Q

MATHEMATICS
1

Thijs Jansen
Christian Kerckhoffs

c July 2019, Maastricht University, School of Business and Economics.

Except as permitted by law, no part of this publication may be reproduced or distributed in any form or by

any means, or stored in a data base or retrieval system, without the prior written permission of the authors.
 CONTENTS

1 FUNCTIONS OF ONE VARIABLE, PART 1 1

1 Functions 1
1.1 Relations 2
1.2 The concept of a function 4
1.3 The graph of a function 6
1.4 Inverting the roles of the variables 8
2 Linear and quadratic functions 12
2.1 An economic decision problem 12
2.2 Linear functions 13
2.3 Quadratic functions 16
End-of-chapter exercises 18

2 FUNCTIONS OF ONE VARIABLE, PART 2 23

1 Broken functions 23
2 Power functions 26
3 Exponential and logarithmic functions 28
3.1 Exponential functions 29
3.2 Logarithmic functions 31
End-of-chapter exercises 35

3 THE DERIVATIVE 41
1 The idea behind the derivative 41
1.1 The average rate of change 41
1.2 The rate of change 44
2 Finding the derivative of some basic functions 48
3 Marginal analysis 51
4 Elasticity 54
End-of-chapter exercises 58

4 FINDING THE DERIVATIVE 63

1 The Product Rule and the Quotient Rule 63
2 The Chain Rule 68
3 The derivative of exponential and logarithmic functions 71
3.1 The derivative of exponential functions 72
3.2 The derivative of logarithmic functions 76
End-of-chapter exercises 78
5 PROPERTIES OF A FUNCTION 83
1 Some characteristic properties of a function 83
2 The role of the first and second derivative 86
2.1 The first derivative 86
2.2 The second derivative 91
2.3 The second-derivative test 93
3 Finding the extreme values of a function 95
End-of-chapter exercises 99

6 FUNCTIONS OF TWO VARIABLES 105

1 Functions of two variables: The basic concepts 106
2 Visualizing a function of two variables 109
3 Level curves for functions of two variables 111
4 Partial derivatives 115
4.1 Determining partial derivatives 116
4.2 A geometrical interpretation of partial derivatives 119
5 The use of partial derivatives in economics 120
End-of-chapter exercises 125

7 APPROXIMATION AND OPTIMIZATION 131

1 Marginality 132
2 An example of two-variable optimization: Minimizing inventory cost 136
3 Finding extreme values 139
3.1 First-order conditions: Stationary points 139
3.2 Second-order conditions: The criterion function 141
4 Minimizing inventory cost 144
End-of-chapter exercises 146
Functions of one variable, part 1 1

1 FUNCTIONS OF ONE VARIABLE, PART 1

In the first two chapters of this reader, we will introduce several economic and business
problems, and show that they can be represented by mathematical models. We will also show
(and get back to this more formally in later chapters) that such models can then be used to
solve the problem at hand, e.g. to find an optimal managerial decision.
The basic building blocks of such mathematical models are various types of functions, e.g.
linear, quadratic, broken and exponential functions. This should sound familiar from sec-
ondary school, and indeed: there are hardly any new concepts in these first two chapters.
Our aim is simply to rehearse some basic mathematical concepts from secondary school, as
far as they are useful to your current study.
In Section 1, we begin by recapitulating the general definition of a function.
In Section 2, we deal with the classes of linear and quadratic functions. As an example, we
present the manager of a restaurant who wants to determine the price for some menu in order
to maximize his profits. As we will see, modeling this problem naturally brings in both linear
and quadratic functions.
In next week’s Chapter 2, we will address several increasingly complex classes of functions.

1 FUNCTIONS
Functions are an important tool in every area where we apply mathematics. For instance, in
economics and business we deal with demand and supply functions, cost functions, production
functions, profit functions, etc. In the first few chapters of this reader we will consider many
economic examples in which functions of one variable play an essential role. In Chapters 6
and 7 we will extend our discussion to functions of two variables.

Before going into detail, Example 1 below reviews most of the concepts that are central in
this chapter.
2 Functions of one variable, part 1

EXAMPLE 1 (INVERSE) DEMAND

The demand for a specific good depends on its price (and on other factors which we will
ignore here). Usually, demand decreases when the price increases. For simplicity one often
assumes a linear relationship between demand, represented by D, and price, represented by
P . An example would be
D = 20 − 4P.

This formula tells us the demand corresponding to a given price. If, for example, the price
equals P = 2, then demand follows as D = 20 − 4 × 2 = 12. The equation above can be
represented graphically as a straight line: the demand curve. Our formula expresses D in
terms of P . Alternatively, we say that demand is a function of price. This reflects our starting
point that demand depends on price.
Sometimes a firm may know how much it wants to sell, and wonders what price it can charge.
To address this question, we have to solve the formula above for P , leading to

P = 5 − 0.25D.

Since P has been written in terms of D, price now depends on demand, rather than the other
way around. E.g., if a firm wants to sell D = 8 units, the new formula implies that it can
charge a price of P = 5 − 0.25 × 8 = 3. The new formula can again be represented graphically
by a straight line. As the roles of P and D have been exchanged or inverted compared to the
original situation, we speak of the inverse demand curve.

1.1 RELATIONS

In this section we consider several examples of a relation between two quantities. We describe
each relation by means of a formula or equation.

EXAMPLE 2 A LINEAR RELATIONSHIP

Imagine that you deposit 200 euros in a savings account that offers an annual interest rate of
r percent. After one year the amount has grown to K euro, where
 r 
K = 200 1 + = 200 + 2r.
100
In this formula, K and r are called variables, i.e. symbols that represent a quantity. The
formula describes a relation between these two variables. For example, if the interest rate
equals r = 5 (i.e. 5%), you end up with K = 200 + 2 × 5 = 210 euros after one year.

EXAMPLE 3 A QUADRATIC RELATIONSHIP

If a driver uses his brakes in case of an emergency, the car doesn’t stop immediately. The
distance which a vehicle will travel from the point when its brakes are fully applied to when
it comes to a complete stop is called the braking distance. Given an initial speed of v km/h,
the braking distance B in meters is given by

v2
B= .
2d
Functions of one variable, part 1 3

Here, B and v are the variables, while d is a constant: it measures the rate of deceleration of
the car, which depends on things like its weight and the quality of its brakes and tires. In the
event of a dry road surface, for the average family car, the value of d is around 7.7 m/sec2 .
Taking care of the dimensions of the variables B and v and the constant d (kilometers versus
meters, seconds versus hours) we end up with:

B = 0.005 v 2 .

According to this formula, an initial speed of 50 km/h corresponds with a braking distance
of 12.5 meter (check!).

EXAMPLE 4 A SQUARE ROOT RELATIONSHIP

Imagine that two sides of a triangle have length 9, while the third side (its base) has length s:
s

h
9 9

FIGURE 1 A triangle with base s

The Pythagorean Theorem implies that 92 = ( 21 s)2 + h2 , so the height h of the triangle can
be calculated as q
h= 81 − 14 s2 .
Hence, the area A of the triangle follows as
q
A = 21 sh = 12 s 81 − 14 s2 .

Here the variables are A and s. For instance, if s = 12, the area follows as A ≈ 40.25.

EXAMPLE 5 A ’BROKEN’ RELATIONSHIP

A farmer wants to fence in part of a large piece of grassland. The fenced area should have a
size of 100 m2 . To keep his materials costs as low as possible, the farmer wants to minimize
the length of the fence. If ℓ denotes the length of the fenced area (in meters) then, to get the
desired size, its width w (in meters) must be
100
w= .
ℓ
Hence, the total length L of the fence is given by
200
L = 2ℓ + 2w = 2ℓ + .
ℓ
Here the variables are L and ℓ. To illustrate, if the farmer chooses the length of the fenced
area to be ℓ = 20 meter, then the length of the fence follows as L = 50 meter.

EXAMPLE 6 AN EXPONENTIAL RELATIONSHIP

Imagine again that you deposit 200 euros in the savings account we introduced in Example
2. The interest rate is r = 5. After two years this amount has grown to K euros, where
 5  5 
K = 200 1 + 1+ = 200 · 1.052 .
100 100
 4 Functions of one variable, part 1

Generalizing, after t years this amount has grown to

K = 200 · 1.05t .

Here the variables are K and t. If, for instance, t = 10, then you end up with K = 200·1.0510 =
325.78 euros.

1.2 THE CONCEPT OF A FUNCTION

In each of the examples of the previous section, the value of one of the two variables involved
is completely determined by the value of the other one:

example first variable second variable

2 the final amount K by the interest rate r
3 the braking distance B by the initial speed v
4 the area A of the triangle by the length s of its base
5 the length L of the fence by the length ℓ of the fenced area
6 the final amount K by the number of years t

(in)dependent variable For this reason, we refer to the first variable as the dependent variable and to the second
one as the independent variable. The formula in each example gives us a rule that tells us
how to calculate, for each value of the independent variable, the corresponding value of the
dependent variable.

function In each of the examples, we have expressed the dependent variable in terms of the independent
one. We may also say that the dependent variable is a function of the independent one. For
instance, in Example 2, the final amount K is a function of the interest rate r. This is denoted
as K = k(r), where
k(r) = 200 + 2r.

This notation brings out clearly that K is the dependent variable and r the independent
one. For private investors, the interest rate on a savings account, r, cannot realistically be
negative. So in the formula for k(r), only nonnegative values for r are allowed. Therefore we
domain say that the function k has the domain [0, ∞).
Generalizing, we have:

FUNCTION OF ONE VARIABLE

A function f of one variable x is a rule or formula that can be used to calculate, for each
value of the variable x, a unique number f (x).
domain Of course this only makes sense for those values of x for which it is possible and meaningful
to evaluate the number f (x). The set of all these values of x is called the domain of the
function f .
range The set of all possible values f (x) is called the range of the function f .
Functions of one variable, part 1 5

To get an adequate mental picture, it may be helpful to think of a function f as analogous to

a device or a machine, which converts an input value, say x, into a unique output value f (x).

device
x input output f (x)
f

E.g., picture those cheap electronic odometers which you can attach to your bike. Here, the
input value is the number of rotations of your wheel. The device counts this number and,
using some rule that has been programmed into its chip, converts it into the output value:
the distance traveled. Another example: a parking meter. The input value is the amount
of cash you throw into it; using some rule, the device converts that amount into the output
value, i.e. the allowed amount of parking time.

EXAMPLE 7 BRAKING DISTANCE

Continuing Example 3, remember that the braking distance B (in meters) of a car depends
on its velocity v (in km/h), according to the formula

B = 0.005 v 2 .

Clearly, the braking distance is a function of the car’s velocity: B = b(v), where the function
b is given by
b(v) = 0.005 v 2 .

Its domain is of course (0, ∞). Check that the range of the function is also (0, ∞).

EXERCISE 1 Write down the functions that correspond with the relations described in the
Examples 4, 5 and 6. Also give the domain of each of these functions.

For the ’braking distance function’ b from Example 7 above, we can easily establish that

b(50) = 0.005 × 502 = 12.5.

In other words: as seen before, the braking distance of a car with a velocity of 50 km/h is
12.5 meter. We call 12.5 the value of b at 50. Continuing our ’device’ analogy, in case of
the ’braking distance function’ b, you could think of your pocket calculator as the relevant
device. If you type e.g. the number 50 into it then, after pushing the appropriate buttons,
your calculator will produce the value 12.5.

device
50 input output 12.5
b

EXERCISE 2 Find f (4) if

√ 4
(a) f (x) = 1 + x (b) f (x) = (c) f (x) = x3 − 2.
x2
Is the number 1 contained in the range of each of the foregoing functions?
6 Functions of one variable, part 1

1.3 THE GRAPH OF A FUNCTION

In the previous section we have described a number of relations between two variables by
means of a function. At secondary school you have learned to visualize such a function by
means of a graph, showing the dependent variable along the vertical axis and the independent
one along the horizontal axis.

Let’s use Example 2 to illustrate. As indicated, the interest rate on a private savings account
cannot be negative. Additionally, let’s assume that market conditions imply that the interest
rate cannot realistically exceed 20%. Under these assumptions, the domain of the function k
given by
k(r) = 200 + 2r

is the interval [0, 20].

Since k is a linear function, its graph is a straight line, so it’s enough to determine two
points on its graph. The obvious choices are r = 0 and r = 20, for which we can calculate
the corresponding values K = 200 and K = 240, respectively. Next, we represent the pairs
(0, 200) and (20, 240) as points in a coordinate system. Finally, we draw the line (segment)
between these two points. Doing so, we get Figure 2 below.

K
240

K = 200 + 2r

200

20 r

FIGURE 2 The graph of the function k

Now consider an arbitrary function f given by the formula

y = f (x).

In order to visualize this function f , we consider for each value a in its domain the associated


value f (a). Then we present all pairs a, f (a) as points in a coordinate system. Thus we
obtain a curve which we call the graph of the function f .

GRAPH OF A FUNCTION


The graph of a function f of one variable x consists of all points a, f (a) in a coordinate
system, where a belongs to the domain of f .
The graph of f is also called the curve y = f (x).

In the left-hand panel of Figure 3 you find the graph of an arbitrary function f , where we
have explicitly indicated its domain and range along the axes.
 Functions of one variable, part 1 7
y
y





 y = g(x)

 y = f (x)
range







x
zero
| {z } x
domain

FIGURE 3 Basic features of a function

zero of a function As we will see lateron, the point (or points) where the graph of a function intersects the
horizontal axis are often of special interest. The right-hand panel of the figure above sketches
an arbitrary function g, which has only one such intersection point. You can find the x-
coordinate of that point by solving the equation g(x) = 0. The solution of this equation is
called the zero of the function g. We emphasize that, depending on the shape of their graph,
other functions may have no zero (like the function in the left-hand panel of Figure 3), or
more than one zero.

EXAMPLE 8 BRAKING DISTANCE

In Example 7 we considered the function b defined by

b(v) = 0.005 v 2 (v > 0).

The graph of the function b is sketched below, in Figure 4. We trust you remember from
secondary school that this graph is in fact part of a parabola.

B
B = 0.005 v 2

FIGURE 4 The graph of the function b

Since the braking distance of a car with velocity 50 km/h equals b(50) = 12.5 meter, the pair

(50, 12.5)
corresponds with a point on the graph of the function b above.

EXERCISE 3 Consider the function h given by h(t) = 2 + t.

(a) Are (1, 6) and (2, 4) points on the graph of h?
(b) Find the value(s) for b such that (b, 9) is a point on the graph of h.
(c) Find the zero(s) of the function h.
8 Functions of one variable, part 1

So far, we have seen that a function of one variable can be represented graphically by a curve
in a coordinate system or plane. An obvious question is now, whether every curve in a plane
can be interpreted as the graph of a function of one variable. Crucially, the answer to this
question is NO. To see why, let’s start by considering the point (50, 12.5) on the graph of
the braking distance function b, addressed in Example 8. Now remember the ’input-output’
interpretation of a function which we discussed in Section 1.2. In the case at hand, the input
value of 50 generates an output value of 12.5.

B
b
unique output 12.5

50 v
input

FIGURE 5 An ’input-output’ interpretation of the point (50, 12.5)

Figure 5 illustrates graphically, how to find the output value from our input value. First,
we intersect the vertical line v = 50, which represents the input value, with the graph of the
function b. Then the output value of b(50) = 12.5 can be found by shifting the intersection
point horizontally to the vertical axis. Note how the arrow head visualizes the ’direction’ of
the operation: from input to output. Equivalently: from the independent to the dependent
variable.

Now the crucial thing is that, for any input value, a function delivers a unique output value.
This property implies that a vertical line like v = 50 cannot intersect the graph of the function
more than once. This brings us to the following conclusion.

VERTICAL-LINE TEST

Any vertical line intersects the graph of a function at most once. If a curve does not pass this
test, then that curve is not the graph of a function.

You can easily verify, that the curve in Figure 5 passes this test: for any vertical line drawn
at a positive value of v, we get precisely one intersection with the curve. This implies that
that curve is indeed the graph of a function (as we already knew).

1.4 INVERTING THE ROLES OF THE VARIABLES

In Section 1.2, when illustrating the ’input-output’ view of a function, one of our examples
was a parking meter. We hope you remember the idea: your input value is the amount of
money which you wack into the coin slot, and then the device calculates your output value,
i.e. the amount of parking time. E.g. if the city council has decided that the parking rate is
2.40 euros per hour then, if you throw in coins with a value of 1.20 euro, the display will tell
you that you’re safe for the next 30 minutes.
Functions of one variable, part 1 9

Nice, but sometimes the situation is the exact opposite. Imagine e.g. that you want to park
your car for 45 minutes. Obviously, you now want to know how much money you have to
throw into that meter. Surely, you can check in a second or so that the answer is 1.80 euros.
The point is, that the role of the variables has now been ’turned around’ or ’inverted’. Initially,
we were thinking in terms of a given amount of money, and the question was how much time
it buys us: so money was our input, and time was our output. Now, we are thinking in terms
of a given amount of time, and the question is how much money we need to buy that amount:
so time is our input, and money our output.
Similarly, reconsider Example 3, where we observed that the relation between the braking
distance B of a car and its velocity v is given by the formula

B = 0.005 v 2 .

For example, we have seen that if the velocity of a car is 50 km/h (our input value), then its
braking distance is 12.5 meter (our output value). Now suppose that, after a car accident in
a 30km/h zone, the police have measured a skid mark of 18 meter. The legal issue will be
obvious: how fast was that car driving? We are now thinking in terms of the braking distance
as our input value, while the velocity is our output. That velocity can be found as follows:

B = 18 =⇒ 0.005 v 2 = 18 =⇒ v 2 = 3600 =⇒ v = 60.

Figure 6 below illustrates both calculations. On the left, we see how a velocity of 50 as input
value leads to a braking distance of 12.5 as output value. On the right, we see how a braking
distance of 18 as input value leads to a velocity of 60 as output value. Note how the arrow
heads visualize that the ’direction’ of the calculation, and therefore the role of the variables,
has indeed been ’turned around’ or ’inverted’.

B B
18

12.5

B = 0.005 v 2 B = 0.005 v 2

50 v 60 v

FIGURE 6 The two problems

Generalizing our numerical example, we get

√
B = 0.005 v 2 =⇒ v 2 = 200B =⇒ v = 200 B.
↑ ↑
B expressed in terms of v v expressed in terms of B

In the foregoing calculation we have inverted the roles of the variables. Below we compare
the two ’devices’ corresponding with the two formulas above.
10 Functions of one variable, part 1

device device
v input output B B input √ output v
B = 0.005 v 2 v = 200 B

EXERCISE 4 In Example 2 we explained that after one year an amount of 200 euro has
grown to K euro, where
K = 200 + 2r
and r is the annual interest rate.
(a) Assume that the amount K equals 220 euro. Find the corresponding
interest rate.
(b) Express r in terms of K.

In the following example, we show how to deal with a more complex formula.

EXAMPLE 9 Suppose that the relation between the variables x and y can by described
by the formula
x−1
y= (x 6= 2).
x−2
In this formula y is the dependent variable whereas x is the independent variable.
Now suppose we want to invert the roles of the variables: x must be the dependent variable
and y the independent one. In other words: we need to express x in terms of y.
To do so, we first multiply both sides of the equation by x − 2 (note that x 6= 2), to obtain
y(x − 2) = x − 1 =⇒ xy − 2y − x = −1 =⇒ x(y − 1) = 2y − 1.
If we now divide both sides of the final equation by y − 1, we get:
2y − 1
x= (y 6= 1).
y−1
Of course, since we cannot divide by 0, we must limit ourselves to values y 6= 1.

EXERCISE 5 Express x in terms of y if

√
(a) y = x2 + 1 (x ≥ 0) (b) y = x3 − 1 (c) y = 21 x0.2

Generalizing the preceding analysis, assume that the relation between the variables x and y
can be described by the formula
y = f (x),
with f some unspecified function. That function expresses y in terms of x: in other words,
the value of the variable y is completely determined by the value of the variable x. As we
have seen before, it is sometimes meaningful (and possible) to express the variable x in terms
of the variable y. In a scheme:
y = f (x) ⇐⇒ · · · · · · ⇐⇒ x = g(y),
where the dots represent the steps that are needed to accomplish the rewriting. Since we
have ’inverted’ the roles of the variables here, the resulting function g is called the inverse of
the original function f . This is reflected in the notation g = f −1 .

Not all functions have an inverse function. In this reader, we will usually confine ourselves
Functions of one variable, part 1 11

to functions whose graph is ’connected’. This means that there are no abrupt changes in the
curve: you can draw it in a single stroke, without having to lift the pen off the paper. For
such functions, the inverse will only exist if their curve is either strictly increasing, or strictly
decreasing. A curve is strictly increasing if it rises along the entire domain of the function,
without ever falling (or ’standing still’). Conversely for a curve which is strictly decreasing.
Figure 7 below shows an example of both these types of curves. Observe that any horizontal
line intersects the graphs in Figure 7 at most once! In fact, this ’horizontal line test’ can be
used to check the existence of an inverse function.
y y

increasing curve decreasing curve

x x

FIGURE 7 Two types of functions possessing an inverse

Finally, we will discuss the relation between the graph of a function, and the graph of its
inverse function (assuming that this inverse exists).
√
EXERCISE 6 Consider the two relationships B = 0.005 v 2 and v = 200 B.
(a) Draw a sketch of the two corresponding curves.
(b) Explain why the point (50, 12.5) lies on the curve B = 0.005 v 2 , whereas
√
the point (12.5, 50) lies on the curve v = 200 B.
(c) What catches your eye if you compare the two curves?

Your conclusion at Exercise 6(c) generalizes.

THE GRAPHS OF A FUNCTION AND ITS INVERSE

The graphs of a function f and its inverse f −1 are symmetric about the line y = x.

This property is illustrated in Figure 8 below.

y x
f f y=x

f −1 f −1

x y

FIGURE 8 The graph of a function and the graph of its inverse

12 Functions of one variable, part 1

2 LINEAR AND QUADRATIC FUNCTIONS

In the previous section, you have definitely noticed that Example 2 involved a linear function,
while Example 3 involved a quadratic one. We will now have a closer and more formal look
at these two important classes of functions.

2.1 AN ECONOMIC DECISION PROBLEM

We start with an example where both linear and quadratic functions pop up naturally.

EXAMPLE 10 A RESTAURANT

A restaurant offers a plat du jour for 30 euros. Every week, 100 customers order this menu.
Based on her experience, the manager expects that if she raises the price to 35 euros, the
interest in this menu will decrease to 80 customers per week. The fixed costs of the restaurant
are 1000 euros per week and the variable costs are 12 euros per menu. The manager wants
to know at what price the restaurant’s profit is maximal.
To solve this problem, we start by introducing the variables p for the price per menu and q
for the weekly number of customers. To keep things simple, suppose that the relationship
between these two variables is linear. More specifically, we assume that q = ap + b, where a
and b are constants.

EXERCISE 7 (a) Find a and b.

Since we want to determine which price maximizes the weekly profits, we will have to examine
how those profits depend on the price. Let’s start from the basic notion of profits as the
difference between revenues and costs.
The weekly revenues, say R, are equal to the price per menu p times the number of customers
q.

EXERCISE 7 (b) Express the weekly revenues R in terms of p.

Now for the weekly costs, say C. Remember that these consist of a fixed component of 1000
euros per week, plus a variable component of 12 euros per menu.

EXERCISE 7 (c) Express the weekly costs C in terms of p.

Finally, the weekly profits, say W , follow by subtracting costs from revenues.

EXERCISE 7 (d) Express the weekly profits W in terms of p.

We will consider the formulas which you found in the previous exercises more closely in the
next two sections.
Functions of one variable, part 1 13

2.2 LINEAR FUNCTIONS

According to the equation

q = 220 − 4p
which you (should have) found in Exercise 7 (a), the number of customers is a function of
price only: so we can write q = d(p), where

d(p) = 220 − 4p.

This so-called demand function d is an example of a linear function.

A LINEAR FUNCTION

The general form of a linear function is

f (x) = ax + b,

where a and b are constants. Its graph is a straight line represented by the equation y = ax+b:
a is called the slope of this line and b the intercept.

In the case of our demand function d, we have a = −4 and b = 220. You will agree that the
domain of this demand function consists of all nonnegative values of p such that d(p) ≥ 0.
Hence, its domain (check!) is the interval [0, 55]. The graph of this function is a straight line
represented by the equation
q = 220 − 4p.
Figure 9 below shows our straight line. The number −4 is the slope of the line, while the
number 220 is the intercept. The slope indicates the change in demand if the price rises by
one euro (in our case: a loss of 4 customers). The intercept corresponds with the point where
the straight line hits the vertical axis (in our case: if the menu were for free, 220 customers
would be attracted).

220

q = 220 − 4p

55 p

FIGURE 9 The linear relation between price and demand

In the next example and the next exercise, we consider two other linear relations.

EXAMPLE 11 In Keynesian macroeconomic theory, total consumption expenditure on

goods and services C is often assumed to be a linear function of national income Y :

C = C0 + cY.
14 Functions of one variable, part 1

The slope c is called the marginal propensity to consume. This number tells us by how much
consumption increases if national income increases by one unit. Since it seems likely that
consumers will save (rather than spend) part of the extra income, we would expect that
0 < c < 1. The intercept C0 is called autonomous consumption. Even if income were zero,
consumers would still need to spend a certain amount on necessities like food, clothing and
shelter (and they might finance this by running down their assets). So we would expect that
C0 > 0.

EXERCISE 8 At a certain moment there are 30 employees at work in a factory hall, and each
employee produces 40 units of a certain product per hour. If the number of
employees increases, the productivity per employee falls because the employees
will get in each other’s way. It turns out that when there are 35 employees, the
hourly production falls to 30 units per employee. We assume that the relation
between the number of employees and the hourly production per employee is
linear. Determine this linear relation.

We now want to address an important property of the slope of a line. As an example, consider
the line ℓ that passes through the points A = (2, 1) and B = (3, 3). Of course, the general
equation of ℓ reads as y = ax + b. To find the value of its slope b, we can start in the same
way as you (probably) did in Exercise 7 (a) and in Exercise 8. Since both (2, 1) and (3, 3) are
points on our line, it must hold that
1 = 2a + b
and 3 = 3a + b.

If we now subtract the first equation from the second one, we obtain
3−1
3 − 1 = 3a − 2a =⇒ a(3 − 2) = 3 − 1 =⇒ a = = 2.
3−2
It turns out that the slope a is a quotient in terms of the coordinates of the points A and B.
Figure 10 below provides a geometrical interpretation of this quotient.

y
ℓ
B
3

2 ∆y

1 A
∆x

1 2 3 x

FIGURE 10 The slope of the line ℓ

The numerator of our quotient, 3 − 1, is the vertical ’displacement’ between our two points
A and B. We denote this ’displacement’ as ∆y, where ∆ is the general mathematical symbol
to indicate a ’change’ or a ’difference’. Similarly, the denominator 3 − 2 is the horizontal
displacement, which we denote as ∆x.
Functions of one variable, part 1 15

In other words: the slope of the line ℓ equals

∆y
.
∆x

Generalizing our numerical example:

THE SLOPE AS QUOTIENT OF VERTICAL AND HORIZONTAL DISPLACEMENT

The slope of the straight line through the points (s1 , t1 ) and (s2 , t2 ) (with s1 6= s2 ) equals

∆y t2 − t1
a= = .
∆x s2 − s1

Once you know that the slope of the line ℓ equals 2, finding its equation becomes very simple.
Since the line contains the point (2, 1), an equation of the line is

y − 1 = 2(x − 2) ⇐⇒ y = 2x − 3.

Here we used the following rule:

THE POINT-SLOPE FORMULA

An equation of the line with slope m and containing the point (a, b) is

y − b = m(x − a).

EXERCISE 9 (a) Find the equation of the line with slope −2 that passes through the point
(−2, 6).
(b) Find the points where this line intersects the horizontal and vertical axes.

EXERCISE 10 Find the equation of the straight line that passes through the points (1, 1) and
(0, 4).

EXERCISE 11 Find the intersection of the two lines given by

(a) 3y = x + 9 and y = 7 − x (multiplying the second equation by 3 could
be helpful)
(b) y = 2x + 1 and x = 12 y − 1 (rewriting the second equation in the form
y = · · · could be helpful)

EXAMPLE 12 THE LINEAR SUPPLY AND DEMAND MODEL

In Example 1, we considered the relationship between consumers’ demand for a specific good
D, and the price P which they have to pay. More specifically, we assumed that demand
depends linearly on price, as follows:

D = 20 − 4P.
 16 Functions of one variable, part 1

Similarly, the quantity of the good that producers are willing to supply, say S, depends on
the price P which they receive. An example could be
S = 12 + 12P,
embodying the plausible assumption that supply will increase when the price rises. The
situation where demand is equal to supply represents a market equilibrium. The corresponding
price (quantity) is called the equilibrium price (quantity).

EXERCISE 12 Find the equilibrium price and the equilibrium quantity in Example 12.

2.3 QUADRATIC FUNCTIONS

At Exercise 7 (d), we trust that you managed to find the following equation:
W = −4p2 + 268p − 3640.
According to this equation, the restaurant’s weekly profit is a function of the menu price only:
we can write W = w(p), with
w(p) = −4p2 + 268p − 3640.
This profit function w is an example of a quadratic function. Quadratic functions are among
the simplest non-linear functions.

QUADRATIC FUNCTION

The general form of a quadratic function is

f (x) = ax2 + bx + c,
where a, b and c are constants with a 6= 0.

In the case of our profit function, we have a = −4, b = 268 and c = −3640.
The graph of a quadratic function is always some kind of parabola. We will discuss three crucial
characteristics of a parabola, which you will probably remember from secondary school:
(1) its shape
(2) its highest or lowest point
(3) its zero(s), i.e the point(s) where the parabola intersects the horizontal axis.

the shape As illustrated in Figure 11 below, a parabola is called concave (it opens downward) if a < 0,
and it is called convex (it opens upward) if a > 0.

y y
a<0 a>0

x x

FIGURE 11 The shape of the graph of a quadratic function

 Functions of one variable, part 1 17

highest (lowest) point The parabola y = ax2 + bx + c has its highest or lowest point at x = −b/2a. Of course, the
type of the point is determined by the shape of the parabola (that is: by the sign of a). A
concave parabola has a highest point, while a convex parabola has a lowest point.

intersection points We can find the points where the parabola y = ax2 + bx + c intersects the horizontal axis by
solving the equation
ax2 + bx + c = 0.

You will remember from secondary school that this quadratic equation can have no, one or
two solutions. This depends on the so-called discriminant

D = b2 − 4ac.

The details are summarized in the quadratic formula below.

QUADRATIC FORMULA

Consider the quadratic equation ax2 + bx + c = 0, where a 6= 0. Define D√= b2 − 4ac. Then:
−b ± b2 − 4ac
– if D > 0, the quadratic equation has two solutions, given by x =
2a
−b
– if D = 0, the quadratic equation has a single solution, given by x =
2a
– if D < 0, the quadratic equation has no solutions.

In Figure 12 below, we have sketched the graph of the profit function w from Exercise 7(d),
given by
w(p) = −4p2 + 268p − 3640.

W
849

17 33.5 50 p

FIGURE 12 The graph of the profit function w

In the next exercise we ask you to check the three crucial characteristics of the parabola
W = −4p2 + 268p − 3640 which can be recognized in Figure 12.

EXERCISE 13 (a) Explain why the parabola is concave.

(b) Find the top (i.e. the highest point) of the parabola.
(c) Find the intersection points of the parabola with the horizontal axis. How
do they relate to the parabola’s top?
18 Functions of one variable, part 1

END-OF-CHAPTER EXERCISES

EXERCISE 1 The curve y = x2 + bx + 3 passes through the point (1, −3).

Find b.

EXERCISE 2 Determine an equation for the following straight line:

(a) ℓ1 passes through (1, 3) and has a slope of 2.
(b) ℓ2 passes through (−2, 2) and (3, 3).
(c) ℓ3 passes through the origin and has a slope of − 12 .
(d) ℓ4 passes through (a, 0) and (0, b), with a 6= 0.

EXERCISE 3 A farmer has 1000 meters of fence wire to make a rectangular enclosure.
(a) Find the areas of the three rectangles whose bases are 100, 250, and 350
meters.
(b) Let the base have length 250 + x. Then the height is 250 − x. What choice
of x gives the maximum area?

EXERCISE 4 Consider the curve y = 3x2 + bx + 3. Here b is some number. Find b

(a) if the curve intersects the horizontal axis at x = −1.
(b) if the curve has only one point in common with the horizontal axis.
(c) if the curve has its lowest point at x = 2.
(d) the second coordinate of the lowest point of the curve is −9.

DUTCH GASOLINE TAXES

As in most European countries, the Dutch government levies a sizeable tax on gasoline: in
fact, in the Netherlands, roughly two-thirds of the price you pay for a liter of Euro95 goes to
the taxman. Imagine that, in the absence of a tax, the demand and supply for gasoline are
given by the following equations:
D = 21 − 6P

S = −6 + 30P.
Here D and S refer to demand and supply (in millions of liters per day), and P to the price
(in euro per liter).

EXERCISE 5 At the market equilibrium, the transaction quantity Q is . . .

(a) somewhere between 10 and 12
(b) somewhere between 12 and 14
(c) somewhere between 14 and 16
(d) somewhere between 16 and 18.
Functions of one variable, part 1 19

Now assume that the government levies a tax of 0.90 euro per liter, which the producers have
to pay to the government. In this new situation, if the consumers pay the producers P per
liter, those producers effectively receive only P − 0.9 (as they have to pay 0.90 euro per liter
to the government). Of course, the producers will base their decisions on what they ’really’
receive per liter. Consequently, the demand and supply schedules become

D = 21 − 6P

and S = −6 + 30(P − 0.9).

EXERCISE 6 At the new market equilibrium, the price effectively received by the producers
is . . .
(a) somewhere between 0.50 euro and 0.75 euro
(b) somewhere between 0.75 euro and 1.00 euro
(c) somewhere between 1.00 euro and 1.25 euro
(d) somewhere between 1.25 euro and 1.50 euro.

EXERCISE 7 At the new market equilibrium, the tax revenues enjoyed by the government
are . . .
(a) somewhere between 6 and 8 million euro per day
(b) somewhere between 8 and 10 million euro per day
(c) somewhere between 10 and 12 million euro per day
(d) somewhere between 12 and 14 million euro per day.

Typically, one of the stated objectives of gasoline taxes is to help conserve the environment
by discouraging car use. However, the government may have a more down-to-earth objective
as well: simply to raise money. Now imagine that the government levies an unspecified tax of
τ euro per liter, rather than 0.90 euro. With some effort, you should be able to express the
equilibrium price and quantity as a function of τ . Subsequently, you can express tax revenue
as a function of τ .

EXERCISE 8 In equilibrium the revenues for the government (as a function of τ ) are
(a) 18τ − 6.667τ 2
(b) 16.5τ − 5τ 2
(c) 15τ − 3.333τ 2
(d) 13.8τ − 2τ 2 .

EXERCISE 9 The maximum tax revenue which the government can collect is . . .
(a) somewhere between 10 and 12 million euro per day
(b) somewhere between 12 and 14 million euro per day
(c) somewhere between 14 and 16 million euro per day
(d) somewhere between 16 and 18 million euro per day.
.
20 Functions of one variable, part 1

ICE CREAM SALES

On a Spanish beach, a young man runs an ice cream stand between 3.00 h and 5.00 h in the
afternoon. A math teacher, bored to death while his kids are having fun, decides to track
the movements of our ice cream vendor. Around 3.00 h, 4.00 h and 5.00 h, he monitors how
much time lapses (on average) between consecutive ice cream sales. E.g. around 4.00 h, he
observes that the next sale occurs on average after 1.5 minutes; in other words, at that time
the vendor is selling ice creams at a rate of 40 per hour. On the basis of his observations, our
math teacher expresses the relation between ice cream sales per hour Q and time t with the
following formula:
Q = 10t2 − 120t + 360 (3 ≤ t ≤ 5). (A)

Note that for t = 4 our formula implies Q = 40, as it should be. Now imagine that the vendor
has just opened his stand, at 3.00 h.

EXERCISE 10 At that time, how many seconds pass (on average) between consecutive sales?
(a) Less than 30 seconds
(b) Between 30 and 50 seconds
(c) Between 50 and 70 seconds
(d) More than 70 seconds.

EXERCISE 11 At which time do ice cream sales peak?

(a) Before 3.30 h
(b) Somewhere between 3.30 h and 4.00 h
(c) Somewhere between 4.00 h and 4.30 h
(d) After 4.30 h.

EXERCISE 12 Which of the following formulas correctly expresses t in terms of Q?

p
(a) t = 0.1 Q
p
(b) t = 9 − 0.4 Q
p
(c) t = 6 − 0.1 Q
(d) None of the options (a), (b) or (c) is correct.

At the end of the day, our math teacher wonders how many ice creams the vendor has sold.
To estimate this amount, he approximates the curve which is implied by formula (A) with
the following straight line through the points (3, 90) and (5, 10):

Q
90

10
3 4 5 t
Functions of one variable, part 1 21

EXERCISE 13 Based on the straight line above, what is the teacher’s estimate for the total
number of ice creams sold between 3.00 h and 5.00 h?
(a) Somewhere between 150 and 170
(b) Somewhere between 130 and 150
(c) Somewhere between 110 and 130
(d) Somewhere between 90 and 110.

EXERCISE 14 How does the estimate for total sales based on the straight line above compare
to the actual total sales?
(a) Actual total sales are above the estimate
(b) Actual total sales are below the estimate
(c) Actual total sales are equal to the estimate
(d) There is not enough information to choose between the options (a), (b)
and (c).

On the next day, our math teacher is back at the beach, and he notices right away that the
ice cream vendor enjoys more business than the day before. The reason is obvious: it’s much
warmer! On the beach at hand, the temperature always hovers between 25 oC and 35 o C; and
while yesterday’s temperature was only 25 o C, today scores 35 o C. Realizing that ice cream
sales are influenced by temperature, he adjusts his formula for the relationship between sales
and time, as follows:

Q = 10t2 − 120t + 360 + H (3 ≤ t ≤ 5). (B)

Here H represents the heat bonus, i.e. the extra sales that occur when temperature exceeds
the 25 o C minimum. After some additional measurements, our teacher concludes that H = 60
today. In a table:

temperature T in o C 25 o C 35 o C
value of H 0 60

The teacher assumes that, for temperatures between 25 oC and 35 o C, the value of H is
linearly related to the value of T , as expressed by the following formula:

H = aT + b,

where a and b are constants.

EXERCISE 15 What is the correct value of b?

(a) −150
(b) 0
(c) −10
(d) None of the options (a), (b) or (c) is correct.

On some day around 4.00 h, the vendor is selling ice creams at a rate of around 50 per hour.
22 Functions of one variable, part 1

EXERCISE 16 What is the temperature that afternoon?

(a) Somewhere between 32.5 oC and 35 oC
(b) Somewhere between 30 o C and 32.5 oC
(c) Somewhere between 27.5 o C and 30 o C
(d) Somewhere between 25 o C and 27.5 oC.

Consider the graph that corresponds to formula (A) (say: the ’A-graph’) and the graph which,
for any nonzero value of H, corresponds to formula (B) (say: the ’B-graph’).

EXERCISE 17 Compared to the ’A-graph’, the ’B-graph’ is . . .

(a) shifted to the right by the amount H
(b) shifted to the left by the amount H
(c) shifted up by the amount H
(d) None of the options (a), (b) or (c) is correct.
Functions of one variable, part 2 23

2 FUNCTIONS OF ONE VARIABLE, PART 2

In this chapter we continue the investigation of some specific classes of functions.

In Section 1, we start with broken functions. As an example, we will build a model for the
inventory management of a car dealer. The dealer has to decide, how many cars to order each
time when his warehouse runs out. The larger his order size, the fewer orders are necessary;
this will reduce the ordering costs, but the larger average inventory level will increase the
holding costs. Correctly modeling total costs leads to a broken function.
Section 2 covers power functions. Our example here revolves around constructing a triangular
gutter using a plate of length 100 and width 18. We want to bend the plate in such a way
that the capacity of the gutter is maximal. Modeling this problem involves a root function,
which is an example of a power function.
In Section 3, exponential and logarithmic functions are addressed. These functions play a
crucial role in modeling processes of exponential growth and decay.

1 BROKEN FUNCTIONS

In this section we consider the class of functions which can be written as the quotient of
two functions. Note that the domain of such a broken function is (apart from other possible
considerations) restricted by the requirement that zeros are not allowed for the function that
appears in the denominator.

EXAMPLE 1 A FENCE

In Example 5 of the previous chapter we considered the function given by the equation

200
L = 2ℓ +
ℓ

which can be rewritten as

2ℓ2 + 200
L= .
ℓ
Here, ℓ denotes the length (in meters) of a piece of land with an area of 100 m2 , while L
denotes the total length (in meters) of the fence that is used to encapsulate that piece of
land. The challenge is to choose ℓ in such a way that the total length of the fence L is as
24 Functions of one variable, part 2

small as possible. The rewritten version of the function above is an example of a broken
function. After all, its numerator contains a quadratic function of ℓ, and its denominator a
linear function. Obviously, the domain of the function is the interval (0, ∞), i.e. with zero
excluded.

EXAMPLE 2 A CAR DEALER

A Toyota dealer sells 72 Toyotas of the type Auris per year. Each car has a value of 25 000
euros. Of course, the dealer’s stocking policy involves costs. In order to analyze these costs,
we start from the following assumptions:
• the sale of cars takes place uniformly over time
• every time when the cars kept in stock have been sold, the dealer immediately orders a
number of new Toyotas, which can be supplied on the same day
• each time, the dealer orders the same number of cars.
We can distinguish two different kinds of costs here: ordering costs and holding costs.
• The ordering costs reflect the costs of transportation and administrative handling: they
equal 250 euros per order, irrespective of the size of the order.
• The holding costs occur because cars in inventory require storage space and involve
interest loss: on an annual basis, these costs are 4% of the total value of the cars in stock.
Of course the dealer wants to know the number of cars he should order each time, such that
the total annual costs are minimal.

Before modeling this decision problem, we examine a special case: let’s assume that the
dealer has decided to order 6 cars each time.

EXERCISE 1 (a) Find the annual ordering costs corresponding with this order size.

Due to our assumptions, the number of cars in the showroom will decrease linearly. For the
situation where the dealer decides to order 6 cars each time, Figure 1 sketches the number
of cars in the showroom, against time in months.

stock

1 2 t (in months)

FIGURE 1 The stock level over time.

Functions of one variable, part 2 25

EXERCISE 1 (b) What is the average number of cars in stock?

(c) Find the annual holding costs.
(d) Find the total annual costs.

Let’s now generalize our numerical example. To solve the car dealer’s decision problem, we
will need to determine the total annual costs for an arbitrary order size q.

EXERCISE 2 (a) Find the number of orders per year, in terms of q.

(b) Find the annual ordering costs, in terms of q.
(c) What is the average number of cars in stock, in terms of q?
(d) Find the annual holding costs, in terms of q.
(e) Find the total annual costs, in terms of q. Show that your result can
be rewritten in such a way, that the total annual costs are a broken
function of q.

We trust that the previous exercise led you to the conclusion, that the total annual costs K
are given by
18 000 + 500q 2
K= .
q
This formula defines a broken cost function k: K = k(q). Figure 2 below sketches the graph
of this function.

K
10 000

6000

2 4 6 8 q

FIGURE 2 The total annual costs

Our challenge is to determine the order size q which minimizes costs. In Chapter 5 we will
use the derivative of the function k to show that the car dealer should order 6 cars each
month.

EXERCISE 3 In order to store his latest models, a Toyota dealer wants to build a new
showroom. This showroom will be rectangular and its area should be 500 m2 .
Three of its walls are to be made of reinforced concrete, costing 600 euros per
meter. The fourth wall will be made of thick glass which costs 1200 euros per
meter.
Let ℓ denote the length of the glass wall and let w be the width of the rectangle.
26 Functions of one variable, part 2

(a) Express the total costs of building this showroom in terms of ℓ and w.
(b) Express w in terms of ℓ.
(c) Write the total costs of building this showroom in terms of ℓ.

2 POWER FUNCTIONS
A function f of the form
f (x) = xa ,

where a is some real number, is called a power function. Note that a quadratic function like

q(x) = 2 − x + 5x2

is a ’combination’ of the power functions 1, x and x2 (corresponding with a = 0, a = 1 and

a = 2, respectively).
√ 1
Observe that the functions g and h defined by g(x) = x and h(x) = x, respectively, can
also be written as a power of x. After all, the function g results if we take a = 21 :

√ 1
x = x2 (x ≥ 0),

while the function h results if we take a = −1:

1
= x−1 (x 6= 0).
x

As these two examples show, we must be a bit careful with respect to the domain of this
type of functions.

EXAMPLE 3 POLYNOMIAL FUNCTIONS

By combining the power functions 1, x, x2 , x3 . . ., we obtain so-called polynomial functions.

An example of such a function is

p(x) = 3 + 2x − 5x3 + 8x4 .

As x4 is the highest power in this function, we say that the function p has degree 4. So a
quadratic function is a polynomial function of degree two, and a linear function is a polyno-
mial function of degree one.

EXAMPLE 4 A DEMAND FUNCTION

A widely used specification for demand in empirical work is

D = A P −α ,

where P is the price. Further, A and α are positive constants. In Example 10 of the next
chapter we will show that the demand function given by this equation has an appealing
property.
Functions of one variable, part 2 27

EXAMPLE 5 A CONSTRUCTION PROBLEM

Using a plate of length 100 and width 18, we want to construct a triangular gutter as
presented below. The idea is to bend the plate through the middle, lengthwise, in such a
way that the capacity of the gutter is maximal.

100 s
9

9 9

FIGURE 3 Constructing a gutter with maximal capacity

The capacity of the gutter, say C, is the area of the triangle presented in Figure 3, say A,
multiplied by the length of the plate. So: C = 100A. In Example 4 of the previous chapter,
we have seen that the area A of this triangle equals
q
A = 12 s 81 − 14 s2 ,

where s is the length of the third side of the triangle. Hence, the capacity of the gutter is
given by
q
C = 50s 81 − 41 s2 .

Of course, the capacity is a function of the variable s, say C = c(s). The function c turns
out to involve a root function. Apart from the condition s > 0, the domain of the function c
is determined by the requirement that the quadratic expression under the root sign must be
nonnegative:
81 − 41 s2 ≥ 0 =⇒ s2 ≤ 324 =⇒ s ≤ 18.

C
4050

6 12 18 s

FIGURE 4 The capacity of the gutter

Figure 4 above sketches the graph of the function c. In Chapter 5 we will show that the
capacity of the gutter is maximal for s ≈ 12.73.
28 Functions of one variable, part 2

EXERCISE 4 A restaurant has a main establishment on the beach, situated 100 meters from
the sea at location H, and a playful sea restaurant built on pillars, which lies
200 meters into the sea at location Z. You can find a map picturing the
situation, with the relevant distances, in Figure 5 below.

sea
200 meters

S
100 meters

H
500 meters

FIGURE 5 A map of the restaurant problem

All foods and drinks are prepared in the main establishment at H, and then
have to be transported to the sea restaurant at Z. Transport takes place via
the successive paths HS (using electrical scooters) and SZ (using electrical
boats); the transport speeds over land and over sea are 200 and 100 meters
per minute, respectively.
As you can see, Figure 5 implicitly defines a variable ℓ.
(a) Express the transport time T1 from H to S in minutes in terms of the
variable ℓ.
(b) Express the transport time T2 from S to Z in minutes in terms of the
variable ℓ.
(c) Express the total transport time T from H to Z in minutes in terms of
the variable ℓ.

3 EXPONENTIAL AND LOGARITHMIC FUNCTIONS

In economics, but also in natural sciences like chemistry, many processes involve quantities
that increase or decrease by amounts which are proportional to their size. In this context
we speak of exponential growth (or exponential decay). We will demonstrate that such
processes can be described by means of exponential functions. If we want to find out when
this exponential growth (or decay) has reached a certain level, we will need logarithmic
functions. We will illustrate most concepts on the basis of the following example of an
exponential growth process.
 Functions of one variable, part 2 29

EXAMPLE 6 THE GROWTH OF CHINA’S GDP PER HEAD

In 2017, China’s GDP (gross domestic product) per head, adjusted for purchasing power,
amounted to around 14 000 euros. For the coming ten years, the Chinese planning agency
predicts a trend growth of 6 % annually. Given these data, several interesting questions
present themselves, e.g.:
• If the annual growth is 6 %, then how big is the growth per quarter, or per month?
• Starting from 2017, after how many years will GDP per head have increased by 25 %?

We will find an answer to these questions in Sections 3.1 and 3.2, respectively.

3.1 EXPONENTIAL FUNCTIONS

the first question Let’s address the first of these two questions step by step. In doing so, we will encounter all
kinds of concepts that are linked to exponential functions.
dealing with
Of course, if GDP per head grows by 6 % annually, then the size of this GDP is multiplied
the problem
by a factor 1.06 every year. So if we write GDPt for the GDP after t years, starting in 2017
from GDP0 = 14 (i.e. 14 thousand euros), then the following holds:

GDP1 = 14 · (1.06) = 14.84

GDP2 = GDP1 · (1.06) = 14 · (1.06)2 = 15.73

..
.

GDPt = 14 · (1.06)t .

growth factor We say that the development of China’s GDP per head is characterized by exponential
growth, where the constant number 1.06 is called the annual growth factor. The expression

14 · (1.06)t ,

which we used to describe the growth of GDP , defines an exponential function.

Of course, we are not restricted to natural numbers for t in this function: it also possible
to fill in other positive numbers. E.g., the value of GDP after a quarter is obtained simply
by filling in t = 0.25 in the expression, and the growth factor per quarter therefore equals
(in 4 decimal places) (1.06)0.25 = 1.0147. Note that this is slightly less than 1.015: this
is a consequence of the principle of compounding which you may remember from secondary
school, and will be addressed explicitly in QM2. Verify for yourself that the growth factor
per month (again in 4 decimal places) equals 1.0049. This answers our first question.

EXPONENTIAL FUNCTION
A function f of the form
f (t) = b · at ,

is called an exponential function with base a, where the constant a 6= 1 is positive. The
constant b is the value of f for t = 0.
30 Functions of one variable, part 2

In our GDP example, b = 14 is the initial amount at t = 0, and the base of a = 1.06 is the
growth factor per year.

Generalizing our GDP example, whenever an amount is multiplied by a constant factor per
time unit, say by a, then we speak of exponential growth or decay. That any such process
can indeed be described by the exponential function f (t) = b · at is easily shown:
 
f (t + 1) = b · at+1 = b · at · a1 = a · b · at = af (t).

In other words:
the value of f at any time t + 1 is a times the value of f at the previous time t
This is the defining characteristic of an exponential growth process.

In Figure 6 below, the curve that represents the exponential function, say y = at , has been
sketched for two very distinct cases: a > 1 and 0 < a < 1. In particular, the graph on the
left represents an exponential growth process with growth factor a = 2, while the graph on
the right does so for a = 0.5. In both cases, we have taken b = 1, so that both graphs pass
through the point (0, 1).

y y
 t
t 1
y=2 y= 2

6 6

4 4

2 2

2 t −4 −2 2 t

FIGURE 6 Two exponential functions

Note that the variable represented by the graph on the right is actually not increasing over
time, but decreasing; its value is cut in half whenever a unit of time passes by. The following
example describes a situation where such a process of exponential decay plays a role.

EXAMPLE 7 DEPRECIATION

If you own a car, you will be aware that its resale value drops every year. Let’s assume that
a car’s value depreciates at an annual rate of 15 %. So if its value at t = 0 equals V0 , then its
value after one year equals V1 = V0 − 0.15V0 = 0.85V0 . Generalizing, after t years its value
is Vt = V0 · (0.85)t . For instance, V4 = V0 · (0.85)4 ≈ 0.52V0 : so after 4 years the car’s value
has dropped to little more than half of its original value.
 Functions of one variable, part 2 31

EXERCISE 5 Find the exponential function for which the graph contains the following
points:
(a) (0, 2) and (3, 16) (b) (0, 3) and (3, − 91 ).

For your convenience, we summarize some secondary school rules for manipulating exponen-
tial functions.

PROPERTIES OF EXPONENTS

If a > 0, b > 0 and x and y are numbers, then

1
(1) a0 = 1 (2) ax · ay = ax+y (3) a−x =
ax
ax
y
(4) = ax−y (5) ax = axy (6) (ab)x = ax · bx
ay

the second question We now continue our discussion of the two questions formulated in Example 6. Our second
question was: starting from 2017, when will Chinese GDP have increased by 25 %, i.e. to
1.25 × 14 = 17.5? Let’s start by sketching the relation between GDP and t in a graph.

GDP

18
17.5

16

14

1 2 3 ?4 t

FIGURE 7 The growth of China’s GDP per head

Remembering Section 1.4 from Chapter 1, we hope that many students immediately smell
the notion of an ’inverse’ here. After all, Figure 7 above shows how a value for the input
variable t translates into a value for the output variable GDP . But our second question
turns things around: it starts from a value of 17.5 for GDP as input, and we are looking for
the implied value of t as our output. We’ll formalize this issue in the next section.

3.2 LOGARITHMIC FUNCTIONS

exponential equation To answer the second question from Example 6, we must solve t from the following so-called
exponential equation:
14 · (1.06)t = 17.5 ⇐⇒ (1.06)t = 1.25.

In words: we are looking for the value of t for which the exponential function with base 1.06
equals 1.25.
32 Functions of one variable, part 2

Before presenting the numerical answer to this particular question, consider the more general
exponential equation
ax = c,

where c > 0. Two examples of such an exponential equation are 2x = 8 and 2x = 6. The first
equation is easy to solve: since 23 = 8, it is clear that x = 3 solves the equation. Finding the
solution of the second equation is not that straightforward. Figure 8 illustrates the challenge.

y
y = 2x

? x

FIGURE 8 Solving the equation 2x = 6

To deal with an equation like 2x = 6, we need logarithms. In fact, we will denote the solution
of this equation by
2
log 6

pronounced as ’two-log-six’. Note: ’two-log-six’ is the exponent to which we must raise 2, in

order to get 6. Consequently:
2
2 log 6 = 6

The number 2 is called the base of our logarithm. Using this ’language’, the solution of
the first equation, 2x = 8, can be written as 2 log 8. Apparently, 2 log 8 = 3. But how do
we calculate the other logarithm, 2 log 6? Luckily, your pocket calculator has a ’log key’ for
calculating logarithms with base 10. This may seem useless, because ’our’ logarithm has base
2. But now the following formula (which we will not prove) is helpful:
10
2 log 6
log 6 = 10 log 2
.

Please use your calculator to verify that the result is about 2.58. Alternatively, you can check
that 22.58 ≈ 6. The rule below summarizes our result.

SOLVING AN EXPONENTIAL EQUATION

The solution of the exponential equation

at = c,

with a > 0, is given by

a log c
log c = ,
log a
10
where we follow the usual conventions and write log c instead of log c.
 Functions of one variable, part 2 33

Now back to our original question: when will Chinese GDP have increased by 25 %, if the
annual growth is 6 %? The implied exponential equation of (1.06)t = 1.25 has the following
solution:

1.06 log 1.25 (please verify!) 0.096910

log 1.25 = = = 3.83 (in 2 decimals).
log 1.06 0.025305

logarithmic So far, we have presented the logarithm as just a tool to solve exponential equations. Next,
function we will consider the so-called logarithmic function g, defined by

g(x) = a log x (x > 0),

where a 6= 1 is a positive real number. As an example, consider the case where a = 2.

Like the left-hand side of Figure 6, Figure 9 below contains the graph corresponding to the
function f defined by f (x) = 2x . The careful reader has definitely recognized, that the
2
number log 6 is in fact equal to the value f −1 (6). Generalizing, the function g defined
2
by g(x) = log x is the inverse of the function f . Now remember from Section 1.4 of the
previous chapter that the graphs of a function and its inverse are symmetric about the line
y = x. Consequently, mirroring the graph of the exponential function with base 2 around
the 45-degree line, we obtain the graph of the logarithmic function with base 2. Figure 9
illustrates the idea. Note in particular that, while the graph of the exponential function
contains the point (x, y) = (0, 1), its logarithmic inverse contains the point (x, y) = (1, 0).

y y = 2x

y = 2 log x

1 x

FIGURE 9 The graph of the logarithmic function with base 2

For your convenience, we summarize some secondary school rules for manipulating logarith-
mic functions.

PROPERTIES OF LOGARITHMS

If a > 0, a 6= 1 and x and y are positive numbers, then

1
(1) a log 1 = 0 (2) a log(xy) = a log x +a log y (3) a log = − a log x
x x


(4) a log = a log x − a log y (5) a log xy = y a log x (6) a log a = 1
y
34 Functions of one variable, part 2

In the next two examples, we will show how these properties can be helpful to solve expo-
nential equations, or to simplify expressions involving logarithms.

EXAMPLE 8 To solve the equation

2x = 5x+1

we can start by taking the logarithm of both sides. We then apply properties (5) and (4), to
obtain the following result:
 
log 2x = log 5x+1 ⇐⇒ x log 2 = (x + 1) log 5 ⇐⇒ x log 2 − log 5 = log 5
log 5 log 5
⇐⇒ x = = .
log 2 − log 5 log(2/5)

Using your pocket calculator you will find that x ≈ −1.756.

EXAMPLE 9 The expression

log 5 + log 12 − log 6

can be simplified considerably. Cleverly using properties (2), (4) and (6), we get

5 × 12
log 5 + log 12 − log 6 = log = log 10 = 1.
6

Likewise, we can simplify the expression

100
log 9
10 .

It’s easy to see that

100 log 9 log 9 log 9 1

log 9 = = 2
= = 2 log 9.
log 100 log 10 2 log 10

Hence, we can now develop our expression as follows:

100
log 9 1
h i1/2 1 √
10 = 10 2 log 9 = 10log 9 = 9 2 = 9 = 3.

EXERCISE 6 Express the following numbers as a multiple of log 5.

1 3
√ 1
(a) log 25 (b) log 625 (c) log 25 (d) log 5 − log 25

EXERCISE 7 Solve the following equations.

2
(a) 2x · 3x = 7 (b) 2x −1
=8 (c) log x + log x2 = 9 (d) log x =2 log 12
Functions of one variable, part 2 35

END-OF-CHAPTER EXERCISES

EXERCISE 1 Match each of the graphs A − F with one of the functions (a)-(e). In (f), try
to find a suitable function which has the remaining graph.
√
(a) y = 12 x2 − x − 23 (d) y = 2 2 − x

x
(b) y = − 21 x2 + x + 23 (e) y = 21 − 2
√
(c) y = 2 x − 2 (f) y = · · · .

A B C

D E F

EXERCISE 2 The volume V of a sphere depends on its radius r, according to the formula

V = 34 πr3 .

In other words V is a function of r: V = g(r), where g(r) = 34 πr3 .

(a) Find the volume of a sphere with radius r = 9.
(b) What is the radius of a sphere with volume 27?
(c) Find the inverse function g −1 .
(d) Give a sketch of the graphs of the functions g and g −1 .

EXERCISE 3 The demand D for sugar in the U.S. in the period of 1915 − 1929, as a function
of the price P , was estimated by H. Schultz as
157.8
D = f (P ) = .
p0.3
Here D and P are measured in appropriate units.
Solve the equation for P and so find the inverse of the function f .

EXERCISE 4 The population of Botswana was estimated to be 1.22 million in 1989, and to
be growing at the rate of 3.4% annually. If t = 0 denotes 1989, find a formula
for the population P (t) at date t. What is the doubling time?
36 Functions of one variable, part 2

EXERCISE 5 Two growth processes

(a) Let f (t) = A · 2rt and g(t) = B · 2st , with A > 0, B > 0, and r 6= s. Solve
the equation f (t) = g(t).
(b) In 1990 the GN P (gross national product) of China was estimated to
be 1.2 · 1012 , and the rate of growth was estimated to be r = 0.09.
By comparison, the GN P for the USA was reported as 5.6 · 1012 , with
an estimated rate of growth of s = 0.02. If the GN P of each country
continued to grow exponentially at the rates r = 0.09 and s = 0.02,
respectively, when would the GN P of the two nations be the same?

EXERCISE 6 In 1964, a five-year plan was introduced in Tanzania. One objective was
to double the real per capita income over the next 15 years. What is the
average annual rate of growth of real income per capita required to achieve
this objective?

AN INSTALLATION PROBLEM

A water supply pipe is to be dug in from a residence V , which lies along an asphalted road,
to a country cottage B, which lies in the middle of a forest. Installation of the pipe alongside
the asphalted road costs 40 euros per meter; but installation through the moist forest soil
costs 80 euros per meter. Below, you can see a map of the situation.

200 meters

road
R
V A

200 meters

forest B

ℓ 200 − ℓ

FIGURE 10 A map of the water pipe problem

To illustrate our options, the map does not only indicate the residence V and the cottage B,
but also a point A somewhere along the asphalted road. The idea is that, starting from the
residence, the pipe follows the asphalted road up to that point (path V A), after which we
dig it in along a diagonal line through the forest towards the cottage (path AB).
Functions of one variable, part 2 37

EXERCISE 7 Assume that the pipe follows the asphalted road for the first 200 meters (path
V R), then makes a 90 degree turn at R into the forest, to move straight to
the cottage for another 200 meters (path RB).
What are the installation costs (in euros) of this option?
(a) Somewhere between 21 500 and 22 500 euros
(b) Somewhere between 22 500 and 23 500 euros
(c) Somewhere between 23 500 and 24 500 euros
(d) Somewhere between 24 500 and 25 500 euros.

EXERCISE 8 Assume we install the pipe straight through the forest towards the cottage
(path V B).
What are the installation costs (in euros) for this alternative?
(a) Somewhere between 21 500 and 22 500 euros
(b) Somewhere between 22 500 and 23 500 euros
(c) Somewhere between 23 500 and 24 500 euros
(d) Somewhere between 24 500 and 25 500 euros.

As you can see, the option described in Exercise 8 is cheaper than the alternative from
Exercise 7. But of course, we have a continuum of intermediate alternatives here, where
you follow the road partly (as depicted by the path V AB in Figure 10). To determine the
installation costs for these intermediate cases, we define the variable ℓ as the distance from
V to A.

EXERCISE 9 The total installation costs K can be expressed in terms of ℓ by the following
equation:
p
(a) K = 40ℓ + 8 ℓ2 − 400ℓ + 80 000
(b) K = 40ℓ + 80(ℓ2 − 400ℓ + 80 000)
p
(c) K = 40ℓ + 80 ℓ2 − 400ℓ + 40 000
(d) None of the options (a), (b) or (c) is correct.

Of course, the installation costs are a function of the variable ℓ, say K = k(ℓ).

EXERCISE 10 The domain of the cost function k is given by . . .

(a) the interval [0, 100]
(b) the interval [0, 141.4]
(c) the interval [0, 200]
(d) None of the options (a), (b) or (c) is correct.
Hint: You can of course answer this question by formally investigating the
function k, but also by showing some panache and drawing your conclusion
directly from a critical look at Figure 10.
38 Functions of one variable, part 2

SOLAR PANELS

An installation company specialized in solar panels offers its customers two options. The
first option is that the company installs the panels, while the second option is to install them
yourself. Under the first option, the company offers three different systems. The table below
summarizes the most important information.

number of panels 6 10 14
price (in euro) of the system 3600 5600 7600
annual electricity yield (in kWh) 1500 2500 3500

Under the second option, a customer can order any number of panels: the costs are a fixed
amount of 1500 euros, plus 300 euros per panel. The government, keen to promote the
adoption of sustainable energy sources, offers a 10 % subsidy on the purchasing costs of solar
panels, up to a maximum of 500 euros.

EXERCISE 11 When you buy 10 panels, the subsidy amounts to . . .

(a) less than 500 euro under option 1 and 500 euro under option 2
(b) less than 500 euro under option 1 and less than 500 euro under option 2
(c) 500 euro under option 1 and 500 euro under option 2
(d) 500 euro under option 1 and less than 500 euro under option 2.

A vital piece of information in the installation company’s advertisements is the so called

payback period. This is the number of years which it takes to earn back the purchasing cost
of the panels through your savings on the cost of electricity. Let’s assume that the price of
electricity is, and continues to be, 22 eurocent per kWh.

EXERCISE 12 When you buy 10 panels, the payback period is . . .

(a) more than 9 years under option 1, and less than 7 years under option 2
(b) more than 9 years under option 1, and more than 7 years under option 2
(c) less than 9 years under option 1, and more than 7 years under option 2
(d) less than 9 years under option 1, and less than 7 years under option 2.

We now limit ourselves to option two.

EXERCISE 13 The relation between the payback period T (in years) and the number of
solar panels bought, say z (with z ≤ 11), is given by . . .
1350 + 270z
(a) T =
55z
1500 + 300z
(b) T =
55z
1350 + 270z
(c) T =
250z
(d) None of the options at (a), (b) or (c) is correct.
Functions of one variable, part 2 39

ENDANGERED SPECIES

Two species of animals are facing an acute extinction risk. Of the first species, only 2000
animals are left, and the population decays at a rate of 5 % per year. The second species is
down to 1000 but, fortunately, this population is growing at a rate of 5 % per year.

EXERCISE 14 By how many per cent does the population of the second species grow per
quarter?
(a) Somewhere between 1.20 and 1.22 % per quarter
(b) Somewhere between 1.22 and 1.24 % per quarter
(c) Somewhere between 1.24 and 1.26 % per quarter
(d) Somewhere between 1.26 and 1.28 % per quarter.

EXERCISE 15 After how many years from today will both populations be equally large?
(a) After between 4 and 5 years
(b) After between 5 and 6 years
(c) After between 6 and 7 years
(d) After between 7 and 8 years.

The environmental protection agency has decided to intervene when the population size of the
first species falls below 1000. Such intervention could involve measures like the introduction
of dedicated breeding programs in suitable zoos.

EXERCISE 16 When will this intervention take place?

(a) After between 9 and 11 years from today
(b) After between 11 and 13 years from today
(c) After between 13 and 15 years from today
(d) After between 15 and 17 years from today.
40 Functions of one variable, part 2
The derivative 41

3 THE DERIVATIVE

This chapter introduces the concept of the derivative, and illustrates its use in economics
and business. We hope you still remember from secondary school, that the derivative of a
function measures something like ’the rate of change of that function at a certain point’. In
Section 1, we will have a closer look at the basic idea.
In Section 2, we will present a few rules for determining the derivative (more precisely: the
derivative function) for a limited class of basic functions. Additional rules will be presented
next week, in Chapter 4.
For economists and business scientists, the notion of ’the rate of change of a function at
a certain point’ is of crucial importance. It allows them e.g. to determine, approximately
but quickly, how profit or costs would change if sales volume changes by a small amount.
Economists typically call such a small change a marginal change; by extension, the derivative
function is then called the marginal function. We will illustrate all of this in Section 3, using
among others the ’restaurant’ example from Chapter 1.
In Section 4 we will study another, but related, core concept in economics and business:
elasticity. Using the marginal function, absolute changes can be approximated, e.g. the
change in sales as a result of a price change. But if we wish to answer the question by how
many percent sales will change if price changes by e.g. 3 percent, then we are dealing with
relative changes. This is where the elasticity concept comes in.
Although many of the issues covered in this chapter may be familiar to you, please do
study it very carefully. As much as possible, and unlike in secondary school, the relevant
mathematical concepts are treated from an economics and business perspective; the results
constitute indispensable tools for later chapters and, without exaggeration, for almost all
later courses.

1 THE IDEA BEHIND THE DERIVATIVE

The objective of this first section is to refresh your understanding of the concept of a ’deriva-
tive’. Our approach takes the study of rates of change as its starting point.

1.1 THE AVERAGE RATE OF CHANGE

In economics and business one often analyses the effects of changes. For example, if a firm
considers to produce more, it wants to know how much its production costs will increase. We
will start our discussion by introducing a way of measuring the rate at which one variable
(say, production costs) changes in response to a change in the other (say, output level).
42 The derivative

EXERCISE 1 Consider a company producing paint brushes. The daily costs (in euros) of
producing q units (of 10 000 brushes each) are given by K = q 3 − 12q 2 + 75q.
In other words: K = k(q) with k(q) = q 3 − 12q 2 + 75q.
(a) Currently 80 000 brushes are produced per day, that is: q = 8.
Find the daily production costs corresponding with this production level.
(b) Now imagine that the output level q is increased by 2 units.
Find the daily production costs corresponding with this new production
level.
(c) Find the increase in production costs.
(d) Find the increase in production costs per unit of output.

In the example below, we will generalize our findings from Exercise 1.

EXAMPLE 1 PRODUCING PAINT BRUSHES

Consider our paint brush producer from Exercise 1 again, with production costs given by
K = k(q). But unlike in Exercise 1, we now leave the cost function k unspecified. Figure
1 below, which has output q along the horizontal axis and costs K along the vertical axis,
illustrates a possible shape for (a part of) the graph of the function k. The general shape of
Figure 1 is consistent with the function which we considered in Exercise 1, but we emphasize
that in the current analysis, k is an arbitrary function.
If, at the moment, 80 000 brushes are produced per day so that q = 8, then the corresponding
production costs are k(8) euros. In Figure 1, this starting scenario corresponds with the
lower-left red dot. Now imagine that the output level is increased by ∆q units, to 8 + ∆q. At
this higher output level, production costs are k(8+∆q) euros. The new scenario corresponds
with the upper-right red dot in Figure 1.

K
k
k(8 + ∆q) 






∆K






k(8)

8 8 + ∆q q

FIGURE 1 The change in production costs in response to an increase in production

Comparing both scenarios, the change in production costs measured in euros equals
∆K = k(8 + ∆q) − k(8).
This change is indicated explicitly in Figure 1. Since production costs increase by ∆K if
the firm increases its output by ∆q, the increase in production costs per unit of output
follows as
∆K k(8 + ∆q) − k(8)
= .
∆q ∆q
 The derivative 43

average rate of change This fraction, which is measured in euros per unit, is called the average rate of change of k
over the interval from 8 to 8 + ∆q.
For the specific cost function used in Exercise 1, you should have found that the average
rate of change of k over the interval [8, 10] is

∆K k(10) − k(8) 550 − 344

= = = 103.
∆q 10 − 8 2

In words: over the interval from 8 to 10 units of output, the production costs are rising at
an average rate of 103 euros per unit.

EXERCISE 2 Find, for the function k given in Exercise 1, the average rate of change over
the interval [8, 9]. Relate your answer as explicitly as you can to Figure 1
above. In particular, try to motivate why your answer is less than the 103
euros per unit which we obtained over the interval [8, 10].

After Example 1, it is easy to formally define the concept of ’the average rate of change’ for
an arbitrary function.

AVERAGE RATE OF CHANGE (or DIFFERENCE QUOTIENT)

Let f be a function of the variable x. The average rate of change of the function f over the
interval [a, a + ∆x] equals
∆y f (a + ∆x) − f (a)
=
∆x ∆x
difference quotient This fraction is also called the difference quotient of the function f corresponding with the
change ∆x.

Next, we discuss the crucial geometrical interpretation of our ’average rate of change’.

Figure 2 below repeats Figure 1, with a slight change in emphasis. Remember our two
scenarios in Example 1: output q increases from 8 to 8 + ∆q, and as a result costs K
increase from k(8) to k(8 + ∆q). In Figure 2, these two scenarios correspond with the points
P and Q.

K
Q
k(8 + ∆q)

∆K
k
P
k(8)
∆q

8 8 + ∆q q

FIGURE 2 The change ∆K in K as a consequence of the change ∆q in q

44 The derivative

From P to Q, the horizontal displacement equals the output rise of ∆q; the vertical dis-
placement equals the cost increase of ∆K = k(8 + ∆q) − k(8). Both these displacements are
indicated explicitly in the figure. As discussed in Section 2.2 of Chapter 1, this implies that
the slope of the (green) line through the points P and Q is equal to the quotient

∆K k(8 + ∆q) − k(8)

= .
∆q ∆q
But that is precisely our ’average rate of change’ formula! It will be clear that this ob-
servation generalizes beyond Example 1. Hence, for any arbitrary function f , we conclude
that
the average rate of change of the function f over the interval [a, a + ∆x] is
the slope of the line through the endpoints of the graph of f on this interval.

EXAMPLE 2 THE TEMPERATURE THROUGHOUT THE MORNING

Assume that, on some day, the temperature T (in degrees Celsius) between eight o’clock
and twelve o’clock in the morning is given by the formula

T = 18 − 81 (t − 16)2 ,

where t is a number in the interval [8, 12], representing time (in hours). Apparently, T = f (t),
with f (t) = 18 − 81 (t − 16)2 . You can easily check that the temperature at eight and twelve
o’clock equals, in turn, f (8) = 10 and f (12) = 16. This implies that the average rate of
change of temperature over the time interval [8, 12] is given by

∆T f (12) − f (8) 16 − 10
= = = 1.5.
∆t 12 − 8 4
So during the morning hours, the temperature is rising at an average rate of 1.5 degrees per
hour.

EXERCISE 3 Find the average rate of change of temperature over the time interval [8, 10].

1.2 THE RATE OF CHANGE

In the previous section we have learned how to calculate the average rate of change of a
function over some interval. We will now discuss how a rate of change can be determined
at a single value.

In Example 2 we found that the average rate of change of temperature over the time interval
[8, 12] is equal to 1.5 degrees per hour. But what can we say about the rate at which
temperature is changing at exactly eight o’clock?
To find an answer to this question, let’s start by determining the average rate of change over
the time interval [8, 8 + ∆t] for some values of ∆t that come closer and closer to zero. By
doing so, the endpoint 8 + ∆t of the interval comes ever closer to its starting value of 8. At
Exercise 3 above, you have already seen that when ∆t equals 2 rather than 4, the average
rate of change turns out a bit higher than our original value of 1.5. In Table 1 below, we
have included some even smaller values for ∆t. Make sure you check the calculations!
 The derivative 45

time interval with ∆t = 4 1 0.5 0.1

average rate of change

1.5 1.875 1.9375 1.9875
(in degrees per hour)

TABLE 1 The average rate of change of temperature over some ’shrinking’ time intervals

It turns out that, starting from eight o’clock, the average rate of temperature change over
shorter and shorter time intervals increases from 1.5 through 1.875 and 1.9375 to 1.9875.
These numbers seems to be approaching the value 2. So we may suspect, that the rate of
change (without ’average’ !) at exactly eight o’clock is 2 degrees per hour.
Let’s verify whether this suspicion is correct. In Table 1 we have been calculating the average
rate of temperature change over some specific time intervals. If we leave the length ∆t of
the time interval unspecified, we find that the temperature change follows as

∆T = f (8 + ∆t) − f (8) = 18 − 18 (8 + ∆t − 16)2 − 10 = 8 − 81 (∆t − 8)2


2 
2
= 8 − 18 ∆t − 16∆t + 64 = 2∆t − 81 ∆t .

Consequently, the average rate of temperature change (i.e. the difference quotient) over the
time interval [8, 8 + ∆t] equals

2
∆T 2∆t − 18 ∆t
= = 2 − 81 ∆t.
∆t ∆t

Make sure to understand that all the scenarios in Table 1 are special cases of this equation:
1
e.g., for ∆t = 1, we get 2 − 8 = 1.875. The question is now: what happens if ∆t gets
closer and closer and closer to zero, i.e. if we carry the sequence in Table 1 to its extreme?
Clearly, the average rate of temperature change will then indeed get ever closer to 2. In
mathematics, this observation is captured by the following formal notation:
 
lim 2 − 81 ∆t = 2.
∆t→0

limit This is pronounced as follows: as ∆t approaches 0, the limit of 2 − 18 ∆t equals 2.

Altogether, we have now shown that the exact value of the rate of change at eight o’clock
equals 2. At that moment, temperature is rising at a rate of 2 degrees per hour.

Our example naturally leads us to the following definition of the rate of change for an
arbitrary function.

RATE OF CHANGE (or DERIVATIVE)

Let f be a function of the variable x. The rate of change of the function f at x = a equals
∆y f (a + ∆x) − f (a)
lim = lim
∆x→0 ∆x ∆x→0 ∆x
′
derivative This number will be denoted by f (a), and is usually called the derivative of the function f
at x = a.
 46 The derivative

Let’s get back to our temperature example. As we have just seen, the rate of temperature
change at eight o’clock, i.e. the derivative at that point, equals 2; so f ′ (8) = 2. Now what
does that actually mean? Like so often, a geometrical representation is extremely helpful
here. Figure 3 sketches our temperature function f , over an interval that includes t = 8
(where T = 10) and t = 12 (where T = 16).

16 f

P
10

8 12 t

FIGURE 3 Choosing ∆t closer and closer to zero

As we already saw in Example 2, when we start at t = 8 and take ∆t = 4, the average rate
16−10
of change over the interval [8, 12] follows as 12−8 = 1.5. This value corresponds with the
slope of the line that passes through the points (8, 10) and (12, 16). From the figure, it is
easy to see that if we take smaller values for ∆t, e.g. 3 or 2, the slopes of the implied green
lines become larger. In fact, if we choose ∆t closer and closer to zero, then the (slopes of
the) respective green lines come closer and closer to the (slope of the) blue line that ’touches’
the graph of f at the point P . This implies that the slope of that blue line corresponds with
f ′ (8); so in the case at hand, it has a value of 2. This blue line is called the tangent to the
graph of f at the point P . Using the ’point-slope’ formula from Section 2.2 in Chapter 1,
we find that
T = 2(t − 8) + 10

is an equation of this tangent.

Like before, we generalize our result to arbitrary functions.

TANGENT


The line through the point a, f (a) with slope f ′ (a) is called the tangent to the graph of f


at the point a, f (a) . An equation of this tangent is y = f ′ (a)(x − a) + f (a).

EXAMPLE 3 Consider again the cost function k introduced in Exercise 1:

3 2
k(q) = q − 12q + 75q. Let’s determine the rate of change k ′ (8) of this function at q = 8.
Paraphrasing our analysis for the temperature function, we can proceed in three steps.
step 1 First, we determine the increase in production costs ∆K which results from an unspecified
rise ∆q of the production level (starting from the current level of q = 8). Removing the
brackets and simplifying the resulting expression (check!) leads to:

3
2
∆K = k(8 + ∆q) − k(8) = ∆q + 12 ∆q + 75∆q.
 The derivative 47

step 2 Next, we compute the difference quotient ∆K/∆q:

∆K
2
= ∆q + 12∆q + 75.
∆q

step 3 Finally, we evaluate the limit of this expression when ∆q approaches 0:

∆K
k ′ (8) = lim = 75.
∆q→0 ∆q

In other words: the rate of change of the cost function k at the production level q = 8 equals
75. At this production level, production costs are rising by 75 euro per 10 000 brushes.
Geometrically, this number corresponds to the slope of the tangent to the graph of k at
point P in Figure 2. An equation for that tangent is

K = 75(q − 8) + 344.

EXERCISE 4 Consider the function f given by f (x) = x2 .

(a) Find the difference quotient for the function f on the interval [1, 1 + ∆x].
(b) Determine f ′ (1).
(c) Determine an equation for the tangent to the graph of f at the point
(1, 1).

At Exercise 4(b), we trust you managed to find that f ′ (1) = 2. Of course, since the function
f is nonlinear, the value of e.g. f ′ (2) or f ′ (3) would be different. Generalizing, let’s see
what happens at an arbitrary point in the domain of f , say for x = a.

EXERCISE 5 Consider again the function f given by f (x) = x2 . Let a be some number.
(a) Find the difference quotient for the function f on the interval [a, a + ∆x].
(b) Show that f ′ (a) = 2a.

Typically, we use the symbol x, rather than a, to denote a variable that can take any value.
Sticking to this convention, we can rewrite your result at Exercise 5(b) as

f ′ (x) = 2x.

This notation brings out that, by evaluating the derivative of our function, we have actually
constructed a new function. This new function, denoted by f ′ , is called the derivative
function of f (or just ’the derivative’ for short). At any point on the graph of f , the value
of f ′ tells us the rate of change of f at that point. Equivalently: it tells us the slope of the
tangent to the graph of f at that point.
JARGON ALERT. In many books and articles, alternative notations are used for the deriva-
df dy
tive. Instead of f ′ (x), you will often encounter (x), or simply .
dx dx
Make sure you stay on track here. Until half a page ago, we have discussed how to find the
value of the derivative of some function at a specific point. E.g. for the function in Exercise
4, the derivative at x = 1 turns out to be 2. We have now seen how to find an expression
for the derivative of a function at any point. That expression is then in itself a function.
48 The derivative

E.g. the function f ′ (x) = 2x tells us the derivative of f for any value of x. And of course,
the derivative at x = 1 is then a special case.

Do all functions have a derivative function? No. Sometimes, for some points in the domain
of a function, the value of the derivative cannot be obtained; strictly speaking, in such
cases, the derivative function does not exist. In the next example, we show three typical
manifestations of this phenomenon.

EXAMPLE 4 Consider the graphs of the functions a, j and r defined by

 
x if x ≥ 0 1 if x > 0 √
a(x) = |x| = j(x) = and r(x) = x (x ≥ 0),
−x if x < 0 0 if x ≤ 0
respectively.

r
j
a

FIGURE 4 Three functions for which the derivative function does not exist

The graphs of all three functions contain the origin. And it is precisely at x = 0 that their
derivatives do not exist. We could show this formally by analyzing the relevant difference
quotients. Instead, let’s have a careful look at the graphs.
Consider the graphs of the absolute value function a in the left panel, and the jump function
j in the middle. It should be obvious that in both cases you can draw more than one line
that ’touches’ the graph of the function at the origin. So in both cases there is no unique
tangent at the origin.
The graph of the root function r in the right-hand panel does have a unique tangent at
the origin, but this tangent is vertical. Such a vertical line does not have a well-defined
slope. Altogether, we conclude that none of these three functions has a derivative at x = 0.
Consequently, the derivative functions do not exist; at least, not on the entire domain of the
original functions.

2 FINDING THE DERIVATIVE OF SOME BASIC FUNCTIONS

In this section, we will present a general expression for the derivative of an important class
of basic functions defined in Chapter 2: power functions. Furthermore, we will see how to
find the derivative of very simple combinations of basic functions. In next week’s Chapter
4, we will address the derivatives of other classes of basic functions, and show how to deal
with more complicated combinations of basic functions.

Up to now, we have used the difference quotient to find the derivative of a function, in
particular in Exercise 5(b). Doing so was a rather tedious job! Now suppose that we want
The derivative 49

to calculate the derivative of a much more complex function: the profit function w from
Section 2.3 in Chapter 1, defined by

w(p) = −4p2 + 268p − 3640.

You should recognize from Section 2 in Chapter 2, that this quadratic function is an example
of a polynomial function of degree 2. In other words, it is a ’combination’ of basic power
functions: in particular, the quadratic function p2 , the linear function p and the constant
function 1. So if we knew the derivatives of these basic functions, and if we knew how to
take the derivative of a combination of a number of functions, then finding the derivative
function w′ of our profit function would be a very simple exercise. We will now first discuss
the relevant rules; having done so, we will get back to the derivative of our profit function
at Example 5.

EXERCISE 6 Use the difference quotient to show that the derivative of the constant func-
tion f defined by f (x) = 1 equals 0. How can you interpret your result
geometrically, i.e. in terms of the graph of f ?

The result at Exercise 6 generalizes: the derivative of a constant function equals zero.

Let’s now have a look at the general class of power functions, i.e. functions of the form
f (x) = xr . It will be clear that basic constant, linear and quadratic functions belong to
this class, in particular: for r = 0, r = 1 and r = 2, respectively. However, we repeat that
the exponent r does not have to be a natural number; it can be any real number. E.g. the
1
function f (p) = = p−1 is also a power function, with exponent r = −1.
p
The following rule allows us to find the derivative of a power function quickly.

THE DERIVATIVE OF A POWER FUNCTION

If f is a power function, i.e. f (x) = xr , with r some real number, then f ′ (x) = rxr−1 .

Note that your results at Exercises 6 and 5(b) are actually special cases of the rule above,
for r = 0 and r = 2. In fact, it would be possible to prove the rule by generalizing your
analyses back there. We’ll spare you going through all that. We do emphasize that, when
we apply the rule above, we must be careful with the values of x which we choose. For
√
instance, the root function g(x) = x is defined for all x ≥ 0. By writing
√ 1
g(x) = x = x2 ,

we can see that g is a power function, with r = 21 . So its derivative function follows as

1 1 1
g ′ (x) = 12 x− 2 = 1 = √ .
2x 2 2 x

However, this derivative function is only defined for x > 0, i.e. not for x = 0. In other words:
the domain of the derivative function is a little bit smaller than the domain of the original
function here. We already observed this complication, using a geometrical argument, in
Example 4.
50 The derivative

Next, we come to the problem of finding the derivative of a combination of (basic) functions.
For the moment, we will address two ways of combining functions:
• adding them up, and
• multiplying them with a constant.
For these combinations of functions the following two rules, which you probably remember
from secondary school, are essential.

THE SUM RULE

If the derivatives f ′ and g ′ exist, then

h i′
f (x) + g(x) = f ′ (x) + g ′ (x)

and

THE SIMPLE PRODUCT RULE

If the derivative f ′ exists and c is some constant, then

h i′
cf (x) = cf ′ (x)

EXAMPLE 5 Applying the rules above, which you will probably do almost subconsciously,
we find that the derivative of the profit function w equals
h i′ h i′  ′  ′
−4p2 + 268p − 3640 = −4p2 + 268p − 3640 = −8p + 268.

Clearly, the standard rules above allow us to find this derivative in a much more convenient
way than by applying the definition of the difference quotient.

EXAMPLE 6 In order to differentiate the function f defined by

√ 2
f (x) = 3x 3 x + 2 (x 6= 0),
x

we can start by rewriting the function more explicitly in terms of power functions:

1 4
f (x) = 3x · x 3 + 2x−2 = 3x 3 + 2x−2 .

If we now apply the Sum Rule and the Simple Product Rule, it follows easily that
h 4 i′ 1 √ 4
3x 3 + 2x−2 = 3 · 43 x 3 + 2 · (−2)x−3 = 4 3 x − 3 .
x

EXERCISE 7 Determine the derivative of the following functions:

4 √
(a) f (x) = 2x2 + 2 + 15 (x 6= 0) (b) f (t) = t t + 500t (t > 0)
√ x
(c) f (x) = 2x (x > 0) (d) f (t) = (t − 1)(t + 1).
The derivative 51

EXERCISE 8 RATE OF PRODUCTION

An assembly plant has an 8-hour working day. An analysis of the production

data has shown that, on a typical day, the number of finished units produced
after t hours of work can be described very well by means of the expression

q(t) = 50t + 2t2 + 81 t3 (0 ≤ t ≤ 8).

Note that q(0) = 0, while q(8) = 592. In other words: at the start of a working
day, there are no finished units of output yet, but as the day proceeds the
number gradually increases and by the time the workers go home, 592 units
are ready for shipment.
(a) Find the number of units produced during the period [2, 4].
(b) Find the average number of units produced per hour during that period.
(c) Determine q ′ (2) and give an interpretation of this number.

3 MARGINAL ANALYSIS
Remember that, in Section 1, we have defined the concept of a derivative; next, in Section
2, we saw how to determine that derivative for a limited class of functions. In the current
section and the next, we will address two crucial applications of the derivative in economics
and business: marginal analysis and elasticity.
In general terms, marginal analysis refers to the study of the change of an economic quantity,
in response to a small change in one of its determinants. Two examples may illustrate why
this is an important topic.
• A brand manager will obviously have a fair idea of how many units his product sells
at the price which is currently being charged. But typically, he will also be interested
to know how much demand would increase (or decrease) when he lowers (or raises) its
price by a certain amount.
• A manufacturer is not only interested in the total costs corresponding to the current
production level. Typically, he also wants to know how much these costs will change if
the production level changes.

EXAMPLE 7 SENSITIVITY WITH RESPECT TO PRICE CHANGES

In Example 10 of Chapter 1, we discussed a restaurant manager who considers to increase

the current price of 30 euros of the plat du jour. In this example a major role was played
by the linear demand function d and the quadratic profit function w. Our analysis back
then implied that, for a general price level of p euros, each week the menu is ordered by q
customers and the profits are W (in euros), where

q = 220 − 4p

and W = −4p2 + 268p − 3640,

respectively. So q = d(p), where d is the linear demand function as defined by the right-hand
side of the first equation above and W = w(p), where w is the quadratic profit function as
defined by the right-hand side of the second equation above.
 52 The derivative

Before the manager takes a decision, he wants to know how ’sensitive’ demand and profits
are with respect to changes in the price level. So let’s assume that the price changes by an
amount ∆p (starting from the current price of 30 euros).
the customers Then the number of customers changes by
 
∆q = d(30 + ∆p) − d(30) = 220 − 4(30 + ∆p) − 100 = −4 ∆p.

So increasing the price by 25 cents will decrease the number of customers by 4 × 0.25 = 2.
Since d′ (30) = −4, we find that
∆q = d′ (30) ∆p
the profits To find the effect on the weekly profits, we assume that ∆p is small. Then the difference
quotient
∆W w(30 + ∆p) − w(30)
=
∆p ∆p
is bound to be close to its limiting value, i.e. the derivative w′ (30). We can denote this as
follows:
∆W
≈ w′ (30).
∆p
As a consequence,
∆W ≈ w′ (30) ∆p
Since w′ (30) = 28 (check!), increasing the price by 25 cents will increase the weekly profits
by approximately 28 × 0.25 = 7 euros. Make sure to check that in this case the exact increase
in profits, i.e. w(30.25) − w(30), equals 6.75 euros. So our approximation is pretty close
here.

To carry our findings over to a more general setting, first consider an arbitrary (nonlinear)
function f describing a relation between the variables x and y: y = f (x). If the derivative
of f exists at x = a, then, for a small change ∆x (starting from a), we have
∆y f (a + ∆x) − f (a)
= ≈ f ′ (a).
∆x ∆x
This implies that
∆y ≈ f ′ (a) ∆x.
We have now obtained the following important result.

∆-FORMULA
If y = f (x) and the derivative of the function f at x = a exists, then
∆y ≈ f ′ (a) ∆x
Here ∆x is a small change in the variable x (w.r.t. the value a) and ∆y is the resulting
change in the variable y.

In words:
the derivative translates a small change in the independent variable into
an approximation for the resulting change in the dependent variable.

Of course for a linear function ℓ we have ∆y = ℓ′ (a) ∆x: in other words, the ∆-formula will
give the exact result.
 The derivative 53

Summarized in a table:

a linear function ℓ a nonlinear function f

x : a → a + ∆x ∆y = ℓ′ (a) ∆x ∆y ≈ f ′ (a) ∆x

TABLE 2 Functions of one variable: the response to a small change in x

Figure 5 provides a geometrical interpretation of the ∆-formula and, not surprisingly, the
tangent plays a crucial role here. The issue at hand will be clear: we are interested in the
increase ∆y of the graph of the function f when we increase x from a to a + ∆x. The left
panel of the figure gives us the exact answer: ∆y corresponds with the distance between the
point P (on the graph of f ) and the horizontal blue (dotted) line.
Geometrically, the idea behind our ∆-formula is
to approximate the actual graph of the function f by its tangent at x = a.
As we learned in Example 7, the exact increase of the tangent (which is the graph of a linear
function!) equals the product of the slope of the tangent – which is f ′ (a) – and ∆x. The
right panel of the figure shows that the the increase f ′ (a) ∆x of the tangent corresponds
with the distance between the point Q (on the tangent to the graph of f ) and the horizontal
blue (dotted) line. As long as the points P and Q are close together (i.e. if ∆x is small),
these two distances will be similar, and the formula gives us an acceptable approximation.
Note that the ∆-formula gives the exact result in case of a linear function because for such
a function its graph and the tangent coincide.

y P  y P



 tangent
f  f Q
∆y )




 f ′ (a) ∆x
to get a clear picture
f (a) f (a)
our ∆x is quite large

a a + ∆x x a a + ∆x x

FIGURE 5 The (blue) tangent is close to the (red) graph.

EXERCISE 9 Assume that the management of the restaurant discussed in Example 7 wants
to increase its weekly profits by 24 euros.
(a) Find an approximation for the new price which would be needed to gene-
rate this profit increase.
(b) What is the exact new price that would be needed to generate this profit
increase?
54 The derivative

EXAMPLE 8 MARGINALITY

In Example 7, we found that a change of the menu price by ∆p (starting from an initial
price level of 30) leads to a change in profits equal to
∆W ≈ w′ (30) ∆p.
Economists often want to know by how much profit changes if the price rises by precisely
one unit. In that case we have ∆p = 1, so our approximation implies a profit rise of around
w′ (30) = 28.
Economists refer to w′ (30) as the marginal profit at p = 30. This concept of marginality
is absolutely essential in economics: in later courses, you will constantly run into concepts
such as marginal profit, marginal revenue, marginal cost, marginal utility, the marginal
propensity to consume, etcetera. The basic idea is always the same: we use the derivative
to approximate the change in some dependent variable (e.g. cost) in response to a one-unit
change in some independent variable (e.g. sales volume).

In the next exercise, you can apply what you have just learned. Be sure to pay close attention
to the units in which production costs and production level are measured!

EXERCISE 10 A company located on the Sand Algarve in Southern Portugal produces salt
using salt pans. The company economist has used the weekly production
data for the past year to estimate the following cost function:
k(q) = 0.001q 3 − 0.05q 2 + 1.2q + 12 (0 ≤ q ≤ 60).
Here q denotes the weekly salt production in tons (i.e. thousands of kilos),
while the weekly production costs are given in thousands of euro.
(a) Starting from a weekly production level of 50 tons, use the concept of
marginal cost (M C) to approximate the cost of producing 100 kilograms
of salt extra. How good is this approximation?
(b) Like most firms, our salt producer faces demand fluctuations: in some
weeks demand falls, in others it rises. The firm’s economist is convinced
that demand increases should not be met if marginal cost exceeds 0.4. To
what interval is production restricted, if it is required that M C ≤ 0.4?

4 ELASTICITY
In Section 3, we have seen that the marginal or derivative of a function (e.g. a profit
or a cost function) can be used to approximate the changes in the dependent variable, if
the independent variable changes by a small amount. This approach is all about absolute
changes: e.g. if, at some point, the derivative of a demand function equals −3.4, then a
price increase by 1 unit will approximately lead to a decline in sales volume by 3.4 units.
But often economists are more interested in relative changes: e.g., by how many percent will
sales volume fall if the price is raised by 1%? The answer to this question is provided by the
elasticity of a function. Like ’marginality’, ’elasticity’ is another core concept in economics
and business, and (after the next exercise) we will illustrate it by means of an example.
The derivative 55

EXERCISE 11 RELATIVE CHANGES

Imagine that some price p is raised by 3 units. So the absolute price change
is ∆p = 3.
(a) Find the size of the relative price change if the initial price is p = 15.
(b) Find the relative price change in percentages if the initial price is p = 15.
(c) Write down an expression for the relative price change if the initial price
p and the absolute price change ∆p are unspecified.

EXAMPLE 9 A FERRY

Consider a ferry which car drivers can use to cross a river. Currently, the ferry-man charges
a fare of 2.50 euro per car for a one-way crossing. Of course, the daily number of cars that
make a crossing, say q, will depend on the fare p according to some demand function d, so
q = d(p). An economic consultant has estimated that the function d can be modeled by
means of the following equation:

q = 1000(1 − 21 p + 1 2
16 p ) (0 ≤ p ≤ 4).

The ferry-man considers raising his fare, e.g. to 2.75 euro per crossing, because he hopes to
increase his income. In response, some car drivers will then inevitably stay away (e.g. to
make a detour and use a bridge instead). The ferry-man understands that, if too many car
drivers do so, his income will actually decrease rather than increase. So ultimately, he will
somehow have to compare the changes in the related variables q and p.

The ferry-man from Example 9 is clearly interested in the impact which a price change, say
by 0.25 euro, will have on the demand for his services. How to measure that impact? An
obvious first idea is to determine the difference

d(2.75) − d(2.50) = 97.65625 − 140.625 ≈ −43.

In words: if we increase the price from 2.50 to 2.75 euro i.e. by 0.25 euro, then demand will
fall by around 43 crossings. Undeniably, this number does give us some idea of how demand
responds to changes in the price. But crucially, the above difference is an absolute change.
To get a clear idea of the sensitivity of the number of crossings with respect to the fare, it
is probably better to think in terms of relative changes. After all, an absolute price increase
of e.g. 0.25 euro can mean many things. A price increase from 1 to 1.25 euro (say: for a
candy bar) constitutes a massive 25% rise, whereas the same absolute increase starting from
1000 euro (say: for a flatscreen TV) only means a rise of 0.025%; almost imperceptible! In
the example at hand, with an original fare of 2.50 euro, those 25 cents correspond to a 10%
price rise. Similarly, an absolute change in demand by 43 crossings doesn’t mean anything in
isolation; such a number can only be understood within its appropriate context, i.e. against
the background of the number of crossings before the price rise. In the case of the ferry-man,
the relative change of demand follows as:

d(2.75) − d(2.50) 43
=− = −0.306
d(2.50) 140.625
 56 The derivative

or minus 30.6%. So we conclude that a 10% price rise will lead to a decline in demand
by no less than 30.6%. That sounds like a massive response; something which would have
been much less obvious from the original absolute change. In fact, your intuition should now
suggest that raising the fare may not be such a good idea here.

Our numerical example illustrates, that it is often more informative to measure the sensitivity
of demand to price on a relative rather than an absolute basis. Let us therefore generalize
our numerical analysis. If a price of p changes by ∆p then, as we saw in Exercise 11, the
relative price change is given by
∆p
.
p
And of course, the corresponding relative change in demand is
∆q
.
q
To find out how these relative changes are related, the obvious starting point is our approx-
imation for the absolute changes as described by the ∆-formula:

∆q ≈ d′ (p) ∆p.

By dividing both sides by q, we find that

∆q 1 ∆q p ∆p
≈ d′ (p) · ∆p =⇒ ≈ d′ (p) · .
q q q q p
Finally, by using q = d(p), we obtain that
∆q p ∆p
≈ d′ (p) · .
q d(p) p
The (red) factor which premultiplies the relative price change,
p
Eld (p) = d′ (p) · ,
d(p)
elasticity is called the elasticity of the demand function d. Using this notation, we get
∆q ∆p
≈ Eld (p)
q p
Apparently,

the elasticity translates a small relative change in the price level

into an approximation for the resulting relative change in demand.

The elasticity of the demand function should give us a reliable picture of the sensitivity of
demand to price in the neighborhood of some price p. Moreover, the elasticity can quickly
be calculated for all values of p, using the derivative function. Do bear in mind though
that the elasticity only gives us an approximation for the relative change in the value of a
function, just like the derivative gives you an approximation for the absolute change in the
value of a function (remember Section 3).

In economics, the elasticity concept is of fundamental importance: not just to characterize

the sensitivity of demand to price, but also e.g. the sensitivity of supply to price, of demand
to consumer income, of demand w.r.t. the price of substitute products, etcetera, etcetera.
Let us therefore define the elasticity concept for an arbitrary (positive) function.
The derivative 57

ELASTICITY

If y = f (x), where f is a positive function, then the elasticity at some value x > 0 is defined
as
x
Elf (x) = f ′ (x) ·
f (x)
∆y
The relative change in the variable y is approximately equal to Elf (x) times the relative
y
∆x
change in the variable x:
x
∆y ∆x
≈ Elf (x)
y x

We can use the definition above to obtain a general expression for the elasticity of our
ferry-man’s demand function, d(p) = 1000(1 − 12 p + 1 2
16 p ). Applying the rules from Section
2 to find the derivative of d, and plugging all the components correctly into the elasticity
definition, we end up with
p p
Eld (p) = d′ (p) · = 1000(− 21 + 81 p) ·
d(p) 1000(1 − 12 p + 1 2
16 p )

− 21 p + 18 p2 2p2 − 8p
= = .
1 − 12 p + 16 1 2
p p2 − 8p + 16
You can easily verify that at a price of 2.50 euro, the formula above implies an elasticity
of Eld (2.5) = −3 31 . In words: starting from an original price of 2.50 euro, a price increase
by 1% causes an approximate decrease in the number of crossings by 3 31 %. Consequently,
a price increase by 10%, from 2.50 to 2.75 euro, causes a percentage change in crossings by
roughly Eld (2.5) · 10 = −33.3%. This 33.3% decrease is a reasonable approximation for the
exact decrease of 30.6% which we found before.

To round off this chapter, we will briefly apply the concept of ’the price elasticity of demand’
to another example.

EXAMPLE 11 A DEMAND FUNCTION WITH CONSTANT ELASTICITY

In Example 4 of Chapter 2, the relation between demand D and price P was given by
D = d(P ), where
d(P ) = A P −α .
Note that d is a power function, with positive constants A and α. Since the derivative of
this demand function is
d′ (P ) = −αA P −α−1 ,

its elasticity equals

P P
Eld (P ) = d′ (P ) · = −αA P −α−1 · = −α.
d(P ) A P −α
Note: the price elasticity of this demand function is constant. This is a nontrivial result: e.g.
in our ferry-man example, the price elasticity was visibly not constant (to convince yourself,
you may try to determine the elasticity for p = 3 rather than p = 2.5).
58 The derivative

END-OF-CHAPTER EXERCISES

EXERCISE 1 Find the derivative of the function f defined by

(a) f (x) = 5 (b) f (x) = x4 (c) f (x) = 9x10 (d) f (x) = π 7
√ 1 x+1
(e) x−1 (x2 + 1) x (f) √ (g) (h) 3x−11 .
x3 x5

EXERCISE 2 Find the (relative) rate of change and the elasticity of the function f defined
by
1√
(a) f (r) = 4πr2 (b) f (y) = Ay b+1 (c) f (A) = A2 A
.

EXERCISE 3 Compute the elasticity of the power function f defined by f (x) = xa .

EXERCISE 4 Prove that El2f (x) = Elf x).

EXERCISE 5 Find an equation of the tangent at the point (0, 1) to the graph of the function
f defined by
(a) f (x) = (1 + x)−1 (b) f (x) = (1 + x)5 (c) f (x) = (1 − x)1/4 .

PRODUCTIVITY

An assembly plant has an 8-hour working day. An analysis of the production data has
shown that, on a typical day, the productivity P in units per hour can be represented by
the following graph:

120

112

8 t (in hours)

Apparently, at the start of the working day, output is being produced at a rate of 120
units per hour, i.e. 2 units per minute. As the working day progresses, productivity is not
constant: it fluctuates. So productivity can be regarded as a function of time t, say P = p(t).

EXERCISE 6 The total production during a working day can reasonably be estimated by
p(0) + p(8)
8×
2
units. This estimate will be . . .
(a) smaller than the real total production
(b) larger than the real total production
(c) exactly equal to the real total production
(d) There is not enough information to choose between the options (a), (b)
and (c).
The derivative 59

The fluctuations in productivity during the working day, as displayed by our graph, can be
described by the formula
p(t) = − 41 t2 + t + 120.

EXERCISE 7 The average rate of change of productivity during an entire working day is
given by
p(0) + p(8)
(a)
2
p (0) + p′ (8)
′
(b)
2
p (0) − p′ (8)
′
(c) .
8
(d) None of the options at (a), (b) or (c) is correct.

EXERCISE 8 At the end of a working day, productivity is . . .

(a) decreasing at a rate of p′ (8) units per minute
(b) decreasing at a rate of p(8) units per hour
(c) decreasing at a rate of p′ (8) units per hour
(d) decreasing at a rate of −p′ (8) units per hour.

Management wants to know, how many units of output have been produced after t hours
into the working day, for any t between 0 and 8. It can be shown that this number is given
by the following expression:

Q = at3 + 21 t2 + 120t + b (0 ≤ t ≤ 8).

EXERCISE 9 In this expression for Q, it must hold that . . .

(a) a < 0 and b > 0
1
(b) a = − 12 and b < 0
1
(c) a = − 12 and b = 0
(d) a = − 34 and b = 0.

GASOLINE

A gas station faces continuous price changes for its best-selling fuel type: Euro 95. The
owner of the station has found out that the relation between the number of liters Q sold per
day, and the price p in euros per liter, is given by

Q = 1000p2 − 6000p + 12 750.

EXERCISE 10 If the price p (in euros per liter) increases by one cent, the resulting decrease
in daily sales can be approximated by b − ap liters. In this expression,
(a) a < 15 and b < 50
(b) a > 15 and b < 50
(c) a < 15 and b > 50
(d) a > 15 and b > 50.
60 The derivative

EXERCISE 11 Currently, the price of a liter of Euro 95 is 150 cent, i.e. p = 1.5. How big is
the price elasticity of demand?
(a) Somewhere between −0.2 and −0.4
(b) Somewhere between −0.4 and −0.6
(c) Somewhere between −0.6 and −0.8
(d) Somewhere between −0.8 and −1.0.

There has been a massive eruption of violence in the Middle East. The station owner expects
that, as a result, the price of Euro 95 will shoot up from 150 to 175 cent next week.

EXERCISE 12 Judged on the basis of the price elasticity of demand, what will happen to
his revenue next week?
(a) Compared to this week, his revenue will go up
(b) Compared to this week, his revenue will go down
(c) His revenue will remain the same as this week
(d) There is not enough information to choose between the options (a), (b)
and (c).

The station owner still expects the price of Euro 95 to rise from 150 to 175 cent per liter.
He uses the concept of marginality to approximate the implied sales decrease next week. He
then wonders how many customers he will lose per day, knowing that his average customer
refuels around 50 liters of Euro 95 at a time.

EXERCISE 13 According to the station owner’s calculation, the number of customers will
decrease by around . . .
(a) 16 per day
(b) 15 per day
(c) 14 per day
(d) 13 per day.

AT WORK IN A FACTORY

The daily output Q of some factory is given by

Q = 900 L1/3 (800 ≤ L ≤ 1200),
where L denotes the input of labour hours. Currently, 1000 labour hours are used each day.

A crucial feature of the function above is the so-called marginal product of labour; while we
will not define this concept formally here, its meaning should be obvious from our discussion
in the main text.

EXERCISE 14 Currently, the marginal product of labour is around . . .

(a) 2
(b) 3
(c) 4
(d) 5.
The derivative 61

EXERCISE 15 Consider the following two claims concerning the elasticity of the factory’s
output with respect to its labour input:
(I) Currently, this elasticity is around one-third.
(II) This elasticity has the same value for any level of the labour input.
(a) Claim (I) is correct, claim (II) is incorrect.
(b) Claim (I) is incorrect, claim (II) is correct.
(c) Both claims (I) and (II) are incorrect.
(d) Both claims (I) and (II) are correct.
62 The derivative
Finding the derivative 63

4 FINDING THE DERIVATIVE

In Section 1 of the previous chapter, we used a geometrical argument to explain the meaning
of the derivative of a function f at some point a in its domain: it is the rate of change at that
point, and it corresponds with the slope of the tangent to the curve of the function at that
point. Initially, we used the difference quotient to calculate that derivative. This operation
can obviously be performed for several (although maybe not all) points in the function’s
domain. In fact, we found that the derivative of a function, say f , was a new function: the
derivative function f ′ . Of course, you would prefer to determine that derivative function (also
called: to differentiate the original function) by using rules with general applicability, rather
than to repeat the laborious ’difference quotient’ approach. In Section 2, we reviewed three
such rules which should be familiar from secondary school. First, we addressed an important
standard derivative, allowing us to differentiate the class of power functions quickly, without
re-inventing the wheel each time. Next, we covered two important arithmetic rules: the Sum
Rule which allows us to take the derivative of a sum of functions, and the Simple Product
Rule which enables us to differentiate a multiple of a function.

This week, we will discuss some more complicated arithmetic rules, and some additional
standard derivatives.

We start with three crucial arithmetic rules: the (general) Product Rule and the Quotient
Rule in Section 1, and the infamous Chain Rule in Section 2. Of course, many of you have
already learned how to apply these rules in secondary school. But possibly your skills have
become a bit rusty, while some of you may not have learned those rules at all. That’s why
we will carefully review them. At the end of Section 2, we will provide a complete list of all
the arithmetic rules which are relevant for us.

In Section 3, we will discuss the standard derivative of two important classes of functions,
introduced in Chapter 2: exponential and logarithmic functions. Again, for many of you
this will boil down to a secondary school rehearsal. At the end of the section, we will provide
a complete list of all the standard derivatives which are relevant for us.

1 THE PRODUCT RULE AND THE QUOTIENT RULE

Functions that can be regarded as the product or the quotient of basic functions can be
differentiated by applying the Product Rule and the Quotient Rule. As already mentioned,
you will probably recognize these rules from secondary school. If not, don’t worry: we will
provide plenty of examples. We start with the Product Rule.
64 Finding the derivative

THE PRODUCT RULE

If the derivatives f ′ and g ′ exist, then

 ′
f (x) · g(x) = f ′ (x) · g(x) + f (x) · g ′ (x)

For those students who are not familiar with the derivative of a function, this rule may come
somewhat ’out of the blue’; in particular, if you never saw the Product Rule before, it may
strike you that there is no term involving f ′ (x) · g ′ (x). To motivate the structure of the rule,
let’s use the ∆-formula: if a is a number in the domain of a function h, then for a small
change ∆x (with respect to a) we have ∆y ≈ h′ (a) ∆x. Since ∆y = h(a + ∆x) − h(a), we
get
∆y ≈ h′ (a) ∆x =⇒ h(a + ∆x) − h(a) ≈ h′ (a) , ∆x

=⇒ h(a + ∆x) ≈ h(a) + h′ (a) ∆x

Now consider a function which can be regarded as the product of two basic functions, say,
f (x) · g(x). Applying our result above to this product, starting from some number a in the
domain of both the functions f and g, we obtain:
  
f (a + ∆x) g(a + ∆x) ≈ f (a) + f ′ (a) ∆x g(a) + g ′ (a) ∆x

2
= f (a) g(a) + f ′ (a) g(a) ∆x + f (a) g ′ (a) ∆x + f ′ (a) g ′ (a) ∆x .

If we now bring the term f (a) g(a) to the left-hand side of the ≈ sign and divide both sides
by ∆x, we end up with

f (a + ∆x) g(a + ∆x) − f (a) g(a)

≈ f ′ (a) g(a) + f (a) g ′ (a) + f ′ (a) g ′ (a) ∆x.
∆x
Note: the left-hand side of this expression is the relevant difference quotient for finding the
derivative of the function f (x) · g(x) at x = a! By choosing ∆x closer and closer to zero, the
last term in the right-hand side vanishes, and we are left with two terms which correspond
with the formula in the Product Rule.

EXAMPLE 1 Consider the following function f , which is the product of two functions:

f (x) = (2x + 1)(4x − 3).

We can easily find the derivative of f using the Product Rule:

f ′ (x) = 2 · (4x − 3) + (2x + 1) · 4 = 16x − 2.

Please note that we could also start by eliminating the brackets in the expression (2x +
1)(4x − 3), to obtain f (x) = 8x2 − 2x − 3; the derivative then follows (without the Product
Rule!) again as f ′ (x) = 16x − 2.
So in this example, the Product Rule does not offer much of an advantage.

We now want to show an interesting application of the Product Rule. In Chapter 3’s Section
4, we introduced the elasticity concept. In Example 9 and its sequel, we used this concept to
Finding the derivative 65

investigate the consequences of a price increase by a ferry-man. By calculating the elasticity

for the ferry-man’s demand function, we obtained an estimate for the percentage decrease
in demand caused by a percentage price increase. But what about the consequences for his
revenue? We already speculated a bit about this issue last week: now is the right time to
formalize things.

EXAMPLE 2 THE PRICE ELASTICITY OF REVENUE

As we have seen, the price elasticity of demand measures how sensitive demand is to price
changes: in particular, it measures the percentage change in demand in response to a percent-
age change in the price. By analogy, the price elasticity of revenue measures the percentage
change in revenue in response to a percentage change in the price. Let’s see what we can
say about the price elasticity of revenue.
Of course, revenue R equals price times demand, which is itself a function of price, say d(p);
revenue is therefore also a function of price, say R = r(p), with r(p) = p · d(p). Using the
elasticity definition from Chapter 3, the price elasticity of revenue is given by
p
Elr (p) = r′ (p) · .
r(p)

Of course, this expression contains the derivative of revenue with respect to the price, r′ (p).
Since r(p) = p · d(p), we can use the Product Rule to find that r′ (p) = 1 · d(p) + p · d′ (p).
Substituting this into the elasticity definition and rearranging a bit, we get:
p p
Elr (p) = [1 · d(p) + p · d′ (p)] = 1 + d′ (p) · = 1 + Eld (p).
p · d(p) d(p)
We conclude: the price elasticity of revenue equals the price elasticity of demand, increased
by 1. This is a pretty interesting result. Consider e.g. our ferry-man again. In last week’s
Example 9 and its sequel, we saw that at a price of 2.50 euro, the price elasticity of his
demand equals Eld (2.5) = −3 31 . We now conclude that, at that price, the price elasticity of
his revenue equals Elr (2.5) = 1 + Eld (2.5) = −2 31 . Consequently, a price increase by 10%
will cause a revenue decrease of around 23 31 %. This result formalizes our earlier ’common
sense’ conclusion: raising his fare is not a good idea here.

EXAMPLE 3 For a certain good, the relationship between the price p (currently 3 euro)
and the demand in liters is aptly described by the demand function d defined by
12
d(p) = 2 + (p > 0).
p
For this demand function, the price elasticity follows as
p −12 p 12
Eld (p) = d′ (p) · = 2 · =− .
d(p) p 2 + 12/p 2p + 12

So if the current price is p = 3, we have Eld (3) = − 32 . Now how will a price increase of e.g.
0.15 euro affect demand and revenue? This price increase amounts to 5%, so the percentage
change in demand can be approximated by Eld (3) · 5 = − 32 · 5 = −3 31 %. Moreover,
with Elr (3) = 1 + Eld (3) = 31 , the percentage change in revenue should be approximately
1
3 · 5 = 1 32 %. Note that, since the price elasticity of revenue is positive now, revenue rises
when the price is increased.
66 Finding the derivative

Please verify that the exact revenue increase (from 18 to 18.3) equals 1.666% here. The fact
that our approximation coincides with the exact value is a matter of pure coincidence.

Note: In our ferry-man example, the price elasticity of demand was smaller (in the sense of
’more negative’) than −1; consequently, the price elasticity of revenue was negative. So if
the price is raised, the relative decline in demand will be larger than the relative price rise,
and the net result is that revenue will decrease. In Example 3 above, the price elasticity of
demand was larger (in the sense of ’less negative’) than −1; consequently, the price elasticity
of revenue was positive. So in case of a price increase, the relative decline in demand will be
smaller than the relative price rise, and the net result is that revenue will increase.
An interesting borderline scenario occurs when, at some price level, the price elasticity of
demand is exactly equal to −1. In this so-called ’unit elasticity’ case, the price elasticity
of revenue equals 0. This implies that small price changes, triggering proportional demand
changes, will leave revenue approximately unchanged; so apparently, revenue reaches its
maximum at that price.
Having addressed the Product Rule, and its application to the price elasticity of revenue,
we now move to the Quotient Rule.

THE QUOTIENT RULE

If the derivatives f ′ and g ′ exist, then

 ′
f (x) g(x) · f ′ (x) − f (x) · g ′ (x)
=
g(x) g(x)2

Again, in the absence of vivid memories from secondary school, this rule will come ’out
of the blue’, so we want to motivate its structure. As it turns out, the Quotient Rule can
actually be derived from the preceding Product Rule. Our interest is now in the derivative of
a function which can be regarded as the quotient (rather than the product) of two functions,
say h(x) = f (x)/g(x). When we multiply both sides by g(x), this translates into

f (x) = h(x) · g(x).

If we now apply the Product Rule, we obtain:

f ′ (x) = h′ (x) · g(x) + h(x) · g ′ (x).

This equation can be solved for the sought-after derivative of h:

f ′ (x) h(x) · g ′ (x)
h′ (x) = − .
g(x) g(x)
Finally, if we replace h(x) in the right-hand side by its definition f (x)/g(x) then, after some
rewriting, we indeed end up with the Quotient Rule:
f ′ (x) f (x) · g ′ (x) g(x) · f ′ (x) − f (x) · g ′ (x)
h′ (x) = − = .
g(x) g(x)2 g(x)2

In the following example, as in the previous Example 1, there are again (at least) two ways
of dealing with the problem.
Finding the derivative 67

EXAMPLE 4 Let
3x2 − 5x
h(x) = √ (x > 0).
x
√
We can view this function as the quotient of f (x) = 3x2 − 5x and g(x) = x = x1/2 .
Applying the Quotient Rule leads to:

x1/2 · (6x − 5) − (3x2 − 5x) · 12 x−1/2

h′ (x) =
x
9√ 5
= 6x1/2 − 5x−1/2 − 32 x1/2 + 52 x−1/2 = 2 x − √ .
2 x

Alternatively, we can rewrite our function as h(x) = (3x2 − 5x) · x−1/2 = 3x3/2 − 5x1/2 , so
that it can be differentiated without using the Quotient Rule:

9√ 5
h′ (x) = 29 x1/2 − 52 x−1/2 = 2 x − √ .
2 x

As you can see, both approaches again give the same result (of course!).

EXERCISE 1 Determine the derivative of the function h in two different ways: first by
using the Product or Quotient Rule, and next by eliminating the brackets or
developing the ratio. Which method is easier and/or faster here?
√ √
1  1
(a) h(x) = x + x x 2√ (b) h(x) = (1 − x) x3 + 3 .
x x x

EXERCISE 2 Remember Chapter 2’s Example 2, where the total cost of the Toyota dealer
could be expressed as a broken function k of order size q, with

18 000 + 500q 2
k(q) = (q > 0).
q

Find the derivative of the function k in two different ways: first by using the
Quotient Rule, and next by developing the ratio. Which method is easier
and/or faster here?

From Examples 1 and 4 and Exercises 1 and 2 we can conclude that, even when a function can
be viewed as a product or a quotient of functions, it may not be necessary (or desirable) to
use the Product Rule or the Quotient Rule. But in other cases, those rules are indispensable.
The next example shows this for the Quotient Rule.

EXAMPLE 5 Consider the function h defined by

x+1
h(x) = (x 6= 0 and x 6= −2).
x2 + 2x

This time, we have to use the Quotient Rule: there is nothing to gain from ’developing the
ratio’ here. The function h can be viewed as the quotient of the two functions f and g,
defined by f (x) = x + 1 and g(x) = x2 + 2x. Since f ′ (x) = 1 and g ′ (x) = 2x + 2, the
derivative of h follows as:

g(x) · f ′ (x) − f (x) · g ′ (x) (x2 + 2x) · 1 − (x + 1) · (2x + 2) −x2 − 2x − 2

h′ (x) = = = .
g(x)2 (x2 + 2x)2 (x2 + 2x)2
68 Finding the derivative

EXERCISE 3 Find the derivative of the function f defined by f (x) =

1  √

x+1 x+1
(a) + 1 x2 + 1 (b) (c) √ .
x x−1 x−1

2 THE CHAIN RULE

In Chapter 2’s Example 5, we studied the construction of a triangular gutter with maximal
capacity out of a plate of length 100 and width 18. We found that, if x (instead of the
original symbol s) represents the length of the base of the triangle, then the capacity of the
gutter is given by C = c(x), where
q
c(x) = 50x 81 − 14 x2 (0 < x ≤ 18).

Your secondary school memory (which will be recapitulated in the next chapter) should tell
you that, in order to find the value of x which maximizes the capacity of the gutter, we
will need the derivative of the function c. Of course, c can be regarded as the product of
two functions of x: this suggests the use of the Product Rule. However, there is an obvious
complication here. In the last factor of the function c, under the square root sign, we do not
simply have x, but a (quadratic) function of x. As will be defined formally below, such a
structure is called a composite function. To differentiate such a function, we need the Chain
Rule.

To bring out the issues as clearly as possible, let us focus on the ’tricky’ part of our capacity
function, i.e. its last factor, to which we will refer as function k:
q
k(x) = 81 − 14 x2 .

Now have a very careful look at the structure of this function. Imagine that you want to
calculate k(x) for a given value of x, using your pocket calculator. You will definitely do this
in two successive steps: first you calculate 81 − 14 x2 , and then you take the square root of
your previous result. Generalizing, the value k(x) is obtained by first calculating the value
g(x) = 81 − 41 x2 , and then applying to this result the square root function f defined by
√

f (y) = y. Consequently, the function k can actually be written as k(x) = f (g(x) . A
function which can be written in this way is called a composite function.

COMPOSITE FUNCTIONS

A function h with the structure h(x) = f (g(x)) is called a composite function. Since g is
applied first, and f is then applied to the result, we also speak of the function ’f after g’.
The function g is called the inner function, while f is called the outer function.

EXAMPLE 6 The expression h(x) = (2x+3)4 can be written as f (g(x)), with g(x) = 2x+3
and f (y) = y 4 . The function h is therefore composite.
√ √
The expression m(x) = x3 can be written as f (g(x)), with g(x) = x3 and f (y) = y. The
function m is therefore also composite.
 Finding the derivative 69

The input-output interpretation of a function which we presented in Chapter 1 can be

helpful to understand the idea of a composite function. Consider the first composite function
discussed in Example 6: h(x) = (2x + 3)4 . We have seen that we can write this function as
h(x) = f (g(x)), with ’inner’ function g(x) = 2x + 3, and ’outer’ function f (y) = y 4 . In the
remainder, we will refer to the ’output’ of the function h as z, and to the ’output’ of the
function g as y.

device device
input x
g
output y 2x + 3 input y
f
output z (2x + 3)4

device h

The function h can be interpreted as a device, or machine, which converts an input value x
into an output value of z = (2x + 3)4 . This conversion can actually be understood as the
result of two successive operations:
– first, machine g converts the input x into an intermediate output of y = 2x + 3;
– then machine f converts y = 2x + 3 into the final output of z = y 4 = (2x + 3)4 .
In the figure above, we have put in line the two machines that correspond with the functions
f and g. The crucial observation is that the output of the first machine, 2x + 3, serves as the
input for the second one. Note that the order of the two machines is relevant here! Imagine
that we interchange the two machines in our ’assembly line’; i.e. f is applied first, and then
g is applied to the result. In that case, we get a very different expression (check!):

x → x4 → 2x4 + 3.

Now how to find the derivative of such a composite function? As already stated, this requires
the dreaded Chain Rule. We will motivate this rule in a similar way as for the Product Rule
in the previous section. Conceptually, we take the picture above as our starting point.
So: consider some value for x, say a, in the domain of an (unspecified) composite function
h(x) = f (g(x)). Let’s examine the effect of a small change in x on the value of h, i.e. on z.
We can think of this effect in two alternative (but ultimately equivalent) ways.

the direct way First, let us not ’open up’ machine h. Of course, the derivative of h translates the change
in x into an approximate change in z. This is the ∆-formula from Chapter 3’s Section 3,
which we can formally express as follows:

∆z ≈ h′ (a) ∆x. (1)

in two steps Next, we do ’open up’ machine h. We can now determine the change in z in two steps.
First, the presence of machine g implies that, starting from x = a, a small change in x will
trigger a change in y. And of course, the derivative of g determines how large that change
will be, approximately. Formally:
∆y ≈ g ′ (a) ∆x.
70 Finding the derivative

Second, the presence of machine f implies that, starting from y = b, a small change in y will
create a change in z. The approximate size of the change depends on the derivative of f :

∆z ≈ f ′ (b) ∆y.

Since the input of machine f is the output of machine g, we have b = g(a). Successive
substitution leads to



∆z ≈ f ′ (b) · ∆y ≈ f ′ g(a) · ∆y ≈ f ′ g(a) · g ′ (a) · ∆x. (2)

If you now compare equations (1) and (2), then the formal rule stated below should not
come as a surprise.

THE CHAIN RULE

If the derivatives f ′ and g ′ exist, then


′

f g(x) = f ′ g(x) · g ′ (x)

So although the Chain Rule may initially look very abstract, it actually makes perfect
intuitive sense:
– a change in x by one unit triggers a change in y by roughly g ′ units;
– every one unit change in y then triggers a change in z by roughly f ′ units;
– the ultimate change in z is therefore the product of those two derivatives.
Now how to apply the chain rule in practice? As you may remember from secondary school,
an obvious approach is to apply the rule in two steps:
step 1: determine f ′ (y) and g ′ (x).
step 2: in the expression f ′ (y) · g ′ (x), replace y by g(x).

EXAMPLE 7 Consider again the composite functions introduced in Example 6. As we

have seen, the expression h(x) = (2x + 3)4 can be written as f (g(x)), with g(x) = 2x + 3
and f (y) = y 4 . So
step 1: f ′ (y) = 4y 3 and g ′ (x) = 2
step 2: according to the Chain Rule we have

h′ (x) = f ′ (g(x)) · g ′ (x) = 4(g(x))3 · 2 = 4(2x + 3)3 · 2 = 8(2x + 3)3 .

√ √
The expression m(x) = x3 can be written as f (g(x)), with g(x) = x3 and f (y) = y. So
1
step 1: f ′ (y) = √
2 y and g ′ (x) = 3x2
step 2: according to the Chain Rule we have

1 1
m′ (x) = f ′ (g(x)) · g ′ (x) = p · 3x2 = √ · 3x2 .
2 g(x) 2 x3
Finding the derivative 71

You may find the second part of Example 7 a bit unconvincing; after all, there is an obvious
temptation to rewrite the function m as m(x) = x3/2 , and then we don’t need the Chain Rule
anymore to obtain its derivative! But in many cases the Chain Rule is truly indispensable,
as demonstrated in Example 8 below, and in Exercises 4 and 5.

EXAMPLE 8 A CONSTRUCTION PROBLEM

Let’s return to the problem which actually triggered our Chain Rule discussion: the con-
struction of a triangular gutter with maximal capacity. Remember: with x representing the
length of the base of the triangle, the gutter’s capacity is given by C = c(x), where
q
c(x) = 50x 81 − 14 x2 (0 < x ≤ 18).

We have already established that, to differentiate this capacity function, we will need to use
both the Product Rule and the Chain Rule. The result is:

25x2
q q
1
c′ (x) = 50 81 − 41 x2 + 50x · q · − 21 x = 50 81 − 41 x2 − q .
2 81 − 14 x2 2 81 − 14 x2

EXERCISE 4 Determine the derivative of the function h defined by √

p
2 6 1 − 2x
(a) h(x) = 2x + 1 (b) h(x) = (3x−1) (c) h(x) = √ (x > 0).
1 + 2x

EXERCISE 5 Let f be a function for which the derivative exists. Find the derivative of
 2 x p
(a) f (x2 ) + x2 (b) xf (x) (c) f (x) (d) (e) f (x).
f (x)

To conclude this section, we reiterate all the arithmetic rules which have been discussed here
and in the previous chapter.

LIST OF ARITHMETIC RULES

name of the rule

 ′
sum rule f (x) + g(x) = f ′ (x) + g ′ (x)
 ′
simple product rule cf (x) = cf ′ (x)
 ′
product rule f (x) · g(x) = f ′ (x) · g(x) + f (x) · g ′ (x)
 ′
f (x) g(x) · f ′ (x) − f (x) · g ′ (x)
quotient rule =
g(x) g(x)2

′
chain rule f g(x) = f ′ (g(x)) · g ′ (x)

3 THE DERIVATIVE OF EXPONENTIAL AND LOGARITH-

MIC FUNCTIONS

In Chapter 2 we have seen examples of growth (or decay) at a constant rate, like the growth
of China’s GDP per head, and the depreciation of the value of a car. In all these cases we
72 Finding the derivative

were able to describe the process by means of exponential functions. Our discussion then
naturally carried us to their inverse: logarithmic functions. In this section we will study the
derivatives of both exponential and logarithmic functions.

3.1 THE DERIVATIVE OF EXPONENTIAL FUNCTIONS

Let’s consider the exponential function f of the form f (x) = ax , with a > 0. Since f (0) =
a0 = 1, (0, 1) is a point on the graph of this function, for any base value a. Although our
ultimate objective is to find the derivative of the entire function, we start by investigating
its derivative at this specific point, i.e. for x = 0.

THE DERIVATIVE OF EXPONENTIAL FUNCTIONS AT x=0

As an example, take the exponential function with base 2. What can we say about its
derivative at x = 0? Well, if g(x) = 2x , the relevant difference quotient is

g(0 + ∆x) − g(0) 2∆x − 20 2∆x − 1

= = .
∆x ∆x ∆x

In Table 1 below, we evaluate this difference quotient for a number of values for ∆x which
come closer and closer to zero.

2∆x − 1
∆x
∆x
0.1 0.71773
0.01 0.69556
0.001 0.69339
0.0001 0.69317
0.00001 0.69315

TABLE 1 The behaviour of the difference quotient as ∆x gets smaller

This table suggests that the limiting value of our difference quotient is

2∆x − 1
lim ≈ 0.69315,
∆x→0 ∆x

implying that the derivative of g at x = 0 is given by

g ′ (0) ≈ 0.69315 < 1.

EXERCISE 6 Consider the exponential function h defined by h(x) = 3x .

Make a table similar to Table 1 and show that h′ (0) ≈ 1.09862 > 1.

As you know, the number g ′ (0) represents the slope of the tangent (say ℓ) to the curve
y = 2x at the point (0, 1). Similarly, the number h′ (0) represents the slope of the tangent
(say ℓ′ ) to the curve y = 3x at that same point (0, 1). Both these tangents are sketched in
Figure 1 below.
Finding the derivative 73

y y
x
y=2 y = 3x

6 6
ℓ′

4 4
ℓ

2 2

2 x 2 x

FIGURE 1 The tangents at the point (0, 1)

If you compare the two tangents ℓ and ℓ′ in Figure 1, it is visible that the angle of tangent
ℓ with the horizontal axis is smaller than 45 degrees, while tangent ℓ′ has an angle with
the horizontal axis which is larger than 45 degrees. This corresponds with the fact that the
slope of the first tangent (around 0.69315) is smaller than 1, while the slope of the second
tangent (roughly 1.09862) is larger than 1. Apparently, when the base of our exponential
function increases (here: from 2 to 3), then the slope of the tangent to its graph at the point
(0, 1) increases. But then there must be a number somewhere between 2 and 3, such that
the corresponding exponential function has a tangent with slope equal to 1. This number
will be denoted by e (named after the famous mathematician Leonhard Euler). The number
e, which is approximately equal to 2.718, plays an important role in a lot of economic
applications.

Summarizing: the number e is defined such, that the slope of the tangent to the curve y = ex
at the point (0, 1) equals 1. Formally:

If f (x) = ex , then
e∆x − 1
f ′ (0) = lim = 1.
∆x→0 ∆x

THE DERIVATIVE OF THE EXPONENTIAL FUNCTION WITH BASE e

The result which we have just obtained may seem quite useless: after all, it concerns the
derivative of one very specific exponential function (with base e), at one very specific value
of x (i.e. x = 0). But on the basis of this result, it is actually possible to find the derivative
of any simple exponential function.

To show this, we start with the exponential function with base e: f (x) = ex . If we want
to obtain the derivative of this function at some value of x (other than x = 0), then the
relevant difference quotient is

f (x + ∆x) − f (x) ex+∆x − ex ex · e∆x − ex e∆x − 1

= = = ex · .
∆x ∆x ∆x ∆x

As always, we find the derivative of f by evaluating what happens to this difference quotient
e∆x − 1
if ∆x gets closer and closer to zero. Using the fact that → 1 as ∆x → 0, we can
∆x
 74 Finding the derivative

easily conclude that

f ′ (x) = ex .

We have now found the following important result:

THE DERIVATIVE OF THE EXPONENTIAL FUNCTION WITH BASE e

 x ′
e = ex

By using the Arithmetic Rules discussed before, it is possible to differentiate more complex
exponential functions with base e, as the following example illustrates.

EXAMPLE 9 Let the function h be defined by

2
h(x) = ex +x
.


Since h(x) can be written as f g(x) with g(x) = x2 + x and f (y) = ey , we conclude that h
is a composite function. We can therefore find its derivative by applying the Chain Rule:

2 2
h′ (x) = ex +x
· (2x + 1) = (2x + 1) ex +x
.

EXERCISE 7 Find the derivative of the function h defined by

2 x √
(a) h(x) = x ex (b) h(x) = x (c) h(x) = 5 e5 x
.
e

THE DERIVATIVE OF GENERAL EXPONENTIAL FUNCTIONS

Now consider the derivative of the general exponential function f defined by

f (x) = ax ,

where the base doesn’t have to be the number e: it can be any a > 0. As it turns out, the
derivative of f can be obtained by cleverly rewriting the function. To do so, we need some
of the properties of logarithms which were discussed in Chapter 2’s Section 3.2. Remember
that the solution of the exponential equation

bx = c

(with c > 0) is given by b log c. We also learned how to express this logarithm with base b in
terms of the quotient of two logarithms with base 10. This result was important, because
the logarithm with base 10 can be found under the log key of your pocket calculator. But
in many cases, it is actually preferable to use logarithms with the base number e (instead of
natural logarithm 10). This so-called natural logarithm is denoted by ln (so ln x = e log x), and can be found
on your pocket calculator under the ln key. For example, you can use that button to check
that ln 2 ≈ 0.69315. So, by definition, the solution of the exponential equation

ex = a
Finding the derivative 75

is given by e log a = ln a. Make sure to remember what this means in words: ln a is ’the
exponent to which we must raise e, in order to obtain a’. In other words:

eln a = a

Let’s now return to our general exponential function f . If we replace its base a by the
equivalent eln a , we get:
 x
f (x) = ax = eln a = ex ln a .

Using this ’trick’, we have managed to rewrite the general exponential function f as an
exponential function with base e. Applying the Chain Rule, it’s now trivial to find its
derivative:
f ′ (x) = ex ln a · ln a = ax · ln a.

We have now found the following property:

THE DERIVATIVE OF EXPONENTIAL FUNCTIONS WITH ANY BASE

For a > 0,
 ′
ax = ax ln a

EXAMPLE 10 Let the function h be defined by

2
h(x) = 2x +x
.


Since h(x) can be written as f g(x) with g(x) = x2 + x and f (y) = 2y , the function h is
composite. We can therefore obtain its derivative by applying the Chain Rule:

2 2
h′ (x) = 2x +x
ln 2 · (2x + 1) = (2x + 1) 2x +x
ln 2.

EXAMPLE 11 EXPONENTIAL GROWTH

In Chapter 2’s Section 3, we observed that exponential growth refers to processes involving,
quote, ’quantities that increase or decrease by amounts which are proportional to their size’.
If an exponential function, say f (t) = b · at , is used to describe such an exponential growth
process then, following Chapter 3, its derivative f ′ represents the rate of change of that
process. Hence, we obtain:

f ′ (t) = b · at ln a = (ln a) · f (t).

It should be clear that this result is totally consistent with the defining characteristic of
exponential growth as embodied in our quote from Chapter 2 above! To make things a bit
more specific:
If a quantity grows exponentially, then the rate of change is at every
moment proportional to the value of the quantity at that moment.
76 Finding the derivative

EXERCISE 8 On January 1st 2015, a certain area is home to a population of 100 badgers.
On January 1st 2018, their number has increased to 108. A biologist assumes
that the number of badgers in the area grows exponentially. Let’s denote the
number of badgers after t years from January 1st 2015 (t = 0) by N (t).
(a) Determine the exponential function N .
(b) What is the growth factor of the badger population per 10 years?
(c) Determine the ’time to double’, i.e. the amount of time it takes for the
badger population to become twice as large.
(d) Determine the rate of change of the number of badgers on January 1st
2018.

EXERCISE 9 A machine cost 10 000 euro at the time it was purchased (t = 0). Two years
later, its value has dropped to only 7000 euro. We assume that the decrease
in value evolves over time according to an exponential function. Let’s denote
the machine’s value precisely t years after the time of purchase by V (t).
(a) Determine the exponential function V .
(b) What is the growth factor per half year?
(c) Determine the ’halflife value’, i.e. the amount of time it takes for the
machine’s value to be cut in half.
(d) Determine the rate of change of the machine’s value, 5 years after its
purchase.

3.2 THE DERIVATIVE OF LOGARITHMIC FUNCTIONS

As our last topic this week, we will show you how to determine the derivative of the loga-
rithmic function with base a > 0. So: consider the function f defined by

f (x) = a log x (x > 0).

By definition, it holds (for any positive x) that

af (x) = a log x
a
= x.

If you do not see this immediately, then remember that a log x is ’the exponent to which we
must raise a, in order to obtain x’. Now let’s see what happens when we differentiate both
the left- and the right-hand side of this equation with respect to x. To find the derivative
of the left-hand side, we need to apply the Chain Rule. The result is:
 ′
af (x) = af (x) ln a · f ′ (x) = a log x
a
ln a · f ′ (x) = x ln a · f ′ (x).

The derivative of the right-hand side, x, is of course 1. Since these two derivatives must be
equal, we get:
x ln a · f ′ (x) = 1.

We can now solve this equation to obtain the derivative of f :

1
f ′ (x) = .
x ln a
Finding the derivative 77

THE DERIVATIVE OF LOGARITHMIC FUNCTIONS

For a > 0 and x > 0,

a ′ 1
log x =
x ln a

Finally, we consider an important special case of our result: choose a = e. Using the fact
that ln e = 1, we now find that, for x > 0,
 ′ 1
ln x =
x

EXERCISE 10 Determine the derivative of the function f defined by

√
(a) f (t) = 10t (b) f (t) = 5 log t (c) f (K) = K · 30.1K
1000
(d) f (t) = ln 6t (e) f (t) = 1000(1 − e−0.2t ) (f) f (t) =
1 + 500 e−0.2t

EXERCISE 11 Find the derivative of the function f defined by f (x) =

x + 1 x
(a) x + ln x (b) ex ln x (c) ln (d)
x−1 ln x

To conclude this section, we reiterate all the standard derivatives which have been discussed
here and in the previous chapter.

LIST OF STANDARD DERIVATIVES

f (x) = f ′ (x) =
c (c is some number) 0
r
x (r is some real number) rxr−1
ex ex
ax (a > 0) ax ln a
1
ln x
x
a 1
log x (a > 0)
x ln a
78 Finding the derivative

END-OF-CHAPTER EXERCISES

EXERCISE 1 Differentiate the following expressions

x−1 x−1 3x − 1
(a) (b) 2 (c)
x+1 x −1 x2
+x+1
ax + b √
1
(d) (e) xn a x + b (f) 2
cx + d ax + bx + c
EXERCISE 2 Find a formula for h′ (x) if


(a) h(x) = f (x2 ) (b) h(x) = f xn g(x) .

EXERCISE 3 Find Elx f (x) if

(a) f (x) = ln x
(b) f (x) = xp eax (here a and p are constants).

EXERCISE 4 The demand for D for apples in the U.S. as a function of income r for the
period 1927 to 1941 was estimated as D = A r1.23 , where A is a constant.
Find and interpret the elasticity of D with respect to r, ie. the income elas-
ticity of demand.

EXERCISE 5 A wine farmer sells wine to customers who visit the farm. Suppose that the
daily demand d in liters as function of the price p in euros per liter is given
by
40
d(p) = p (p ≥ 0).
2
4p + 4p + 1
At a price of 6 euros per liter, the price is decreased with 10 cents. With help
of the (point) price elasticity we can conclude that, as a consequence of the
price decrease, the demand increases or decreases with how many percent?

AN INSTALLATION PROBLEM

In a number of end-of-chapter exercises of Chapter 2, we studied the installation costs of

a water supply pipe running from a residence to a country cottage. We found that, if x
(rather than ℓ) represents the length of the pipe alongside the asphalted road, these costs
amount to K = k(x), where
p
k(x) = 40x + 80 x2 − 400x + 80 000 (0 ≤ x ≤ 200).

EXERCISE 6 Taking the derivative of this cost function leads to k ′ (x) =

80(x − 200)
(a) 40 + √
x2− 400x + 80 000
80(2x − 400)
(b) 40 + √
x2 − 400x + 80 000
40(x − 200)
(c) 40 + √ .
2
x − 400x + 80 000
(d) None of the options (a), (b) or (c) is correct.

Next week we will see that, when we want to find the minimum of the function k, the zero(s)
of its derivative function k ′ are of great interest.
Finding the derivative 79

EXERCISE 7 The zero(s) of the function k ′ can be found by solving the quadratic equation
(a) 3x2 + 1200x + 80 000 = 0
(b) 3x2 − 1200x + 80 000 = 0
(c) 3x2 + 1200x − 80 000 = 0
(d) 3x2 − 1200x − 80 000 = 0.

A FENCE

In Example 5 of Chapter 1 and its sequel, Example 1 of Chapter 2, we studied the problem
of fencing a piece of land, where the fenced area should be 100 m2 large. We found that,
if x (rather than ℓ) represents the length of the fenced area, then the total length L of the
fence is given by L = ℓ(x), where

2x2 + 200
ℓ(x) = (x > 0).
x

The challenge was to minimize the length of the fence.

EXERCISE 8 Taking the derivative of this function leads to ℓ′ (x) =

200
(a) +6
x2
200
(b) 2 − 2
x
200
(c) 2 + 2
x
(d) 4x.

EXERCISE 9 If a is the unique zero of the function ℓ′ , then ℓ(a) =

(a) 10
(b) 20
(c) 30
(d) 40.

LOGISTIC GROWTH

The way in which the sales S of a certain product develop over time can be described by a
logistic growth function. That is: S = s(t), with

1
s(t) = c + (t ≥ 0),
d + e−2t

where c > 0 and d ∈ (0, 1) are constants.

EXERCISE 10 In the long run (that is: for large values of t), sales S will approach . . .
(a) c
1
(b) c +
d+1
1 + cd
(c)
d
c
(d) .
d
80 Finding the derivative

EXERCISE 11 The derivative s′ (t) is . . .

(a) positive for all values of t ≥ 0
(b) first positive and then negative
(c) first positive, then negative and then positive again
(d) negative for all values of t ≥ 0.

Obviously, the derivative s′ which you determined in the previous exercise is itself a function
of time. Therefore, it is possible to calculate the derivative of s′ .

EXERCISE 12 The derivative of the function s′ is zero for

1
(a) t = 2 ln d
(b) t = − 21 ln d
(c) t = 12 d
(d) t = − 21 d.

EXERCISE 13 At t = 0, the sales level S is 6, and its rate of change is 1 81 . What are the
values of c and d?
(a) c = 5 and d = 0
(b) c = 5.2 and d = 0.25
1
(c) c = 5.25 and d = 3
(d) c = 5.5 and d = 1.

THE DEMAND FOR GASOLINE

Suppose that the demand D (in liters) for gasoline at price p (in euros) is given by

D = b e−0.2p ,

where b is a constant. Clearly, D is a function of p: D = d(p).

EXERCISE 14 The demand function d satisfies d′ (p) = c d(p), with c =

(a) 0.2b
(b) −0.2
(c) −0.2b
(d) 0.2.

Remember that, for p > 0 and a small number ∆p, the following approximation holds:

d(p + ∆p) ≈ d(p) + d′ (p) ∆p

EXERCISE 15 Suppose that the price of gasoline rises by 0.1 euro. Using the approximation
above, we find that demand decreases by around . . .
(a) 2 %
(b) 0.2 %
(c) 0.2 units
(d) 0.02 units.
Finding the derivative 81

EXERCISE 16 The price elasticity of demand is given by

(a) −0.2
(b) −0.2D
(c) −0.2 p
(d) None of the options (a), (b) or (c) is correct.

Suppose an oil cartel controls the entire supply of gasoline and can price it as it wishes. Its
revenues R are a function of this price p: R = r(p).

EXERCISE 17 The derivative r′ of the revenue function r is zero if

(a) p = 2
(b) p = 3
(c) p = 4
(d) p = 5.
82 Finding the derivative
Properties of a function 83

5 PROPERTIES OF A FUNCTION

In the previous two chapters, we have seen what the derivative of a function is, and how we
can determine it for several important classes of functions.

In this chapter, we discuss how the basic properties of a function can be established. As we
will see, derivatives play a crucial role in such an examination.

In the introductory Section 1, we will use an example to illustrate some characteristic prop-
erties of functions. One crucial property is the possible occurrence of extreme values, i.e.
maxima or minima; these concepts (and some others) will be defined formally. We will
also see that, apart from the first derivative (i.e. ’the’ derivative as defined so far), we can
typically determine a function’s second derivative as well.

In Section 2, we limit ourselves to functions defined on a single open interval. We will

carefully examine what the first and second derivatives of such a function can tell us about
the properties of (the graph of) that function. Finding and characterizing the extreme values
will be a major element of the discussion.

In Section 3, we drop the ’single open interval’ restriction. We will see that this may bring
in additional extreme values, of a different nature than the ones discussed in Section 2.

1 SOME CHARACTERISTIC PROPERTIES OF A FUNCTION

In secondary school you learned how derivatives can be used to find the minima and maxima
of a function. Of course we are going to refresh your knowledge in this field. After all,
you must know how to find out where, for instance, an average cost function reaches its
minimum or where a profit function reaches its maximum. However, as the title of this
chapter suggests, extreme values are not the only relevant characteristic of a function (and
its graph): other properties may be interesting as well. We will use the ’product life cycle’
to illustrate this.

EXAMPLE 1 THE PRODUCT LIFE CYCLE

After a new video game was launched, its sales (per day) in a big department store developed
over time according to the following pattern:
84 Properties of a function

sales

300

200

100

1 2 3 4 5 6 7 8 t (in months)

FIGURE 1 The ’rise and fall’ of sales

As you can see, the graph above exhibits a number of characteristic features:
1 It has a lowest and a highest point.
2 In some areas the graph goes up, elsewhere it goes down.
3 Sometimes the graph ’opens upward’, sometimes it ’opens downward’.
4 The graph is ’connected’: you can draw it in a single stroke, without taking your pen off
the paper.
5 The graph is ’smooth’: there are no kinks.
In your Marketing class, you must have been introduced to the ’Product life cycle theory’,
and you probably see right away that most of the properties listed above correspond with
features of this theory. Immediately after the launch of the product, sales only develop
slowly, but after a while they start to gather steam. We can see that after one month, sales
are at around 80 items per day. This first phase is called the introduction stage. Please note
that, during this stage, the graph ’opens upwards’: it rises ever faster, i.e. the local slope is
increasing. After this stage, the expansion stage starts: sales increase even further, but after
a while this increase starts to slow down. This more moderate development, where the graph
’opens downward’, signals the beginning of the maturity stage. This stage corresponds with
the period somewhat before and after the sales maximum. Crucially, that sales maximum
cannot be maintained. Once the initial ’fad’ starts to fade, the declining stage sets in. Sales
start to go down ever faster, but after a while this decline ’bottoms out’. Now the saturation
stage begins, in which sales remain at a stable level. From our graph it can be seen that
this level is about 60 items per day.

In the previous two chapters, we have seen how to differentiate a function f to obtain its
derivative f ′ . This derivative is often referred to as the first derivative of f . As you know,
the first derivative of a function f of x is itself a function of x. Consequently, it would be
possible to differentiate f ′ (x) with respect to x (in fact, we have already done this in an
end-of-chapter exercise last week). This derivative of the derivative is denoted as f ′′ (x);
for obvious reasons, the function f ′′ is called the second derivative of the original function
f . Surely you remember that the first derivative f ′ tells us, how fast the original function
f is increasing or decreasing at any point of interest in its domain. Similarly, the second
derivative f ′′ tells us, how fast the first derivative f ′ is increasing or decreasing; this, in
turn, provides vital information about the behavior of the original function f !
 Properties of a function 85

In this chapter, we will discuss techniques to find the characteristic properties of the graph
of some given function. The first and second derivatives play a central role in this process.
In general, these derivatives are indispensable to determine where the graph of a function
increases, decreases, opens downward, opens upward and where we can find the highest and
lowest point(s) of the graph.

Before addressing the role of the first and second derivatives in the next section, we will
provide the definitions of some concepts which we will use later on.
For a start, we observe that the domain of a function of one variable generally consists of a
finite number of intervals. To illustrate, consider the following examples.
The broken function h defined by
1
h(x) =
x2 − 1
is defined for all x except x = −1 and x = 1. Consequently, the domain of the function h
consists of three intervals: (−∞, −1), (−1, 1) and (1, ∞). Observe that the endpoints −1
and 1 of the interval (−1, 1) are not included in that interval; this is denoted by the ( and
open interval ) brackets. For this reason, this interval is called open. The other two intervals (−∞, −1)
and (1, ∞) are also open. After all, since these intervals are ’endless’ at one side, they each
have only a single endpoint, −1 and 1, respectively; and these endpoints are not included.
The quadratic function q defined by q(x) = x2 is defined for all x. Its domain is the set of all
real numbers. In math this set, which is sometimes denoted by (−∞, ∞), is also considered
as an (open) interval.
√
The domain of the root function r defined by r(x) = x consists of one interval: [0, ∞).
This interval is not open because its endpoint of x = 0 is included in the interval; this is
denoted by the square [ bracket. An interval such as [0, 2) is also not open, because one of
its two endpoints, x = 0, is included in the interval.

In the rest of this chapter, we will exploit the fact that the domain of a ’decent’ function
consists of a finite number of intervals. In the theory and examples which we are going to
discuss, we will only consider functions defined on a single interval. This is not really a
limitation: after all, functions with a domain consisting of more than one interval can then
be examined ’interval by interval’.

The following figure shows the graph of a function f defined on the single interval [−1, 3].

y E
C
f
H
G
A
D F

−1 3 x
B

FIGURE 2 The graph of a function defined on the interval [−1, 3]

 86 Properties of a function

Note that the graph of the function f reaches its highest point at E. We say that the function
has a global maximum at this point. Similarly, the function has a global minimum at B. In
addition to the global maximum occurring at E, the graph also peaks at the points A, C
and H. We say that the function has a local maximum at each of these points. Similarly,
there is a local minimum at D, and also at F . We give the formal definitions below.

GLOBAL AND LOCAL MAXIMUM

Let f be a function defined on an interval I containing c.

The function f has a global maximum at x = c if its graph reaches its highest point at


c, f (c) . In that case, it holds that

f (c) ≥ f (x),

for all x in I.
If the foregoing only holds for all x close to c, i.e. in a ’neighborhood’ of c, then the function
has a local maximum at x = c.

We trust that you are able to write down the corresponding definition of a global or local
minimum yourself.
extreme value In the rest of this chapter, we will use the term extreme value to refer to either a maximum
or a minimum.

2 THE ROLE OF THE FIRST AND SECOND DERIVATIVE

In this section we will consider functions defined on a single open interval I. This could
be a ’bounded’ interval like (0, 1) or an ’unbounded’ one like (0, ∞). As observed before,
the defining feature of such an open interval is that its endpoints are not included in the
interval. Furthermore, we assume that the graph of the function is ’connected’, and that
the first and second derivatives f ′ and f ′′ exist on the entire interval I.
For such functions, we will address the following two questions in turn:
(1) What can we conclude about the graph of the function if its first (or second) derivative
is positive or negative on (parts of) the interval I?
(2) What can we conclude about the graph of the function if its first (or second) derivative
changes sign at some point in the interval I?

2.1 THE FIRST DERIVATIVE

We begin by discussing the questions (1) and (2) above for the first derivative. The second
derivative will be addressed in Section 2.2.

question (1) We start with question (1). By means of example, let’s assume that f ′ is positive on the
entire interval I. As we learned in Chapter 3, this means that the rate of change of the
original function f is positive at any point along its graph. Consequently, the function f is
strictly increasing. Figure 3 below illustrates two such strictly increasing functions.
 Properties of a function 87

y y

x x

FIGURE 3 Strictly increasing functions

The formal definitions are as follows.

INCREASING AND DECREASING FUNCTIONS

A function f is strictly increasing on the interval I if x2 > x1 (x1 , x2 in I) implies that

f (x2 ) > f (x1 ).

A function f is strictly decreasing on the interval I if x2 > x1 (x1 , x2 in I) implies that

f (x2 ) < f (x1 ).

A function which is either strictly increasing or strictly decreasing is called monotonic.

The fact that a positive first derivative corresponds with a strictly increasing function is, we
trust, intuitively obvious to you. Still, let’s state things formally.

MONOTONICITY AND THE SIGN OF THE FIRST DERIVATIVE

Let f be a function of one variable defined on an open interval I.
If f ′ (x) > 0 for all x in the interval I, then f is strictly increasing on I.
If f ′ (x) < 0 for all x in the interval I, then f is strictly decreasing on I.

EXAMPLE 2 Consider the function g defined by

g(x) = 2x3 − 9x2 + 12x.

To find out where this function is strictly increasing or strictly decreasing, we have to
examine where its derivative is positive or negative. To do so, we first determine the zeros
of that derivative (this is a general rule in mathematics: inequalities can be examined by
stationary point first solving the corresponding equality). These zeros are called the stationary points of the
function g. Since the derivative of g can be written as

g ′ (x) = 6x2 − 18x + 12 = 6(x2 − 3x + 2) = 6(x − 1)(x − 2),

we conclude that g has two stationary points: its derivative will be zero at x = 1 and at
x = 2.
 88 Properties of a function

We can represent these stationary points on a straight line, where the values underneath
the line refer to x and those above it to g ′ :

0 0
g′
1 2

Next, we determine the sign of the derivative at the other (i.e. the non-stationary) points.
To start, we may choose an arbitrary number smaller than 1, say 0. As g ′ (0) = 12 > 0, the
derivative is positive everywhere to the left of 1. Similarly, consider an arbitrary number
larger than 2, say 3. As g ′ (3) = 12 > 0, the derivative is positive everywhere to the right of
2. Finally, take an arbitrary number between 1 and 2, say 1.5. As g ′ (1.5) = −1.5 < 0, the
derivative is negative everywhere between 1 and 2.
sign diagram We can add this information to the straight line above, in the form of plus and minus signs.
This leads us to the following sign diagram for the derivative g ′ :

+ + + + 0 − − 0 + + + +
g′
1 2

From this sign diagram for the derivative g ′ , we can now conclude the following about
the behavior of the original function g: that function is strictly increasing on the intervals
(−∞, 1) and (2, ∞), and strictly decreasing on the interval (1, 2). This conclusion can be
presented in another diagram, as follows (note the label g rather than g ′ !):

ր ց ր
g
1 2

The meaning of the symbols ր and ց will be self-evident.

EXAMPLE 3 Consider the function a defined by


x if x ≥ 0
a(x) = |x| =
−x if x < 0.

Note that this ’absolute value’ function is not differentiable at x = 0 (if you don’t remember
why, please review Chapter 3’s Example 4). Of course, a′ (x) = 1 for x > 0 and a′ (x) = −1
for x < 0. All this leads to the following sign diagram for the derivative a′ :

− − − − × + + + +
a′
0

Note that we have put the symbol × above the point corresponding with x = 0, to indicate
that the derivative a′ doesn’t exist at this point. It will be clear that the function a is
strictly decreasing on the interval (−∞, 0), and strictly increasing on the interval (0, ∞).
The diagram below visualizes these conclusions.

ց ր
a
0
 Properties of a function 89

EXERCISE 1 Let m be the function defined by

1
m(x) = 2 (x 6= 0).
x
Construct a sign diagram for the derivative m′ , and explain where the function
m is strictly increasing and/or decreasing.

question (2) Remember the two questions which we posed at the start of this section. The first one, ’what
can we conclude about the graph of a function if its first derivative is positive or negative on
(parts of) the interval I?’, has now been answered: the function will be strictly increasing
or decreasing on those parts of I. We can now move to the second question: ’what can we
conclude about the graph of a function if its first derivative changes sign at some point in
the interval I?’
Let’s reconsider the function g introduced in Example 2. According to the sign diagram
which we constructed in that example, the derivative g ′ changes sign twice: at x = 1 and at
x = 2. Let’s have a closer look at both these sign changes in turn. At the stationary point
x = 1, the sign of the derivative goes from plus to minus. Consequently, as we have seen,
the graph of g is strictly increasing to the left of x = 1, and strictly decreasing to its right.


This implies that the graph of g reaches a maximum at 1, g(1) , at least locally. Similarly,
the graph of g is strictly decreasing to the left of the stationary point x = 2, and strictly


increasing to its right. This implies that the graph of g reaches a minimum at 2, g(2) , at
least locally. Figure 4 sketches the graph of the function g, using the fact that g(1) = 5 and


g(2) = 4. Note that the maximum at 1, g(1) is indeed local (rather than global), because
in another area of the graph the function produces even higher values. A similar conclusion


holds for the minimum at 2, g(2) .

g
5
4

1 2 3 x

FIGURE 4 The graph of the function g

For the function g, the local minimum and maximum on its graph correspond precisely with
the stationary points of the function, i.e. the points where its derivative equals zero. In
geometrical terms: at these points on the graph, the tangent is horizontal. This should make
excellent intuitive sense: so much so, that you may be tempted to equate ’stationary points’
with ’extreme values’. Unfortunately, that overstates the case. In Example 4, we will show
that a stationary point does not always correspond with an extreme value of the function.
Additionally, we will show that an extreme value does not always occur at a stationary point.
90 Properties of a function

EXAMPLE 4 Consider the function n defined by

n(x) = x3 .

Since its derivative is given by n′ (x) = 3x2 , we conclude that x = 0 is the only stationary
point of the function n. The derivative is strictly positive both for values to the left of x = 0,
and to its right. So we get the following sign diagram:

+ + + + 0 + + + +
n′
0
ր ր
n
0

Clearly, the function n must be strictly increasing to the left and to the right of x = 0. So
although the function has a stationary point at x = 0, its graph does not have a minimum
or maximum there! However, it is still true that the graph of the function n has a horizontal
tangent at x = 0. Figure 5 clarifies how this horizontal tangent can coincide with the absence
of a minimum or maximum.

3 y = x3

−2 −1 1 2 x

−1

−2

−3

FIGURE 5 The graph of the function n

Example 4 illustrates that, as announced, a stationary point of a function does not necessarily
correspond with an extreme value of that function!

We also mentioned that the opposite holds as well: not every extreme value of a function
occurs at a stationary point. Just consider the function a introduced in Example 3. Accor-
ding to the sign diagram constructed there, the derivative a′ changes sign at x = 0: it goes
from negative to positive. Of course, this implies that the graph of a has a minimum at the
origin. However, x = 0 is not a stationary point of the function a: after all, the function
is not even differentiable at x = 0, so its derivative cannot be zero there! This example
illustrates that a function can have an extreme value at a non-stationary point.
 Properties of a function 91

The ’First-Derivative Test’ brings our results together.

THE FIRST-DERIVATIVE TEST

Let f be a function defined on an open interval I containing the point c. Assume that the
function is differentiable on the entire interval I, with the possible exception of x = c.
If f ′ switches sign at x = c, then the function f has an extreme value at x = c.
If f ′ goes from plus to minus, then f has a maximum at x = c.
If f ′ goes from minus to plus, then f has a minimum at x = c.

Make sure to understand exactly what is being claimed here. The crucial observation is that
an extreme value does not occur simply when the derivative of a function equals zero i.e. at
a stationary point; instead, it occurs when the derivative exhibits a sign change. Typically,
such a sign change will occur at a stationary point (cf. Example 2), but it may also occur at
a point where the derivative is not defined (cf. Example 3). In the ’First-Derivative Test’,
this qualification is captured by the subtle phrase ’with the possible exception of x = c’.

EXERCISE 2 According to Exercise 1, the derivative of the function m defined by

1
m(x) = 2 (x 6= 0)
x
switches sign at x = 0. Does the graph of m have a maximum or a minimum
there?

EXERCISE 3 Use a sign diagram for its derivative to find the extreme value(s) of the func-
tion p defined by
x
(a) p(x) = 3x2 − 2x3 (b) p(x) = .
1 + x2

2.2 THE SECOND DERIVATIVE

Remember the two questions that were formulated at the beginning of Section 2? In the
previous subsection, we addressed them for the first derivative. Now, we want to answer
them for the second derivative.
question (1) So our first question is: what can we conclude about the graph of a function f if its second
derivative is positive or negative on (parts of) the interval I?
For a start, let’s assume that the second derivative f ′′ is positive on the interval I. Note that
f ′′ is the derivative of f ′ . So we can apply our previous results to the function f ′ , instead of
f ! In particular, since the derivative of f ′ is positive, it follows that f ′ is a strictly increasing
function. And with f ′ being the derivative of the original function f , we conclude:
if we move to the right, the slope of the tangents to the graph of f increases
By consequence, if we move along the graph of f from left to right, the tangent to the graph
turns in a counterclockwise direction. Intuitively, this means that the graph of the function
opens upward: it falls ever slower, or it rises ever faster. In analogy with our terminology
for parabolas, we call the function f convex.
92 Properties of a function

y y

x x

FIGURE 6 A convex function A concave function

The left panel of Figure 6 provides an illustration.

On the other hand: if f has a negative second derivative f ′′ on the interval I, then f ′ is a
strictly decreasing function, so we can conclude:
if we move to the right, the slope of the tangents to the graph of f decreases
This means that the tangent will rotate in a clockwise fashion. We can now say that the
graph opens downward: it rises ever slower, or it falls ever faster. Continuing our analogy
with parabolas, the function f is called concave, as in the right panel of Figure 6. Putting
it all together:

CONVEX AND CONCAVE FUNCTIONS

Consider a function f defined on an open interval I.

• If f ′′ (x) > 0 for all x in I, then f ′ is strictly increasing, and f is called convex.
• If f ′′ (x) < 0 for all x in I, then f ′ is strictly decreasing, and f is called concave.

EXAMPLE 5 In Example 2, we introduced the function g defined by

g(x) = 2x3 − 9x2 + 12x.

So far, we have only studied the first derivative g ′ , and its implications for the properties
of the original function g. Let’s now have a closer look at the second derivative g ′′ . Since
g ′ (x) = 6x2 − 18x + 12, we find that g ′′ (x) = 12x − 18. In order to construct a sign diagram
for g ′′ , we first determine its zeros. Obviously, this second derivative is zero for x = 1.5 only,
so we get the following sign diagram:

− − − − 0 + + + +
g ′′
1.5

According to this diagram, the function g is concave on the interval (−∞, 1.5) and convex
on the interval (1.5, ∞). This information can be presented in the following diagram:

⌢ ⌣
g
1.5
 Properties of a function 93

The meaning of the symbols ⌣ and ⌢ will be self-evident. Note that, of course, the implied
shape of the function g corresponds with Figure 4.

In the following exercise we will see that, just like the first derivative before, the second
derivative does not necessarily change sign at a zero.

EXERCISE 4 Let the function p be defined by p(x) = x4 .

Show, by constructing a sign diagram, that p′′ does not change sign at its
unique zero at x = 0.

EXERCISE 5 Use a sign diagram for their second derivatives to determine where the func-
tions defined in Exercise 3 are convex and/or concave.

question (2) Now for the second question: what can we conclude about the graph of a function if its
second derivative changes sign at some point in the interval I?
Let’s continue Example 5, i.e. analyze the function g that was introduced in Example 2.
According to the sign diagram which we constructed a minute ago, the second derivative
g ′′ changes sign at x = 1.5: it goes from minus to plus. As a consequence, at x = 1.5 the
function changes from being concave to being convex. We say that the function g has an
inflection point at x = 1.5. Compare Figure 4 again! Generalizing:

INFLECTION POINT

Let f be a function defined on an open interval I such that f ′′ exists on this interval.


If f ′′ changes sign at x = c, then the point c, f (c) is called an inflection point for the
function f . At such a point, the function changes from being concave to being convex (if
f ′′ changes from minus to plus), or from being convex to being concave (if f ′′ changes from
plus to minus).

Next to Figure 4, Figure 5 provides another example of an inflection point. As you can
see, the origin is an inflection point of Example 4’s function n, defined by n(x) = x3 . By
contrast, as we saw in Exercise 4, the apparently somewhat similar function p defined by
p(x) = x4 does not have an inflection point at the origin.

EXERCISE 6 Find the inflection points (if any) for the functions defined in Exercise 3.

2.3 THE SECOND-DERIVATIVE TEST

In Section 2.1 we explained how the sign survey of the first derivative of a function can be
used to investigate the stationary points of that function. Using our results from Section 2.2,
we now discuss how (some of) those stationary points can be characterized by evaluating
the second derivative of the function.

To illustrate, we continue our analysis of the function g, first introduced in Example 2:

g(x) = 2x3 − 9x2 + 12x.
94 Properties of a function

Remember that its first derivative is given by

g ′ (x) = 6x2 − 18x + 12,

from which it follows that the function has the stationary points x = 1 and x = 2, with
g(1) = 5 and g(2) = 4, respectively. Differentiating another time, we obtained its second
derivative as
g ′′ (x) = 12x − 18.

As a consequence, at the first stationary point x = 1, we have g ′′ (1) = −6 < 0. Now note:
g ′′ is a linear function. So the fact that g ′′ is negative at x = 1 implies that g ′′ is also
negative in the neighborhood of x = 1; e.g. in the open interval ( 12 , 1 21 ). As we learned in
Section 2.2, this implies that the function g is concave on that open interval. So close to
the point (1, 5), the graph of the function g must have roughly the shape that is shown in
the left panel of Figure 7 below:

y y

5 g
g 4

1 x 2 x

FIGURE 7 The graph of the function g in the neighborhood of x = 1 and x = 2

One look at this graph should suffice to convince you, that the function g has a (local)
maximum at x = 1.
Similarly, since g ′′ (2) = 6 > 0, there is an open interval around x = 2 where the second
derivative is positive. But then, the function g must be convex on that interval. The right
panel in Figure 7 reveals that g has a (local) minimum at x = 2.
After this example, our next result should not come as a surprise.

THE SECOND-DERIVATIVE TEST

Let f be a function defined on an open interval I.

If f ′ (c) = 0, i.e. if x = c is a stationary point, then f has
(1) a minimum at x = c if f ′′ (c) > 0
(2) a maximum at x = c if f ′′ (c) < 0
(3) a minimum, a maximum or an inflection point at x = c if f ′′ (c) = 0.

EXAMPLE 6 In Exercise 4, we considered the function p defined by

p(x) = x4 .

We can see right away that p(0) = 0. Moreover, since x4 is a square (i.e. the square of
x2 ), it holds that for all x 6= 0, we have p(x) > 0. So without any differentiation, we can
Properties of a function 95

conclude that the function p has a global minimum at x = 0! Now let’s see whether our
’Second-Derivative Test’ delivers the same conclusion here. Since p′ (x) = 4x3 , the function
obviously has a unique stationary point at x = 0. But with p′′ (x) = 12x2 , we have p′′ (0) = 0.
In terms of our test, we run into scenario (3): so the second derivative does not allow us to
determine the nature of the stationary point here. Further investigation (in particular: the
sign pattern of the first derivative) would be necessary.

EXERCISE 7 Use the ’Second-Derivative Test’ to find and characterize the extreme value(s)
of the functions defined in Exercise 3.

3 FINDING THE EXTREME VALUES OF A FUNCTION

In the previous Section 2, we strictly confined ourselves to functions defined on an open
interval. We will now drop this restriction. It turns out that in this more general situation,
a new category of extreme values comes into play; however, the sign diagram of the first
derivative is still indispensable to determine the extreme values.

EXAMPLE 7 Consider the root function r defined by

√
r(x) = x (x ≥ 0).

At the start of the current Chapter 5, we have already seen that this function is defined on
the interval [0, ∞). Since this interval is not open, the theorems discussed in Section 2 are
not directly applicable. Let us nonetheless have a look at the derivative of r, defined by

1
r′ (x) = √ (x > 0).
2 x

Clearly, this derivative is not defined at x = 0 (which we denote by a × above the number
0), and positive for all x > 0, so its sign diagram looks as follows:

× + + + + +
r′
0
ր ր
r
0

We conclude that the function r is strictly increasing on the interval (0, ∞). As a conse-
quence, r must have a global minimum at the endpoint x = 0 of its domain.
Of course, for this standard secondary school function, the preceding analysis contains an
element of ’overkill’: we don’t really need a sign diagram to find this function’s extreme
value. After all, r(0) = 0, and for all x > 0 we have

√
r(x) = x > 0,

which immediately proves that r has a global minimum at x = 0.

 96 Properties of a function

Despite the ’overkill’ element, Example 7 does illustrate a fundamental point: it shows that
an endpoint of the domain of a function may correspond
with an extreme value of that function

Such a type of extreme value could not occur in Section 2 because, back then, those endpoints
were not included in the domain. Bringing together everything we have seen in our examples
so far, it turns out that there are really three types of extreme values. Let’s list the bits and
pieces.

THREE TYPES OF EXTREME VALUES

If a function f has an extreme value at c, then

(1) c is a stationary point of f , that is: f ′ (c) = 0, OR
(2) c is a point where f is not differentiable, OR
(3) c is an endpoint of the domain of the function.

The function g from Example 2 illustrated type (1). The function a from Example 3 illustra-
ted type (2). The root function r from Example 7 above illustrates the new type (3): r has
an extreme value at x = 0, which is an endpoint of the function’s domain [0, ∞). You may
also note that, in this example, r happens to be not differentiable at x = 0, as for type (2).

EXERCISE 8 Find the extreme value(s) of the functions defined in Exercise 3 if their domain
is the interval [0, 1].

In economics and business textbooks, you often have to find the extreme value of some func-
tion, e.g. the minimum of a cost function, or the maximum of a profit function. Typically,
the analysis in such textbooks focuses on finding the stationary points of that function, on
the assumption that one of these stationary points will necessarily constitute the extreme
value of the function. In the language of the previous theorem: it is assumed that the
extreme value has to be of type (1). Admittedly, this approach will work for many, if not
most, problems that are relevant in economics or business. But still, the theorem above and
the discussions preceding it remind us that we have to be careful: the approach described
here is only guaranteed to work if we are looking for an extreme value of a differentiable
function f (so: no type 2) that is defined on an open interval (so: no type 3).
first-order condition In the type (1) scenario, the condition f ′ (x) = 0 is often called the first-order condition for
an extreme value. The points satisfying this first-order condition are of course the stationary
points of the function. In Section 2.1, we have seen that we can establish the character of
these stationary points by examining the sign diagram of the first derivative. Alternatively
second-order condition (and often faster), we can apply a so called so-called second-order condition. Here our
’Second-Derivative Test’ from Section 2.3 comes in: the sign of the second derivative can
be used to determine the character of a stationary point. However, if the second derivative
equals zero, the ’Second-Derivative Test’ from Section 2.3 is useless (remember Example 6!).
In such a case, we have to check explicitly whether the first derivative switches sign at that
stationary point, using a sign diagram as in Section 2.1.
Properties of a function 97

We will end this section with two examples where we solve optimization problems that were
introduced in Chapter 2.

EXAMPLE 8 A CAR DEALER

In Chapter 2’s Example 2, we analyzed the ordering and holding costs of a Toyota dealer
who always orders q cars at a time. The total costs K were given by K = k(q), where
18 000
k(q) = 500q + (q > 0).
q
Of course the Toyota dealer wants to know which amount q will minimize these total costs.
In order to find this optimal order size, we can use the derivative which you (should have)
found in Chapter 4’s Exercise 2:
18 000
k ′ (q) = 500 − (q > 0).
q2

The next step is, of course, to determine the zero(s) of k ′ , i.e. the stationary point(s) of k.
Remembering that q > 0, we find:
18 000 18 000
k ′ (q) = 0 ⇐⇒ 500 − = 0 ⇐⇒ 500 = ⇐⇒ q 2 = 36 ⇐⇒ q = 6.
q2 q2
So, on its domain, the function k has just one stationary point. This easily leads to the
following sign diagram:

× × − − − − 0 + + + +
k′
0 6
× × ց ր
k
0 6

As you can see, the car dealer’s costs are minimal if q = 6. In other words: he should
order 6 cars at a time. With annual sales of 72 cars, this implies that he will have to place
such an order 12 times per year, i.e. once a month. The minimal costs equal k(6) = 6000
euro. We could have come to the same conclusion using the ’Second-Derivative Test’: you
can easily verify that, at the stationary point q = 6, the second derivative k ′′ is positive,
implying a minimum location. This approach would have saved us the effort of drawing the
sign diagram for the first derivative.

EXAMPLE 9 A CONSTRUCTION PROBLEM

In Chapter 2’s Example 5, we analyzed the construction of a triangular gutter with maximal
capacity. This capacity was given by C = c(x), where
q
c(x) = 50x 81 − 14 x2 (0 < x ≤ 18).

In order to find the value of x that maximizes the capacity, we can use the derivative which
was determined in Chapter 4’s Example 8:

25x2
q
c′ (x) = 50 81 − 14 x2 − q .
2 81 − 14 x2
98 Properties of a function

The stationary point(s) of c can be found as follows:

25x2
q
c′ (x) = 0 ⇐⇒ 50 81 − 41 x2 − q =0
2 81 − 41 x2

25x2
q
⇐⇒ 50 81 − 41 x2 = q
2 81 − 14 x2

⇐⇒ 100(81 − 41 x2 ) = 25x2 ⇐⇒ 324 − x2 = x2

√
⇐⇒ 2x2 = 324 ⇐⇒ x = 162 = 12.73.

You can easily check that we obtain the following sign diagram:

+ + + + 0 − − − −
c′
0 12.73
ր ց
c
0 12.73

This sign diagram implies that we obtain a gutter with maximal capacity if we choose the
base of the triangle equal to 12.73. Like in Example 8, the ’Second-Derivative Test’ offers
an alternative (and maybe faster) approach to the sign diagram analysis above. Note that
the maximal capacity equals c(12.73) = 4050.

EXERCISE 9 The average costs AC of a producer of meat-replacing products (on the basis
of tofu) depend on the weekly production q, so AC = k(q). Average costs
are measured in thousands of euro per unit of production, and production is
measured in thousands of kilograms per week. The average cost function k is
given by
k(q) = 2q 3 − 12q 2 + 100 (q ≥ 0).

Note that, in the terminology of Chapter 2’s Example 3, k is a polynomial

function of degree 3: such functions are quite flexible and therefore often used
by economists. Determine the weekly production for which these average costs
are minimal. Also sketch the graph of this average cost function. How would
you describe its shape?

EXERCISE 10 Use the tools discussed in this chapter to sketch the graph of the function h
defined by
h(x) = x ex .
Properties of a function 99

END-OF-CHAPTER EXERCISES

EXERCISE 1 Find the intervals where the following curves are strictly increasing.
x2
(a) y = (b) y = ex − e3x
e2x

3
(c) y = ln ex + e−x (d) y = x − 2 ln(x2 + x).

EXERCISE 2 Consider the function f on the interval [0, ∞) defined by

2x2
f (x) = .
x4+1

(a) Show that the derivative of the function f is given by

4x(1 + x2 )(1 + x)(1 − x))

f ′ (x) = .
(x4 + 1)2

(b) Find the maximum of the function f .

EXERCISE 3 In an economic model, the proportion of families whose income is no more

than x, and who have a home computer, is given by


p(x) = a + k 1 − e−cx .

Here a, k and c are positive constants.

Determine p′ (x) and p′′ (x). Does p have a maximum? Sketch the graph of p.

EXERCISE 4 Find the values of x that maximize/minimize the function given by the fol-
lowing equation:
(a) y = ex + e−2x
(b) y = 9 − (x − a)2 − 2(x − b)2
(c) y = ln(x) − 5x (x > 0).

EXERCISE 5 Find numbers a and b such that the graph of the function f defined by

f (x) = ax3 + bx2

passes through the point (−1, 1) and has an inflection point at x = 12 .

EXERCISE 6 A company produces q units of its only product at cost C(q) to get revenue
R(q) and make a profit π(q). There is a maximum quantity q̄ that can be
produced. Let R(q) = 80q and C(q) = q 2 + 10q + 900. The firm can produce
at most 50 units.
(a) Draw the graphs of R and C in the same coordinate system.
(b) Answer computationally and graphically: how many units must be pro-
duced to make a profit?
(c) Answer computationally and graphically: how many units must be pro-
duced to maximize profit?
100 Properties of a function

EXERCISE 7 Let f be the function on the interval (−1, 2] given by

f (x) = ln(x + 1) − x + 21 x2 − 31 x3 .
x2 − x3
(a) Prove that f ′ (x) = .
2(x + 1)
(b) Find possible extreme points and inflection points.
(c) Sketch the graph of the function f .

AN INSTALLATION PROBLEM

In a number of end-of-chapter exercises of Chapter 2, we studied the installation costs of

a water supply pipe running from a residence to a country cottage. We found that, if x
(rather than ℓ) represents the length of the pipe alongside the asphalted road, these costs
amount to K = k(x), where
p
k(x) = 40x + 80 x2 − 400x + 80 000 (0 ≤ x ≤ 200).

In the end-of-chapter exercises of last week’s Chapter 4, we found that the derivative of this
cost function is given by
80(x − 200)
k ′ (x) = 40 + √ ,
x2 − 400x + 80 000
and that its stationary point(s) can be found by solving the equation
3x2 − 1200x + 80 000 = 0.

EXERCISE 8 The function k has

(a) one stationary point, one global minimum and one global maximum
(b) one stationary point, one global minimum and no maxima
(c) two stationary points, one minimum and one maximum
(d) one stationary point, and two minima.

EXERCISE 9 The function k

(a) is first convex and then concave
(b) is convex, first decreasing, then increasing
(c) is convex and decreasing
(d) is concave and increasing.

BOUNDED GROWTH

The development over time of the sales S of a certain product can be described by the
function
s(t) = 32 − 16 e−3t (t ≥ 0).

EXERCISE 10 In the long run (that is: for large values of t) sales are about
(a) 16
(b) 80
(c) 32
(d) None of the options (a), (b) or (c) is correct.
Properties of a function 101

EXERCISE 11 The function s is

(a) convex and strictly increasing
(b) first convex and then concave
(c) first concave and then convex
(d) concave and strictly increasing.

EXERCISE 12 Sales are

(a) minimal for no value of t and maximal for no value of t
(b) minimal for one value of t and maximal for no value of t
(c) minimal for one value of t and maximal for one value of t
(d) minimal for no value of t and maximal for one value of t.

THE SPREADING OF INFORMATION

In last week’s end-of-chapter exercises 10 to 12, you have been introduced to the phenomenon
of logistic growth. A nice example of logistic growth is provided by the process of spreading
information within a group. Imagine a group of 30 people. Initially, only one of them has
a certain piece of information (e.g.: ’Julie is pregnant’). Once a day, anyone who ’knows’
can share this information with one other group member, but without knowing whether
that person already knows it. In the beginning, when only few people have the information,
those who ’know’ will mostly run into people who ’don’t know’. Consequently, the number
of people who ’know’ will increase, and ever more quickly. Later on, as more people ’know’,
more and more of their encounters will involve group members who also ’know’; as a result,
the increase in the number of ’knowers’ will start to stagnate. It can be shown that this
logic is captured by the logistic function n(t), where
30
n(t) = (t ≥ 0)
1 + e−(t−3.37)
denotes the number of people having the information after t days. You can easily verify
that, at t = 0, we start with n(0) = 1.

EXERCISE 13 For any number N between 1 and 30, how many days t does it take to spread
the information among that number of group members?
30 − N
(a) t = 3.37 − ln
N
N − 30
(b) t = 3.37 − ln
30
N − 30
(c) t = 3.37 − ln
N
30 − N
(d) t = 3.37 − ln .
30

The first and second derivatives of the function n are given by

30 e−(t−3.37)
n′ (t) =  2
1 + e−(t−3.37)
30 e−(t−3.37) h i
and n′′ (t) =  3 e
−(t−3.37)
−1 .
1 + e−(t−3.37)
102 Properties of a function

EXERCISE 14 The function n

(a) is increasing and convex
(b) is concave
(c) is first concave and then convex
(d) None of the options at (a), (b) or (c) is correct.

A GREEN BOX

A storage box should have a volume of 72 cm3 . The dimensions of the objects to be stored
in the box imply, that its length ℓ must be twice its width w. The box will be decorated with
metallic green paint. Since this paint is quite expensive, we want to choose the dimensions
of the box in such a way, that its surface area is minimized.

w
ℓ

EXERCISE 15 Which of the equations below correctly expresses the height h of the box in
terms of its length ℓ?
144
(a) h =
ℓ2
36
(b) h =
ℓ
72
(c) h = 2
ℓ
18
(d) h = .
ℓ

EXERCISE 16 Which of the equations below correctly expresses the surface area A of the
box in terms of its length ℓ?
144
(a) A = ℓ2 +
ℓ
432
(b) A = 12 ℓ2 +
ℓ
108
(c) A = ℓ2 +
ℓ
432
(d) A = ℓ2 + .
ℓ

EXERCISE 17 According to the previous exercise, the area A is a function of the length ℓ
of the box: A = a(ℓ). On the interval (6, 8), this function a is
(a) convex and decreasing
(b) convex and increasing
(c) concave and decreasing
(d) concave and increasing.
Properties of a function 103

EXERCISE 18 If the optimal (i.e. cost-minimizing) dimensions of the box are denoted by
ℓopt , wopt and hopt , then it holds that . . .
(a) ℓopt = hopt
(b) wopt > hopt
(c) wopt < hopt < ℓopt
(d) wopt = hopt .
104 Properties of a function
Functions of two variables 105

6 FUNCTIONS OF TWO VARIABLES

In the preceding chapters, we have limited ourselves to functions of one variable. Such func-
tions always describe a relationship between two economic variables: they relate a dependent
(or ’output’) variable to a single independent (or ’input’) variable. As a simple economic
example, think about posting an inland letter: the cost of the necessary stamps (the depen-
dent variable) is completely determined by a single factor, i.e. the weight of the letter (the
independent variable).

However, most economic or business phenomena that might interest us are more complex.
To illustrate, let’s reconsider our Example 2 from Chapter 1. This time, imagine that you
deposit an unspecified amount of A (rather than 200) euros in a savings account, which
offers an annual interest rate of r. After one year the amount has grown to K euros, where
 
r
K =A 1+ .
100

We are now dealing with a relation between three economic variables: K, A and r. To
clarify the logic of our expression: if you deposit A = 200 euro at an interest rate of r = 5
(i.e. 5%), you end up with K = 200(1 + 0.05) = 210 euro after one year. Since the value
of K is completely determined by the values of A and r, we refer to K as the dependent
variable, and A and r as the independent variables. In this case, K is a function of the two
variables A and r.

Such functions of two variables are central in this chapter. In Section 1, we will use two
examples to illustrate the basic concept of a function of two variables and its domain, and
formally define the class of linear functions. Section 2 shows how to sketch such functions.
Section 3 continues this discussion by introducing the important concept of a level curve.
In Section 4, we address the crucial notion of a partial derivative. The final Section 5
shows some important applications of these partial derivatives, among others to determine
elasticities. This listing of topics should suggest, that the analysis of functions of two
variables is in many ways a generalization of our previous discussion for functions of one
variable. But as we will see in the current chapter and the next, some truly new issues do
present themselves. Once you thoroughly understand these new issues, further generalization
(to functions of three variables, four etcetera) is straightforward.
 106 Functions of two variables

1 FUNCTIONS OF TWO VARIABLES: THE BASIC CONCEPTS

We start with an example of a linear function of two variables.

EXAMPLE 1 THE DEMAND FOR PORK

Imagine a hungry student, fed up with those Domino pizzas. Instead, he decides to cook a
proper meal, with pork as its major ingredient. At first glance, you may suspect that his
demand for pork only depends on its price. A moment of reflection will suggest that, realis-
tically, other factors may be relevant here as well. E.g., once he is inside the supermarket,
our student may notice that chicken is much cheaper than pork. This could induce him
to buy less pork than planned, and to go for chicken instead. If so, then his demand for
pork does not only depend on its own price, but also on the price of chicken. But then, the
demand for pork may also depend on e.g. the price of beef, or even the price of tofu. To keep
the analysis manageable, we will from now on assume that the student’s quantity demanded
for pork in kilograms (say q), depends only on the price of pork in euro per kilogram (say
p), and on the price of chicken in euro per kilogram (say r). For each specific pair of price
values, one specific quantity of pork will be demanded; consequently, we can view q as a
function, say d, of the two variables p and r:

q = d(p, r).

(in)dependent variable Obviously, q is the dependent variable here, while p and r are the independent variables.

Usually, the relationships between economic variables are described by means of equations.
As an example of such an equation for the case at hand, we take

q = 8 − 0.2p + 0.1r.

In order to indicate that q represents the dependent variable and p and r the independent
ones, we have isolated q in the left-hand side of the equation above, just as we would do
if there was only a single independent variable. The right-hand side of this equation can
be considered as an arithmetic rule that must be applied to the values of the independent
variables p and r, in order to determine the implied value of the dependent variable q. This
rule defines the demand function d for the variables p and r:

d(p, r) = 8 − 0.2p + 0.1r.

Of course, this function only makes sense for nonnegative values for p and r, and then only
for those combinations of values for p and r that lead to nonnegative values for q.

Our next example discusses an important nonlinear function of two variables.

EXAMPLE 2 THE COBB-DOUGLAS PRODUCTION FUNCTION

A production function describes the relationship between the inputs and the output of a
production process in mathematical terms. Inputs are production factors such as labour,
capital goods and raw materials. Output consists of the goods and services that are the
result of the production process. Schematically:
 Functions of two variables 107

inputs → production process → output

Think e.g. about a car plant, where the output is the number of physical units, say Q,
produced over a certain period (e.g. a year). Many different kinds of capital goods are
needed to produce cars: the building which houses the plant, steel presses, robots, tools
etcetera. Likewise, various kinds of labour are needed: process operators, engineers, me-
chanics etcetera. Let’s ignore raw materials, and assume that it is possible to combine all
these different types of capital and labour into an aggregate measure K for the annual input
of capital services, and an aggregate measure L for the annual input of labour services.
Under the stated assumptions, Q will be a function, say p, of these two variables K and L:

Q = p(K, L).

Cobb-Douglas pro-
A special kind of production function with a proven track record of describing real-life
duction function
production processes is the Cobb-Douglas production function, named after its ’inventors’
C.W. Cobb and P.H. Douglas. Such a function is determined by the equation

Q = γK α L1−α ,

where 0 < α < 1 and γ > 0. If we take the values α = 0.5 and γ = 5 as a special case for
the constants, then the equation for Q, K and L becomes

Q = 5K 0.5 L0.5 .

Using the previously introduced symbol p to denote the production function that is deter-
mined by this equation, we have:
√
p(K, L) = 5 KL (K ≥ 0, L ≥ 0).

Of course, this function only makes sense for nonnegative values of the inputs K and L:
after all, there is no such thing as a negative amount of input.

After these two examples, let’s continue our discussion in more general terms. We start by
briefly reviewing how we arrived at a function of two variables in Example 1. This example
describes a relationship between three economic variables: one dependent variable (q) and
two independent variables (p and r). Mathematically, this relationship can be described
by an equation, which expresses the dependent variable (in the left-hand side) in terms of
the independent variables (in the right-hand side). In our example, we assumed that the
equation reads as q = 8 − 0.2p + 0.1r. We already stated that ’The right-hand side of this
equation can be considered as an arithmetic rule that must be applied to the values of the
independent variables p and r, in order to determine the implied value of the dependent
variable q’. That rule defines our ’function of two variables’. It will be clear that the
preceding text can easily be paraphrased in terms of Example 2. Formalizing:
 108 Functions of two variables

A FUNCTION OF TWO VARIABLES

A function f of two variables x and y is a rule or formula that can be used to calculate,
for each pair of numbers (x, y), a unique number f (x, y).
domain Of course this only makes sense for those pairs of numbers (x, y) for which it is possible and
meaningful to evaluate the number f (x, y). The set of all these pairs is called the domain
of the function f .

EXERCISE 1 Find f (4, 9) if f is the function defined by

(a) f (x, y) = x3 + 2xy + y 3

2/3
(b) f (x, y) = x3/2 + y 3/2 (x ≥ 0, y ≥ 0).

Let’s have a closer look at the domain of a function, defined above as ’those pairs of numbers
(x, y) for which it is possible and meaningful to evaluate the number f (x, y)’.
√
EXAMPLE 3 Consider the function f defined by f (x, y) = x + y. Without further in-
formation about the nature of the independent variables x and y, and the implied dependent
variable, we cannot judge when it is meaningful to evaluate f (x, y). But one thing is clear:
since the square root is not defined for negative numbers, it is only possible to evaluate
f (x, y) for those pairs (x, y) for which x + y ≥ 0. So in this case, the domain is the set
{(x, y) | x + y ≥ 0}. This notation can be verbalized as follows: ’the domain of the function
f consists of those pairs of values for x and y, for which it holds that their sum is at least
zero’. Observe e.g. that f is defined at (2, 2) and that
√ √
f (2, 2) = 2 + 2 = 4 = 2.

EXERCISE 2 Find the domain of the function (of x and y) given by the equation
√ √ 1
(a) z = ln xy (b) z = x + y (c) z = 4 .
x + y4

Finally, linking up with Examples 1 and 2, we provide a formal definition of linear (as
opposed to nonlinear) functions of two variables.

A LINEAR FUNCTION
A function f of two variables x and y is called a linear function if it has the form

f (x, y) = ax + by + c.

Here, a, b and c are constants; a is called the coefficient of x, b the coefficient of y, and c
the constant term. Analogous to linear functions of a single variable, a and b are sometimes
referred to as slope coefficients (or just slopes for short), while c is often called the intercept.

We trust you can easily verify that Example 1’s demand function d satisfies the template
above and is therefore linear; by contrast, Example 2’s production function p does not satisfy
this template and is therefore nonlinear.
Functions of two variables 109

EXERCISE 3 Determine which of the functions below is a linear function of x and y.

(a) f (x, y) = 3 + x
(b) f (x, y) = 2x + 3xy + y
(c) f (x, y) = ln(10 − x − y).

2 VISUALIZING A FUNCTION OF TWO VARIABLES

In this section we discuss how you can visualize a function f of two variables. Remember
how, in Chapter 1, we represented a function of one variable as a curve in a two-dimensional
coordinate system. We will generalize this approach to functions of two variables.

Consider a function f of two variables, say z = f (x, y). In order to represent f geometrically,
we will first show how the combination of a specific pair (a, b) and the implied value f (a, b)
can be represented as a point in a three-dimensional coordinate system. Next, we generalize
our approach to all pairs (a, b) in the domain of f .
Since three variables (x, y and z) are involved now, we obviously need a system with three
coordinate axes; along two of them we locate the (values of the) independent variables x
and y, and along the third one we locate the (values of the) dependent variable z. The
coordinate axes of this so-called (x, y, z)-coordinate system are perpendicular to each other.
Unfortunately, on a piece of paper or a computer screen, this three-dimensional system has
to be represented in two dimensions. To do so, we draw the positive x-axis pointing forwards,
the positive y-axis pointing to the right and the positive z-axis pointing upwards. In fact,
we will in principle always show the dependent variable along the axis that points upwards
(unless there is a specific reason to deviate from this).

f (a, b)

b y

a
x

FIGURE 1 The point corresponding with the triple a, b, f (a, b)


Figure 1 above shows how the triple a, b, f (a, b) can be represented in our three-dimensional
coordinate system as the point with first coordinate a, second coordinate b and third coor-
dinate f (a, b).


If we now represent each triple a, b, f (a, b) with (a, b) in the domain of the function f as
a point in our (x, y, z)-coordinate system, then we obtain the so-called graph of f . Figure 2


sketches an arbitrary example which contains our original triple a, b, f (a, b) .
110 Functions of two variables

f (a, b)

b y

a (a, b)

x
FIGURE 2 The graph of a function f

Summarizing:

THE GRAPH OF A FUNCTION

Let f be a function of two variables x and y. The graph of f consists of all points a, b, f (a, b)
in an (x, y, z)-coordinate system, where (a, b) belongs to the domain of the function f .
The graph of f is also called the surface z = f (x, y).

EXERCISE 4 Sketch the graph of the linear function f defined by

f (x, y) = 10 − x − y (x ≥ 0, y ≥ 0, 0 ≤ x + y ≤ 10).

As your answer to Exercise 4 should show, the graph of a linear function is always a (part
of a) plane.

EXAMPLE 4 In Example 2 we considered the Cobb-Douglass production function p

defined by
√
p(K, L) = 5 KL,

where K ≥ 0 and L ≥ 0. The graph of this function is represented by the equation

√
Q = 5 KL.

Since p(10, 10) = 50, the point (10, 10, 50) lies on the graph of this function. And with
p(8, 2) = 20, so does the point (8, 2, 20). But since p(2, 2) = 10, the point (2, 2, 0) is not a
point on the graph of this function.

EXERCISE 5 Which of the points (1, 1, 2) and (2, 3, 10) lie on the graph of the function f
defined by f (x, y) = x2 + y 2 ?
Functions of two variables 111

In general it is quite complicated to sketch the graph of a nonlinear function of two variables.
Practically speaking, it is almost inevitable to use some dedicated drawing software to do
the job. Below you find the graph of the Cobb-Douglas production function p from Example
4, generated with the WolframAlpha software. Note that the lower-left point of the graph
is the origin, with output measured upwards as before. This is also where both the capital
and labour axes start; but for presentational purposes, the software has moved the scale of
the labour axis to the right. On our instruction, only input values of at the most 10 are
considered; the red dot in the upper-right corresponds with the point (10, 10, 50).

FIGURE 3 The graph of the Cobb-Douglas function p for K and L both between 0 and 10

Note that our software has traced out some curves along the graph of the function p. These
so called level curves connect points on the graph that have the same value for the dependent
variable. Geometrically, this means that they have the same height above the horizontal
plane at level zero. In the next section, we will have a closer look at these curves.

3 LEVEL CURVES FOR FUNCTIONS OF TWO VARIABLES

If you go for a walk in a mountainous area, you will probably make use of a topological map
(or its digital equivalent on your smartphone). Such a map contains curves that connect
points of equal elevation. These so-called contour lines are useful because they illustrate the
up-and-down shape of the landscape on the flat surface of your map. In the left panel of
Figure 4 you find an example of a map containing such contour lines. The idea of connecting
points which share the same level for some quantity is also used in meteorology, where an
isobar is a curve that connects points of equal atmospheric pressure. Every decent weather
forecast comes with a picture similar to the right panel in Figure 4: the light-blue curves
are the isobars.
112 Functions of two variables

FIGURE 4 Contour lines and isobars

The idea of connecting points where some variable of interest takes on a given constant value
can also be used in economics and business. In our first example, the variable of interest
is the output level of some production process, and the points-to-be-connected correspond
with input combinations of capital and labour.

EXAMPLE 5 ISOQUANTS FOR A COBB-DOUGLAS PRODUCTION FUNCTION

Let’s continue the analysis of our special case Cobb-Douglas production function from Ex-
ample 2:
√
p(K, L) = 5 KL (K ≥ 0, L ≥ 0).

In Example 4, we have already seen that the input combination (K, L) = (8, 2) will yield an
output of Q = 20. But then, you should see in a split second that the input combination
(K, L) = (2, 8) will lead to the same result. This simple observation brings out a crucial
feature of our production function: there are different ways of producing the same output
level! Our technological knowledge allows us to substitute capital and labour: we can produce
those 20 units with eight machines and just two workers, or with eight workers operating
only two machines. For a manager balancing a factory in Germany (where labour is very
expensive) against an offshore facility in Bangladesh (were labour is very cheap), this is
obviously very useful information. But these input pairs are just two examples. The input
pair (K, L) = (4, 4) would do just as well. In the left panel of Figure 5, we have represented
these three input combinations as red dots in a two-dimensional (K, L)-coordinate system.
Now let’s try to find all input combinations (K, L) that yield an output of 20 units. To find
these combinations, we will have to solve the equation p(K, L) = 20:

√ √
p(K, L) = 20 ⇐⇒ 5 KL = 20 ⇐⇒ KL = 4 ⇐⇒ KL = 16 ⇐⇒ L = 16K −1 .

We conclude: all input combinations lying on the curve defined by the equation L = 16K −1
will yield an output of Q = 20. This curve, which can be drawn in a (K, L)-coordinate
system, is called an isoquant (iso = equal, quantitas = quantity). We have done so in the
Functions of two variables 113

left panel of Figure 5: and of course, this curve contains the three input combinations which
we marked before.

L L

10 10

8 8

6 6

4 p(K, L) = 20 4 Q = 30

2 2 Q = 20
Q = 10
2 4 6 8 10 K 2 4 6 8 10 K

FIGURE 5 Some isoquants for the Cobb-Douglas function p

By drawing several distinct isoquants for different output levels, as we have done in the right
panel of Figure 5, you can get a pretty good idea of the properties of our Cobb-Douglas
production function p. To see how the 3-dimensional graph of the function p is related to
our set of level curves, have a closer look at Figure 3. Note that the horizontal plane at
output level zero is in fact our (K, L)-coordinate system. Now remember the curves that
have been drawn along the graph of p. If we drop some of those curves onto that horizontal
plane, you obtain the curves in the right panel of Figure 5. Make sure to understand that
the situation as depicted in the left panel of Figure 4 is completely similar: its upper-left
corner shows a 3-dimensional representation of the actual shape of the landscape, and by
’dropping’ the ’virtual curves in the landscape’ onto a horizontal plane, we get the map with
the contour lines.

EXERCISE 6 For the Cobb-Douglas production function in Example 5, determine the iso-
quants for the output values Q = 10 and Q = 30.

EXAMPLE 6 INDIFFERENCE CURVES

We now present a second example of the use of level curves in economics and business. In
economics, the satisfaction or utility which a consumer derives from consuming goods is
measured by a utility function, say U : so the utility generated by a bundle of goods (x, y) is
given by U (x, y). An indifference curve connects points which provide the same amount of
utility to the consumer. Here, a point represents the quantities of both goods. With utility
being equal, the consumer is said to be indifferent between such points: this explains the
name of the curves.

To illustrate, consider the satisfaction which a consumer derives from a goods bundle (L, R),
where L represents the number of left shoes in his possession, and R the number of right
114 Functions of two variables

shoes. Let’s assume that the utility function looks as follows:

U (L, R) = minimum{L, R},

which means that utility is equal to the lowest of the two numbers L and R. You can easily
verify that the utility of the bundle (2, 2) is equal to 2. But the bundle (2, 3) yields the same
utility! When you think about it, this makes perfect sense. After all, shoes intrinsically
come in pairs. Now imagine that you own two pairs of shoes, and then someone donates a
third right shoe to you. Although you may appreciate that gesture, your satisfaction from
shoes will not increase: since you have only two left shoes, that third right shoe is useless.
The same would be true if that benefactor had instead given you a third left shoe, or even a
fourth: utility would stay put. Consistent with this, our utility function delivers a common
utility of 2 for all bundles of the types (2, 3), (2, 4), . . . and (3, 2), (4, 2), . . .. In Figure 6,
we have connected these bundles in an (L, R)-coordinate system: together, they form the
indifference curve U = 2.

2 U (L, R) = 2

2 4 6 8 L

FIGURE 6 An indifference curve

In Examples 5 and 6 we have considered, for a given function of two variables, the points
where this function takes on a specific value. In the first example, we examined where
the Cobb-Douglas production function p took on the value 20; in the second example, we
examined where the utility function U assumes the value 2.

We now extend these ideas to an arbitrary function of two variables.

LEVEL CURVE

For a function f and a given number c, the curve consisting of all domain points (x, y) for
which it holds that f (x, y) = c is called the level curve of f at level c.

So when an economist looks at the left panel of Figure 5, he sees an isoquant; using math-
ematical terminology, it’s actually the level curve at level 20 of the production function p.
Similarly, the indifference curve in Figure 6 is a level curve of the utility function U , at level
2.
Functions of two variables 115

EXAMPLE 7 For the function f defined by f (x, y) = x2 + y 2 , the level curve f (x, y) = 4
is the set of points (x, y) in an (x, y)-coordinate system that satisfy the equation x2 + y 2 = 4
or
x2 + y 2 = 22 .

This set can be represented by a circle of radius 2, centered at the origin. You can easily check
that the intersections of that circle with the axes occur at the points (2, 0), (0, 2), (−2, 0)
and (0, −2). Make sure you understand the meaning of the circle: all the points on it have
in common that, when we fill those (x, y) pairs into the function f , a value of 4 results!

EXERCISE 7 (a) Determine and sketch the level curve for the function f (x, y) = x2 + y 2
at level 1.
(b) For which values of c is there no level curve f (x, y) = c?

EXERCISE 8 The contour lines in the left panel of Figure 4 can be seen as the level curves of
a function of two variables. Write down the general definition of this function,
and explain the meaning of the two variables.

EXERCISE 9 Remember the linear demand function from Example 1:

d(p, r) = 8 − 0.2p + 0.1r.

Determine the level curves for the function d at the levels 5.5, 6 and 6.5. Then
draw these curves in a (p, r)-coordinate system.

We trust your picture for Exercise 9 shows three parallel lines. This result illustrates a
general property: for linear functions of two variables, the level curves are always lines (or,
more precisely, line segments), and these are parallel to each other.

4 PARTIAL DERIVATIVES

In Chapter 3, for functions of one variable, we introduced the concept of the ’derivative’: the
rate of change of the function at some point. In the current section, we generalize our dis-
cussion to functions of two variables. The basic idea behind the resulting ’partial derivative’
is that we analyze the rate of change of the function w.r.t. one of its independent variables,
while keeping the other independent variable constant. By treating one independent variable
as a constant, we effectively obtain a function of one variable only; the derivative of that
function of one variable is a partial derivative of the original function of two variables. If
that all sounds a bit abstract, then just wait and see: in Section 4.1 below we will show you
how to determine those partial derivatives, and in Section 4.2 we will provide a geometrical
interpretation. Why is all that stuff useful? Well: basically for the same reasons as with
functions of only one variable, as we will show in Section 5, and in Chapter 7 next week.
 116 Functions of two variables

4.1 DETERMINING PARTIAL DERIVATIVES

Consider the function g of two variables x and y, defined by the equation

z = x2 y + y 2 + y.

For this function, as a first step, we will determine the values of its partial derivatives at a
single point in its domain: the point (0, 1). In a second step, we will show how to determine
the partial derivatives of g at any point in its domain, i.e. as functions of the variables x
and y. Remember that, in Chapter 3, we went through similar steps for a function of one
variable.
the first step As mentioned, we take the point (0, 1) in the domain of g as our benchmark. Now we
have already seen that the basic idea behind a ’partial’ derivative is that one of the two
independent variables is held constant. So let’s say we keep y fixed at 1. Under this
assumption, the function g simplifies to

g(x, 1) = x2 + 2.

Of course, this is a function of the single variable x only! Taking its derivative gives 2x. At
x = 0, this derivative is equal to zero, which we denote as follows:

gx′ (0, 1) = 0.

The value 0 is called the partial derivative of g with respect to x at the point (0, 1). ’Partial’,
because we are only interested in the role of one of the variables, here: x, while y = 1. In
Section 4.2, we will use geometrical tools to explain what this value of 0 really means.
Similarly, keeping x fixed at 0, g simplifies to

g(0, y) = y 2 + y.

This is a function of y only, with derivative 2y + 1. At y = 1, this derivative is equal to 3,

which we denote as
gy′ (0, 1) = 3.

The value 3 is the partial derivative of g with respect to y at the point (0, 1).

the second step Of course, our second step is now to obtain the partial derivatives of g at any point in its
domain. In order to find them, we proceed as follows.
Let us, for the moment, again treat y as a constant rather than a variable. But, unlike in
the first step, we do not impose a specific value for y. Still, whatever value we might choose,
once y is constant the equation z = x2 y + y 2 + y effectively defines a function of x only! So
we can find the derivative of this function (obviously with respect to x) by using the rules
familiar from previous chapters. This derivative is denoted gx′ (x, y), and it follows as:

gx′ (x, y) = 2xy + 0 + 0 = 2xy.

↑ ↑ ↑
derivative derivative derivative
of x2 y w.r.t. x of y 2 w.r.t. x of y w.r.t. x
 Functions of two variables 117

The function gx′ is called the partial derivative of g with respect to x. Note that gx′ is a
function of the variables x and y now, rather than a number. Note also, that our previous
result that gx′ (0, 1) = 0 follows as a special case.
Similarly, imagine treating x as a constant rather than a variable. If so, then the equation
z = x2 y + y 2 + y defines a function of y only. Its derivative with respect to y is denoted by
gy′ (x, y), and it follows as:

gy′ (x, y) = x2 + 2y + 1 = x2 + 2y + 1.
↑ ↑ ↑
derivative derivative derivative
of x2 y w.r.t. y of y 2 w.r.t. y of y w.r.t. y

Of course, the function gy′ is called the partial derivative of g with respect to y. Again, it is
a function of x and y; and again, gy′ (0, 1) = 3 follows as a special case.

PARTIAL DERIVATIVES

Let f be a function of x and y.

The derivative of f (x, y) with respect to x, treating y as a constant, is called the partial
derivative of f with respect to x. It is denoted by fx′ (x, y).
The derivative of f (x, y) with respect to y, treating x as a constant, is called the partial
derivative of f with respect to y. It is denoted by fy′ (x, y).

partial derivatives Since we differentiate only once, the functions fx′ and fy′ are called partial derivatives of the
∂f ∂f
of the first order first order. In economic literature, you will also find the expressions for fx′ and for
∂x ∂y
′
fy . The symbol ∂ has a meaning similar to that of d in the case of functions of one variable,
as discussed in Chapter 3. It indicates that we are dealing with a derivative of a function of
more than one variable and that the derivative with respect to exactly one variable is being
referred to, treating the other variable(s) as constant. In the rest of this reader, whenever
we mention a partial derivative of a function, we will implicitly assume that this derivative
actually exists.

EXERCISE 10 Determine the partial derivatives of the functions f defined by

(a) f (x, y) = 2xy − 5x2 + y − 1
(b) f (x, y) = x3 − 5x2 y − 5xy 2 + y 3
√ x
(c) f (x, y) = x + .
y

As we have seen, fx′ and fy′ are themselves functions of x and y, so it is possible to deter-
mine their partial derivatives. The resulting ’partial derivatives of the partial derivatives’
are called the second-order partial derivatives of the original function f . E.g., partially
differentiating fx′ with respect to x, we find the second-order partial derivative of f with
respect to x:
∂2f
′′
fxx .
or
∂x2
Note the different positions of the number 2 in the numerator and the denominator of the
second expression: this is just a notational convention and nothing more. Similarly, if we
118 Functions of two variables

partially differentiate fx′ with respect to y, we find the second-order partial derivative of f
with respect to x and y:
′′ ∂2f
fxy or .
∂y ∂x
Now for the (first-order) partial derivative fy′ . Partially differentiating it with respect to x,
we find the second-order partial derivative of f with respect to y and x:

′′ ∂2f
fyx or .
∂x ∂y

Partially differentiating it with respect to y leads to the second-order partial derivative of f

with respect to y:
′′ ∂2f
fyy or .
∂y 2

Generally, for most functions f which are relevant in an economic context, it holds that

′′ ′′
fxy = fyx .

In other words, regardless of the order in which the partial differentiations with respect to
x and y are performed, the result will be the same.

Like first-order partial derivatives, second-order partial derivatives play an important role
in determining the extreme value(s) of a function of two variables. This will be the main
topic of next week’s Chapter 7; you probably anticipate that this boils down to generalizing
the analysis in Chapter 5 from one to two variables.

EXAMPLE 8 Let’s determine the second-order partial derivatives of the function g defined
by
z = x2 y + y 2 + y.

We have already established that gx′ (x, y) = 2xy and that gy′ (x, y) = x2 + 2y + 1. We can
now determine the partial derivatives of gx′ with respect to x and y in turn, to obtain:

′′ ′′
gxx (x, y) = 2y and gxy (x, y) = 2x.

Next, we determine the partial derivatives of gy′ with respect to x and y, leading to:

′′ ′′
gyx (x, y) = 2x and gyy (x, y) = 2.

′′ ′′
Notice that indeed gxy = gyx , as it should be.

EXERCISE 11 Determine all the second-order partial derivatives of the Cobb-Douglas pro-
duction function given by

p(K, L) = 5K 0.5 L0.5 (K > 0, L > 0).

Functions of two variables 119

4.2 A GEOMETRICAL INTERPRETATION OF PARTIAL DERIVATIVES

So far, we have basically done little more than telling you, how to determine the partial
derivatives of a function of two variables. But what do these partial derivatives really
mean? To get your intuitive understanding right, it is again useful to ’make a picture’. Let’s
continue with our function g defined by

g(x, y) = x2 y + y 2 + y.

Consider again the point (0, 1) in its domain. You can easily check that g(0, 1) = 2. So in
the usual (x, y, z)-coordinate system, the graph of g would contain the point (0, 1, 2). In
the upcoming analysis, this point will again serve as our ’benchmark’. In particular, we will
discuss a geometric interpretation of the partial derivatives gx′ (0, 1) and gy′ (0, 1). Remember
that the values of these partial derivatives were 0 and 3, respectively.
In our ’first step’ at the start of Section 4.1, we observed that gx′ (0, 1) is the derivative of
the function g(x, 1) = x2 + 2 at x = 0. This function is obtained from the original function
g, by keeping y constant at the value 1. In Figure 7, the green square visualizes all points
with y = 1. The graph of our function g(x, 1) is the intersection of that green square with
the graph of the original function g. In other words: the graph of g(x, 1) is a vertical ’slice’
of the graph of the original function g, for y = 1. The red parabola, contained in the green
square, represents that slice. We hope the bottom line is clear: our interest is in that red
parabola. Along that parabola, y is constant at 1, so we are thinking in the x-direction.

z = g(x, 1)

1 y=1

x
FIGURE 7 The graph of g(x, 1)

Since gx′ (0, 1) is the derivative of a function of one variable, it has the normal geometric
interpretation: it is the slope of the tangent to the point (0, 1, 2). Because gx′ (0, 1) = 0, this
tangent has a slope of 0; the dashed line in Figure 7 visualizes this.
We conclude:
the partial derivative of g with respect to x at some point corresponds with
the slope of the tangent to the graph of g at that point in the x-direction.
 120 Functions of two variables

Similarly, gy′ (0, 1) = 3 is the slope of the tangent to the graph of g at the point (0, 1, 2) in
the y-direction. It might be helpful to sketch the corresponding analogue to Figure 7!

Generalizing our discussion above, we have:

TANGENTS IN THE x- AND y-DIRECTION

Let f be a function of x and y, and let (a, b) be a point in its domain.


The tangent to the graph of f at the point a, b, f (a, b) which is contained in the vertical
plane y = b is called the tangent in the x-direction. The number fx′ (a, b) represents the
slope of this tangent.


The tangent to the graph of f at the point a, b, f (a, b) which is contained in the vertical
plane x = a is called the tangent in the y-direction. The number fy′ (a, b) equals the slope of
this tangent.

tangent plane Note that there exists a unique plane containing both the tangent in the x-direction and
the tangent in the y-direction. This plane is called the tangent plane to the graph of the


function f at the point a, b, f (a, b) . We will briefly get back to this concept in next week’s
Chapter 7.

5 THE USE OF PARTIAL DERIVATIVES IN ECONOMICS

So far, we have addressed the determination and the geometric interpretation of the partial
derivatives of a function of two variables. We will now demonstrate why those partial
derivatives are so useful. Our discussion will be continued in next week’s Chapter 7.

Previous chapters have discussed at length the various uses of the derivative for a function
of one variable. In the remaining part of this reader, we will replicate most of this, but now
in terms of the partial derivatives of functions of two variables. E.g., generalizing Chapter
3’s Section 3, the current Section 5 demonstrates how the partial derivatives can be used
to easily estimate the increment of the dependent variable resulting from a small change in
one of the independent variables (we will continue this topic next week). Also, generalizing
Chapter 3’s Section 4, we will see how these partial derivatives can be used as building
blocks for the definition of elasticities.

Most important of all, Chapter 5 has shown that derivatives play an important role in de-
termining the extreme values of functions of one variable. In particular, when the derivative
equals zero, this will often indicate that the function has an extreme value at that point.
Not surprisingly, when you want to find the extreme value(s) of a function of two variables,
a very similar approach can be used in terms of its partial derivatives. This crucial topic
will be addressed next week.

So we start by generalizing the topic from Chapter 3’s Section 3: marginality. We trust
you remember that, for a function f of one variable, the derivative tells us the approximate
effect which small changes in the independent variable, say x, will have on the value of the
Functions of two variables 121

dependent variable, say y. More specifically, by using the ∆-formula we found that

∆y ≈ f ′ (a) ∆x.

Here, ∆y indicates the change in the variable y if x is changed from a to a + ∆x, and ∆x is
supposed to be small. To illustrate the idea: if we start from some point where the derivative
of our function equals 2, and x changes by 0.1, then as a result y will change by roughly
2 · 0.1 = 0.2.
An analogous result holds for the partial derivatives of functions of two variables. This is
illustrated in the following example.

EXAMPLE 9 THE MARGINAL PRODUCT OF CAPITAL

Let’s examine our ’special case’ Cobb-Douglas production function p from previous examples
more closely:
√
p(K, L) = 5K 0.5 L0.5 = 5 KL (K > 0, L > 0).

Treating L as a constant, it should be easy to establish (and you should have done this for
Exercise 11!) that the partial derivative of this function with respect to K is given by
r
0.5 L
p′K (K, L) = 2.5K −0.5
L = 2.5 .
K

Now imagine that we are currently using 4 units of capital and 16 units of labor, and we
are thinking about increasing the number of machines. Applying the principles discussed in
Section 4.1, p′K (4, 16) is the derivative of p(K, 16) at K = 4. Since p(K, 16) is a function of
one variable, the ∆-formula from Chapter 3 should hold! So we have:

∆Q ≈ p′K (4, 16) ∆K.

And since p′K (4, 16) = 5 (check!), we end up with ∆Q ≈ 5 ∆K. So if we increase the capital
input by 1 unit, i.e. from 4 to 5, while continuing to use 16 units of labour, then output will
increase by approximately 5 units. By extension: 2 extra units of capital, and we expect an
output rise by roughly 5 · 2 = 10. Etcetera: surely, you get the idea.
This analysis implies that the partial derivative p′K measures the (approximate) change in
output that is caused by a one unit change in the capital input, while keeping the labour
input constant. In economics, this change in output is called the marginal product of capital.
Of course, an entirely symmetrical argument can be developed for the marginal product of
labour.
Next week in Chapter 7, we will see how we can measure (or at least: approximate) the
changes in the output volume that occur if the input values of labour and capital change
simultaneously.

To generalize our findings, we now consider an arbitrary function f .

122 Functions of two variables

CHANGING ONE OF THE INDEPENDENT VARIABLES

If the relation between the variables x, y and z can be described by means of the equation
z = f (x, y), where f is a nonlinear function of two variables and (a, b) is a point in its
domain, then
∆z ≈ fx′ (a, b) ∆x.
Here ∆z indicates the change in the variable z if x is changed from a to a + ∆x (with ∆x
small), while y = b.
Of course, the partial derivative fy′ (a, b) has a similar interpretation.

In words:
the partial derivative translates a small change in the corresponding independent
variable into an approximation for the resulting change in the dependent variable.
Our result above generalizes the ∆-formula from Chapter 3. That formula was about ap-
proximating the change of a function of one variable. Back then, we also noted that for linear
functions, the approximation is actually exact. This result carries over to linear functions of
two variables, as Example 10 below will show.

EXAMPLE 10 Remember the linear demand function d from Example 1:

d(p, r) = 8 − 0.2p + 0.1r.
This function expresses how the demand for pork q depends on the pork price p and the
chicken price r. The partial derivatives of this function should be easy to find:
d′p (p, r) = −0.2 and d′r (p, r) = 0.1.
Note that both partial derivatives are constant: this is true for all linear functions. The
number d′p (p, r) = −0.2 indicates that an increase in the pork price by ∆p euro causes a
fall in the demand for pork by exactly 0.2 · ∆p kilogram, assuming that the chicken price
remains unchanged. After all,
d(p + ∆p, r) = 8 − 0.2(p + ∆p) + 0.1r = 8 − 0.2p + 0.1r − 0.2 ∆p = d(p, r) − 0.2 ∆p,
which implies that the change ∆q in the demand for pork satisfies
∆q = d(p + ∆p, r) − d(p, r) = −0.2 ∆p.
Similarly, d′r (p, r) = 0.1 implies that pork demand will increase by exactly 0.1 · ∆r kilogram
if chicken becomes ∆r euro more expensive, while the price of pork stays the same.

We summarize our findings in the following table, which generalizes the similar Table 2 from
Chapter 3:
a linear function ℓ a nonlinear function f

x : a → a + ∆x
∆z = ℓ′x (a, b) ∆x ∆z ≈ fx′ (a, b) ∆x
y:y=b

x:x=a
∆z = ℓ′y (a, b) ∆y ∆z ≈ fy′ (a, b) ∆y
y : b → b + ∆y

TABLE 1 Functions of two variables: the response to small changes in x or y

Functions of two variables 123

EXERCISE 12 Continuing Example 9, determine the partial derivative of p with respect to

L at (K, L) = (4, 16), and give an economic interpretation of the result.

So far, we have seen that the marginal products of capital and labour tell us how much
output will (approximately) rise if we use one more unit of capital or labour, while keeping
the other input constant. Note that the changes in an input, and the resulting change in
output, are both measured in absolute terms here. But this may not always be the most lucid
measure. In many cases, it will be more informative to measure the changes in inputs and
the implied change in output in relative terms, i.e. in terms of percentage changes. Just like
in Section 4 of Chapter 3, this is where the elasticity concept comes in. Below, we provide
the formal definition of this concept for the case of a function of two variables. We trust
you will recognize that this definition is an obvious generalization of our previous definition
for functions of one variable. Note in particular, that the definition of the elasticity with
respect to a specific variable involves the partial derivative with respect to that variable.

ELASTICITY

If f is a positive function of x and y, then the elasticity Elx (x, y) of f (x > 0 and y > 0)
with respect to x is defined as
x
Elx (x, y) = fx′ (x, y) ·
f (x, y)
and the elasticity Ely (x, y) of f with respect to y as
y
Ely (x, y) = fy′ (x, y) · .
f (x, y)

For the elasticity of Example 9’s production function p with respect to capital, we have
r
′ K L K
ElK (K, L) = pK (K, L) · = 2.5 · √ = · · · (check!) · · · = 0.5.
p(K, L) K 5 KL
The interpretation of this number is as follows: an increase of the capital input by one
percent, while keeping the labour input constant, results in an output rise of around 0.5
percent. In a similar way, one can show that the elasticity of p with respect to labour also
equals 0.5. Notice, that both these elasticities are constant; unlike the underlying marginal
products, their values do not depend on the specific values of the variables K and L. This
constancy is obtained for every Cobb-Douglas function.

EXERCISE 13 Determine the elasticity of the general Cobb-Douglas function p defined by

p(K, L) = γK α L1−α (K > 0, L > 0)

with respect to K (here 0 < α < 1 and γ > 0). Having done so, what do
you expect for the elasticity with respect to L?

We hope you experienced an ’Aha-Erlebnis’ at Exercise 13: the elasticities of output with
respect to K and L are not just constant, they turn out to be equal to the corresponding
124 Functions of two variables

exponents in our Cobb-Douglas function. This is an obvious generalization of the result in

Chapter 3’s Example 10 (’A demand function with constant elasticity’), and the property
again holds for every Cobb-Douglas function.

In Example 10, we observed that both partial derivatives of the linear demand function d
introduced in Example 1 are constant. However, the elasticities which are implied by these
partial derivatives are not constant, as the next exercise will show.

EXERCISE 13 (a) Calculate the elasticity Elp of the demand for pork with respect to the
variable p at (10, 10), (15, 10) and (20, 10).
(b) Show that the elasticity Elp (p, 10) is strictly decreasing for 10 < p < 20.
(c) Is the elasticity Elr (15, r) strictly increasing or strictly decreasing for
5 < r < 15?
Functions of two variables 125

END-OF-CHAPTER EXERCISES

EXERCISE 1 Find all first-order partial derivatives of the function f of x and y defined by
the equation
(a) z = 2x + 3y (b) z = x2 + y 3
(c) z = x3 y 4 (d) z = (x + y)2
p
(e) z = x/y (f) z = x2 + y 2 .

EXERCISE 2 We can estimate lobster production by the function below, where S is the
stock of lobsters, E is the harvesting effort, and F (S, E) is the catch. Assume
that
F (S, E) = 2.26 S −0.44E 0.48 .

(a) Find FS′ (S, E) and FE′ (S, E).

(b) Show that S FS′ (S, E) + E FE′ (S, E) = kF (S, E) for a suitable k.

EXERCISE 3 The demand for money M in the United States for the period 1929 − 1952
has been estimated as

M = 0.14Y + 76.03(r − 2)−0.84 ,

where Y is the annual national income, and r > 2 is the interest rate in
percent per year.
∂M ∂M
Find and and discuss their signs.
∂Y ∂r

EXERCISE 4 The demand for a product depends on the price p of the product and the
price q charged by a competing producer:

D(p, q) = a − b pq −α .

Here a, b and α are positive constants with α < 1.

Find the partial derivatives of D and comment on their signs.

EXERCISE 5 Find the partial elasticities for the function f with regard to x and y, where
f is defined by the equation
(a) z = xy (b) z = x2 y 5
(c) z = xn ex y n ey (d) z = x + y.
3 −4
(e) z = x y (f) z = ln(x2 + y 2 ).
126 Functions of two variables

EXERCISE 6 Consider the function f defined by

f (x, y) = (x − 1)x2 + y(1 − x) + 1.

The level curve for value 1 consist of

(a) the point (1, 1)
(b) the curves x = 1 and y = x2
(c) the curve x = 1
(d) the curves x = 1 and y = 1.

A CONSTRUCTION PROBLEM

In Example 5 of Chapter 2, we discussed the problem of constructing a gutter with maximal

capacity. Our starting point was a plate of length 100 and width 18. The gutter was
obtained by bending the plate through the middle. We now consider an alternative (and
possibly more efficient) way to construct the gutter. Instead of bending the plate through
the middle, we bend the plate in three parts, where the left and right parts have equal
length. A cross-cut of this gutter is shown in Figure 8 below.

x x

FIGURE 8 Constructing a gutter with maximal capacity

The challenge is again to construct the gutter in such a way that its capacity is maximal.
As a first step to solve this problem, we introduce the variables x and y, as shown in the
left-hand panel of Figure 8. The capacity of the gutter, say C, is the area of the green
cross-cut, say A, multiplied by the length of the plate. So C = 100A.

EXERCISE 7 Which of the following equations correctly expresses the area A in terms of x
and y?
p
(a) A = (18 − y) x2 − y 2
p
(b) A = (18 − 2x + y) x2 − y 2
p
(c) A = (18 − 2x + 0.5y) x2 − y 2
(d) None of the options (a), (b) or (c) is correct.

Of course, the area A is a function of the variables x and y, say A = a(x, y).
Functions of two variables 127

EXERCISE 8 Which of the following shaded areas correctly represents the domain of the
function a?

y y

9 9

(a) (b)

9 x 9 x

y y

9 9

(c) (d)

9 x 9 x

EXERCISE 9 Consider the following two claims concerning the two partial derivatives of
the function a:
p x(18 − 2x + y)
(I) a′x (x, y) = −2 x2 − y 2 + p
x2 − y 2
p 2y(18 − 2x + y)
(II) a′y (x, y) = x2 − y 2 + p
x2 − y 2
(a) claim (I) is correct, claim (II) incorrect
(b) claim (I) is incorrect, claim (II) correct
(c) both claims (I) and (II) are incorrect
(d) both claims (I) and (II) are correct.

A BLACK BOX

A black box should have a volume of 72.

EXERCISE 10 Let’s suppose that the box has length ℓ = 6 and width w = 2. If the length
ℓ of the box changes slightly by ∆ℓ while its width w stays the same then, in
order to keep the volume of the box the same, the height h of the box must
change by about
(a) ∆ℓ
(b) −6 ∆ℓ
(c) 6 ∆ℓ
(d) −∆ℓ.
128 Functions of two variables

EXERCISE 11 If the length ℓ of the box changes slightly while its width w stays the same,
then the relative change of the height h is approximately given by
(a) minus the relative change in ℓ
(b) the relative change in ℓ
(c) twice the relative change in ℓ
(d) minus twice the relative change in ℓ.

THE RELATIVE STRENGTH INDEX (RSI)

One of the indicators which investors relying on ’technical analysis’ use to determine the
optimal timing for the purchase and sale of a stock is the so called relative strength index, or
RSI for short. This indicator is based on the stock’s ups and downs. An up (down) is the
number of points that the stock price has risen (fallen) compared to the previous trading day.
The sum U of all ups and the sum D of all downs is calculated over a predetermined period,
the so-called ’look-back setting’. Below, we provide an example for a five-day period (14
days is actually the most commonly used value, but we want to keep the example simple).
First, we have determined the sums U and D for trading days 1 to 5, and then the quotient

U
RS = .
D

P P U
day price ups downs U= ups D = downs RS =
D
0 24.05 − −
1 24.25 0.20 − − − −
2 24.50 0.25 − − − −
3 24.65 0.15 − − − −
4 24.45 − 0.20 − − −
5 24.15 − 0.30 0.60 0.50 1.20
6 23.95

TABLE 2 Stock price movement: an example

The RSI is now calculated according to the following formula:

100
RSI = 100 − .
1 + RS

In the example at hand, the RSI for day 5 equals

100
100 − = 54.55.
1 + 1.20

EXERCISE 12 The RSI for day 6 (which is based on the trading days 2, 3, 4, 5 and 6)
(a) lies below 35
(b) lies above 35 and below 36
(c) lies above 36 and below 37
(d) lies above 37.
Functions of two variables 129

As time evolves, the RSI for a stock will fluctuate between 0 and 100. When its value falls
below 30 (the ’oversold zone’), an investor will typically decide to sell the stock; if it rises
above 70 (the ’overbought zone’), he should purchase the stock.

RSI
100

70

30

5 6 7 8 t (in days)

FIGURE 9 The movement of the RSI over time

EXERCISE 13 Use Table 2 to determine the highest day 7 stock price at which the investor
would decide to sell the stock.
(a) 24.40
(b) 24.10
(c) 23.95
(d) 23.70.

Obviously, the RSI for a given day is a function of the U and D for that day: RSI = r(U, D).
In the remaining exercises, we will consider this function r of the two variables U and D.

EXERCISE 14 Expressing RSI in terms of U and D leads to RSI =

D
(a) 100
D+U
h U i
(b) 100 1 −
D+U
D−U
(c) 100 .
D+U
U
(d) 100 .
D+U

EXERCISE 15 Imagine a two-dimensional coordinate system with U along the horizontal

axis and D along the vertical axis. In this coordinate system, the level curve
RSI = L, for 0 < L < 100, is a straight line . . .
100
(a) through the origin and with slope
L
100 − L
(b) through the origin and with slope
L
100
(c) with intercept 100 and slope
L
100 − L
(d) with intercept 100 and slope .
L
130 Functions of two variables

EXERCISE 16 Imagine that, at the end of a trading sequence of 5 days, the stock price is
higher than at the start. If so, then the pair (U, D) will lie on the level curve
RSI = L, where L is . . .
(a) smaller than 30
(b) between 30 and 50
(c) equal to 50
(d) larger than 50.

Let’s have another look at the price movements on days 1 to 5, as reported in Table 2.
Remember that the RSI for day 5 turned out to be 54.55. We now want to investigate how
sensitive this value of RSI is to changes in the stock price on day 5.

EXERCISE 17 If the price on day 5 had been δ higher, with 0 < δ < 0.1, then . . .
(a) U would have been larger and D smaller
(b) U would have been larger while D would remain the same
(c) both U and D would have been the same
(d) U would remain the same while D would have been smaller.

EXERCISE 18 If the price on day 5 had been δ higher, with 0 < δ < 0.1, then RSI would
be higher by around . . .
(a) 45δ
(b) 50δ
(c) 55δ
(d) 60δ.
Approximation and optimization 131

7 APPROXIMATION AND OPTIMIZATION

In the current Chapter 7, we will have a closer look at functions of two variables, as introduced
in Chapter 6. We cover two main topics.

First, in Section 1 we will generalize our discussion in Chapter 6’s Section 5. In particular, we
will show how the partial derivatives of a function of two variables can be used to approximate
the change of the value of the dependent variable if small changes occur in the values of both
the independent variables simultaneously, rather than one-at-a-time.

Second, in the remaining sections, we will examine how to determine the extreme value(s) of
functions of two variables.

In Chapter 5, we have already seen how we can use derivatives to determine the extreme
value(s) of functions of one variable. We learned that an extreme value of a differentiable
function on an open interval (’type 1’) can be found by setting the derivative of the function
equal to zero and solving the resulting equation. The solutions are called the stationary
points of the function. Stationary points come in three different types: such a point can be
either a maximum, a minimum, or a point of inflection (in the latter case, the stationary
point does not constitute an extreme value). If the second derivative of a function exists,
then this second derivative can be used to examine the nature of each stationary point.

In the current chapter, we will show how these techniques can be generalized to tackle
functions of two variables. Throughout, we will only consider functions of two variables
for which all first- and second-order partial derivatives exist, and we will refer to
such functions as being (twice) differentiable.

In Section 2, we illustrate the economic and business relevance of the issue at hand by
modeling an inventory management problem which involves two decision variables. The
question is, which decision will minimize total inventory costs.

In Section 3, we will show that the extreme value(s) of such a function can be found using very
similar procedures as in the one-variable case. We begin by setting both partial derivatives
equal to zero. This yields two equations in terms of two variables. Pairs of numbers which
satisfy both these equations are again called stationary points. Again, each such stationary
point may correspond to a maximum or a minimum, or it doesn’t correspond to an extreme
value at all. Their nature can be examined by means of a so-called criterion function. This
function is made up of all the second-order partial derivatives of the original function, and
it plays a similar role as the (single) second derivative in the one-variable case.
 132 Approximation and optimization

Finally, in Section 4, we will use our toolkit to solve the inventory management problem from
Section 2.

1 MARGINALITY
In this section, we will show how the partial derivatives of a function of two variables can
be used to approximate the change in the value of the dependent variable if small changes
occur in the values of both the independent variables simultaneously.

one variable In Chapter 3’s Section 3, we addressed the very same issue for nonlinear functions of only
one variable. Imagine such a function g, and a point a where its derivative exists. Then,
according to the ∆-formula, for a small change ∆x (with respect to a), we have

∆y ≈ g ′ (a) ∆x

We also saw that, if g is a linear function of one variable, this result holds exactly, so we have
∆y = g ′ (a) ∆x. Below, we repeat the associated Table 2 from Chapter 3:

a linear function ℓ a nonlinear function f

x : a → a + ∆x ∆y = ℓ′ (a) ∆x ∆y ≈ f ′ (a) ∆x

TABLE 1 Marginality for functions of one variable

two variables Last week, in Section 5 of Chapter 6, we generalized this result to functions of two variables.
If f is a nonlinear function of two variables and (a, b) is some point in its domain then, for a
small change ∆x in the value a of the independent variable x, while y = b, the corresponding
change in the dependent variable z follows as

∆z ≈ fx′ (a, b) ∆x

Similarly, a small change ∆y in the value b of the independent variable y, while x = a, implies
a change in z of
∆z ≈ fy′ (a, b) ∆y
And again, if f is a linear function of two variables, these results hold exactly, so we can
replace the ≈ symbol by an equality sign. Below, we repeat the relevant Table 1 from Chapter
6:

a linear function ℓ a nonlinear function f

x : a → a + ∆x
∆z = ℓ′x (a, b) ∆x ∆z ≈ fx′ (a, b) ∆x
y:y=b

x:x=a
∆z = ℓ′y (a, b) ∆y ∆z ≈ fy′ (a, b) ∆y
y : b → b + ∆y

TABLE 2 Marginality for functions of two variables

Approximation and optimization 133

We emphasize that the results above refer to a situation where one of the independent
variables changes, while the other one stays constant. We now consider the situation where
a change occurs in the values of both independent variables simultaneously. We start with
an example involving a linear function of two variables.

EXAMPLE 1 A LINEAR (DEMAND) FUNCTION

Remember the linear demand function d from Example 1 in Chapter 6:

d(p, r) = 8 − 0.2p + 0.1r.

This function expresses how the demand for pork q depends on the pork price p and the
chicken price r. In Example 10, we then showed how the partial derivatives

d′p (p, r) = −0.2 and d′r (p, r) = 0.1

can be used to determine the change in pork demand if the price of pork changes, or the
price of chicken. Let’s now see what happens to pork demand if both the prices of pork and
chicken change simultaneously. If the pork price per kilo changes by ∆p euro and the price
of a kilo of chicken changes by ∆r euro, then the new demand for pork is given by

d(p + ∆p, r + ∆r) = 8 − 0.2(p + ∆p) + 0.1(r + ∆r) = 8 − 0.2p + 0.1r − 0.2 ∆p + 0.1 ∆r

= d(p, r) − 0.2 ∆p + 0.1 ∆r.

In other words: the change in demand, caused by these simultaneous price changes, is given
by
d(p + ∆p, r + ∆r) − d(p, r) = −0.2 ∆p + 0.1 ∆r = d′p (p, r) ∆p + d′r (p, r) ∆r
| {z } | {z }
effect of the effect of the
change in p change in r

Apparently,
the effect on the demand for pork equals the sum of the effect
caused by a change in the pork price and the effect caused by a
change in the chicken price.

This conclusion generalizes to any linear function of two variables, say ℓ. Starting from a
point (a, b) in its domain, it holds that

∆z = ℓ′x (a, b) ∆x + ℓ′y (a, b) ∆y.

Let us now consider a nonlinear function of two variables, say f . Suppose again that,
starting from a point (a, b) in its domain, small changes ∆x and ∆y occur in the values of
the independent variables simultaneously. What will happen to the value of the dependent
variable, say z? From Tables 1 and 2, the answer should be intuitively obvious: we get the
same result as for linear functions, but replacing the = sign by a ≈ sign. Doing so, we get:

∆z ≈ fx′ (a, b) ∆x + fy′ (a, b) ∆y

134 Approximation and optimization

In words: the effect on the dependent variable z is roughly the sum of the effect caused by
a change in independent variable x, and the effect caused by a change in the independent
variable y. Table 3 below summarizes our conclusions.

a linear function ℓ a nonlinear function f

x : a → a + ∆x
∆z = ℓ′x (a, b) ∆x + ℓ′y (a, b) ∆y ∆z ≈ fx′ (a, b) ∆x + fy′ (a, b) ∆y
y : b → b + ∆y

TABLE 3 Marginality for functions of two variables, simultaneous changes

EXAMPLE 2 Consider the function h given by the equation

z = x2 − y 3 + 2xy 2 ,

and the point (x, y) = (2, −1) in its domain. If x is changed from 2 to 2 + ∆x and y from
−1 to −1 + ∆y (where ∆x and ∆y are small), then the resulting change ∆z satisfies

∆z ≈ 6 ∆x − 11 ∆y,

where we used the fact that h′x (2, −1) = 6 and h′y (2, −1) = −11 (check!). So if ∆x = −0.01
and ∆y = 0.02, this leads to an approximate change of ∆z ≈ 6 × (−0.01) − 11 × 0.02 = −0.28.
Compare this approximation with the exact change:

∆z = h(2 + ∆x, −1 + ∆y) − h(2, −1) = h(1.99, −0.98) − h(2, −1)

= 8.723684 − 9 = −0.276316.

As you can see, our approximation is pretty adequate, to put it mildly.

EXAMPLE 3 Continuing Example 2 from Chapter 6, we again consider the production

function p given by the equation
Q = 5K 0.5 L0.5 .

Starting from K = 8 and L = 2, where Q = 20, we are interested in the effect of input
changes ∆K and ∆L on output Q. Please verify that p′K (8, 2) = 1.25 and p′L (8, 2) = 5. So
it follows that
∆Q ≈ 1.25 ∆K + 5 ∆L.

E.g., for the values ∆K = 0.5 and ∆L = 0.1, the approximate change in output follows as

∆Q ≈ 1.25 × 0.5 + 5 × 0.1 = 1.125,

while the exact change is given by

∆Q = p(8 + ∆K, 2 + ∆L) − p(8, 2) = p(8.5, 2.1) − p(8, 2)

= 5 · 8.50.5 2.10.5 − 20 ≈ 1.1246.

Again, the approximation can be considered as extremely good.

Approximation and optimization 135

EXERCISE 1 Consider the production function p, given by the equation

Q = −(2L − K)(2L − 3K).

Determine the expression for the approximate change in output in response to

changes ∆K and ∆L, starting from the levels K = 8 and L = 7.

EXAMPLE 4 QUALITY CONTROL

As a quality control engineer at a plant which manufactures cylindrical cans, you have to
check whether those cans have the correct volume. Of course, the volume of a can (say V )
depends on its radius (say r) and its height (say h). In particular, that volume equals the
base area of the can, given by πr2 , times its height h:

V = πr2 h.

So V is a function of the can’s radius r and height h: V = v(r, h), where v(r, h) = πr2 h.
You want to know how sensitive the volume is to small variations in radius and height.
Let’s use the techniques just discussed. You can easily check that vr′ (r, h) = 2πrh and
vh′ (r, h) = πr2 . Consequently, starting from some (r, h), the change in volume caused by
small changes ∆r and ∆h in radius and height is approximately

∆V ≈ vr′ (r, h) ∆r + vh′ (r, h) ∆h = 2πrh ∆r + πr2 ∆h.

E.g. for cans which should have r = 5 and h = 10, we would get ∆V ≈ 100π ∆r + 25π ∆h.
All of this is about the absolute production error in volume V . Conceivably, you are more
interested in the relative error. That relative error equals the absolute error ∆V , divided by
the intended value V , preferably multiplied by 100 to obtain an interpretation in percentages.
Using the formulas for V and ∆V developed so far, we get the following expression for the
approximate percentage error in volume:

∆V 2πrh ∆r + πr2 ∆h ∆r ∆h
× 100 ≈ × 100 = 2 × 100 + × 100 .
| V {z } πr2 h | r {z } | h {z }
percentage error in V percentage error in r percentage error in h

To illustrate the use of this result, imagine that the tolerances of the plant’s machinery
imply that the radius and the height of those cans can both deviate at the most 1% from the
programmed values. We can now estimate the maximum percentage error in the volume as
2 × 1% + 1% = 3%. Note that you may want to pay special attention to the quality control
of the radius: after all, the can’s volume is twice as sensitive to a percentage error in the
radius, as it is to an equal percentage error in the height.
Clever students should immediately suspect, that the latter result is due to the fact that
the radius r occurs with an exponent of 2 in the equation for V , while the height h has an
exponent of only 1. Let’s go one step further: very clever students might have been able to
reason their way to our final formula without any formal derivation, simply by exploiting
the analogy with Chapter 3’s Example 10 (’A demand function with constant elasticity’).
After all, the equation for V defines a power function in terms of r and h. So for given h,
136 Approximation and optimization

the elasticity of V w.r.t. r immediately follows as the exponent of r, i.e. 2. This implies
that a percentage error in r will indeed create roughly 2 times that percentage error in V .
Similarly, for given r, the elasticity of V w.r.t. h follows as 1; so a percentage error h will
roughly cause the same percentage error in V . Adding up both these effects, we arrive at
the formula above!

EXERCISE 2 Revenue R equals the product of price p and quantity q: R = pq. Show that
the relative change in R, caused by small changes in price p and quantity
q, is approximately equal to the sum of the relative change in price and the
relative change in quantity.

N.B.: This result obviously generalizes to other products. In words: the

percentage change in a product is approximately equal to the sum of the
percentage changes in its components.

2 AN EXAMPLE OF TWO-VARIABLE OPTIMIZATION:

MINIMIZING INVENTORY COST

You are not studying mathematics, but economics or business. To see why it is nonetheless
important to know a bit about the optimization of functions of two variables, we reconsider
the Toyota dealer which we first encountered in Example 2 of Chapter 2. Five successful
business years have passed, and our dealer is now selling the Auris at a rate of 120 per
year, for 30 000 euros per car. In order to satisfy this demand, he still periodically orders
a fixed number of cars. He therefore still faces the two types of costs discussed before:
ordering costs and holding costs. The ordering costs, i.e. the costs of placing an order (e.g.
due to administrative handling) amount to 225 euros per order, irrespective the size of the
order. The holding costs, i.e. the annual costs involved in stocking the cars in the dealer’s
warehouse, equal 5% of the average value of the stock in inventory.

So far we have assumed that, once the dealer’s stock has been depleted, he immediately
orders a new batch of Toyotas, which is then delivered to his warehouse on the same day. So
when the next customer shows up, he can basically drive his new car home right away. We
now add a new consideration to our inventory problem: supplying on order. Since it is quite
expensive to keep a large number of cars in stock, the dealer investigates the possibility of
supplying a few cars to his customers on order. What this effectively means is that he does
not immediately order a new batch when his stock has run out. As a consequence, his next
few customers (the so-called backlog) have to wait a while, i.e. until he does order the next
batch, before they can take possession of their desired car. Such a strategy allows the dealer
to economize on the holding costs, but it does lead to a third type of costs: backlogging costs.
E.g., he may have to give impatient customers a small discount, throw in a few options for
free, etc. We assume that these costs are 500 euros per car per year: so if e.g. the average
backlog during the year is 3 cars, then the dealer faces an additional cost of 1500 euros.

The addition of this new element makes the car dealer’s decision problem more complicated:
 Approximation and optimization 137

he now has to determine not only the number of cars that he is going to order each time,
but also how many customers he is going to serve on order. Before modeling the adjusted
decision problem, we examine a special case: let’s assume that the dealer has decided to
order 10 cars each month (note that 10 times 12 equals 120!), and that 2 cars are supplied on
order at the end of each month. Figure 1 below sketches this scenario: like before, it depicts
the number of cars in the showroom, against time in months. When the dealer orders 10 cars
at the start of the first month, he has an accumulated backlog of 2 orders from the previous
month. These customers are served first, so only 8 cars actually enter the showroom. During
that first month, 10 cars are sold at regular intervals, as visualized by the downward-sloping
solid line in the figure. By the time the first 8 cars have been sold, the stock has fallen to
zero. The remaining 2 customers in that month are then served ’on order’; in the picture,
this shows up as a negative stock, which accumulates into −2 at the end of the month. The
dealer then grabs the phone, orders a fresh batch of 10 cars, and the cycle repeats itself in
the second month. Etcetera, etcetera.

stock

1 2 t
−2

FIGURE 1 The stock level over time.

The three types of costs involved in this inventory policy can now be calculated as follows.
type 1 Since the dealer orders 10 cars each time, implying that he places 12 orders per year, the
annual ordering costs are equal to
12
|{z} × 225
|{z} = 2700 euros.
number of orders costs per order

This calculation is the same as in the original scenario. The calculation of the other two
types of costs is more involved. The key thing to see is that the dealer has a positive stock
during a certain fraction of the year, but a negative stock (i.e. a positive backlog) during
the rest of the year. Take a careful look at Figure 1 above, in particular: at developments
8
during the initial month. During the first fraction of 10 of that month, his stock is positive;
2
during the remaining fraction of 10 , his stock turns negative. The repetitive nature of the
process implies that these fractions also apply on an annual basis. Having established this,
we can now have a closer look at the holding and backlogging costs.
type 2 The holding costs are equal to 5% of the value of the cars kept in stock (i.e. 5% of 30 000
euro per car), computed on a yearly base. Obviously, we now need to determine the average
number of cars in stock during the year. Our previous analysis implies that, during a fraction
8 2
10 of the year, there are on average 4 cars in stock; during the remaining fraction 10 of the
 138 Approximation and optimization

year, his physical stock is zero (implying zero holding costs). Consequently, the annual
holding costs are given by:
8
10 × 4
|{z} × (0.05 · 30 000) = 4800 euros.
|{z} | {z }
part of average number 5% of the
the year of cars in stock value per car

type 3 The backlogging costs are equal to 500 euro per car on order, again computed on a yearly
base. So now, we need to determine the average size of the backlog during the year. We
8
trust you see right away that, during a fraction 10 of the year, the backlog is zero (implying
2
zero backlogging costs); during the remaining fraction 10 of the year, the average backlog is
1 unit. As a result, the annual backlogging costs are given by
2
10 × 1
|{z} × 500
|{z} = 100 euros.
|{z}
part of average number of costs per car
the year cars out of stock per year

Let’s now generalize our numerical example. To solve the car dealer’s decision problem,
we will need to determine the total annual costs for an arbitrary order size q and arbitrary
backlog s.

EXERCISE 3 (a) Sketch an adjusted version of Figure 1 above, which visualizes this more
general scenario. Hint : If you can replace all the numbers along the
axes of the original Figure 1 by expressions in terms of q and s, you’re in
business!
(b) How many times per year does the dealer have to place a new order? Use
your result to express the annual ordering costs in terms of q and s.
(c) Using the fact that the fraction of the year with a ’positive’ amount of
stock equals (q − s)/q, express the annual holding costs in terms of q and
s.
(d) Which fraction of the year is characterized by a ’negative’ stock? Use your
result to express the annual backlogging costs in terms of q and s.

Adding up the three cost items which you found (we hope) in Exercise 3 above, we can
conclude that the dealer’s total annual costs are given by the following function:
27 000 (q − s)2 s2
k(q, s) = + 750 + 250 .
q q q

The example above is of course stylized and relatively simple but, like the original Toyota
example from Chapter 2, it does contain some of the basic trade-offs which are at stake in
real-life inventory management decisions. The sale of commercial software to help firms take
such decisions is a multi-million euro industry.
Many decision problems in economics or business are similar to our example, in the sense
that you start out from an essentially verbal description of the problem; the first step is
then to convert it into a mathematical optimization problem. For the problem at hand,
we have completed this step above. The second step is to actually solve that mathematical
optimization problem. The next section will discuss how to do so, in general terms. In
Section 4, we will get back to the problem above and solve it.
Approximation and optimization 139

3 FINDING EXTREME VALUES

In Section 3 of Chapter 5 (review it if necessary!), we have argued that a function of one
variable can have three different types of extreme values. Although we will not state it
explicitly, this theorem actually generalizes to functions of two variables. But in this chapter,
we deliberately reduce the palette of possibilities. As mentioned in the introduction, we will
only discuss differentiable functions, i.e. functions for which all first- and second-order partial
derivatives exist. Consequently, ’type (2)’ extreme values at non-differentiable points cannot
occur. Moreover, we will choose our functions, or limit their domains, in such a way that
’type (3)’ extreme values at endpoints of the domain (now called: boundary points) cannot
occur either. That leaves us with ’type (1)’ extreme values only. In the introduction to this
chapter, we have already described how the search for such extreme values has to start, in
case of a function of two variables: by setting both the first-order partial derivatives of that
function equal to zero. The points satisfying these first-order conditions are the stationary
points of our problem. In Section 3.1, we will argue that if a function has an extreme value,
i.e. a maximum or a minimum, it must occur at such a stationary point. But we will also
show that, like in the one-variable case, not every stationary point constitutes an extreme
value: occasionally, we find a so-called saddle point. To evaluate the nature of a stationary
point (i.e. maximum, minimum or saddle point), we can bring in the second-order partial
derivatives of our function: the implied second-order conditions will be discussed in Section
3.2. Sometimes, these second-order conditions will fail to establish the nature of a stationary
point. We will address this phenomenon in the Exercises 7 and 8.

3.1 FIRST-ORDER CONDITIONS: STATIONARY POINTS

Consider the function g of x and y defined by

g(x, y) = x2 + y 2 .

It will be clear right away that g(0, 0) = 0: at the origin, our function is zero. But as soon
as either x or y differs from 0, it holds that x2 + y 2 > 0, i.e. g(x, y) > 0. So at any point
other than the origin, our function is strictly positive. Clearly, the lowest point on the graph
of g must occur at the origin: this is where the global minimum of the function g is located.
The way in which our example illustrates the concept of a minimum is an obvious generaliza-
tion of the one-variable case; and of course, the concept of a maximum could be illustrated
in a similar way (compare Chapter 5, Section 1).

Where do the partial derivatives come into this? We already know that our extreme value
occurs at (0,0). Now let’s keep y fixed at 0. If so, our function simplifies to g(x, 0) = x2 ,
which is a function of x only. Of course, g(x, 0) must reach its minimum at x = 0. This
implies that its derivative at x = 0 must be 0. But in the previous Chapter 6, we have
seen that this derivative is precisely the partial derivative gx′ (0, 0). We conclude that at our
extreme value, it must hold that gx′ (0, 0) = 0. An entirely symmetrical argument (’Now let’s
keep x fixed at 0...’) will show that gy′ (0, 0) = 0 must hold as well.
 140 Approximation and optimization

In principle, the previous argument applies to any differentiable function f with an extreme
value at some point (a, b): at that point, both its partial derivatives have to be 0. There is
one caveat: (a, b) must be an interior point of its domain. That is: (a, b) is contained in a
circular disc that lies entirely inside the domain of f . This situation is sketched in Figure 2
below, where the domain of f consists of all pairs (x, y) on or inside the solid red shape.

b
Df

a x

FIGURE 2 An interior point in the domain of a function of two variables

A differentiable function can have a ’type (1)’ extreme value at an interior point of its domain
(like (a, b) in Figure 2), but also a ’type (3)’ extreme value at a boundary point of its domain
(i.e. somewhere along the solid red shape in Figure 2). As already indicated, we ignore the
latter possibility in this chapter, and confine ourselves to interior points only. Subject to this
limitation, we obtain the following result.

FIRST-ORDER CONDITIONS FOR AN EXTREME VALUE

If a differentiable function f has a maximum or minimum at an interior point (a, b) of its

stationary point domain, then (a, b) must be a stationary point of f , i.e.

fx′ (a, b) = fy′ (a, b) = 0.

To illustrate the geometrical significance of stationary points, let (a, b) be such a point of an
arbitrary function f . According to our findings in Section 4.2 of the previous chapter, fx′ (a, b)
is the slope of the tangent to the graph of f in the x-direction. Since this slope equals zero at
our stationary point, that tangent is horizontal. Similarly, the tangent in the y-direction is
also horizontal. Now remember the notion of the tangent plane, i.e. the (unique) plane that
contains both these two tangents. Since both these tangents are horizontal, we can conclude:
at our stationary point, the tangent plane to the graph of f is horizontal. This result is
entirely analogous to functions of one variable: at a stationary point of such a function, the
tangent line to their graph is also horizontal.

EXAMPLE 5 To find the stationary points of the function h defined by

h(x, y) = 3x2 + y 3 − 6xy

we have to set both the partial derivatives of h equal to zero.

 Approximation and optimization 141

These partial derivatives are given by

h′x (x, y) = 6x − 6y

and h′y (x, y) = 3y 2 − 6x.

So we must solve the two equations h′x (x, y) = 0 and h′y (x, y) = 0 simultaneously. This is
indicated by placing a curly bracket in front of these two equations:
 ′  
hx (x, y) = 0 6x − 6y = 0 x=y
⇐⇒ 2
⇐⇒
′
hy (x, y) = 0 3y − 6x = 0 y 2 = 2x.
Unfortunately, there are no fixed rules for solving such systems of two equations: they must
be considered case by case, and your intuition and common sense are usually indispensable.
In the case at hand, an obvious idea is to substitute y = x into the second equation, so that
we obtain
x2 = 2x ⇐⇒ x2 − 2x = 0 ⇐⇒ x(x − 2) = 0 ⇐⇒ x = 0 or x = 2.
Using y = x, we end up with the following two stationary points: (0, 0) and (2, 2).
In Example 6, we will show that h has a minimum, i.e. an extreme value, at the stationary
point (2, 2). In Example 8, it will be shown that the function h has no extreme value at the
other stationary point, (0, 0).

EXERCISE 4 Determine the stationary points of the functions f defined by

(a) f (x, y) = 1 + 2x2 − x + 4y 2 − y
(b) f (x, y) = x3 − 12x − 18y + 3y 2 (x ≥ 0)
2 2
(c) f (x, y) = −x + xy − y + y − 8x
(d) f (x, y) = x3 + y 3 − 12x − 3y + 10.

3.2 SECOND-ORDER CONDITIONS: THE CRITERION FUNCTION

Let’s continue the analogy with the analysis of a function f of a single variable x. Imagine
that such a function has a stationary point at x = a, in other words: f ′ (a) = 0. As we have
seen in Chapter 5, we can examine whether or not this is an extreme value by determining
the second derivative f ′′ (a), and applying the ’Second-derivative test’. There are now three
possibilities. If f ′′ (a) is negative, then f has a maximum at x = a; if f ′′ (a) is positive, then
f has a minimum at x = a; and if f ′′ (a) is zero, then further examination of the nature of
the point x = a is necessary (typically by examining the sign diagram of the derivative f ′ in
a neighborhood of x = a, i.e. by applying the ’First-derivative test’).
A similar ’test’ in terms of the second-order derivatives exists for the investigation of the
stationary points of a function f of two variables x and y. Of course, we now have three
′′ ′′ ′′ ′′
distinct types of second-order partial derivatives: fxx , fyy and fxy (remember that fxy =
′′
criterion function fyx ). From these three, we can construct the so called criterion function C, defined by
′′ ′′
 ′′ 2
C(x, y) = fxx (x, y) · fyy (x, y) − fxy (x, y) .
To decide whether f has an extreme value at a stationary point (a, b), we need the criterion
value C(a, b) at (a, b). When a stationary point has a nonzero criterion value, its character
can be determined right away. If the criterion value is zero, further investigation will be
necessary.
142 Approximation and optimization

SECOND-ORDER CONDITIONS FOR AN EXTREME VALUE

Let (a, b) be a stationary point of a differentiable function f . Then f has

′′
(a) a minimum at (a, b) if C(a, b) > 0 and fxx (a, b) > 0
′′
(b) a maximum at (a, b) if C(a, b) > 0 and fxx (a, b) < 0
(c) a saddle point (i.e. no extreme value) at (a, b) if C(a, b) < 0
(d) a minimum, a maximum, or a saddle point at (a, b) if C(a, b) = 0.

In the next three examples, we will have a closer look at the cases (a), (b) and (c); we will
then also define what a saddle point is. Case (d) will be examined in Exercises 7 and 8.

EXAMPLE 6 A STATIONARY POINT WITH A POSITIVE CRITERION VALUE

According to Example 5, (2, 2) is a stationary point of the function h defined by

h(x, y) = 3x2 + y 3 − 6xy.

As h′′xx (x, y) = 6, h′′xy (x, y) = −6 and h′′yy (x, y) = 6y, the criterion function follows as
 2
C(x, y) = 6 · 6y − −6 = 36y − 36.

So at the stationary point (2, 2), we have

C(2, 2) = 36 · 2 − 36 = 36 > 0.

Since h′′xx (2, 2) = 6 > 0, we are dealing with case (a) of the Second-order Conditions here.
We conclude that h has a minimum at (2, 2).

Let’s have a closer look at the scenario discussed in Example 6: a stationary point (a, b)
with a positive criterion value. Since the second term of the criterion function C is always
negative (check!), a positive criterion value can only occur if its first term, i.e. the product
′′ ′′
of fxx (a, b) and fyy (a, b), is positive. This implies that those two second-order derivatives
must have the same sign. Obviously, this leaves us with two possibilities:
′′ ′′
– in case (a), both fxx (a, b) and fyy (a, b) are positive;
′′ ′′
– in case (b), both fxx (a, b) and fyy (a, b) are negative.
We can now use a geometrical argument to understand case (a) better. Starting from the
stationary point (a, b), let’s keep y constant at b, and think about the one-variable function
′′
f (x, b). We know from the previous chapter that fxx (a, b) is the second derivative of f (x, b)
at x = a. The fact that this second derivative is positive implies, that f (x, b) is convex on an
′′
interval around x = a. Using a similar argument, the fact that fyy (a, b) is positive implies
′′ ′′
that f (a, y) is also convex on an interval around y = b. So: if both fxx (a, b) and fyy (a, b)
are positive, then this means that at that stationary point, f (x, y) is both convex in the
x-direction (i.e. for given y = b), and convex in the y-direction (for given x = a). This
explains why, in case (a), the function has a minimum at (a, b).
′′ ′′
A similar geometrical argument can be applied to case (b): if both fxx (a, b) and fyy (a, b)
are negative, then at the stationary point (a, b) our function is concave in both the x- and
y-directions. Consequently, that stationary point is a maximum.
 Approximation and optimization 143

EXAMPLE 7 A STATIONARY POINT WITH A NEGATIVE CRITERION VALUE

The partial derivatives of the function g defined by

g(x, y) = y 2 − x2

are given by gx′ (x, y) = −2x and gy′ (x, y) = 2y. Clearly, (0, 0) is the only stationary point
′′ ′′ ′′
of g. Since gxx (x, y) = −2, gxy (x, y) = 0 and gyy (x, y) = 2, the criterion function follows as
C(x, y) = −4, so that C(0, 0) < 0. According to case (c) of our Second-order Conditions, we
conclude that the function g has no extreme value at the stationary point (0, 0). So what is
′′
happening at that point? We can again use a geometrical argument. The fact that gxx is
negative implies, that the graph of g(x, 0) is concave, with a maximum at x = 0. Similarly,
′′
the fact that gyy is positive implies, that the graph of g(0, y) is convex, with a minimum
at y = 0. Combining these observations, the graph of g must look something like Figure 3
below. Our stationary point (0, 0) constitutes a maximum in the x-direction, and a minimum
in the y-direction, so it is not an extreme value. Instead, we say that our stationary point
saddle point is a saddle point of the function g. And indeed, with some stretch of imagination, the shape
of the graph below resembles a saddle: it is concave in the direction in which the cowboy’s
legs clasp the saddle, and convex in the direction of the spine of his horse.

FIGURE 3 The graph of the function g given by g(x, y) = y 2 − x2

This example illustrates what we already claimed at the start of Section 3: not every sta-
tionary point constitutes an extreme value.

The situation in Example 7 above is typical for a stationary point (a, b) with a negative
′′ ′′
criterion value: the partial derivatives fxx (a, b) and fyy (a, b) now have opposite signs. For
the geometrical reasons explained above, in this case (c), a stationary point will always be a
saddle point.

EXAMPLE 8 Continuing Example 6, remember that the function h from Example 5

actually had another stationary point: (0, 0). At this stationary point, the criterion value is

C(0, 0) = 36 · 0 − 36 = −36 < 0.

Consequently, the function h does not have an extreme value at (0, 0), but a saddle point.
144 Approximation and optimization

EXERCISE 5 Check whether the functions from Exercise 4 have extreme values.

EXERCISE 6 We try to find the extreme values of a function f of two variables x and y. By
setting the first-order partial derivatives equal to zero, five stationary points of
f are found: (a1 , b1 ), . . ., (a5 , b5 ). When calculating the second-order partial
derivatives, it turns out that . . .
′′ ′′ ′′
at (a1 , b1 ): fxx < 0, fyy = −1 and fxy =0
′′ ′′ ′′
at (a2 , b2 ): fxx = fyy = fxy =1
′′ ′′ ′′
at (a3 , b3 ): fxx = 0, fyy = −5 and fxy =2
′′ ′′
 ′′
2 ′′ ′′
at (a4 , b4 ): fxx > 0, fyy > 0 and fxy < fxx · fyy
′′ ′′ ′′ ′′
at (a5 , b5 ): fxx = −2, fxy = 1 and fxx · fyy = 4.
Use this information to determine which stationary points constitute the max-
ima and minima of f .

In the following two exercises, we will consider stationary points with a zero criterion value.
In one of them, this stationary point actually corresponds with an extreme value. In the
other exercise, the function does not have an extreme value.

EXERCISE 7 Consider the function f given by the formula

z = x4 + y 4 .

(a) Show that (0, 0) is a stationary point with criterion value zero.
(b) Explain why the function f has a minimum at this stationary point.

EXERCISE 8 Consider the function g given by the formula

z = x3 y.

(a) Show that (0, 1) is a stationary point with criterion value zero.
(b) Explain, by considering the function g(x, 1), why the function g has no
extreme value at this stationary point.

4 MINIMIZING INVENTORY COST

Let us now apply the tools from Section 3 to solve the inventory cost problem introduced in
Section 2.

In order to find a possible minimum of the cost function k defined by

27 000 (q − s)2 s2
k(q, s) = + 750 + 250 ,
q q q

we start by applying the first-order conditions to find the stationary point(s?) of k.

Approximation and optimization 145

We have, for q > 0 and s > 0,

27 000 q · 2(q − s) − (q − s)2 · 1 s2

kq′ (q, s) = − 2
+ 750 2
− 250 2
q q q
1

= 2 −27 000 + 750 2q 2 − 2qs − q 2 + 2qs − s2 − 250s2
q
1 
= 2 −27 000 + 750q 2 − 1000s2
q

and
2(q − s) · (−1) s 1 
ks′ (q, s) = 750 + 500 = −1500(q − s) + 500s
q q q
1 
= −1500q + 2000s .
q
Next, we set these partial derivatives equal to zero, to obtain the first-order conditions of
our problem:  
kq′ (q, s) = 0 −27 000 + 750q 2 − 1000s2 = 0
⇐⇒
ks′ (q, s) = 0 −1500q + 2000s = 0.
The second equation immediately implies that s = 43 q. Plugging this into the first one, we
get:
750q 2 − 562.5q 2 = 27 000 =⇒ 187.5q 2 = 27 000 =⇒ q = 12.

We can conclude that the pair (12, 9) is the only stationary point of k.
Now for the second-order conditions. Since

′′ q 2 · 1500q − (−27 000 + 750q 2 − 1000s2) · 2q

kqq (q, s) =
q4
1500q 2 + 54 000 − 1500q 2 + 2000s2 54 000 + 2000s2
= 3
= ,
q q3
′′ 2000s
kqs (q, s) = − ,
q2
′′ 2000
and kss (q, s) = ,
q

the criterion value at the point (12, 9) follows as

C(12, 9) = 125 × 166 32 − (−125)2 > 0.

′′
Since the partial derivative kqq (12, 9) is also positive, we have case (a) here: our stationary
point is a minimum, as desired.
Altogether, we conclude that the inventory costs are minimized if the dealer places orders for
12 cars each time, and lets his backlog accumulate to 9 cars before placing the next order.
146 Approximation and optimization

END-OF-CHAPTER EXERCISES
EXERCISE 1 A firm produces two different kinds A and B of a commodity. The daily cost
of producing x units of A and y units of B is

C(x, y) = 2x2 − 4xy + 4y 2 − 40x − 20y + 514.

(a) Suppose that the firm sells all its output at a price per unit of 24 euro
for A and 12 euro for B.
Find the daily production levels for both kinds of the commodity that
maximize profit.
(b) The firm is required to produce exactly 54 units per day of the two kinds
combined. What is now the optimal production plan?
(c) The firm produces 5 units of commodity A and 6 units of commodity B.
Assume this production changes with ∆x units and ∆y units, respectively.
Find an approximation ∆C for the change in the production cost.

EXERCISE 2 A firm produces and sells a product in two separate markets. When the price
in market A is pA per ton, and the price in market B is pB per ton, the demand
in tons per week in the two markets are, respectively,

qA = a − bpA
qB = c − dpB .
The cost function is given by

C(qA , qB ) = α + β(qA + qB ),

and all constants are positive.

(a) Find the firm’s profit as a function of the prices pA and pB , and then find
the pair (p∗A , p∗B ) which maximizes profit.
(b) Suppose it becomes unlawful to discriminate by price, so that the firm
must charge the same price in the two markets. What price p∗ will now
maximize profits?

EXERCISE 3 Let f be the function defined by

f (x, y) = x2 − y 2 − xy − x3 .

(a) Find the stationary points of the function f .

(b) Classify the stationary points.

EXERCISE 4 Consider the function f defined by

f (x, y) = 21 x2 − x + ay(x − 1) − 13 y 3 + a2 y 2 ,

where a is a constant.
(a) Prove that (x∗ , y ∗ ) = (1 − a3 , a2 ) is a stationary point of f .
(b) Classify this stationary point.
Approximation and optimization 147

EXERCISE 5 Consider the function f defined by

f (x, y) = xy + 2xy 2 + xy 3 .

(a) Prove that (a, −1) is a stationary point of the function f . Here a can be
any number.
(b) Find the criterion function C for this function f .
(c) Classify the stationary points (a, −1) for all a.

A BLUE BOX

A box without lid should have a storage capacity, i.e. a volume, of 32. The figure below
sketches this box: note that its upper side is indeed open.

w
ℓ

EXERCISE 6 Assume that, initially, w = h = 2 and ℓ = 8; with these dimensions, the

volume of the box indeed equals the desired 32. Now imagine that its length
ℓ and its width w change by small amounts ∆ℓ and ∆w. Clearly, to keep the
volume of the box constant at 32, its height h has to adjust appropriately.
The implied change ∆h is roughly given by
(a) − 41 ∆ℓ − ∆w
(b) − 41 ∆ℓ − 1
4 ∆w
1
(c) −∆ℓ − 4 ∆w
(d) −∆ℓ − ∆w.

EXERCISE 7 Consider the same thought experiment as for Exercise 6. If the length ℓ and
the width w of the box change slightly by ∆ℓ and ∆w, respectively, then the
relative change of its height h is approximately
(a) minus twice the relative change in ℓ, minus twice the relative change in w
(b) minus the relative change in ℓ, minus the relative change in w
(c) minus twice the relative change in ℓ, minus the relative change in w
(d) minus the relative change in ℓ, minus twice the relative change in w.

The plan is to decorate the outside of the box with an expensive blue paint. In order to
minimize the cost of the paint, the dimensions of the box must be chosen in such a way that
the surface area to be painted, say A, will be as small as possible.
148 Approximation and optimization

EXERCISE 8 Which of the following equations correctly expresses the painted area A of the
box in terms of its length ℓ and width w?
32 32
(a) A = + + 2ℓw
w ℓ
64 32
(b) A = + + ℓw
w ℓ
32 64
(c) A = + + 2ℓw
w ℓ
64 64
(d) A = + + ℓw.
w ℓ

According to Exercise 8, the area to be painted, A, can be expressed as a function of the

length ℓ and the width w of the box: say A = a(ℓ, w). It can be shown that the function a
has a unique stationary point.

EXERCISE 9 At this unique stationary point, the criterion value is

(a) smaller than 0
(b) equal to 0
(c) larger than 0, but smaller than 10
(d) equal to 10 or larger.

EXERCISE 10 Let’s denote the dimensions which minimize the painted area A by ℓopt , wopt
and hopt . Which of the following (in)equalities concerning these optimal
dimensions is correct?
(a) ℓopt = hopt
(b) wopt > hopt
(c) wopt < hopt < ℓopt
(d) ℓopt < hopt .

A CONSTRUCTION PROBLEM

In a number of end-of-chapter exercises of Chapter 6, we discussed the problem of construct-

ing a gutter with maximal capacity. Our starting point was a plate of length 100 and width
18. The gutter was obtained by bending the plate in three parts, where the left and right
parts have equal length. As a reminder, a cross-cut of the gutter is shown below:

x x

In Exercise 7, we asked you to prove that the capacity of the gutter equals C = 100A, where
p
A = (18 − 2x + y) x2 − y 2 .

Here x and y are the variables as introduced in the figure above.

Approximation and optimization 149

Clearly, A is a function of x and y, say A = a(x, y). In the subsequent Exercise 9, the partial
derivatives of this function turned out to be
p x(18 − 2x + y)
a′x (x, y) = −2 x2 − y 2 + p
x2 − y 2
p y(18 − 2x + y)
and a′y (x, y) = x2 − y 2 − p .
x2 − y 2

EXERCISE 11 At any stationary point of the function a, it must hold that

(a) x = 1.5y
(b) x = 2y
(c) x = 2.5y
(d) x = 3y.

EXERCISE 12 The unique stationary point of a is given by

(a) x = 9 and y = 3
(b) x = 7.5 and y = 3
(c) x = 6 and y = 3
(d) x = 4.5 and y = 3.

In the problem at hand, the second-order partial derivatives are extremely complex, so we
don’t want to ask you to verify the second-order conditions. Just take our word: it can be
shown that our unique stationary point corresponds with a maximum, as intended.

EXERCISE 13 The maximal capacity of the gutter is

(a) less than 3500
(b) between 3500 and 4000
(c) between 4000 and 4500
(d) more than 4500.

EXERCISE 14 Assume that, initially, x = 5 and y = 3. If these amounts change slightly

by ∆x and ∆y, respectively, then the relative change of the capacity C, in
terms of the relative changes in its determinants x and y, is approximately
given by
∆x ∆y
(a) 0.65 − 0.29
5 3
∆x ∆y
(b) 0.50 + 0.40
5 3
∆x ∆y
(c) −0.35 + 0.54
5 3
(d) None of the options at (a), (b) or (c) is correct.