Pemberton and Rau
Pemberton and Rau
Pemberton and Rau
LINEAR EQUATIONS
q S = cp + d.
15 3
11 31
= 5, b = 31 + 54 = 51, c =
= 3 and d = 3 38 = 21,
Then a =
84
12 8
so
q D = 5p + 51, q S = 3p 21.
51 + 21
= 9,
3+5
q = 39 21 = 6.
3y 2z =
2
11y + 5z =
4
11y + 5z = k 6.
When k = 4, the last two equations are inconsistent and the system has no solution.
When k = 2, the last two equations are identical and hence the third equation can be
dropped. Then assigning z = s and solving for y and then x gives the solution as
x=
5s + 4
7s + 10
, y=
, z = s.
11
11
13. Substitute the expression for T into that for C and the resulting expression for C into
that for Y . Solving the resulting equation for Y gives
Y = 3.33 + 2.78(I + G), C = 3.33 + 1.78(I + G), T = 1.67 + 0.56(I + G).
If G increases by x units, Y , C and T increase by 2.78x, 1.78x and 0.56x respectively.
14. The gross outputs x, y of X, Y satisfy
x 0.1x 0.2y = a, y 0.7x 0.4y = b.
These equations lead to x = 1.5a+0.5b, y = 1.75a+2.25b. Since a and b are assumed to
be positive numbers, and their coefficients in the equations for x and y are all positive,
x and y are positive.
LINEAR INEQUALITIES
21. Let
A = c0 c1 t0 , B = I + G,
c = c1 (1 t1 ).
Substituting the expression for T into that for C gives C = A + cY . Substituting this
expression for C into that for Y gives Y = A + B + cY . Hence
Y =
A+B
,
1c
C =A+
c
A + cB
(A + B) =
,
1c
1c
1
T = t0 +
t1
(A + B).
1c
The answers to the last two parts are Yes and No. Since c is the product of two
1
> 1. If G
numbers which are strictly between 0 and 1, 0 < c < 1, whence
1c
increases by x units, where x > 0, B increases by x units with A remaining unchanged,
cx
x
> x units. C increases by
units, which is less than x if
so Y increases by
1c
1c
1
and only if c < 1 c, i.e. c < 2 .
22. The budget line has equation p1 x1 + p2 x2 = m when x1 z. Hence the budget line has
slope p1 /p2 to the left of z. Since the price of good 1 is p1 + t for all consumption in
excess of z, the budget line has slope (p1 + t)/p2 to the right of z.
(i) When t < 0, the budget line is less steep to the right of z than to the left.
(ii) When the consumption of good 1 is rationed at z, the budget line becomes vertical
at z.
23.
(i) Total usage of labour is 7x + 6y, where x and y are the gross outputs of of X and
Y respectively. Using the expressions for x and y given in the answer to Problem
14, total usage of labour is
7(1.5a + 0.5b) + 6(1.75a + 2.25b) = 21a + 17b.
Similarly, total usage of land is
3(1.5a + 0.5b) + 2(1.75a + 2.25b) = 8a + 6b.
(ii) Using the answer to (i), the conditions are the labour constraint 21a + 17b 800,
the land constraint 8a + 6b 300 and the non-negativity constraints a 0, b 0.
The corners of the feasible set in the abplane are (0, 0), (37.5, 0), (30, 10) and
(0, 47.06).
24. Let x and y be the amounts of FB and KC consumed each day by Oleg. Then the cost
minimisation programme is to minimise 2x + y subject to
10x + 4y 20, 5x + 5y 20, 2x + 6y 12, x 0, y 0.
Once the feasible set has been drawn, it is clear that costs are minimised at the intersection of the calcium and protein borders. Here the slope of the isocost lines (2) lies
between the slopes of the two borders ( 52 and 1). The required point of intersection
is ( 23 , 10
3 ) and the least cost is
2 10
14
2 +
= .
3
3
3
(i) The slope of the isocost lines is still 2 so the optimal combination is still ( 23 , 10
3 );
the least cost is now 28
.
3
(ii) The slope of the isocost lines is now 23 which still lies between 25 and 1. So
26
the optimal combination is still ( 23 , 10
3 ); the least cost is now 3 .
(iii) The slope of the isocost lines is now 3, so these lines are now steeper than the
calcium border and the optimal combination is (0, 5). The least cost is now 5.
The solution is not unique when the isocost lines are parallel to one of the borders.
Denoting the prices of FB and KC by p1 and p2 respectively, this will happen when
p1 /p2 is 25 , 1 or 13 .
31. f (g(x)) = 1 if g(x) 1, which happens if and only if |x| 1. If |x| < 1 then
g(x) = x2 < 1, so f (g(x)) = |x2 | = x2 .
g(f (x)) = 4 if f (x) 2, which happens if and only if x 2. If 2 < x < 1 then
f (x) = |x| < 2, so g(f (x)) = |x|2 = x2 . If x 1 then f (x) = 1, so g(f (x)) = 12 = 1.
Summarising,
f (g(x)) =
x2
1
4 if x 2,
g(f (x)) = x2 if 2 < x < 1,
1 if x 1.
if |x| < 1,
if |x| 1;
32. The function takes the form Y = X 2 . Hence the graph in the XY plane is U-shaped
with the bottom of the U at the origin. When X = 0, x = 3; when Y = 0, y = 2.
Therefore (3, 2) in the xyplane corresponds to (0, 0) in the XY plane, and the graph
in the xyplane is U-shaped with the bottom of the U at (3, 2).
A similar argument shows that the second graph is V-shaped with a right angle at
(2, 5).
3-3.
X + s(P X)
F (X) = X
X t(X E)
if 0 X < P ,
if P X < E,
if X E.
The graph consists of three line segments: the first has slope 1 s and intercept
sP , the second starts at the right end of the first and has slope 1, the third starts
at the right end of the second and has slope 1 t.
if 0 X < E,
X + s(P X)
F (X) = X t(X E) + s(P X) if E X < P ,
X t(X E)
if X P .
The graph consists of three line segments: the first has slope 1 s and intercept
sP , the second starts at the right end of the first and has slope 1 s t, the third
starts at the right end of the second and has slope 1 t.
(iii) When E < P and s + t > 1, the graph is similar to that in (ii) except that the
middle segment now has negative slope.
34.
y = f (x)
y = f (g(x))
y = g(x)
y = g(f (x))
T
41. The graph of the first equation is -shaped with vertex at (0, 5); the graph of the second
equation is a straight line with slope 2 and intercept 3. Eliminating q between the two
equations gives 5p2 = 2p3, i.e. p2 +2p8 = 0. This factorises to (p+4)(p2) = 0 so
p is 4 or 2. When p = 4, substituting back into either of the original two equations
equation gives q = 11; similarly, when p = 2, q = 1.
(i) Suppose K and L both increase by 1%. Let old value of Y be Y0 , new value Y1 .
Then
Y1 = 2(1.01 a)2/3 (1.01 b)1/3 = 2a2/3 b1/3 (1.01)2/3 (1.01)1/3 = Y0 1.01,
so Y increases by 1%. Similar argument holds if 1.01 is replaced by 1.10 or by
x
1+
for any x > 0 (or, more generally, any x > 100). Thus if K and L both
100
increase by 10% (or x%), then Y increases by 10% (resp. x%).
(ii) Y = 18L1/3 , log Y = c + 13 log L, where c = log 18 = 1.255 to 3 decimal places.
Here, as in (iii) below, logarithms are to base 10.
log Y
Y
36
18
3c
1
8
L
p
(iii) K = 1000/L, log K = g 21 log L, where g =
log L
1
2
1.5
10
0
10
log L
4-4. The graph of q = 41 p4 is a U shape with the bottom of the U at the origin. The part
in the non-negative quadrant is the graph of the supply function. Since p is positive,
1/p decreases as p increases; hence q = 8p1 decreases as p increases. The equilibrium
occurs when 14 p4 = 8p1 , i.e. p5 = 32; thus the equilibrium price and quantity are 2
and 4.
The supply and demand functions in log-linear form are
log q = log 4 + 4 log p,
1
5
log 25 = log 2
51. un = 3
p
N
102
103
103
104
104
105
105
106
3
. When n is large, the term n2 is negliglible; so for large n the sequence
n + 3n2
{vn } behaves similarly to {3/n}. Therefore vn 0 as n .
vn =
5-2.
r =
12s
1+
12
12
1 = (1 + s)12 1.
When r = 0.20, s = 0.0153 and when r = 0.25, s = 0.0188. So the interest rate
per month increases from 1.53% to 1.88%.
5-3.
(i) From the second equation, the further increase of c1 units in Y causes a further
increase of c21 units in C which then, by the first equation, causes a further increase
of c21 units in Y. Continuing in this way, the total increase in Y is 1 + c1 + c21 + . . .
Since 0 < c1 < 1, we may apply the geometric series formula: the sum is 1/(1c1 ).
(ii) Substituting the expression for C into that for Y and solving the resulting equation
for Y , we have
c0 + I + G
c0 + c1 (I + G)
Y =
, C=
.
1 c1
1 c1
From the expression for Y , if G increases by one unit, Y will increase by 1/(1c1 ),
which agrees with the answer obtained in (i).
5-4.
(i) The profit obtained at time T is pf (T ), so the value of the forest at time 0 is
pf (T )/(1 + r)T .
(ii) The value of the forest at time 0 is
pf (T )
pf (T )
pf (T )
+
+
+ ...
T
2T
(1 + r)
(1 + r)
(1 + r)3T
This is a GP with first term xpf (T ) and common ratio x, where x = (1 + r)T .
Since 0 < x < 1, the sum is xpf
(T )/(1 x). The value of the forest at time 0 is
therefore pf (T ) (1 + r)T 1 .
6
61.
INTRODUCTION TO DIFFERENTIATION
dy
dy
= 3x2 . If x = 2 then y = 8 and
= 12, so the equation of the tangent is
dx
dx
y 8 = 12(x 2),
or y = 12x 16.
= h2 (6 + h).
If h is small then h2 is very small and 6 + h 6, so LHS RHS is indeed very small.
If x = 2 + h, the value of the function is (2 + h)3 and the value of y given by the
tangent is 8 + 12h. The error of approximation is the same as RHS LHS in the small
increments formula and is therefore equal to h2 (6+ h). The ratio of the absolute value
of the error to the true value of the function is
h2 (6 + h)
,
(2 + h)3
which is
6.01
7
if h = 1.
104 if h = 0.01,
3
2.01
27
The required percentages are (i) 0.0074%, (ii) 25.93%.
62. Let f (x) = x5 + 3x 12. Then f (x) = 5x4 + 3, which is positive for all x. The curve
y = f (x) is therefore upward-sloping, with slope 3 at the point (0, 12) and the slope
increasing as we move away from the yaxis in either direction. Thus the curve cuts
the xaxis exactly once, at a point (a, 0) such that a > 0. One can see without using
a calculator that a is slightly less than 1.5: for if x = 1.5 then x5 = 243/32, which is
slightly greater than 7.5, while 3x 12 = 7.5, so f (x) is positive but very small.
The curve y = f (x) has the same general U-shape as the curve y = x4 , but with the
vertex at the point (0, 3). [The shapes of the power functions were introduced without
explanation in Chapter 4, but we can now see why the curve y = x4 looks as it does.
Obviously the curve passes through the origin. Also, since the slope is 4x3 , the curve
is downward-sloping where x < 0, upward-sloping where x > 0, and the absolute value
of the slope increases as we move away from the origin in either direction.]
f (0) < 0 < f (0) and f (x) > f (x) if x is positive and very large, so the curves y = f (x)
and y = f (x) cross at least once. Since f (x) 0 < f (x) if x a, the crossing-point(s)
must be such that x > a. In fact,
f (x) f (x) = x5 5x4 + 3x 15 = (x4 + 3)(x 5),
which is zero if and only if x = 5. Thus there is exactly one crossing-point, namely the
point (5, 55 + 3).
y
y = f (x)
355
y = f (x)
12
2
3
2 0.01 =
1
q
p
=
, then
:
p0
100
q0
100
METHODS OF DIFFERENTIATION
dx v
dv v
dx
v dx
Writing
u
1
= u , we have by the product rule:
v
v
1 du
d 1
u dv
1
dv
du
1 du
d u
=
+u
2
= 2 v
u
=
.
dx v
v dx
dx v
v dx v dx
v
dx
dx
dv
1
v
= (2 + t)2/3 1 =
.
dt
3
3(2 + t)
11 + 5t
dQ
=
.
dt
3(5 + 2t)1/2 (2 + t)2/3
73.
(i) y 1/3 = cx1/2 , so y = c3 x3/2 . dy/dx = bx5/2 , where b is the positive constant
3c3 /2.
(ii) The equation of a typical isoquant is
K 1/2 L1/3 = c
(K, L > 0)
= 0.7 +
+
,
Y
Y
Y (Y + 2)
Y
Y +2
using the Hint, which enables us to check that C/Y is a monotonic function of Y without
further messy differentiation. In fact, the graph against Y of C/Y (the APC), like that
of dC/dY (the MPC), is downward-sloping, with the same horizontal asymptote: both
APC and MPC tend to 0.7 as Y . The vertical axis is also an asymptote of APC.
C
APC
MPC
2.20
1.95
APC
0.95
0.70
0.80
0.30
0
MPC
81. dy/dx = 5x4 (2 x)4 4x5 (2 x)3 = x4 (2 x)3 (10 9x); this is zero if x is 0, 10/9
or 2. So the critical points are (0, 0), (1.111, 1.057) and (2, 0).
When x = 0, dy/dx = (+)(+)(+) = +; when x = 0+, dy/dx = (+)(+)(+) = +;
hence (0, 0) is a critical point of inflexion. Similarly (1.111, 1.057) is a maximum point
and (2, 0) is a minimum point.
d2 y/dx2 = 20x3 (2 x)4 40x4 (2 x)3 + 12x5 (2 x)2
= x3 (2 x)2 20(4 4x + x2 ) + (80x + 40x2 + 12x2 ) .
10
2
Simplifying, d2 y/dx2 = 8x3 (2 x)2 (9x2 20x + 10) = 8x3 (2 x)2 ([3x 10
3 ] 9 ). It
follows that d2 y/dx2 is 0 at four values of x, namely 0, (10 10)/9, (10 + 10)/9
and 2, but changes sign only at the first three. So the points of inflexion are (0, 0),
(0.760, 0.599) and (1.462, 0.559). The function is
10
dz
tf (t) f (t)
f (t)
then
=
,
t
dt
t2
which is zero when tf (t) = f (t). However, the geometric argument given above provides
the easiest way of seeing that the critical point must be the global maximum.
83. From the production function,
L(Q K) = KQ.
Hence the isoquant Q = Q0 may be written
L=
KQ0
.
K Q0
=
dK
K Q0 (K Q0 )2
K Q0
and
2Q20
d2 L
=
.
2
dK
(K Q0 )3
From the production function, Q must be less than K if L > 0. Hence K Q0 > 0, so
d2 L/dK 2 > 0.
Since dL/dK < 0 and d2 L/dK 2 > 0, the isoquant is downward-sloping and convex. If
K is very large and Q = Q0 then L is close to Q0 . Similarly, if L is very large and
Q = Q0 then K is close to Q0 . Hence the asymptotes of the isoquant are the lines
K = Q0 and L = Q0 . Note also that the isoquant crosses the line L = K at the point
(2Q0 , 2Q0 ) and its slope at that point is 1. The diagram shows two isoquants, Q = Q0
and Q = Q1 , where Q0 < Q1 < 2Q0 .
11
Q1
Q0
0
84.
Q0 Q1
+
1200
= 23 x 100
3
3
100 100
6 18 < 0,
so AR is monotonic.
d
(px) = 36000 3600x + 150x2 2x3 , so
(ii) MR =
dx
d
MR = 3600 + 300x 6x2 = 6(x 20)(x 30).
dx
Thus MR is not monotonic, being a decreasing function of x for 0 x < 20 and
x > 30, and an increasing function for 20 < x < 30.
(iii) Both graphs meet the vertical axis at 36000. AR is monotonic decreasing with a
point of inflexion at x = 100/3, where the curve changes from convex to concave.
MR has a minimum at x = 20 and a maximum at x = 30. AR is always above
MR: this follows from the fact that dp/dx < 0. Both AR and MR are negative
for all sufficiently large x: AR = 0 when x = 60, MR is positive when x = 40 but
negative when x = 45.
9
91.
which always has the same sign as 2ax2 1. Let b = (2a)1/2 : the function is convex
for |x| > b, concave for |x| < b, and the points of inflexion occur where x = b.
Since ab2 = 41 , the points of inflexion are ((2a)1/2 , e1/4 ) and ((2a)1/2 , e1/4 ).
(iii) Bell-shaped, asymptotic to the xaxis, with global maximum given by (i) and
points of inflexion given by (ii).
92.
(i) Suppose the original sum of money is P . Then T is such that P erT = 2P . Therefore rT = ln 2, and
ln 2
100 ln 2
69
T =
=
,
r
R
R
since ln 2 = 0.69315 to 5 decimal places.
(ii) Let the APR be S%, and let s = 0.01S. By the answer to Exercise 9.2.1, ln(1 + s)
is what we called r in part (i) of this problem. Hence
T =
ln 2
S
A
S ln 2
ln 2
=
= , where A =
.
ln(1 + s)
S
ln(1 + 0.01S)
S
ln(1 + 0.01S)
The values of A for different values of S are given in the following table:
S
A
93
2
70.0
4
70.7
6
71.4
8
72.1
10
72.7
b0
13
y
b=2
b=1
b=
1
2
b=0
1
b = 1
1
2
b = 2
x
12
1
94.
(i) Let the value of the forest at time 0 be v(T ). Since the profit obtained at time T
is pf (T ), v(T ) = pf (T )erT . Therefore
v (T ) = pf (T )erT pf (T )rerT = (f (T ) rf (T ))perT ,
which is zero when f (T )/f (T ) = r.
(ii) In this case, let the value of the forest at time 0 be V (T ). Then
V (T ) = pf (T ))erT + pf (T )e2rT + pf (T )e3rT + . . . ,
the sum of the geometric progression whose first term is v(T ) (as defined in (i))
and whose common ratio is erT . Since 0 < erT < 1,
V (T ) =
v(T )
pf (T )
= rT
.
rT
1e
e 1
(iv) Differentiating v (T ) gives v (T ) = (f (T ) 2rf (T ) + r 2 f (T ))perT . Using condition (9.7) we see that, at the critical point, v (T ) = (f (T ) r 2 f (T ))perT .
Therefore, if the value of T for which v (T ) = 0 is in the region for which f is
concave, the critical value is a local maximum. If we assume further that there is
only one critical point, it must be the global maximum.
Similarly, differentiating V (T )/V (T ) gives
V (T ) 2 f (T )
f (T ) 2
d
V (T )
=
r
1 + (erT 1)1 .
V (T )
V (T )
f (T )
f (T )
dt
2
rerT
f (T )
V (T )
r 2 erT
=
rT
,
+ rT
V (T )
f (T )
e 1
(e 1)2
whence
V (T ) =
V (T )
r 2 erT V (T )
p
pr 2 erT
f (T )
= rT
f (T ) rT
f (T ).
rT
f (T )
e 1
e 1
(e 1)2
The argument concerning the global maximum is similar to that for v(T ).
10
APPROXIMATIONS
101.
(i) f (x) = 0 has at most one root. For if there were two distinct roots, say a and
b where a < b, then by Rolles theorem there would be a real number c, with
a < c < b, which is a root of f (x) = 0.
(ii) f (x) = 0 has at most two roots. For if there were three distinct roots, say a, b, c
where a < b < c, then by Rolles theorem there would be real numbers p and q,
with a < p < b < q < c, which are roots of f (x) = 0.
(iii) f (x) = 0 has at most three roots, by a similar argument to (i) and (ii).
General result: if f (x) = 0 has n distinct roots, then f (x) = 0 has at most n + 1 roots.
102.
(i) Let f (x) = x5 5x + 2; then f (x) = 5x4 5 and f (x) = 20x3 . f (x) = 0 when
x = 1; using f (x), it follows that (1, 6) is a maximum point and (1, 2) is a
minimum point. As x , f (x) ; as x , f (x) .
(ii) Since f (2) = 20 and f (1.5) 2, x1 is between 2 and 1.5. Since f (1) = 2
and f (1.5) 2, x3 is between 1 and 1.5. Taking 1.6 as the initial approximation
to x1 and using Newtons method, V (1.6) = 0.01749 and so the second approximation is 1.58251. Applying the method again, V (1.58251) = 0.00047, so
x1 = 1.582 to 3 decimal places. Similarly, taking 1.2 as the initial approximation
to x3 and carrying out 4 iterations yields the successive approximations 1.48161,
1.39093, 1.37258 and 1.37188; thus x3 = 1.372 to 3 decimal places.
15
(iii) The equation f (x) = 0 has 2 roots at x = 1. Hence, by the result of Exercise
10.2.3, the equation f (x) = 0 has at most 3 roots. On the other hand, f (2) =
20, f (1) = +6, f (1) = 2 and f (2) = +24. Hence, by the intermediate value
theorem, the equation f (x) = 0 has at least one root between 2 and 1, at least
one between 1 and 1, at least one between 1 and 2 and therefore at least three
roots in all. It follows that f (x) = 0 has exactly 3 roots.
103.
(i) Suppose g (x) = c for all x, where c is a constant. Let f (x) = g(x) cx; then
f (x) = 0 for all x, so f (x) is a constant, say b. Hence g(x) = b + cx for all x.
(ii) Let ln y = g(t). By assumption, g (t) is a constant, say c. By the result of (i) with
x replaced by t, there is a constant b such that g(t) = b + ct for all t. Let A = eb .
Then for all t,
y(t) = exp g(t) = exp(b + ct) = Aect .
104. Da = x/(1 x), Dc = ln(1 x). Since 0 < x < 1, the expressions given for Da and Dc
follow from the power series expansions for (1 x)1 and ln(1 x) respectively. Since
x > 0, the fact that Db < Dc < Da can be read from the coefficients in these series.
11
MATRIX ALGEBRA
111.
2 = 3 = 0,
3 = 0.
x 3 = 3 .
0 1
1
0
112. A =
, B=
. A is an anticlockwise rotation through a right angle,
1
0
0 1
B is a reflexion in the xaxis.
0 1
0 1
AB =
, BA =
. AB is the composite mapping consisting of
1 0
1
0
the reflexion B followed by the rotation A; this amounts to interchanging the two
coordinates. BA is the composite mapping consisting of the rotation A followed by
the reflexion B; this amounts to interchanging the two coordinates and then reflecting
in the origin.
113.
(ii) Here we can use forward substitution: solve for x1 , then for x2 and finally for
x3 . We have x1 = y1 /4, x2 = (3y1 4y2 )/8, x3 = (5y1 4y2 8y3 )/8.
(i)
y1 = x1 a11 x1 a12 x2 . . . a1n xn
12
121.
2
1
5 2 t
0 1/2 3/2 2 t/2 .
0
0
0 0 0
The system has been reduced to one in which the coefficient matrix is a Type 4
echelon matrix. The third equation has been reduced to 0 = 0 and hence may be
ignored. The system has a solution for all values of t.
The rank of A is the number of nonzero rows in the echelon form and hence is 2.
Similarly, the rank of [A b] is 2: notice that A and [A b] have the same rank.
1 6 7
0 3 1
0 0 0
0 0 0
3
1
2
0
5 1
4 1
.
1 2
0 t7
The system has been reduced to one in which the coefficient matrix is a Type 4
echelon matrix. The third equation has been reduced to 0 = t 7, so it may be
17
B1 O
For the second part, denote the two matrices by B and C. We may write B =
O B2
1
B1
O
1
where B1 and B2 are 22 matrices. Hence B1 =
, where B1
1 and B2
O B1
2
are calculated by the inversion formula. Also,
C1 0
C=
0 2
where C1 is a 33 matrix; hence
C1 =
0
C1
1
,
0 12
where C1
1 is obtained from C by GaussJordan. Performing the calculations,
3/19 2/19 0
0
1
9 4 0
2/19 5/19 0
0
, C1 = 2 15 7 0 .
B1 =
0
0
1/3 2/3
2 17 8 0
0
0
1/6 1/6
0
0 0 12
123.
(i) Since B is invertible, the only 2vector z with the given property is B1
a3
.
b3
(ii) If we assign the arbitrary value to x3 , the given system can be written as
x1
a3
B
=
.
x2
b3
This holds if and only if
x1
z
1 a3
1 a3
=B
= B
= 1 .
x2
b3
z2
b3
Hence the given vector equation holds if and only if
z1
x = z2 for some scalar .
1
18
()
(iii) A is singular if and only if there is a non-zero x such that Ax = 0. From (ii), the
first two equations of this system hold if and only if () is true; and, for x to be
non-zero, must be non-zero. It follows that A is singular if and only if this x
satisfies the third equation of the system. Therefore, A is singular if and only if
c1 z1 + c2 z2 c3 = 0 for some non-zero ,
which happens if and only if c1 z1 + c2 z2 = c3 .
(iv) If A is singular then c3 must be as in (iii). So if we replace the (3, 3) entry of A
by any number other than c3 , then A becomes invertible.
124. From Problem 114, Bx = y, where B = I A. If there is to be a unique x for any y,
then B must be invertible and x = B1 y.
In addition, it is given that y has non-negative components. To ensure that x has
non-negative components for every such y it is necessary that all entries of B1 be
non-negative. For suppose that B1 had some negative entry, say the (2, 3) entry. By
taking y to be the vector with third component 1 and zeros elsewhere, we see that the
second component of x is negative.
13
131.
(ii) Let A be a singular 22 matrix and let C be the matrix obtained by adding x
to each of its diagonal entries. Since det A = 0, det C = tx + x2 , where t is the
sum of the diagonal entries of A. If t = 0, det C > 0 for any non-zero x; if t 6= 0,
det C > 0 whenever x has the same sign as t; in each case, |x| can be as small as
we please.
Now suppose we have a singular 3 3 matrix A. As in (i), we denote the 2 2
leading principal submatrix of A by B. If B is invertible then, as in (i), we can
make A invertible by an arbitrarily small change to its (3, 3) entry. If B is singular
we can apply the proposition in the 22 case, making B invertible by arbitrarily
small changes to its diagonal entries; we can then use (i) as before. This proves
the proposition for 33 matrices.
For the 44 case, if necessary we apply the proposition for the 33 case to ensure
that the leading principal submatrix of order 3 is nonsingular. Then, by a similar
argument to (i), the 4 4 matrix can be made invertible by an arbitrarily small
change to its (4, 4) entry. In the same way, the proposition for the 44 case can
then be used to prove it for the 55 case, and so on.
(iii) It is easy to see from the expansion formulae that small changes in the entries of a
matrix cause only small changes in the determinant. Therefore, arbitrarily small
changes in diagonal entries are not enough to transform a matrix with nonzero
determinant into a singular matrix.
19
14
141.
20
f g f
H
f g
H
=
+
, (2)
=
.
i
Y i
i
u
Y u
In the particular case given, H(i, u) = ABe(a+b)i uc . Then
H/i = (a + b)H(i, u) = (a + b)M and H/u = (c/u)H(i, u) = cM/u.
Also f /Y = f (y, i)/Y = M/Y , f /i = af (y, i) = aM ,
g/i = bg(i, u) = bY and g/u = (c/u)g(i, u) = cY /u. Hence
M
(bY ) aM = (a + b)M = LHS(1),
RHS(1) =
Y
M cY
cM
RHS(2) =
=
= LHS(2).
Y
u
u
(1)
143.
Q
F
= A(K 1 )L et = F (K, L, t) =
.
K
K
K
Q
F
F
=
and
= Q. Hence by (14.10),
Similarly,
L
L
t
dQ
Q
Q
=
(mK) +
(nL) + Q = (m + n + )Q,
dt
K
L
so the rate of growth of output is m + n + .
(ii) In this special case, dK/dt = nK, dL/dt = nL and Q/t = Q, so by (14.10)
F
F
dQ
=n K
+L
+ Q.
dt
K
L
By Eulers theorem the expression in square brackets is equal to rQ, so the rate
of growth of output is nr + .
144.
= f (k) + Lf (k)
= f (k) + Lf (k) 2 = f (k) kf (k).
L
L
L
(iii) Multiplying K and L by leaves k unchanged; hence by (i) and (ii), the average
and marginal products of labour and capital are left unchanged.
21
15
IMPLICIT RELATIONS
151. Let c > 0. Since V (x, y) = ln U (x, y), any point on the indifference curve U (x, y) = c
satisfies V (x, y) = ln c. Conversely, since U (x, y) = exp V (x, y), any point satisfying
V (x, y) = ln c must lie on the indifference curve U (x, y) = c. The curve U (x, y) = c is
therefore identical to the curve V (x, y) = ln c. Similarly, the curve V (x, y) = k is the
same as the curve U (x, y) = ek . Thus V and U give rise to the same indifference curve
diagrams. Since the natural logarithm is a monotonic increasing function, the ordering
of the curves is also the same.
The indifference curve diagram for W is the same as that for V except that the lines
x = a and y = b correspond to the axes.
152.
Q
dL
=
=
.
dK
(1 )A (Q/L)1
1 K
Since 0 < < 1 the isoquants are negatively sloped. Now consider moving along
an isoquant in the direction of K and L . Since K and Q remains the same,
Q/K ; since < 1, it follows that |dL/dK| decreases. Hence the isoquants are
convex.
(ii) Let < 0. We may write the the equation of the isoquant Q = Q in the form
K + (1 )L = (Q/A) .
()
(iii) Let 0 < < 1. The equation of the isoquant Q = Q is still (). Since we
now have > 0, L = 0 when L = 0; thus the isoquant meets the Kaxis
where K = (Q/A) . Hence the isoquant Q = Q meets the Kaxis at the point
(d1 Q/A, 0), where d1 is defined as in (ii). Similarly, the isoquant Q = Q meets
the Laxis at the point (0, d2 Q/A), where d2 is defined as in (ii).
From the formula for dL/dK, the slope of the isoquant is 0 at the first point and
at the second. The isoquant therefore meets the two axes tangentially.
m()
Q
, where
(iv) ln =
A
m() = ln[K + (1 )L ].
Since m(0) = 0, we infer from lHopitals rule (or the definition of a derivative)
d x
that ln(Q/A) m (0) as 0. Using the fact that
(a ) = ax ln a, we have
dx
m () =
K ln K + (1 )L ln L
.
K + (1 )L
and
lim Q = AK L1 .
The diagram shows a typical isoquant in each of the three cases < 0, = 0 (Cobb
Douglas) and 0 < < 1
L
153.
<0
=0
0<<1
So long as the optimal value of T is in the region for which f is concave, then
dT /dr < 0.
(ii) Define the function F (r, T ) = f (T )(erT 1) f (T )rerT . The Faustmann rule can
be written in the form F (r, T ) = 0, so by implicit differentiation
dT
F
F
=
.
dr
r
T
By definition of the function F ,
F
= [T f (T ) (1 + rT )f (T )]erT ,
r
It follows that
F
= f (T )(erT 1) r 2 f (T )erT .
T
dT
[T f (T )/f (T )] 1 rT
= 2
.
dr
r + [f (T )/f (T )](1 erT )
23
()
So long as the optimal value of T is in the region for which f is concave, the
denominator on the right-hand side of () is positive. By the Faustmann rule, the
1 + rT erT
, which is
numerator on the right-hand side of () can be written as
erT 1
x
easily seen to be negative (use the series for e ). Hence dT /dr < 0.
Part (ii) of this problem could be solved by the method given in the text for comparative
statics of optima, which directly exploits the second order condition at the optimum.
The above method, which starts by transforming the rule into a form not involving
quotients, is simpler in this case.
154.
=
< 0.
Y
r
(r)
Notice that det J = (1 f (Y ))
Y
r
Y
r
A typical diagram of the ISLM model in the non-negative quadrant of the (Y, r)
plane shows the graphs of the two relations sloping as we have just indicated and
intersecting at a unique point.
Since J is invertible at the given equilibrium, there is a unique local solution for
Y and r in terms of G and M which may be differentiated as follows:
Y /G
/G
Y /M
/M
= J1
,
= J1
.
r/G
/G
r/M
/M
/G /M
Since
= I,
/G /M
1
L/r
(r)
Y /G Y /M
1
.
=J =
r/G r/M
det J L/Y 1 f (Y )
r
Y Y
,
and
are all positive, while
Since det J < 0, the partial derivatives
G M
G
r
< 0.
M
Y
given in the answers to (i) and (iii) re(iv) Using the expressions for det J and
G
spectively, we see that
Y
1
L L
=
,
where
s
=
.
G
1 f (Y ) s (r)
Y
r
24
As we saw in the answer to (ii), s is positive, and is in fact the slope of the LM
relation. If s is small, or if (r) 0, then Y /G [1 f (Y )]1 : this is the
expression for dY /dI in Exercise 15.2.2, and is known as the Keynesian multiplier.
Notice that in this case Y /G > 1. In the general case, where (r) < 0 < s,
Y /G is less than the Keynesian multiplier and may be less than 1.
16
with centre the origin and radii (2 k)1/2 . Note the extreme cases: when k = 0,
the two circles coincide; when k = 4, one of the circles reduces to the origin. For
k > 4, each contour
has one branch consisting of the circle with centre the origin
and radius (2 + k)1/2 . The global minimum of (x2 + y 2 2)2 is 0 which occurs
at each point of the circle x2 + y 2 = 2.
(ii) The contour (x2 + y 2 2)3 = k can be expressed as x2 + y 2 = 2 + k1/3 where,
for k < 0, k1/3 is interpreted as |k|1/3 . For k = 8, the contour is the origin.
For k > 8, the contour is a circle with centre the origin. The global minimum of
(x2 + y 2 2)3 is 8 which occurs at the origin.
In case (ii), unlike case (i), the global minimum occurs at the same point as the global
minimum of x2 + y 2 2. The reason for the difference is that H(w) = w3 is a strictly
increasing function, whereas H(w) = w2 is not. This difference is also illustrated by
the ordering of the contours.
The contour (x2 + y 2 2)1 = k can be expressed as x2 p
+ y 2 = 2 + k1 . For k > 0,
the contours are circles with centrethe origin and radius (2k + 1)/k. As k increases
the radius decreases, approaching 2 as k . Now consider the case where k 0.
For 21 < k 0, the contour is undefined; for k = 21 , the contour is the origin; and
for k < 21 , the contour is a circle with centre the origin. As k decreases from 12 the
From now on, assume that + 1. The firms problem is to maximise (K, L)
subject to K 0, L 0. Since (K, L), a solution to the the first order conditions, if
it exists, will give a global maximum.
25
The first-order conditions for a solution with K > 0 and L > 0 are
pAK 1 L = r,
pAK L1 = w.
Taking natural logarithms and rearranging, we may write this pair of equations as
1
ln K
ln(r/) ln(pA)
=
.
()
1
ln L
ln(w/) ln(pA)
If + < 1, the coefficient matrix is invertible, and () has the unique solution
1
1
ln(r/) ln(pA)
ln K
.
=
1
ln(w/) ln(pA)
ln L
1
Hence the profit-maximising inputs are
"
K = pA
Setting
1
r
#1/(1)
"
Z = pA
"
L = pA
r
1
r
#1/(1)
#1/(1)
w
1
1
we see that case (c) occurs if and only if p = p, in which case the (K, L) pairs which
wL
rK
=
. Case (a) occurs if p > p, and case
maximise profit (at zero) are given by
1
(b) if p < p; the reasons for this will become clear when you have read Section 17.3 and
done Exercise 17.3.2.
163 Denote the expression to be minimised by Q(b1 , b2 ). Then
n
Q X
(2x1i )(yi b1 x1i b2 x2i ),
=
b1
i1
26
Q X
(2x2i )(yi b1 x1i b2 x2i ).
=
b2
i1
It follows that
n
X
2
x21i
D 2 Q(b1 , b2 ) = i=1
X
n
x1i x2i
2
2
2
n
X
i=1
n
X
i=1
i=1
x1i x2i
= 2XT X.
2
x2i
Since the columns of X are linearly independent, XT X is positive definite. This shows
that the function, Q(b1 , b2 ) has positive definite Hessian and is therefore convex.
Now DQ(b1 , b2 ) = 0 when
n
X
i=1
x21i +
n
X
x1i x2i =
i=1
n
X
x1i yi ,
n
X
x1i x2i +
i=1
i=1
n
X
i=1
x22i =
n
X
x2i yi .
i=1
This may be written as (XT X)b = XT y, where y is the nvector whose ith component
is yi . Since XT X is positive definite, it is invertible. It follows that
b = (XT X)1 XT y.
Since Q is convex, this gives the global minimum.
164. Denote the given utility function by W (c, y). The first-order conditions are
W
U
=
(1 + r)V (p) = 0,
c
c
W
U
=
+ (1 + r)V (p) = 0,
y
y
where p = (1+r)(yc). [p stands for pension.] The Jacobian matrix J of (W/c, W/y)
with respect to (c, y) is the Hessian D 2 W (c, y); therefore
2U
2U
c2 + s c y s
, where s = (1 + r)2 V (p).
J=
2U
2U
s
+
s
c y
y 2
Assume that for a given value of r there is a unique pair of optimal values c and y which
satisfy the first-order conditions. If in addition J is invertible at the given optimum,
then c and y may be differentiated with respect to r using the implicit function theorem;
dc/dr
t
1
=J
,
dy/dr
t
where t is the partial derivative with respect to r of (1 + r)V ((1 + r)(y c)), considered
as a function of c, y, and r. Calculating J1 by the inversion formula for 22 matrices,
and t by partial differentiation, we see that
2
2
U
U
t
2U
t
2U
dy
dc
=
+
=
+
,
,
dr
det J y 2
c y
dr
det J c2
c y
where t = V (p) + pV (p).
27
To discuss the signs of dc/dr and dy/dr, notice that det J 0 by the second-order
conditions for a maximum. These second-order conditions will be met, with J invertible,
if at the optimum 0 < c < y (so that p > 0), U is concave and V (p) < 0. From now
on, assume these further conditions are met; in particular det J > 0. Let
=
pV (p)
> 0,
V (p)
and let A, B denote the expressions in square brackets in the solutions just given for
dc/dr and dy/dr respectively. Then dc/dr has the sign of (1 )A, and dy/dr has the
2U
0
sign of ( 1)B. By the concavity of U , A and B cannot both be positive. If
c y
both A and B will be non-positive; in this case
dc
dy
0
if 1,
dr
dr
dc
dy
0
if 1.
dr
dr
2U
> 0 it is possible, but not inevitable, that dc/dr and dy/dr have the same sign.
c y
Finally,
2
ds
U
dy
dc
t
2U
2U
=
=
+2
+
.
dr
dr dr
det J c2
c y
y 2
If
Since U is concave, the term in square brackets is non-positive, regardless of the sign
2U
. Also, we are assuming that det J > 0. Hence ds/dr has the same sign as t, so
of
c y
ds
0 if 1,
dr
ds
0 if 1.
dr
17
171.
(i) The least-cost combination (K, L) occurs where the isoquant corresponding to the
given output level q is tangential to an isocost line. The result follows from the
fact that this point, and hence also the capital-labour ratio, depend only on q and
the slope of the isocost lines.
(ii) If s increases (i.e. the price of capital increases relative to that of labour) we would
expect the capital-labour ratio to decrease. Thus g/s < 0 and therefore > 0.
(iii) The first-order conditions are r = F/K, w = F/L. Now
F
= AZ (1/)1 K 1 ,
K
F
= AZ (1/)1 (1 )L1 ,
L
K
=
.
w
1 L
It follows that g(s, q) = (s(1 )/)1/(1) , so = (1 )1 .
(iv) Proceed as in (iii). Although the expressions for F/K and F/L are different,
division of one first-order condition by the other yields the same result as in (iii).
Hence g(s, q) and are as in (iii).
28
(v) In both (iii) and (iv), g/q = 0. More generally, let F (K, L) be any homogeneous
function of degree > 0. Then F/K and F/L are homogeneous of degree
1 so their ratio is homogeneous of degree 0. Therefore, r/w is a function of
K/L alone, so K/L depends only on s. The same argument applies to the still
more general case where the production function is G(K, L) = H(F (K, L)), where
F (K, L) is homogeneous of degree > 0 and H is a monotonic increasing transformation. [A function which is a monotonic transformation of a homogeneous
function is said to be homothetic.]
172.
(i) w can be interpreted as the wage rate and t as the firms fixed cost per worker.
(ii) The Lagrangian is
L(h, N, ) = whN + tN (F (h, N ) q),
so the first-order conditions are
wN =
F
,
h
wh + t =
F
.
N
pi
i=1
29
xi
= 1,
m
n
X
xi
= 0.
pj
pi
i=1
174.
1a
a
(wT + N ), =
(wT + N ),
p
w
where a =
.
+
a(wT + N )
,
p1 + wt1
x2 =
(1 a)(wT + N )
,
p2 + wt2
a(p1 T t1 N ) (1 a)(p2 T t2 N )
+
.
p1 + wt1
p2 + wt2
As in (i), h/N < 0 and h/w may have either sign, but here the criterion for
the sign of the latter derivative is much more complicated.
(iii) The model in (ii), with p1 = p, t1 = 0, p2 = 0 and t1 = 1, reduces to that in (i).
18
181.
30
(ii) (p, z) = bp2 + (bc + z)p (cz + k). Completing the square,
(p, z) = b(p t)2 + bt2 (cz + k),
where t = 12 (c + [z/b]). It is clear that t is the optimal value of p, given z. Hence
(z1 ) = bp21 (cz1 + k),
z
2b
z1 2 h2
= .
2b
4b
[A slightly different way of answering part (i) is to approximate D for small |h| by
a quadratic function of h, using the method of Section 10.3; then use the envelope
theorem to show that only the term in h2 does not vanish. Part (ii) exhibits the case
where the approximation is exact.]
182. The Lagrangian for the problem is
L(K1 ,K2 , L1 , L2 , , , p1 , p2 , K, L)
(i) For i = 1, 2, V /pi = L/pi by the envelope theorem, and L/pi = Fi (Ki , Li ).
(ii) V /K = L/K by the envelope theorem, and L/K = . It remains to show
that = Fi /Ki for i = 1, 2. But this follows from the first-order conditions
L/Ki = 0 for i = 1, 2.
(iii) Similar to (ii).
183.
V
am p1
1
.
=
p2
bm p21
(ln x1 ln x2 ) =
(ln A)
(ln B) =
.
m
m
m
m m
Denoting the income elasticities of demand by 1 and 2 , this gives
1 2 = .
(1)
x1
x2
The Engel aggregation condition is p1
+ p2
= 1. This can be written in
m
m
terms of s1 , s2 , 1 , 2 as
s1 1 + s2 2 = 1.
(2)
Solving (1) and (2) simultaneously for 1 and 2 and remembering that s1 + s2 = 1
gives
1 = 1 + ( )s2 , 2 = 1 ( )s1 .
31
(ln x1 ln x2 ) =
(ln A)
(ln p1 ) = .
p1
p1
p1
p1 p1
Denoting the two own-price elasticities by 11 , 22 and the two cross-price elasticities by 12 , 21 , we have
11 21 = ( + 1).
(3)
Similarly
22 12 = ( + 1).
(4)
x2
x1
+ p2
= xi
pi
pi
for i = 1, 2.
(5)
s1 12 + s2 22 = s2 .
(6)
22 = (1 + s1 ).
Since and are positive, the own-price elasticities are negative and greater
than 1 in absolute value, and the cross-price elasticities are positive.
184.
1
x1
F
2
x2
which rearranges to
1 F
1 F
w1 x1 w2 x2
1
2
w1 w2
()
19
INTEGRATION
2)
2
3
3
3 x2
3 x+3
3 x +x6
49
.
which is evaluated as ln 2 ln 1 + 2 ln 7 2 ln 6 = ln 18
(i) The more general version of Rule 2 gives the required integral as
h
ln |x 2|
i4
5
h
i4
= ln 6 ln 7 + 2 ln 1 2 ln 2 = ln 14
+ 2 ln |x + 3|
3 .
5
ln |x 2|
i1
h
i1
= ln 1 ln 3 + 2 ln 4 2 ln 2 = ln 43 .
+ 2 ln |x + 3|
1
(iii) Since 1 and 4 are on opposite sides of 2, the integral is not defined.
(iv) Since 4 and 0 are on opposite sides of 3, the integral is not defined.
(v) In this case there are two reasons why the integral is undefined!
33
q0
bounded by the paxis, the inverse demand function and the horizontal line p = f (q0 )
and is given by
Z q0
f (q) dq q0 f (q0 ).
0
(i) The present value at time 0 of the profit gained during the short time interval
[t, t + h] is approximately ert g(t)h. If we split [0, T ] into a large number of small
sub-intervals, the present value at time 0 of the profit stream up to T can be
approximated by a sum of terms of the above form. Passing to the limit as h 0,
we get
Z
T
V (T ) =
(i) (a) The value of the investment at time t + t is equal to the value at time t plus
the interest gained in the time interval [t, t + t]. Approximating this interest
by that on A(t) at the rate r(t) gives the result as stated.
(b) Rearranging the result of (a) gives
A(t + t) A(t)
= r(t).
A(t) t
Taking the limit as t 0, we obtain A (t)/A(t) = r(t), as required.
Integrating the result of (b) over the interval [0, T ] gives
h
ln A(t)
iT
0
r(t) dt.
()
Since A(0) = P , the left-hand side of () is ln(A(T )/P ), and the result follows.
34
(ii) Let P (t, h) be the present value at time 0 of the income received during the short
time interval [t, t + h]. By the final result of (i), with T replaced by t and t by s,
Z t
r(s) ds .
f (t) h P (t, h) exp
0
number of small sub-intervals, the present value at time 0 of the income stream
up to T can be approximated by a sum of terms of the form P (t, h). Passing to
Z T
eR(t) f (t) dt.
the limit as h 0, the present value of the stream is
0
20
201.
d
(ex ), we may integrate by parts:
dx
Z
Z X
iX Z X
h
(ex )nxn1 dx = X n eX + n
xn ex dx = (ex )xn
that ex =
xn1 ex dx.
X n eX
As X ,
0 (see Exercise 10.4.3), so In = nIn1 . Applying this result
repeatedly gives In = n! I0 . It is easy to show that I0 = 1, so In = n! .
Z A
Z 0
Z A
f (x) dx. Making the substituf (x) dx and J =
f (x) dx = I + J, where I =
203.
0
tion y = x,
I=
f (y) dy =
f (y) dy.
0
In case (i), I = J and the result follows. In case (ii), I = J and the result follows.
(iii) Denote the required integral by K. By result (i),
Z 1
2
ex /2 dx.
K=2
0
2
xex /2
xex
2 /2
Z 0
Z
3
x3 dx diverge.
x dx and
integrals
0
204. V (s) =
(cert ) dt +
0
f rt
rs
rt
rs
rs
V (s) = ce
+
e dt
f (s, t)e dt = ce
f (s, s)e
+
s s
s s
by Leibnizs rule. Therefore, the value of s which maximises V (s) must satisfy the
equation
Z T
f
dt.
er(st)
f (s, s) + c =
s
s
21
INTRODUCTION TO DYNAMICS
211.
(i) All solutions of the differential equation satisfy dy/dt = 0 when t = 0 and t = 2.
It is also clear from the differential equation that, as t increases from 0 to 0+,
dy/dt changes from negative to positive. Hence there is a minimum at t = 0.
Similarly, as t increases from 2 to 2+, dy/dt changes from positive to negative;
there is therefore a maximum at t = 2.
From the differential equation, the slopes of the solution curves when t = 1, 1, 3
are 3, 1, 3 respectively. The directions of the tangents to the curves when t =
1, 0, 1, 2, 3 can be shown as sets of parallel line segments of slopes 3, 0, 1, 0, 3
cutting the lines t = 1, 0, 1, 2, 3 respectively. This enables us to draw the family
of solution curves, each with a minimum at t = 0, a maximum at t = 2 and with
directions of tangents as just stated at t = 1, 1, 3. It is then clear that, as t
increases beyond 2, dy/dt becomes more negative and y decreases. As t ,
y .
(iii) This problem corresponds to the case a = 2, b = 1 of Exercise 21.1.5 (see also
Exercise 21.3.4). There you were asked to find the solution of the differential
equation which satisfies y = y0 when t = 0 where 0 < y0 < a/b. The solution
obtained satisfies 0 < y < a/b for all t and, as t , y a/b. This confirms the
behaviour of the solution curves lying between y = 0 and y = 2.
R
R
212. (i) Separating the variables and integrating, z 1 dz = r dt. Hence ln z = rt + B,
which can be arranged in the form z = Cert .
(ii) Since extraction costs are zero, (i) gives p = Cert . Assuming the market for the
resource clears at each instant of time, we have Cert = q , which can be arranged
in the form q = Ae(r/)t .
Z
q(t) dt = S. But
(iii) Since the total amount of mineral to be extracted is S,
0
q(t) dt = lim
T 0
Aet dt = lim (A/)(1 eT ) = A/,
T
rS
213.
1 dK
= sAk k,
L dt
sA (1)nt
K(t) = L(0)e k(t) = L(0)
e
+ ce(1)t
+n
nt
1/(1)
sA
,
+n
sA
k =
+n
1/(1)
dK
+ K = sAK L .
dt
Setting N = L/(1) we obtain the differential equation
dK
+ K = sAK N 1 .
dt
We may therefore proceed as in (i), with L replaced by N ; notice that the rate
of growth of L must be replaced by the rate of growth of N , so n is replaced by
n/(1 ). The solution for K is therefore
1/(1)
(1 )sA nt
(1)t
e + ce
,
K(t) = L(0)
(1 ) + n
where c is a constant.
214. The discrete-time analogue is yt + ayt = b, or
yt+1 + (a 1)yt = b.
Letting y = b/a, we may write the general solution of the differential equation as
y(t) = y+Aeat , and the general solution of the difference equation as yt = y+A(1a)t .
Both for the differential equation and for the difference equation, the stationary solution
is y = y.
For the differential equation, all solutions approach the stationary solution as t ,
provided limt eat = 0; this occurs if and only if a > 0. For the difference equation,
all solutions approach the stationary solution as t , provided limt (1 a)t = 0;
this occurs if and only if |1a| < 1, i.e. 0 < a < 2. The general solution of the difference
equation exhibits alternating behaviour if a > 1.
Notice that the variety of possible behaviour is greater for the difference equation than
for the differential equation, in that alternating behaviour is possible for the latter. The
qualitative behaviour of the discrete-time analogue is not necessarily the same as that
of the differential equation for the same parameter values.
38
22
221.
(i) (a) y = e3x sin 4x cuts the xaxis where sin 4x = 0, i.e. where x = k/4 for
k = 0, 1, 2, . . .
(b) dy/dx = e3x (4 cos 4x 3 sin 4x). Hence dy/dx = 0 if and only if tan 4x = 34 .
Setting = arctan 43 (= 0.927 to 3 decimal places), we see that dy/dx = 0
if and only if 4x = + k for some integer k. Hence the critical points are
{ (xk , yk ) : k = 0, 1, 2, . . . }, where xk = 14 ( + k) and
yk = exp( 34 ( + k)) sin( + k) = e3/4 e3k/4 (1)k sin .
Thus yk = ar k for all k, where a = e3/4 sin and r = e3/4 : notice that
1 < r < 0.
(c) Knowing the points found in (a) and (b) enables the curve to be sketched. It
is clear from the configuration of the points that the positive critical values
are maxima and the negative critical values are minima, but for a rigorous
demonstration of this see Problem 242. As x , |y| ; as x ,
y 0. The graph is that of a damped oscillation.
Thus Yk = bsk for all k, where b = e3/4 sin and s = e3/4 : notice that
s < 1.
(c) The graph is that of an explosive oscillation.
222.
(i) The graph of 3t + 1 is a straight line of slope 3 and intercept 1. The graph of
2 sin 6t is like that of sin t but magnified by a factor of 2 and with period /3.
The graph of ln Y is the sum of these two and is thus an oscillation of period /3
about 3t + 1.
(ii) The graph of 2t + 5 is a straight line of slope 2 and intercept 5. If > 0, the graph
of 3 sin(6t + ) is like that of sin t but magnified by a factor of 3, with period /3
and shifted to the left through /6 (since 6t + = 6[t + 16 ]). The graph of ln Z is
the sum of these two and is thus an oscillation of period /3 about 2t + 5. Since,
for example, the maximum and minimum points of 3 sin(6t + ) occur at a time
/6 earlier than the corresponding points of 2 sin 6t, the former periodic function
is said to lead the latter by /6. Similarly, if < 0, the maximum and minimum
points of 3 sin(6t + ) occur at a time ||/6 later than the corresponding points
of 2 sin 6t; the former periodic function is then said to lag behind the latter by
||/6.
(a) If = , the periodic component of ln Z leads the corresponding component
of ln Y by /6, i.e. half a period. When the periodic component of ln Y is at
a maximum, the corresponding component of ln Z is at a minimum and vice
versa.
39
p = b sin 6t,
where a and b are positive constants. Then A is obviously right. The periodic
component of the inflation rate at time t is
dp/dt = 6b cos 6t = 6b sin 6t 2 .
Hence the periodic component of the inflation rate at time t +
is also correct.
12
is 6b sin 6t, so B
223. Let (R, ) be the polar coordinates of the point with Cartesian coordinates (A, B).
Then
A cos + B sin = R cos cos + R sin sin
= R(cos cos + sin sin )
= R cos( ).
(ii) G(m) =
2
1
2
arctan 1 = = .
4
2
(iii) Since
arctan([m/x] )
arctan y
= lim
= 1,
x
y0
[m/x]
y
lim
h m i
x x
!,
1+
h m i2
x
2/
.
+ m x1
m x1+
()
Since m and are positive numbers, f (x) > 0 for all x > 0. Also, since > 1,
x1+ 0 and x1 as x 0, while x1+ and x1 0 as x .
Thus the denominator on the right-hand side of () becomes very large both as
x 0 and as x . It follows that limx0 f (x) = limx f (x) = 0.
h (x) = 21 (1 + )(x/m) + (1 )(x/m) .
Then h (m) = 0; also, h (x) < 0 if 0 < x < m and h (x) > 0 if x > m. Since
f (x) always has the opposite sign to h (x), the required properties are satisfied
by x = m.
23
COMPLEX NUMBERS
()
(ii) If A and B are conjugates, A + B and i(A B) are real numbers, so yt is real.
Conversely, if yt is real for all t, then A + B and i(A B) are real numbers, say
C and D respectively. Then 2iA = iC + D, whence A = 12 (C iD) and
B = C A = 12 (C + iD) = A.
for all t.
41
232. Since e(2+5i)t = e2t (cos 5t + i sin 5t) and e(25i)t = e2t (cos 5t i sin 5t),
y = e2t ((A + B) cos 5t + i(A B) sin 5t).
(i) If A and B are real, the real part of y is (A + B)e2t cos 5t and the imaginary part
of y is i(A B)e2t cos 5t.
(ii) Similar to Problem 23-1, part (ii).
(iii) Since y = 0 when t = 0, A + B = 0. Also,
dy/dt = 2y 5e2t ((A + B) sin 5t i(A B) cos 5t) :
setting t = 0 we have 1 = 0 5(0 i[A B]). Thus B = A and A B = i/5.
It follows that A = i/10, B = i/10 and
y=
233.
sin 5t
.
5e2t
.
u = 12 (1 + 3), v = u
3
2
(ii) Let g(z) = z 4z + 14z 20; also let u = 1 3i. Then u2 = 8 6i and
u3 = 1 9i 27 + 27i = 26 + 18i, so
g(u) = 26 + 18i + 32 + 24i 6 42i = 0.
Thus u is a root; since the polynomial g has real coefficients, u
= 1 + 3i is also a
root. Denoting the third root by v, we see that
g(z) = (z u)(z u
)(z v) = ([z 1]2 [3i]2 )(z v).
234.
it
it
te dt = (t/i)e
i/2
0
(1/i)
/2
eit dt.
/2
teit dt.
(iii) Denote the integrals by I and J. Then iI is the imaginary part of the first integral
in (ii), so I = 1. J is the real part of the second integral in (ii), so J = 41 ( 2 8).
42
24
FURTHER DYNAMICS
241. Denote the differential equations by (i) and (ii) and observe that (ii) may be rearranged
as follows:
dz
+ 2z = y.
(ii )
dt
d
Adding (i) to 2(i) and using (ii ),
dt
d2 y
dy
dy
=4
+ 2y 5y.
+2
dt2
dt
dt
Rearranging,
dy
d2 y
2 3y = 0.
2
dt
dt
The characteristic equation is x2 2x 3 = 0, which has roots 3 and 1. The general
solution for y is therefore
y = Ae3t + Bet ,
where A and B are constants. Substituting in (i) gives
z = 51 Ae3t + Bet .
(a) Here A + B = 1 and (A/5) + B = 3; hence A = 5, B = 4.
From the differential equation, d2 y/dt2 > 0 for all t. Hence the function is
strictly convex, with its only critical point at t = 0. Therefore the graph is
U-shaped with vertex at (0, 1).
(c) Integrating twice,
y = 13 t3 12 t2 + At + B.
Imposing the initial conditions, we obtain the required solution:
y = 13 t3 12 t2 + 1.
From the differential equation, d2 y/dt2 < 0 for t < 21 and d2 y/dt2 > 0 for
t > 12 . Hence the function is strictly concave for t < 21 and strictly convex for
t > 12 . The critical points occur where t is a root of the equation t2 t = 0,
i.e. where t is 0 or 1; there is therefore a local maximum at (0, 1) and a local
minimum at (1, 65 ).
(ii) The characteristic equation is p2 + 2ap + (a2 + b2 ) = 0, which has roots a ib.
The general solution of the differential equation is therefore
y = Ceax cos(bx + ),
where C and are arbitrary constants.
From the differential equation, d2 y/dx2 has the opposite sign to y if dy/dx = 0.
Therefore, any critical point (X, Y ) such that Y > 0 is a maximum, and any
critical point (X, Y ) such that Y < 0 is a minimum.
The function y = e3x sin 4x of Problem 222 is the special case of the general
solution with a = 3, b = 4 and the constants C and put equal to 1 and /2
respectively. Therefore, the function satisfies the differential equation and has the
above property concerning critical values. Similarly, y = e3x sin 4x has the same
property.
243. First look for a particular solution of the form y = At + B. Substituting this into the
differential equation gives
0 + bA + c(At + B) = kt + .
Hence cA = k and bA+cB = , so A = c1 k and B = c2 (cbk). The complementary
solution is oscillatory (O) when the roots of the characteristic equation are complex,
i.e. when b2 < 4c. Otherwise the complementary solution is non-oscillatory (N). The
complementary solution tends to 0 as t (S) when both roots of the characteristic
equation have negative real parts, the criterion for which is obtained in the text as
b > 0, c > 0. Otherwise the complementary solution does not tend to 0 as t (U).
We may therefore classify the possible forms taken by the general solution as follows:
SO The general solution oscillates about the particular solution and tends to the particular solution as t .
SN The general solution is non-oscillatory and tends to the particular solution as
t .
UO The general solution oscillates about the particular solution but does not tend to
the particular solution as t .
44
UN The general solution is non-oscillatory and and does not tend to the particular
solution as t .
244. The discrete-time analogue of the differential equation is
2 yt + byt + cyt = u,
which may be written as
(yt+2 2yt+1 + yt ) + b(yt+1 yt ) + cyt = u,
or more simply as
yt+2 + f yt+1 + gyt = u
where f = b 2, g = 1 b + c. For the differential equation, the stationary solution
occurs if d2 y/dt2 = dy/dt = 0 and is therefore y = u/c. For the difference equation,
the stationary solution occurs if 2 yt = yt = 0 and is therefore yt = u/c.
The differential equation exhibits oscillatory behaviour if its characteristic equation has
complex roots, i.e. if b2 < 4c. The difference equation exhibits oscillatory behaviour if
its characteristic equation has complex roots, i.e. if f 2 < 4g: this inequality may be
written
(b 2)2 < 4(1 b + c)
and therefore reduces to b2 < 4c.
For the differential equation, the condition for the stationary solution to be stable has
been obtained in the text as b > 0, c > 0. For the difference equation, the condition
for the stationary solution to be stable is that the roots of the characteristic equation
are are both < 1 in absolute value (or modulus, if the roots are complex). In the case
of real roots, it is therefore necessary that x2 + f x + g > 0 at x = 1. This ensures
that one of the following three cases occurs: (a) both roots between 1 and 1, (b) both
roots < 1, (c) both roots > 1. But the product of the roots is g, so if we assume that
g < 1 then cases (b) and (c) are eliminated and we are left with (a). In the case of
complex roots, we have x2 + f x + g > 0 for all x, and in particular at x = 1. Also the
roots are complex conjugates, so the product of the roots is r 2 where r is the common
modulus: to ensure that r < 1 we must therefore impose the condition g < 1.
To summarise, the criterion ensuring stability in the cases of both real and complex
roots is
1 + f + g > 0, 1 f + g > 0, g < 1.
In terms of b and c these three conditions may be written respectively as c > 0, c >
2b 4, c < b and therefore reduce to the chain of inequalities
0 < c < b < 2 + 12 c.
As in the first-order case, we note that the qualitative behaviour of the discrete-time
analogue is not necessarily the same as that of the differential equation for the same
parameter values.
45
25
251.
3x1 x2 = 4x2 .
3x1 x2 = 5x2 .
6x1 x2 = 4x2 .
6x1 x2 = 5x2 .
(i) zH w = z1 w1 + . . . + z1 w1 and wH z = w
1 z1 + . . . + w
1 z1 . In particular,
zH z = z1 z1 + . . . + z1 z1 = |z1 |2 + . . . + |zn |2 .
For j = 1, . . . , n, zj wj is equal to wj zj , whose conjugate is w
j zj . Therefore zH w
H
and w z are complex conjugates. For each j = 1, . . . , n, |zj | 0, with equality
if zj 6= 0. Therefore zH z > 0 provided at least one of z1 , . . . , zn is non-zero, i.e.
provided z 6= 0.
T has the same diagonal entries as A,
the diagonal entries of AH are the
(ii) Since A
complex conjugates of those of A. Thus, if A is Hermitian, the diagonal entries
must be equal to their conjugates and hence must be real. When all the entries
= A and the condition for A to be Hermitian reduces to
are real numbers, A
T
A = A.
46
a
u
u
b
and
u
u
b
w
v
w ,
c
(iv) Let A be the 22 Hermitian matrix given in (iii) and let z C2 . Multiplying out
the expression for zH Az, we have
zH Az = P + Q,
where P = az1 z1 + bz2 z2 and Q = uz1 z2 + u
z2 z1 . Now a is real and z1 z1 = |z1 |2 , so
az1 z1 is real. A similar argument using the fact that b is real shows that bz2 z2 is
real, and hence that P is real. Also, since uz1 z2 and u
z1 z2 are complex conjugates,
their sum Q is real. Hence P + Q is real, as required.
Now let A be the 33 Hermitian matrix given in (iii) and let z C3 . Reasoning
as in the 22 case, we may write zH Az = P + Q, where P is the real number
a|z1 |2 + b|z2 |2 + c|z3 |2
and Q is twice the real part of
uz1 z2 + v z1 z3 + wz2 z3 .
Hence P + Q is real, as required. The general case is similar.
(v) Suppose is an eigenvalue. Then there is a non-zero vector z such that Az = z.
Therefore
zH Az = zH z = zH z.
Since z 6= 0, zH z > 0 by (i), so we may define the positive real number p = (zH z)1 .
But then = pzH Az, which is real by (iv).
(vi) The (1, 1) entry of A is the value of zH Az when z is the first column of the identity
matrix I; the (2, 2) entry of A is zH Az when z is the second column of I; and so
on. Hence by (iv), the diagonal entries of a Hermitian matrix are real.
Now let
1 w
z1
A=
.
, z=
w
1
z2
Multiplying out as in (iv) we have
zH Az = |z1 |2 + |z2 |2 + Q,
where the real number q is twice the real part of wz1 z2 . On the other hand
|z1 + wz2 |2 = (
z1 + w
z2 )(z1 + wz2 ) = |z1 |2 + Q + |w|2 |z2 |2 .
Subtracting and rearranging,
zH Az = |z1 + wz2 |2 + (1 |w|2 )|z2 |2 .
47
This is clearly positive if |w|2 < 1 and at least one of z1 and z2 is not zero; while
if |w|2 1, we can make zH Az non-positive by setting z1 = w, z2 = 1. Thus A
is positive definite if and only if |w|2 < 1, which happens if and only if det A > 0.
The general result for n = 2 is that a Hermitian matrix is positive definite if and
only if its diagonal entries and its determinant are all positive. The really general
result is that a Hermitian matrix is positive definite if and only if all its principal
minors are positive. [As in the case of real symmetric matrices, it is also true that
a Hermitian matrix is positive definite if and only if all its leading principal minors
are positive.]
(vii) When all the entries of a matrix are real numbers, the Hermitian transpose reduces
to the ordinary transpose. So a unitary matrix whose entries are all real numbers
has as its inverse its transpose and hence is the same thing as an orthogonal matrix.
Denoting the given matrix by S,
"
#
1 i
1
2 i
1
By straightforward matrix multiplication, SH S = I, so SH = S1 as required.
253.
P=
as required.
48
ZT
O
= ZZT ,
2
0
5
1)
2
2 5
=
( 1)(1 + 2 7 + 10 4)
( 1)(2 8 + 7)
2
5
( 1)2 ( 7).
5 1 2
7I A = 1
5 2 .
2
2 2
Solving the system by Gaussian elimination, we see that the solution consists of
all vectors for which z1 = z2 = 12 z3 . We may therefore choose z = [1 1 2]T .
(iv) Let the vectors x, y, z be as in (ii) and (iii). It is easy to see that these vectors
are linearly independent. Arguing as in the proof of Proposition 1 in Section 25.1,
we set D = diag(1, 1, 7) and let S be a matrix whose columns are scalar multiples
of x, y, z. To ensure that ST S = I, each column of S must have length 1; we
therefore define the columns of S to be
(1 + 1)1/2 x, (1 + 1 + 1)1/2 y, (1 + 1 + 4)1/2 z.
Then
1/ 2
S=
1/ 2
0
49
1/ 3
1/ 3
1/ 3
1/ 6
1/ 6
2/ 6
26
DYNAMIC SYSTEMS
261.
1
1
1
u = 1 , v = 0 , w = 2
1
1
1
are corresponding eigenvectors. Therefore, the general solution is
y(t) = 0t c1 u + c2 v + 3t c3 w.
Here and below, 00 should be interpreted as 1.
(ii) Using the given initial condition and our convention that 00 = 1,
c1 + c2 + c3 = 1, c1 2c3 = 2, c1 c2 + c3 = 4.
Solving these equations simultaneously gives c1 = 7/3, c2 = 3/2, c3 = 1/6. The
solution is therefore
1
1
1
t1
3
3
2 ,
y(t) = t 1 0 +
2
2
1
1
1
where 0 = 7/3 and t = 0 for all t > 0.
(iii) Since 1 is an eigenvalue, I A is not invertible and we cannot use the formula
x = (I A)1 b for the particular solution. However, all we need for a constant
particular solution is some vector x such that (I A)x = b, and it is easy to
see that, for example, x = [1 1 0]T does the trick. The general solution is
1
1
1
1
x(t) = 1 + 0t c1 1 + c2 0 + 3t c3 2 .
1
1
0
1
(iv) In this case, the system of linear equations (I A)x = b has no solution, so the
system of difference equations does not have a constant particular solution. We
therefore look for a solution of the form x (t) = tp+q, where p and q are constant
vectors. Then
(t + 1)p + q = tAp + Aq + b
for all t. Equating coefficients of t, we have Ap = p; equating constant terms,
we have Aq + b = p + q. Hence p is an eigenvector of A corresponding to the
eigenvalue 1; we may therefore set p = v, where v = [1 0 1]T as in (i) and
is a constant. It follows that
1
(I A)q = b v = 0 .
2+
50
1
1
t
3
x (t) = 1 0 ,
2
2
0
1
3
2
and we are
1
1
1
1
3
t
x(t) = 1 + 0t c1 1 + c2
0 + 3t c3 2 .
2
2
1
1
0
1
262.
x(t) = x (t) + t
+ t
,
3
3
2
2
where x (t) is a particular solution. To find this, try x (t) = tp + q, where p and
q are constant vectors. Then
(t + 1)p + q = tAp + Aq + b(t)
for all t. Equating coefficients of t and then equating constant terms, we have
1
2
(I A)p =
, (I A)q =
p.
3
0
Hence p2 = 3, p1 = 9/4, q2 = 3(2 + 94 ) = 51/4 and q1 = 34 (0 + 3) = 9/4.
Thus our particular solution is
" #
"
#
3
3t 3
3
x (t) =
4 4
4 17
and the general solution is
"
#
" #
" #
"
#
3
3t 3
5t c2 1
3
c1 1
x(t) =
+ t
.
+ t
4 4
4 17
3
3
2
2
2
(ii) The characteristic polynomial
ofA is
+8 + 15, so the eigenvalues are 5 and
1
3
3. It is easy to show that
and
are corresponding eigenvectors. Thus
1
1
the general solution is
5t 1
3t 3
x(t) = x (t) + c1 e
+ c2 e
,
1
1
51
(a) For a particular solution, try x (t) = et p, where p is a constant vector. Then
d t
t
t
t 1
t
.
e p = (e p) = A(e p) + b(t) = e Ap + e
0
dt
Multiplying by et and rearranging, we obtain
1
(I + A)p =
.
0
Solving for p, we have p1 = 5/8, p2 = 1/8. Hence
t 5/8
x (t) = e
.
1/8
(b) For a particular solution, try x (t) = e2t p, where p is a constant vector.
Reasoning as in (a), we obtain
0
(2I + A)p =
.
2
Solving for p, we have p1 = 1, p2 = 0. Hence
2t 1
x (t) = e
.
0
(c) Let u and v satisfy
du
= Au +
dt
et
0
dv
= Au +
dt
0
e2t
()
Then
d
(4u + 3v) = 4Au + 3Av +
dt
4et
3e2t
x (t) = 4u + 3v =
1 t
2e
5
1
2t
3e
1
0
y = (pqxy cy).
There are therefore three fixed points: (0, 0), (k, 0) and the point given by
x
c
r
1
.
x= , y=
pq
q
k
52
c
r
Setting =
, we may write the coordinates of this third point as k, (1 ) .
kpq
q
This is the only fixed point that could be in the positive quadrant: it will be in the
positive quadrant if < 1, i.e. if c < kpq. From now on we assume that this condition
holds.
Writing the system of differential equations as
x = f (x, y),
y = g(x, y),
we have
f (x, y) rx f
g
g
g(x, y)
f
=
,
= qx,
= pqy,
=
.
x
x
k
y
x
y
y
Thus at the fixed point in the positive quadrant,
f
f
g
g
= r,
= kq,
= pr(1 ),
= 0.
x
y
x
y
The associated linear sytem is therefore
x = r(x k) k(qy r(1 )),
Let
vA =
r
pr(1 )
It follows that both eigenvalues of A have negative real parts: the fixed point is locally
stable. Also, the characteristic polynomial of A is
+ r
kq
= 2 + r + rc(1 ).
pr(1 )
Hence the eigenvalues of A are real and distinct if (r)2 > 4rc(1); they are complex
conjugates if (r)2 < 4rc(1 ). Thus the fixed point is a stable node of the linearised
system if < 0 and a spiral sink if > 0 , where
0 =
264.
rc
r 2
=
.
4c(1 )
4kpq(kpq c)
dh
1 dH
=
= s2 AK H 1 2 .
dt
H dt
(ii) Let bi = ln(si A/i ) for i = 1, 2. Then the set of points in the khplane for which
k = 0 is the straight line
( 1)k + h + b1 = 0.
Since < 1 and > 0, this is an upward-sloping line of slope (1 )/. By a
similar argument, the set of points in the khplane for which h = 0 is an upwardsloping straight line of slope /(1 ). Since + < 1,
1
>1>
.
1
Thus the line k = 0 is steeper than the line h = 0, so the two lines intersect at
exactly one point (k , h ).
(iii) The Jacobian of the autonomous system is
"
( 1)s1 ek Q
s1 ek Q
s2 eh Q
( 1)s2 eh Q
where Q = A exp(k + h). For all k and h, both diagonal entries are negative
and the determinant is
(1 )s1 s2 e(k+h) Q2 > 0.
Therefore, conditions (i) and (iia) of Olechs theorem hold and (k , h ) is globally
stable.
27
271.
9
X
t=0
and the given endpoint conditions. The Hamiltonian is 3x2t 4yt2 + t xt , the
control condition is t = 6xt and the costate equation is 8yt = t1 t .
(ii) Eliminating xt between the state equation and the control condition, we see that
t = 6(yt+1 yt ). Hence from the costate equation,
8yt = 6(yt+1 2yt + yt1 ),
which simplifies to the required equation.
(iii) The characteristic equation of the second-order difference equation has roots 3 and
1/3, so we can write the solution as 3t A + 3t B where A and B are constants.
From the left-endpoint condition, A + B = 1, so
yt = 3t A + 3t (1 A) (t = 0, 1, . . . , 10).
10
From the right-endpoint condition, A = 100 310
3 310 .
54
()
(iv) The solution is again () but with a different value of A. Instead of a right-endpoint
condition, we have the transversality condition 9 = 0. As in (ii), t = 6(yt+1 yt )
for t = 0, 1, . . . , 9. Therefore y10 = y9 , or
310 39 A + 310 39 1 A = 0.
1
Hence A = 319 + 1 .
272. For this problem, it is helpful to use the subscript notation for partial derivatives:
1 (h, s) =
,
h
2 (h, s) =
.
s
2t = 2 (ht , st ).
subject to
ht 0,
st+1 st = g(st ) ht
(t = 0, 1, . . . , T )
and given s0 .
(ii) The Hamiltonian for period t is
(1 + )t (ht , st ) + t (g(st ) ht ).
The control condition is
(1 + )t 1t t , with equality if ht > 0,
and the costate equation is
(1 + )t 2t + t g (st ) = t1 t .
Setting t = (1 + )t t , we may write the costate equation in the form
2t = (1 + )t1 (1 + g (st ))t .
(iii) The control condition for the steady state says that t is constant over time and
s). Substituting this into the costate equation,
equal to 1 (h,
s) = ( g (
s),
2 (h,
s))1 (h,
which rearranges to the required equation.
55
273.
(i) It is convenient to use z = I/K as the control variable; the state variable is
again K. The state equation is Kt+1 = (zt )Kt . Denoting the costate variable
by , the Hamiltonian for period t is
Ht (Kt , zt , t ) = (1 + r)t (t azt bzt1+ )Kt + t (zt )Kt .
The control condition is
(1 + )bzt = (1 + r)t t a
if the right-hand side is positive, zt = 0 otherwise. The costate equation is
(1 + r)t (t azt bzt1+ ) + t (zt ) = t1 t .
The transversality condition is T = 0.
(ii) Let t = (1 + r)t t for t = 0, 1, . . . , T ; also let t = max(t a, 0). Then the
control condition may be written
1/
t
.
zt =
(1 + )b
From the costate equation and the control condition,
(1 + r)t1 = t + (1 )t +
t zt .
1+
()
274.
1
12 8 12 7 4
10 7
9 3 5
1
3 6
5 7 4 1
Underlined entries show the optimal path.
(v)
14
12
3
10
9
6
14
11
5
10
3
7
56
4
5
4
1
1
1
28
281.
i.e. f (h) = . Since f is strictly concave, f is a decreasing function; let its inverse
be the decreasing function . Then the control condition may be written h = (),
and the state equation then has the required form. By definition of ,
= + ,
so the costate equation may be written = [ g (s)]. This and the reformulated state equation form an autonomous system in s and .
(iii) The Jacobian matrix of the autonomous system, evaluated at the unique fixed
point (
s,
), is
g (
s)
(
)
g (
s) g (
s)
The determinant is
2
(
)g (
s).
g (
s)
A
B
C
D
s = 0
s0
s1
57
In the phase diagram, the broken curve is the stable branch. To get an idea of
what optimal paths look like, consider for definiteness the case where s(0) = s0 ,
s(T ) = s1 and s0 < s1 < s. Then the solution may look like AB or CD or CDE
in the diagram, depending on the value of T . There is a unique value of T (say
T ) such that the solution path is along the stable branch. There is also a unique
value of T (say T , where T > T ) such that the solution path has the property
that s(T
) = 0. If T < T and a solution exists, the solution path looks like AB. If
T < T < T , the solution path looks like CD; and if T > T , like CDE.
282.
(i) If the given inequality did not hold, then Marks initial debt would be at least
as great as the present value of his labour income, discounted at the borrowing
rate. Hence it would be impossible for him to have positive consumption at each
moment and die solvent.
An alternative way of making the same point is as follows. Suppose Marks problem
has a solution. If a0 0 the required inequality obviously holds, so suppose a0 < 0.
By the right-endpoint condition, a(t) 0 for some t T ; let be the smallest
such t. Then a( ) = 0 (since a cannot jump), 0 < T and a(t) < 0 for all t
such that 0 t < . For such t, multiplication of the state equation by eit gives
d it
e a(t) = eit (w(t) c(t)).
dt
Setting
v(t) = a(t) +
eis w(s) ds
we see that
v(t)
(0 t )
(0 < t < ).
it
w(t) dt = v(0) +
f (a) =
i if a < 0,
the costate equation is = ( r) if a > 0, = ( i) if a < 0. Since f (a) is
not defined if a = 0, nor is the costate equation. This means that is allowed to
jump when a = 0, though it may not be optimal for it to do so.
(iii) From the control condition and the costate equation,
(
(r )c if a > 0,
c =
(i )c if a < 0.
This and the state equation form an autonomous system in a and c. It follows
from our assumptions about i, r and that c > 0 if a < 0, c < 0 if a > 0. The
phase diagram is as follows.
58
slope r
slope i
(iv) Consider first the case where a0 > 0. We see from the phase diagram that a(t) 0
for all t; and for as long as a > 0, c > w and c is falling at the proportional rate
r. Thus there are two possibilities:
Hence
a0 + (w/r) 1 erT
c0 =
.
1 eT
This will be the solution if a0 is sufficiently large that c0 e(r)T w. Specifically, this will be so if and only if a0 a+ , where
T
1 erT 1
rT e
.
a+ = we
r
(b) If 0 < a0 < a+ , the consumption path is given by c(t) = c0 e(r)t (0
t < ) and c(t) = w if t T . Here c0 and are chosen such that
a( ) = 0, and limt c(t) = w. The latter condition is required because,
with a concave utility function, jumps in consumption are undesirable. [The
same
conclusion is reached by considering the free-time problem of maximising
Z
= a0 .
59
w
1 eit < a0 < 0. From the phase diagram, a(t) 0 for all
Now suppose
i
t; and for as long as a < 0, c < w and c is rising at the proportional rate i .
There are two possibilities:
(e) c(t) = c0 e(i)t (0 t T ), where c0 is chosen such that c(T ) w and
a(T ) = 0. To satisfy the latter condition we must have
a0 + (w/i) 1 eiT
.
c0 =
1 eT
This will be the solution if a0 is sufficiently large that c0 e(i)T w. This
will be so if and only if a0 a , where
T
e 1 eiT 1
a = weiT
< 0.
i
(d) If a < a0 < 0, the consumption path is given by c(t) = c0 e(i)t (0 t < )
and c(t) = w if t T . Here c0 and are chosen such that a( ) = 0, and
limt c(t) = w. Hence c0 = we(i) and is chosen such that
i
e 1
i e 1
we
= a0 .
i
Finally, if a0 = 0 then a(t) = 0 and c(t) = w for all t. This is case (c).
(v) See (iv) above, parts (a) and (e).
283.
C = bC/,
where b = A . Therefore
d
C
K
C
C
ln
=
=
,
dt K
C
K
K
where = 1 1 b + . Hence our dynamic system has a solution where C/K
remains unchanged over time and C and K grow at the same constant rate:
C = K,
C
K
A
=
=
C
K
( 1)(A ) +
.
()
Because > 1 and A > + , and the common growth rate of C and K are
positive. To show that this solution satisfies the transversality condition (28.9)
in the text note that KU (C) = C 1 /, which grows at the negative rate
( 1 1)(A ). Since also > 0, (28.9) is satisfied. Since the utility function
is concave and the differential equations are linear, the path is optimal. On the
60
.
A
()
Given (), KU (C) grows at the positive rate ( 1 1)(A ), and for the
transversality condition we need this to be less than . In fact,
(1 )(A ) = (1 )(A ),
which is positive by (). Thus if () holds, the path along which C/K = for
all t is feasible and satisfies the transversality condition; it is therefore optimal.
284. We assume throughout that
A
> + .
1+
()
In particular, A > . Let , c be constants such that /A < 1 and 0 < c < 1. If we
set z(t) = , C(t) = (A )cK(t) for all t, then C and K grow at the positive rate
(1 c)(A ). Thus endogenous growth is possible. To show that it is not desirable,
we derive the optimal path.
The current-value Hamiltonian is
ln C bAKz 1+ + (AKz C K).
The control conditions, taking into account the constraint z 1, are
= C 1 ,
Setting B =
1
, we see that
(1 + )b
(
z=
1
(B/C)1/
if C B;
if C > B.
A
A
2
= +
,
A+
C dt
C
(1 + )B C
C
(1 + )B
C
while if C > B,
1 dC
A (B/C)1/
+
=
Az
+
=
+
.
C 2 dt
C
1+
C
1+
C
C
61
( + ).
=A
and
=
K
C
K
C
1+ C
if C B,
if C > B,
Let
A
C =
(1 + )( + )
1/
B,
K =
C
.
+ (1 + )( + )
By (), C > B and (K , C ) is the unique fixed point of the autonomous system.
C
K = 0
It is clear from the phase diagram that the fixed point is a saddle point. The optimal
policy is: given K(0), choose C(0) so that (K(0), C(0)) is on the stable branch. the
optimal path then follows the stable branch, which converges to (K , C ). Since C >
B, z(t) < 1 for all sufficiently large t.
How do we know that this path is optimal? If we use the control conditions to express
C and z in terms of K and , we see that they depend only on and not on K. Hence
the maximised Hamiltonian, like the ordinary one, is linear in the state variable. Thus
the standard concavity condition is satisfied; the transversality condition is satisfied
because > 0.
29
INTRODUCTION TO ANALYSIS
291.
(iii) By assumption, 2/u is a positive number such that (2/u)2 > 2. Arguing as in
(i), we obtain a real number t such that 0 < t < 2/u and t2 > 2. Then w = 2/t
satisfies our requirements.
(iv) By the axiom of Archimedes, we may choose a natural number N > (w u)1 .
By the same axiom, the set of non-negative integers k such that k N u is finite;
let K be the greatest member of this set. Then
N u < K + 1 < N u + N (w u) = N w,
so we may let x be the rational number (K + 1)/N . To show that u is not an
upper bound for A, it suffices to show that x A. We have already ensured that
x is positive and rational, so it remains to prove that x2 < 2. This is so because
2 x2 is the sum of the positive numbers 2 w2 and (w + x)(w x).
(v) The proof is by contraposition. Let c be a real number such that c > 0 and c2 6= 2;
we wish to show that c 6= s. If c2 > 2 we may apply (ii) with v = c, inferring that
c > s. If c2 < 2 we may apply (iv) with u = c, inferring that c is not an upper
bound for A; in particular, c 6= s.
(vi) Let a be a real number such that a > 1, and let A = { x Q : x > 0 and x2 < a }.
Then 1 A, so A is non-empty. By the axiom of Archimedes there is a positive
integer M such that M > a > 1; hence M 2 > M > a. Thus if x A then
(M x)(M + x) = M 2 x2 > a x2 > 0,
z = 21 (v v 1 a).
Our assumptions about v imply that y and z are positive, and it is easy to show
that y = v z < v and y 2 = a + z 2 > a. This proves the analogue of (i) when 2 is
replaced by a (except where 2 is used as an index); the analogues of (ii)(v) are
proved exactly as above. Hence s2 = a.
Finally, let b be a real number such that 0 < b 1; we wish to prove that there is
a positive real number such that 2 = b. If b = 1 then = 1. If 0 < b < 1 then,
for reasons just explained, there is a positive real number s such that s2 = 1/b; we
may therefore let = 1/s.
292.
(i) Let an = n1/n 1 n N. Since n > 1 if n > 1 and > 0, an > 0 n > 1. Let
n > 1 and apply the given inequality with a = an :
n>
2
n(n 1) 2
an , whence a2n <
.
2
n1
If > 0 and N is a natural number greater than 2/2 , then 0 < an < n > N .
Therefore limn an = 0, as required.
(ii) Let a = b1 1 > 0. For all n, bn > 21 n(n 1)a2 by the given inequality.
Therefore 0 < nbn < (2/a2 )(n 1)1 n > 1, and nbn 0 by SQ1 in Section
293.
To prove the last part, let c = b1/k . Then 0 < c < 1 and nk bn = (ncn )k for all n. By
(ii), with b replaced by c, limn ncn = 0. But then limn (ncn )k = 0 by repeated
application of SQ2, part (b), in Section 293.
63
293.
(i) Let
sn =
n
X
xr ,
tn =
n
X
ur .
r=1
r=1
r=m+1
r=m+1
(ii) Since limn (xn /un ) = 1, we may choose an integer k such that 0 < xn /un <
2 n > k. Since un > 0, 0 < xn < 2un n > k, and hence |xn | < 2un for such n.
We may therefore repeat the argument of (i), replacing un by 2un and assuming
m > n > k in the chain of inequalities.
2n + 3
2n
For the last part, let xn = n
, un = n . Then
3 +2
3
294.
1 + 32n
xn
=
1 as n .
un
1 + 23n
P
P
2
Now
1 23 = 2. Hence by (ii), the the series nr=1 xr is convergent.
r=1 ur = 3
(i) Let x0 I, > 0. By definition of uniform continuity, > 0 such that x I and
|x x0 | < imply |f (x) f (x0 )| < . Since this argument is valid for all > 0, f
is continuous at x0 . Since this is so for all x0 I, f is continuous on I.
(ii) Suppose f : I R is not uniformly continuous. Then > 0 with the following
property: for any > 0, however small, there exist u, v I such that |u v| <
and |f (u) f (v)| . Using this fact for = 1, 12 , 13 . . . . , we obtain sequences
{un } and {vn } in I such that |un vn | < 1/n and |f (un ) f (vn )| for every
natural number n. Since a vn b for all n, it follows from SQ6 that there is
a subsequence {vnk } of {vn } which converges to a real number x
with the same
property; thus x
I. [This is where we use the assumption that I contains its
endpoints. Without that assumption, the proof doesnt work and the conclusion
may be false: see (iii) below.] We shall prove that f is not continuous at x
.
Since un vn 0 as n , limk unk = limk vnk = x
. But |f (un )f (vn )|
for all n, so the sequences {f (unk )} and {f (vnk )} cannot both converge to f (
x).
The required result now follows from SQ8.
(iii) It suffices to show that for any > 0, there exist real numbers u, v such that
0 < u < v < 1, v < u + and u1 v 1 1. Fix > 0, let v be a real number
such that 0 < v < min(1, 2) and let u = v/2. Then 0 < u < v < 1, v u = 21 v <
and u1 v 1 = v 1 > 1.
30
301.
64
(ii) Let 1 i m; then |ain | han i for all n; therefore, if the sequence {han i}
converges to 0, so does {ain }. Conversely, suppose ain 0 for i = 1, . . . , m. Let
> 0. Then we may choose natural numbers N1 , N2 , . . . , Nm such that
|ain | < n > Ni
(i = 1, . . . , m).
(iii) Let a Rm , and let the integer k be such that 1 k m and |ak | = hai. Then
a2k a21 + . . . + a2m ma2k ,
Taking square roots, we obtain the required result.
(iv) By a basic property of sequences of real numbers (SQ1 in Section 29.3), it suffices
to find positive numbers K and L such that kxn xk Khxn xi for all n and
302. For each n N, let xn = (2n)1 . Then fn (xn ) = 1 and fm (xn ) = 0 m 2n.
If n N and x [0, 1], then |fn (x) f (x)| = fn (x) 1, with equality if x = xn . Thus
d(fn , f ) = 1. If m, n N and x [0, 1], then 0 fm (x) 1 and 0 fn (x) 1, so
|fn (x) fm (x)| 1. If m 2n, fn (xn ) fm (xn ) = 1, whence d(fn , fm ) = 1.
y
y = fn (x)
1
2n
1
n
(i) Since d(fn , f ) = 1 n N, the sequence {fn } does not converge uniformly to f .
To prove pointwise convergence, let x [0, 1]; we show that fn (x) = 0 for all
sufficiently large n. If x = 0 then fn (x) = 0 for all n. If 0 < x 1, we can choose
a natural number N such that N > x1 ; then If n N , x > n1 , so fn (x) = 0.
(ii) Since d(fm , fn ) = 1 whenever m 2n, the sequence {fn } has no subsequence that
is a Cauchy sequence, and hence no convergent subsequence. The sequence {fn }
is bounded because d(fn , f ) = 1 n N.
(iii) Immediate from the proof of (ii) and the definition of compactness.
(iv) (i)(iii) remain true when B[0, 1] is replaced by C[0, 1] because each fn is a continuous function, as is f .
65
303.
(i) Let x I. Then |fm (x) fn (x)| d(fm , fn ) for all m and n, where d is the
metric of B[a, b]. Hence {fn (x)} is a Cauchy sequence of real numbers. By the
completeness of R, the sequence {fn (x)} converges, say to f (x). This defines a
function f : I R such that the sequence {fn } converges pointwise to f .
(ii) Fix n > N and x I; then |fm (x) fn (x)| < m > N . Let > 0. By definition
of the function f we may choose a positive integer M such that |f (x) fm (x)| <
m > M . If m > max(M, N ), then
|f (x) fn (x)| |f (x) fm (x)| + |fm (x) fn (x)| < + .
Since this reasoning is valid for any positive number , however small, |f (x)
fn (x)| . This is so for all x I and all n > N .
(iii) Since the argument of (ii) is valid for every positive number , the sequence {fn }
converges uniformly to f . We may therefore choose a positive integer k such that
|f (x) fk (x)| < 1 x I. Since fk is a bounded function, we may choose a
positive number K such that |fk (x)| < K x I. Then |f (x)| < K + 1 x I.
(iv) Completeness of B[a, b] follows immediately from (i) and (iii). C[a, b] is a closed
set in B[a, b] (Exercise 30.2.7); the completeness of the metric space C[a, b] now
follows from the completeness of B[a, b] and the result of Exercise 30.2.8.
304.
(i) Since we are using d in the conventional manner of integral calculus, we denote
the standard metric on C[a, b] by . We shall show that for all V1 , V2 C[a, b],
(Ve1 , Ve2 ) (V1 , V2 ).
(1)
The required result then follows from the fact that 0 < < 1.
Let V1 , V2 C[a, b] and let y [a, b]. Then we may choose x1 X such that
Z b
(x1 , y, z) V1 (z) dz.
Ve1 (y) = f (x1 , y) +
a
Hence
(2)
For all z [a, b], V1 (z) V2 (z) (V1 , V2 ) and (x1 , y, z) 0, so the integrand on
the right-hand side of (2) cannot exceed (V1 , V2 )(x1 , y, z). Thus (2), together
Rb
with the fact that a (x1 , y, z) dz = 1, implies that
Ve1 (y) Ve2 (y) (V1 , V2 ).
(3)
A similar argument with the roles of V1 and V2 reversed shows that (3) remains
true when its left-hand side is replaced by its absolute value. Since this is so for
all y in [a, b], (1) is also true.
(ii) There is a unique function V C[a, b] such that
Z b
(x, y, z) V (z) dz y [a, b].
V (y) = max f (x, y) +
xX
66