Uses of Elementary Calculus

Arising in Economics
KCB

Contents
1 Continuous compounding of interest 2
2 Continuously variable interest rates 4
3 First order linear dierential equations 9
4 The CobbDouglas Model of Production 11
5 Constant Returns to Scale 13
6 Eulers Theorem for homogeneous functions 14
7 The basic Solow Model of capital accumulation 17
8 Economic Growth in a Solow model 19

When I took calculus, the books were chock full of examples and applications from physics
and engineering, but they had very few applications to economics or other social sciences. As a
result, many students believed that calculus, and more generally, mathematics, is useful only in
the natural sciences and engineering. Current textbooks are better in this regard, but nonetheless
many of my undergraduates are delighted to discover that they can use their mathematical aptitude
in other ways. I have jotted down a few examples for use by calculus teachers who are looking for
something dierent to show their students. These examples are mostly well-known to economists.
1
1 Continuous compounding of interest
When a dollar is invested at the annual rate of interest of r, after a year it is worth
(1 + r). But most nancial institutions compound interest more frequently than
annually. For instance, with monthly compounding, the interest rate is divided by
12, and after the rst month, the amount r/12 is credited to your savings account,
and after two months the 1+
r
12
has earned another months interest (1+
r
12
)(r/12),
so you have (1 +
r
12
)
2
= (1 +
r
12
) + (1 +
r
12
)(r/12) in your account. By the end of
the year, you have (1+
r
12
)
12
. You might think that if interest is compounded daily,
the annual rate would be divided by 365, so the total return would be (1 +
r
365
)
365
,
but banking is an industry where it was once acceptable to use a 360-day year, so
your bank may have payed (1 +
r
360
)
360
instead.
If interest rate r is compounded n-fold annually, at the end of t years you have
C(r, t, n) = (1 +
r
n
)
nt
for every dollar invested.
Lets examine the properties of the function C(r, t, n).
If r
1
t
1
= r
2
t
2
0, write
r
2
r
1
=
t
1
t
2
= a, then C(r
1
, t
1
, n) = C(r
2
, t
2
, an).
To see this, write
C(r
1
, t
1
, n) = (1 +
r
1
n
)
nt
1
= (1 +
r
2
an
)
ant
2
= C(r
2
, t
2
, an).
For r > 0, and t > 0, the limit lim
n
C(r, t, n), if it exists, depends only on
the product rt.
Let r
1
t
1
= r
2
t
2
, where
r
2
r
1
=
t
1
t
2
= a > 0. Then
lim
n
C(r
1
, t
1
, n) = lim
n
C(r
2
, t
2
, an) = lim
n
C(r
2
, t
2
, n).
The limit lim
n
C(r, t, n) exists for each r, t.
In fact, we shall compute the limit in Proposition 1.
Continuous compounding is the limit of n-fold compounding as n .
Proposition 1
lim
n
C(r, t, n) = lim
n
(1 +
r
n
)
nt
= e
rt
.
2
Proof : Now
_
1 +
r
n
_
nt
=
_
e
ln
_
1+
r
n
_
_
nt
= e
nt ln
_
1+
r
n
_
so we rst nd lim
n
nt ln
_
1 +
r
n
_
. Using Taylors Theorem
_
1 +
r
n
_
= ln(1) + ln

(1)
r
n
+ R(n) =
r
n
+ R(n),
where the remainder R(n) satises R(n)/(
r
n
) = nR(n)/r
n
0. Thus
lim
n
nt
_
1 +
r
n
_
= lim
n
nt
_
r
n
+ R(n)
_
= lim
n
rt + tnR(n) = rt.
Since the exponential function is continuous,
lim
n
(1 +
r
n
)
nt
= lim
n
e
nt
_
1+
r
n
_
= e
lim
n
nt
_
1+
r
n
_
= e
rt
.
The next proposition explains why a bank prefers to use a 360-day year for
calculating interest on deposits.
Proposition 2 For r > 0, t > 0, and n > 0, the compounded return C(r, n, t) is an
increasing function of n. That is, more frequent compounding leads to a higher
return.
Proof : Again let us rewrite C as
C(r, n, t) =
_
1 +
r
n
_
nt
=
_
e
ln
_
1+
r
n
_
_
nt
= e
nt ln
_
1+
r
n
_
.
While the economic interpretation of n is as the number of subdivisions per an-
num, there is no mathematical reason that in the formula above that n needs to
be an integer. In fact we can go ahead and dierentiate with respect to n. By the
chain rule,
d
dn
e
nt ln
_
1+
r
n
_
= e
nt ln
_
1+
r
n
_
d
dn
_
nt ln
_
1 +
r
n
__
= te
nt ln
_
1+
r
n
_
_
ln
_
1 +
r
n
_
+ n
1
1 +
r
n
r
n
2
_
= te
nt ln
_
1+
r
n
_
_
ln
_
1 +
r
n
_

r
n
1 +
r
n
_
> 0.
3
The last inequality follows from Lemma 3 immediately below. Since dC/dn > 0,
the function C is an increasing function of n.
Lemma 3 For x > 0,
ln(1 + x) >
x
1 + x
.
Proof : Note that ln(1 + x) = ln(1 + x) ln(1). By the Fundamental Theorem of
Calculus,
ln(1 + x) ln(1) =
_
1+x
1
d
dt
ln(t) dt.
Now
d
dt
ln(t) = 1/t, and for 1 t < 1 + x we have 1/t > 1/(1 + x), so
_
1+x
1
1
t
dt >
_
1+x
1
1
1 + x
dt.
But 1/(1 + x) is independent of t, so
_
1+x
1
1
1 + x
dt =
1
1 + x
_
(1 + x) 1
_
=
x
1 + x
.
Putting it all together
ln(1 + x) =
_
1+x
1
1
t
dt >
_
1+x
1
1
1 + x
dt =
x
1 + x
gives the desired conclusion.
Proof of Lemma 3 based on convex analysis: It is well known that the logarithm
function is strictly concave (its second derivative is strictly negative) and that
strictly concave functions lie below their tangent lines. That is, if f is strictly
concave and dierentiable, f (y) < f (z) + f

(z)(y z) for all y z. Letting y = 1

and z = 1+x, for the log function we get ln(1) < ln(1+x) ln

(1+x) x for x 0,
which rearranges to give the conclusion.
2 Continuously variable interest rates
For this well just consider a unit length of time, so t [0, 1]. Let r(t) denote the
instantaneous rate of interest at time t. By this I mean that the value at time
4
t = 1 of a dollar invested at time t = 0 is gotten by taking the limit as the number
of compounding periods tends to innity of the following simple procedure.
Over the interval between time t = (k 1)/n and t = k/n, the investment earns
at a rate of interest equal to r
_
(k 1)/n
_
if the rate is set prospectively or r(k/n)
if the rate is set retrospectively. But since the length of the compounding period
is 1/n, the interest rate per compounding period must be divided by n. Thus the
ratio of the value at time t = (k 1)/n and t = k/n is just
1 +
r(k/n)
n
in the retrospective case. Therefore the value at time t = 1 of a single dollar
invested at time t = 0 is
n
_
k=1
1 +
r(k/n)
n
.
I claim that if r is continuous, then the limit is
lim
n
n
_
k=1
1 +
r(k/n)
n
= e
_
1
0
r(t) dt
.
Note that this generalizes the continuous compounding case of the previous
section, which corresponds to the function r being constant.
Note too that the continuous varying interest rate r delivers the same return
as continuous compounding of a constant rate equal to its simple time average
_
1
0
r(t) dt. Thus continuous time compounding of an instantaneous rate of return
converts the complex geometric averaging of growth rates to a simple time aver-
age rate of growth.
Formally we have the following.
Proposition 4 (Continuously variable rate of return) Let r : [0, 1] Rbe con-
tinuous. Then
lim
n
n
_
k=1
1 +
r(k/n)
n
= lim
n
n1
_
k=0
1 +
r(k/n)
n
= e
_
1
0
r(t) dt
.
Before we begin the proof, we have another lemma about the logarithm func-
tion. See Figure 1.
5
-1 1 2 3
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
Figure 1. h(x) = ln(1 + x) x.
Lemma 5 Dene the function h on the interval (1, ) by
h(x) = ln(1 + x) x,
so that
ln(1 + x) = x + h(x).
Then h(0) = 0, h(x) < 0 for x 0, h is twice continuously dierentiable, h

(x) > 0
for x < 0, h

(x) < 0 for x > 0, h

(0) = 0, h

(x) < 0 for all x, and

lim
x0
h(x)
x
= 0.
Proof : Clearly h is twice continuously dierentiable as
h

(x) =
x
1 + x
and h

(x) =
1
(1 + x)
2
< 0.
Thus
h

(0) = 0.
(In fact h has continuous derivatives of all orders.) In particular, h is strictly
concave, which means a lot to me, but maybe not to you.
6
To see that h(x) < 0 for x 0, I would normally appeal to strict concavity
and the fact that h

(0) = 0, which means it is a unique maximizer. But lets do

something more fundamental:
By the Fundamental Theorem of Calculus,
h(x) =
_
x
0
h

(t) dt =
_
x
0
t
1 + t
dt < 0.
(Note that if x < 0, integrating from 0 to x reverses the sign, so it is still < 0.)
Finally, to see that h(x)/x 0 as x 0, observe that
lim
x0
h(x)
x
= lim
x0
h(0 + x) h(0)
x
= h

(0) = 0.
(Actually, I could have used Taylors Theorem here.)
Now back to continuously variable rates of return.
Proof of Proposition 4: I will deal with the retrospective case. The prospective
case is similar.
Start by observing that since r is continuous, it is bounded in absolute value.
(This is Weierstrasss theorem on the maximum and minimum of a continuous
function on a compact interval.)
Next we proceed by taking logarithms.
ln
_

_
n
_
k=1
1 +
r(k/n)
n
_

_
=
n

k=1
ln
_
1 +
r(k/n)
n
_
but by the denition of h we have
=
n

k=1
__
r(k/n)
n
_
+ h
_
r(k/n)
n
__
(let me raise a warning ag here, which I will discuss below)
=
n

k=1
r(k/n)
n
+
n

k=1
h
_
r(k/n)
n
_
.
The warning is this: the function h is dened only for x > 1, and there is no
guarantee that
r(k/n)
n
> 1. However, since r is bounded, for large enough n, we
7
must have
r(k/n)
n
> 1. Since we are interested in the limit as n , this is not an
issue.
Thus
ln
_

_
n
_
k=1
1 +
r(k/n)
n
_

_

n

k=1
r(k/n)
n
=
n

k=1
h
_
r(k/n)
n
_
0.
We now obtain a lower bound on the right-hand side. Temporarily x n. Since h
is increasing for x < 0 and decreasing for x > 0, the minimum value of h
_
r(t)/n
_
occurs when t is either the maximum or the minimum value of r. Denote this
value of r by M, which is not necessarily positive. Then h
_
r(t)
n
_
h(M/n) for all Oops! It may
switch between
the max and min
as n changes.
Need to show this
doesnt happen
for large n.
t [0, 1]. Thus we have
0 ln
_

_
n
_
k=1
1 +
r(k/n)
n
_

_

n

k=1
r(k/n)
n

n

k=1
h
_
M
n
_
= nh
_
M
n
_
. ()
But h(x)/x 0 as x 0, so (unxing n),
nh
_
M
n
_
0 as n .
But r is Riemann integrable so we have the limit
n

k=1
r(k/n)
n

_
1
0
r(t) dt as n .
Combining this with () gives
lim
n
ln
_

_
n
_
k=1
1 +
r(k/n)
n
_

_
= lim
n
n

k=1
r(k/n)
n
=
_
1
0
r(t) dt.
Exponentiating, and using the continuity of the exponential function, we have
lim
n
n
_
k=1
1 +
r(k/n)
n
= e
_
1
0
r(t) dt
.
Quod erat demonstrandum. I hope you can see now how to deal with the prospec-
tive case.
8
3 First order linear dierential equations
The following theorem is a standard statement on the solution to a rst order linear
dierential equation. I took it from Apostol [1, Theorems 8.2 and 8.3, pp. 309
310].
Theorem 6 (First order linear dierential equation) Assume P, Q are continu-
ous on the open interval I. Let a I, b R.
Then there is one and only one function y = f (x) that satises the initial value
problem
y

+ P(x)y = Q(x) (3.1)

with f (a) = b. It is given by
f (x) = be
A(x)
+ e
A(x)
_
x
a
Q(t)e
A(t)
dt
where
A(x) =
_
x
a
P(t) dt.
The theorem appears a bit mysterious in this form, but I can give it an eco-
nomic interpretation that makes sense (at least to me). Its easier if we change
some notation. The rst thing we will do is change the variable on which y de-
pends from x to time, t.
Let y(t) denote the value of a savings account at time t. At each point of time it
earns an instantaneous rate of return r(t). Moreover, we add a ow of additional
savings to the account at the rate s(t). Thus the rate of change of the value of the
account is
y

(t) = r(t)y(t) + s(t). (3.2)

This is just (3.1) with t replacing x, r(t) replacing P(x), and s(t) replacing Q(x).
Moreover, lets rewrite the initial condition as y(t
0
) = y
0
. If the instantaneous rate
of return at time t is r(t), then by Proposition 4, the average rate of return r(t) over
the interval [t
0
, t] is just
r(t) =
1
t t
0
_
t
t
0
r() d.
Now if we add nothing to the initial investment over time, that is, if s(t) = 0
for all t, then the value of the account at time t is given by
y(t) = y
0
e
r(t)(tt
0
)
. (3.3)
9
That is, earning the varying rate of return r over the interval [t
0
, t] is the same
as earning the average rate of return r over the interval. We can verify this by
showing that y given by (3.3) solves (3.2).
dy
dt
=
d
dt
y
0
e
r(t)(tt
0
)
= y
0
e
r(t)(tt
0
)
d
dt
( r(t)(t t
0
))
= y
0
e
r(t)(tt
0
)
d
dt
_
t
t
0
r() d
= y
0
e
r(t)(tt
0
)
r(t)
= r(t)y(t),
which is (3.2) with s = 0.
But in general, the additional savings s(t) is not zero. In order to deal with the
general case, we introduce the incredibly useful notion of present value. If you
invest $1 at time t
0
it will be worth $e
r(t)(tt
0
)
at time t, so the value at time t
0
, that
is, the present value, of $1 at time t is e
r(t)(tt
0
)
= 1/e
r(t)(tt
0
)
. For if you invest
e
r(t)(tt
0
)
at t
0
, you will have e
r(t)(tt
0
)
e
r(t)(tt
0
)
= 1 dollar at time t.
The present value of the ow s(t) is s(t)e
r(t)(tt
0
)
. The present value S (t) of all
the additional savings up to time t is thus
S (t) =
_
t
t
0
s()e
r()(t
0
)
d.
But at time t this present value will be worth an additional
S (t)e
r(t)(tt
0
)
.
Thus the total value of the savings account at time t is given by
y(t) = (y
0
+ S (t)) e
r(t)(tt
0
)
.
This then is the solution of the dierential (3.2), which we shall now verify.
Dierentiating with respect to time yields
y

(t) =
_
(y
0
+ S (t)) e
r(t)(tt
0
)
d
dt
( r(t)(t t
0
))
_
+
_
S

(t)e
r(t)(tt
0
)
_
=
_
(y
0
+ S (t)) e
r(t)(tt
0
)
r(t)
_
+
_
s(t)e
r(t)(tt
0
)
e
r(t)(tt
0
)
_
= r(t)y(t) + s(t).
10
The initial condition y(t
0
) = y
0
is satised, since S (t
0
) = 0. Note too that this is
the same solution as given by Apostol, when translated as above. In other words,
we have the following translation.
Theorem 7 (First order linear dierential equation) Assume r, s are continu-
ous on the open interval I. Let t
0
I, y
0
R.
Then there is one and only one function y = f (t) that satises the initial value
problem
y

= r(t)y + s(t)
with f (t
0
) = y
0
. It is given by
f (t) =
_
y
0
+ S (t)
_
e
r(t)(tt
0
)
where
r(t) =
1
t t
0
_
t
t
0
r() d
and
S (t) =
_
t
t
0
s()e
r()(t
0
)
d.
4 The CobbDouglas Model of Production
In the 1920s, the economist (and later U.S. Senator) Paul Douglas was trying to
model the relationship between national output Y, the level of labor employment
L, and the amount of capital equipment (machinery, etc.) K for the U.S. as a
whole. That is, he was looking for a function F that satised
Y = F(K, L)
for all the years that he had data on. Such a function is called a production
function. Using data from the National Bureau Of Economic Research for the
decade 19091918, he observed that the fraction of Y paid to labor each year was
fairly constant at .74, with the remainder going as a return to capital, despite the
fact the wages of labor uctuated, as did the amount of capital and labor employed.
He consulted with Charles Cobb for a function F that might account for this. They
were successful and published A Theory of Production in 1928 [2].
I imagine that they approached the problem this way: Assume that manufac-
turers maximize their prot when producing output. Each laborer is paid an annual
11
wage w (measured in units of output) and capital is rented at the annual rate r per
unit. The prot is then, output less wage payments less returns to capital,
F(K, L) wL rK.
At a maximum, the derivative of this with respect to each of K and L must be zero.
That is,
dF
dK
r = 0 and
dF
dL
w = 0.
In addition, it was observed that for = .74,
wL
F(K, L)
= and
rK
F(K, L)
= 1 .
Let us take this observation one step further, and assume that these latter rela-
tions will always hold. Then combining the zero derivative condition for prot
maximization with the xed shares, we must have
dF
dK
= r = (1 )
F(K, L)
K
and
dF
dL
= w =
F(K, L)
L
.
This gives us a pair of dierential equations:
dF
dK
= (1 )
F
K
and
dF
dL
=
F
L
.
We would write the second equation more traditionally as
Y

=
Y
L
and
Y

Y
=

L
which is a fairly simple equation. Noting that
Y

Y
= (ln Y)

, we have
(ln Y)

L
,
so
ln Y =
_
1
L
dL + C = ln L + C
1
,
where C
1
is a constant of integration.
12
Similar reasoning implies
ln Y = (1 ) ln K + C
2
,
so we conclude that
ln Y = ln L + (1 ) ln K + C
3
.
Exponentiating, we get
Y = AL

K
1
,
where A = e
C
3
. A function F of the form
F(K, L) = AL

K
1
is known to economists as a CobbDouglas production function. When A = 1,
it is known to mathematicians as a weighted geometric mean. It is still used as
a model of national economies, but in the U.S., the coecient of total factor
productivity, A, increases by 2.54.0 percent per year due to technical change.
(See, e.g., Solow [5].)
5 Constant Returns to Scale
The CobbDouglas production function F(K, L) = AL

K
1
exhibits a property
that economists refer to as constant returns to scale. What this means is that if
we multiply both capital and labor by the same positive number t, then output also
gets multiplied by t. That is,
F(tK, tL) = tF(K, L),
for all nonnegative values of t, K, L. We can verify that the CobbDouglas pro-
duction function has this property:
F(tK, tL) = A(tk)
1
(tL)

= t
1
t

AK
1
L

= tF(K, L).
More generally, mathematicians will say that a function g of mvariables x
1
, . . . , x
m
is (positively) homogeneous of degree k if
g(tx
1
, . . . , tx
m
) = t
k
g(x
1
, . . . , x
m
)).
13
Thus constant returns to scale is just homogeneity of degree 1.
One of the most useful results about homogeneous functions is Eulers Theo-
rem, which is stated and proven below. But there are some useful observations to
be made directly.
Dene the output per capita to be Y/L, that is, total output divided by total
employment. In the, U.S., total employment is generally about 2/3 of the popu-
lation, so output per capita is a measure of average economic well-being, or as
economists, say welfare. Under constant returns to scale, we have
Y
L
=
1
L
F(K, L) = F
_
K
L
,
L
L
_
= F
_
K
L
, 1
_
.
That is, output per capita depends only on the quantity of capital employed by
each worker. This suggests that a major source of economic progress is capi-
tal accumulation, the increase over time of the quantity of capital used by each
worker. We know now that there is more to progress than capital accumulation
technological improvement of machines and the improved skills of workers are
both important factors. In Section 7, we will examine a simple model of capital
accumulation.
6 Eulers Theorem for homogeneous functions
Eulers theorem Let f : R
n
+
R be continuous, and also dierentiable on R
n
++
.
Then f is homogeneous of degree k if and only if for all x R
n
++
,
k f (x) =
n

i=1
f (x)
x
i
x
i
. ()
Proof : (=) Suppose f is homogeneous of degree k. Then for all 0,
g
x
() := f (x)
k
f (x) = 0.
Therefore
g

x
() = 0
for all > 0. But by the chain rule,
g

x
() =
n

i=1
f (x)
x
i
x
i
k
k1
f (x).
14
Evaluate this at = 1 to obtain ().
(=) Suppose
k f (x) =
n

i=1
f
i
(x)x
i
for all x. Fix any x 0 and set
g
x
() = f (x)
k
f (x)
and note that g
x
(1) = 0. Then
g

x
() =
n

i=1
f
i
(x)x
i
k
k1
f (x)
=
1
_

_
n

i=1
f
i
(x)x
i
_

_
k
k1
f (x)
=
1
k f (x) k
k1
f (x),
so
g

x
() = k
_
f (x)
k
f (x)
_
= kg
x
().
Since is arbitrary, g
x
satises the following dierential equation:
g

x
()
k

g
x
() = 0
and the initial condition g
x
(1) = 0. By Theorem 6,
g
x
() = 0 e
A()
+ e
A()
_

1
0 e
A(t)
dt = 0
where, irrelevantly, A() =
_

1
k
t
dt = k ln . This implies g
x
is identically zero,
so f is homogeneous on R
n
++
. Continuity guarantees that f is homogeneous on
R
n
+
.
Corollary 8 Let f : R
n
+
R be continuous and dierentiable on R
n
++
. If f is
homogeneous of degree k, then f
j
is homogeneous of degree k 1.
15
Proof if f is twice dierentiable: By the rst half of Eulers theorem,
n

i=1
f
i
(x)x
i
= k f (x)
so
d
dx
j
_

_
n

i=1
f
i
(x)x
i
_

_
= k

x
j
f (x)
or
n

i=1
f
i j
x
i
+ f
j
= k f
j
or
n

i=1
f
i j
(x)x
i
= (k 1) f
j
. (6.1)
Thus f
j
is homogeneous of degree (k 1) by the second half of Eulers theorem.
Proof without twice dierentiability: The dierence quotients satisfy
f (x + h) f (x)
h
=

k
f (x + h)
k
f (x)
h
=
k1
f (x + h) f (x)
h
whenever > 0. Thus f is dierentiable at x if and only if it is dierentiable at
x and f
i
(x) =
k1
f
i
(x) for all i = 1, . . . , n.
Corollary 9 If f is homogeneous of degree k, then
f
i
(x)
f
j
(x)
=
f
i
(x)
f
j
(x)
for > 0 and x R
n
++
.
Proof : By Corollary 8 each f
i
satises f
i
(x) =
k1
f
i
(x), so
f
i
(x)
f
j
(x)
=

k1
f
i
(x)

k1
f
j
(x)
=
f
i
(x)
f
j
(x)
.
16
Corollary 10 If f is homogeneous of degree 1 and twice dierentiable, then the
Hessian matrix [ f
i j
(x)] is singular for all x R
n
++
.
Proof : By (6.1),
n

i=1
f
i j
(x)x
i
= (k 1) f
j
.
When k = 1 this becomes [ f
i j
(x)]x = 0 in matrix terms, so for x 0 we conclude
that [ f
i j
(x)] is singular.
7 The basic Solow Model of capital accumulation
This family of models of capital accumulation is more or less due to Robert
Solow [4]. Dierent versions make various simplifying assumptions for ease of
exposition. Here are the basic variations.
The labor force L(t) at time t is assumed to grow at a constant rate n.
The ow of output Y(t) at time t is given by by a constant returns to scale
production function F. That is,
Y(t) = F
_
K(t), L(t)
_
,
where K(t) is the total capital stock at time t.
In some versions, technological progress is incorporated through the growth
of a coecient A(t) that enters as either a coecient of total factor produc-
tivity,
Y(t) = A(t)F
_
K(t), L(t)
_
,
or a coecient of labor augmenting technological change,
Y(t) = F
_
K(t), A(t)L(t)
_
.
In most versions, A is assumed to grow at a xed rate x. More recent work
allows the growth rate of progress to respond to economic conditions.
A constant fraction s of output is saved and is added as ow to the capital
stock:
K

(t) = sY(t).
17
More sophisticated models allowthe savings rate to depend on interest rates,
but such models are not generally referred to as Solow models.
In some versions, capital depreciates at rate , so
K

(t) = sY(t) K(t).

For concreteness, F may be assumed to be a CobbDouglas production
function,
F(K, L) = K

L
1
.
This is not just for simplicity, it actually ts the data rather well.
To analyze this model, it is convenient to work in per capita terms. We shall
use output per capita as our proxy for economic welfare. So dene
y(t) =
Y(t)
L(t)
, k(t) =
K(t)
L(t)
.
Then
y =
F(K, L)
L
= F(K/L, L/L
.,,.
=1
) = f (k)
With depreciation,
K

(t) = sY(t) K(t).

We want to translate this to per capita terms. We assumed that population growth
is constant,
L

(t)
L(t)
= n or L(t) = L
0
e
nt
.
Using this and the denition of k, we have
K(t) = k(t)L(t) = k(t)L
0
e
nt
which implies by the chain rule that
K

(t) = k

(t)L
0
e
nt
+ nk(t)L
0
e
nt
.
But on the other hand, our savings assumption (which is an assumption on behav-
ior) says that
K

(t) = sY(t) K(t) = sL

0
e
nt
f
_
k(t)
_
k(t)L
0
e
nt
.
18
Combining these gives
k

(t)L
0
e
nt
+ nk(t)L
0
e
nt
= sL
0
e
nt
f
_
k(t)
_
k(t)L
0
e
nt
or, dividing by L
0
e
nt
and rearranging,
k

(t) = s f
_
k(t)
_
(n + )k(t). (7.1)
This dierential equation describes the process of capital accumulation. It is a
rst order nonlinear dierential equation.
I can solve (7.1) for the CobbDouglas case. When F(K, L) = K

L
1
, then
f (k) = F(k, 1) = k

and (7.1) becomes

k

= sk

(n + )k. (7.2)
There are two solutions to this, depending on k
0
= k(0). If k
0
= 0, then k(t) = 0
for all t. The more interesting case is when k
0
> 0, in which case
k(t) =
__
k
1
0

s
n +
_
e
(n+)(1)t
+
s
n +
_
1
1
. (7.3)
Consequently, the capital/labor ratio tends to a nite limit, independent of the
starting point for k
0
> 0, namely,
k(t)
_
s
n +
_
1
1
as t .
This in turn implies that per-capita income y(t) = k(t)

also tends to a limit,

y(t)
_
s
n +
_

1
as t .
But this is at odds with two centuries of sustained growth of per-capita income at
a rate of about two percent per year. So where does economic growth come from?
8 Economic Growth in a Solow model
The generally accepted explanation for economic growth is not, as Marx [3] hy-
pothesized, simply accumulation of capital. Instead growth in per-capita income
19
comes from growth in productivity. The simplest model of this is to take produc-
tivity growth as exogenously given, and write the CobbDouglas case as
y(t) = e
t
k(t)

,
is the exogenously given rate of productivity advance. In this case, (7.2) be-
comes
k

(t) = se
t
k(t)

(n + )k(t), (8.1)
which has the solution
k(t) =
_
_
k
1
0

s
n +
_
e
(n+)(1)t
+
(1 )s
(n + )(1 ) +
e
t
_
1
1
. (8.2)
Observe that for = 0, this reduces to (7.3), but k(t) does not tend to a limit. But
lets look at the growth rate k

(t)/k(t). From (8.1)

k

(t)
k(t)
= se
t
k(t)
1
(n + )
Substituting the solution (8.2) into the right-hand side gives
k

(t)
k(t)
=
se
t
_
k
1
0

s
n+
_
e
(n+)(1)t
+
(1)s
(n+)(1)+
e
t
(n + ),
from which it is easy to see that
k

(t)
k(t)


1
as t .
Since y(t) = e
t
k(t)

, we have y

(t) = e
t
k(t)

+ e
t
k(t)
1
k

(t), so (y

/y) =
+ (k

/k), so
y

(t)
y(t)
+

1
=

1
as t .
In other words, with a CobbDouglas technology, the only thing that determines
the long run rate of growth of the per-capita income is , the rate of growth of
productivity, and , the marginal productivity of capital. In particular, the savings
rate s, and the population growth rate n have no impact on the long run rate of
growth of per-capita income!
20
References
[1] Apostol, T. M. 1967. Calculus, 2d. ed., volume 1. Waltham, Massachusetts:
Blaisdell.
[2] Cobb, C. W. and P. H. Douglas. 1928. A theory of production. American
Economic Review 18(1):139165. Supplement, Papers and Proceedings of
the Fortieth Annual Meeting of the American Economic Association.
[3] Marx, K. and F. Engels. 1847. Manifesto of the communist party. English
edition, 1888.
[4] Solow, R. M. 1956. A contribution to the theory of economic growth. Quar-
terly Journal of Economics 70(1):6594.
[5] Solow, R. M. 1957. Technical change and the aggregate production function.
Review of Economics and Statistics 39:312320.
21