for solving (and setting) examination questions! Next, it is shown that the KuhnTucker conditions
discussed in Chapter 14 can be applied to LP problems, and imply complementary slackness. This is important
in linking the results of Chapters 17 and 14. Finally, there is a brief discussion of what happens when some
constraints in the primal must hold with equality. Then the signs of the corresponding dual variables are no
longer constrained to be nonnegative.
17.1
2. (a) A graph shows that the solution is at the intersection of the lines 2x1 + 3x2 = 6 and x1 + x2 = 5.
Hence max = 98/5 for (x1 , x2 ) = (9/5, 16/5).
(b) The solution satisfies 2x1 + 3x2 = 13 and x1 + x2 = 6. Hence max = 49 for (x1 , x2 ) = (5, 1)
(c) The solution satisfies x1 3x2 = 0 and x2 = 2. Hence max = 10/3 for (x1 , x2 ) = (2, 2/3)
4. (a) No maximum exists. Consider Fig. M17.1.4. By increasing c, the dashed level curve x1 + x2 = c
moves to the north-east and so this function can take arbitrarily large values.
(b) Yes, the maximum is at P = (1, 0). The level curves are the same as in (a), but the direction of
increase is reversed. x2
x1 + x2 = 1
4
x 1 + x2 = c
x1 + 3x2 = 3
3
x1
1 1 2 3 4 5 6
1
Figure M17.1.4

3x + 5y 3900
6. The LP problem is: max 700x + 1000y subject to x + 3y 2100 x 0, y 0

2x + 2y 2200
A figure showing the admissible set and an appropriate level line for the criterion function will show that
the solution is at the point where the two lines 3x + 5y = 3900 and 2x + 2y = 2200 intersect. Solving
these equations yields x = 800 and y = 300. The firm should produce 800 sets of type A and 300 of
type B.
17.2
u1 + 2u2 + u3 8
2. min 8u1 + 13u2 + 6u3 subject to u1 0, u2 0, u3 0
2u1 + 3u2 + u3 9
17.3

10x1 + 25x2 10 000
2. max 300x1 + 500x2 subject to x1 0 , x2 0
20x1 + 25x2 8 000
The solution can be found graphically. It is x1 = 0, x2 = 320, and the value of the criterion function is
160 000, the same value found in Example 17.1.2 for the optimal value of the primal criterion function.

CHAPTER 17 LINEAR PROGRAMMING 81
17.5
2. (a) Figure M17.5.2 shows the relevant portion of the feasible set with the different constraints numbered,
as well as two dashed level curves for the criterion function Z = y1 + 2y2 . We see that the minimum is
attained at the point (y1 , y2 ) = (3, 2).
y2
(4)
10
(3)
5
(2) y1 + 2y2 = Z0
(3, 2)
(1)
y1
5 10 15
Figure M17.5.2
x1 + x2 x3 + x4 1
(b) The dual: max 15x1 + 5x2 5x3 20x4 s.t. , xj 0, j = 1, . . . , 4.
6x1 + x2 + x3 2x4 2
Because y1 and y2 are both positive, both constraints are satisfied with equality. Furthermore, the two last
constraints in the primal problem are satisfied with strict inequality at the optimum. Hence x3 = x4 = 0.
The system therefore reduces to the system x1 + x2 = 1, 6x1 + x2 = 2, so x1 = 1/5 and x2 = 4/5.
The maximum is thus at (x1 , x2 , x3 , x4 ) = (1/5, 4/5, 0, 0).
(c) If the constraint is changed to y1 + 6y2 15.1, the solution to the primal is still at the intersection of
the lines (1) and (2) in Fig. M17.5.2, but with (1) shifted up slightly. The solution to the dual is completely
unchanged. The optimal values in both problems increase by (15.1 15) x1 = 0.02.
4. (a) For x3 = 0, the solution is x1 = x2 = 1/3. For x3 = 3, the solution is x1 = 1 and x2 = 2.
(b) Let zmax denote the maximum value of the objective function. If 0 x3 7/3, then zmax (x3 ) =
2x3 + 5/3 for x1 = 1/3 and x2 = x3 + 1/3. If 7/3 < x3 5, then zmax (x3 ) = x3 + 4 for x1 = x3 2
and x2 = 5 x3 . If x3 > 5, then zmax (x3 ) = 9 for x1 = 3 and x2 = 0. Because zmax (x3 ) is increasing,
the maximum is 9 for x3 5.
(c) The solution to the original problem is x1 = 3 and x2 = 0, with x3 as an arbitrary number 5.

x1 + 2x2 16

x1 2x2 6
2. (a) max x1 + x2 s.t. x1 0, x2 0

2x 1 x2 8

4x1 5x2 15
The maximum value is 8 at (x1 , x2 ) = (0, 8).
(b) (y1 , y2 , y3 , y4 ) = ( 21 (b + 1), 0, b, 0) for any b satisfying 0 b 1/5.
(c) The maximand for the dual becomes kx1 + x2 . The solution is unchanged provided that k 1/2.

MEA Solutions

Загружено:

Сведения о документе

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

MEA Solutions

Загружено:

Авторское право:

Доступные форматы

8 CHAPTER 3 INTRODUCTORY TOPICS III: MISCELLANEOUS

Answers to Even-Numbered Problems

Using the properties of sums and the definition of aj , we have

Knut Sydster and Peter Hammond 2006

2. F B C is the set of all

4. (a) B M (b) F B C 6= (c) T B = (d) F \ (T C) B

Knut Sydster and Peter Hammond 2006

Review Problems for Chapter 3

Knut Sydster and Peter Hammond 2006

Chapter 4 Functions of One Variable

Knut Sydster and Peter Hammond 2006

Answers to Even-Numbered Problems

Figure M4.2.12a The area Figure M4.2.12b The area

14. (a) Defined for x 6 = 2, i.e. Df = (, 2) (2, ) (b) f (8) = 5

See Fig. M4.3.4.

See Fig. M4.3.6.

Knut Sydster and Peter Hammond 2006

Figure M4.3.4 Figure M4.3.6

Figure M4.4.2a Figure M4.4.2b Figure M4.4.2c

4. L1 is y = x + 2, with slope 1; L2 is y = 53 x + 3, with slope 3/5; L3 is y = 1, with slope 0;

8. (a) See Figs. M4.4.8a, M4.4.8b, and M4.4.8c.

Figure M4.4.8a Figure M4.4.8b Figure M4.4.8c

Knut Sydster and Peter Hammond 2006

Knut Sydster and Peter Hammond 2006

2x 1/8 1/4 1/2 1 2 4 8

2x 8 4 2 1 1/2 1/4 1/8

Knut Sydster and Peter Hammond 2006

Review Problems for Chapter 4

Knut Sydster and Peter Hammond 2006

Chapter 5 Properties of Functions

Answers to Even-Numbered Problems

Figure M5.1.2a Figure M5.1.2b Figure M5.1.2c

Knut Sydster and Peter Hammond 2006

Figure M5.2.2a Figure M5.2.2b Figure M5.2.2c

4. f (f (x)) = f (3x + 7) = 3(3x + 7) + 7 = 9x + 28. Then 9x + 28 = 100 for x = 8.

Figure M5.3.8a Figure M5.3.8b Figure M5.4.2

10. (a) x = ln y 4, y > 0 (b) x = ey+4 , y (, ) (c) x = 3 + ln(ey 2), y > ln 2

Knut Sydster and Peter Hammond 2006

Figure M5.4.4 Figure M5.5.2 Figure M5.R.10

Review Problems for Chapter 5

Knut Sydster and Peter Hammond 2006

Knut Sydster and Peter Hammond 2006

Answers to Even-Numbered Problems

Knut Sydster and Peter Hammond 2006

4. (a) 5 (b) 1/5 (c) 1 (d) 2 (e) 3x 2 (f) h2

Knut Sydster and Peter Hammond 2006

Review Problems for Chapter 6

Knut Sydster and Peter Hammond 2006

Chapter 7 Derivatives in Use

Answers to Even-Numbered Problems

Knut Sydster and Peter Hammond 2006

Figure M7.3.2 Figure M7.4.10

Knut Sydster and Peter Hammond 2006

and the answer is correct to 7 decimal places.

T 36.3 35.0 33.9 32.4 24.7 24.2

ln n 5.04 4.89 4.70 4.54 3.64 3.58

10. (a) Elx A = (x/A)(dA/dx) = 0

Knut Sydster and Peter Hammond 2006

Figure M7.8.2a Figure M7.8.2b

4. See Fig. M7.8.4; y is discontinuous at x = a.

Knut Sydster and Peter Hammond 2006

4. (a) y = x 1 (x = 1 is a vertical asymptote). (b) y = 2x 3 (c) y = 3x + 5 (x = 1 is a vertical

Review Problems for Chapter 7

Knut Sydster and Peter Hammond 2006