Mathematics for the Science of Complex Systems

5th March 2010

Session 3. Power Laws in the Science of Complex Systems

Prepared for the course team by Jeffrey Johnson

Contents
3.1 The Aims and Objectives of Session 3 ..................................................................... 2
3.2 Representing functions graphically........................................................................... 3
3.3 Linearity and linear relationships.............................................................................. 6
3.4 Quadratic Functions .................................................................................................. 7
3.5 Cubic Functions, Quartic Functions and Polynomials.............................................. 8
3.6 Scale Invariance ...................................................................................................... 10
3.7 Measuring the slopes of functions .......................................................................... 11
3.8 The Exponential and Logarithmic Functions.......................................................... 14
3.9 Logarithms, polynomials, and straight lines ........................................................... 16
3.10 Graphing on logarithmic paper ............................................................................. 18
3.11 Zipf’s Law............................................................................................................. 19
3.12 Power Laws .......................................................................................................... 21
3.13 The Pareto Distribution......................................................................................... 22
3.14 Examples............................................................................................................... 24
3.14.1 Earthquakes.................................................................................................... 24
3.14.2 Allometric laws.............................................................................................. 25
3.14.3 Power-law distributions in financial markets ................................................ 26
3.15 Of Mice and Elephants: A Matter of Scale........................................................... 27
3.16 Power Laws and Prediction .................................................................................. 32
3.17 Conclusions........................................................................................................... 35
3.18 Multiple choice questions ..................................................................................... 36
3.19 Answers to the SAQs ............................................................................................ 42

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 1 of 48
3.1 The Aims and Objectives of Session 3

Sessions 1 and 2 laid the foundations for our studies. Set theory gives the basic language to talk
about things mathematically. Hopefully it is obvious that for making observations it is essential to
understand the distinction between nominal, ordinal, interval and ratio observations, and to
understand that there are different kinds of number systems. In order to tie down the concept of
infinity we introduced the concept of cardinality defined by one-to-one mappings.
Combined with sets, mappings lie at the heart of mathematics, since they provide very precise
ways of establishing relationships between things. In this lesson we will look at some
fundamental types of mappings that occur repeatedly in mathematical theory and its applications.
You have probably heard that complex systems are ‘non-linear’, but it’s not as simple as that.
Most complex systems have linear and non-linear subsystems. In ordinary language the term
‘linear’ can mean sequential, with one thing following another, which is different from the usual
mathematical meaning applied to mappings between number systems. So, the first task in this
session is to use mathematics to clarify things.
Mappings between sets of numbers are often called functions, and that is how the term will be
used on this course. One of the simplest class of functions f : X → Y, the ‘straight line’ functions,
have the property that f (x) = m y + c, where m and c are constants. The expression x2 means x
multiplied by itself, x2 = x  x. The term x2 is referred to as ‘x squared’, and one says that x is
‘raised to the power of 2’. x3 = x  x  x, is referred to as ‘x cubed’, and is ‘x raised to the power
of 3’. The function f (x) = x2 and f (x) = x3 are special cases of polynomials, which have the
general form f (x) = c0 + c1 x + c2 x2 + c3 x3 + … . , where c0, c1, etc are constants. An empirical
relationship of the form f(x) = cxn is called a power law. A remarkably large number of systems
across many domains exhibit power laws for some value of n.
The number n is also called an exponent. Negative exponents are allowed by the definition x-n =
1/xn. Exponents can be added by the rule xm+n = xm  xn . For example x2 + x3 = (x  x)  (x  x 
x) = (x  x  x  x  x) = x5, and subtracted by the rule xm – n = xm/xn.
The inverse of the ‘quadratic’ function f : x → x2 is the square root function g: x → √x,
encountered in Session 2 when we considered irrational numbers like √2, the square root of 2.
Thus the quadratic and square root functions are paired by being the multiplicative inverses of
each other.
The ‘exponential’ and ‘logarithmic’ functions form another very important pair of mutual
inverses. They are the target of this session because they are so widely used in complex systems
science. In particular, the logarithm allows power laws to be plotted as straight lines.
The aims of this session are to define and investigate a number of important mathematical
functions and to see how these functions are used in complex systems science. The objectives of
this session are to

 to establish the mathematics of linear systems and polynomials

 to introduce the exponential and logarithmic functions
 to see how these are combined and applied in scale free relationships and power laws
 to see some of the wide variety of power laws across many natural and social domains
 to see how these mathematical ideas are used in complex systems science

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 2 of 48
3.2 Representing functions graphically

Many observations in science involve mapping things into numbers, such as temperature in
climate change studies, share prices in economics, blood cell counts in medicine, populations in
geography, and so on. In Session 2 you saw that there are different kinds of data represented on
nominal, ordinal, interval and ratio scales. You also saw that there are different number systems
that can be used as these scales. Of particular importance are the integers (positive and negative
whole numbers including zero), the rational numbers of the form p/q where p and q are integers,
and the real numbers which contains irrational numbers such as  and √2. If the numbers are
represented by dots on a page and laid out next to each other, they soon merge together to form a
line, as shown in Figure 1.

0 1 2 3 4 5 6 7 8 9… 0 1 2 3 4 5 6 7 8 9… 0 1 2 3 4 5 6 7 8 9… 0 1 2 3 4 5 6 7 8 9…
(a) the integers (b) integers and halves (c) integers and quarters (d) all numbers make a line

Figure 1. Representing number sets by a line

The possibility of laying out numbers along a line gives a geometric interpretation to the concept
of number. Descartes (Figure 2(a)) took this a major step forward using algebraic methods to
solve geometric problems, and this enabled the development of what is called Cartesian geometry
in which two dimensional space is represented by paired axes (Descartes did not explicitly do
this). Of course the Cartesian representation generalises from two dimensions to three and then,
more abstractly, to four and higher multidimensional spaces. Platonists would say that Cartesian
geometry has always existed, and Descartes just discovered it. Be that as it may, Descartes gave
the human race an incredibly powerful way of representing systems, and the Cartesian
representation is extremely widely used in the science of complex systems.

(x, y)
y

(0, 0) x X

(a) René Descartes, 1569-1650 (b) the Cartesian plane

Figure 2 Descartes’ work using algebra in geometry led to the Cartesian way of representing space.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 3 of 48
 Usually for a mapping f : X → Y, the domain of the mapping, X, is represented by the
horizontal line and the codomain, Y, is represented by the vertical line. These lines are
called axes, the x-axis being horizontal and the y-axis being vertical.
As you have probably seen many times before, the Cartesian plane provides an excellent
graphical way of representing functions. Let f : X → Y be a function. Then the graph of the
function is defined to be the set of points G( f ) = { (x, f(x) ) | for all x belonging to X }. In the
functions considered here, the points in the graph merge together to form continuous lines.

To illustrate this, consider the function f : x → ½ x + 1. When x = –6, f(x) = ½  –6 + 1 = –3 + 1

= –2, giving the pair (–6, –2). Other pairs include (–5, –1.5), (–4, –1), (–3, –0.5), (–2, 0), (–1,
0.5), (0, 1), (2, 2), (3, 2.5) (4, 3), (5, 3.5), (6, 4). These points can be plotted on the Cartesian
plane, as shown in Figure 3(a). If more points in the graph were calculated and included in Figure
3(a) it would begin to ‘fill in’ to become a line. Thus the graph of this function is a continuous
line, as illustrated in Figure 3(b).

6 6
5 5
4 4
3 3
2 2
1 1

-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
-1 -1
-2 -2
-3 -3
-4 -4
-5 -5
-6 -6

(a) plotting points on the Cartesian grid (b) joining the points to make a line

Figure 3. Plotting a graph on the Cartesian grid

The intercept of the graph with the axes is the point at which it cuts them. So, for example, the
intercept with the x-axis is the point (–2, 0), while the intercept with the y-axis is the point (0, 1).

The gradient or slope of the line is the rate of change of f (x) with x. It can be calculated by
forming a triangle with horizontal length denoted by x and the length of the vertical side being
denoted by y. Then the gradient is defined to be y/x, and measures the slope of the line, as
shown in Figure 4.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 4 of 48
 6
5
y intercept 4

3
y = 3.5 – 1.5 = 2
x intercept 2

1
x = 5 – 1 = 4 gradient = y/x
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
-1
-2

-3

-4
-5
-6

Figure 4. The intercepts and gradient of the function f : x → ½ x + 1

SAQ 1
(a) At which (x, y) point does the graph of the function f : x → ½ x + 1 cut the vertical y-axis?
(b) At which (x, y) point does the graph of the function f : x → ½ x + 1 cut the x-axis?
(c) what is the gradient of the function f : x → ½ x + 1?
■

SAQ 2
(a) The function in SAQ 1 function illustrates a wide class of functions that have the general form
f (x) = mx + c, where m and c are constants. On the grids below draw the graphs of the functions
f (x) = mx + c for m = 1, c = 0 and for m = – 2, c = 2.
(b) what are the gradients of these two graphs?
(c) What are the x- and y-intercepts of these two graphs?
(d) What is the relationship between the y-intercept and the constant c?

6 6
5 5

4 4

3 3

2 2

1 1

-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
-1 -1

-2 -2

-3 -3

-4 -4

-5 -5
-6 -6

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 5 of 48
As you have seen in SAQ 2, the gradient of the function f (x) = mx + c is the constant m and the
intercept with the y-axis is the point (0, c). When m gets larger the slope of the line gets greater.
When m is positive the line slopes from bottom left to top right. When m is negative the line
slopes ‘backwards’, from top left to bottom right.

3.3 Linearity and linear relationships

A function f: X → Y is said to be linear if

(1) f(a + b) = f(a) + f(b) (additivity)
(2) f (  a) =  f (a), for all  (homogeneity)

SAQ 3.
Show that the real valued function f: R → R with f(x) = mx is linear for all values of m in R.
■

To show that f is linear, consider f(a + b) = m(a + b). By the distributive law of multiplication
over addition (see Session 2) m(a + b) = ma + mb = f (a) + f (b). So, f(a + b) = m(a + b). Also f (
a) = m(a) = ma = ma =  f (a). Thus f is linear.
The word linear comes from the Latin word linearis, which means created by lines 1 . If is f: R →
R is linear, then f (0) = 0. To see this, suppose f is linear. Then 0 = f (a) – f (a) = f ( 0 + a ) – f (a)
= f (0) + f (a) – f (a) = f (0). So 0 = f (0), or f (0) = 0.

6
Figure 5 shows a family of
5 f(x) = x/6 + 4 functions given by f(x) = x/6 + c.
4 These lines all have the same slope,
3 f(x) = x/6 + 2 1/6, and they only vary by the
constant c which just moves the
2
lines up and down. The underlying
1 f(x) = x/6
relationship is linear for all the
lines, since any change in x, say
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
-1
f(x) = x/6 – 2 x, results in a change of x /6 to
f (x).
-2

-3
f(x) = x/6 – 4
In general, given f (x) = mx + c, the
-4 difference due to adding x to x is
-5 f (x + x) – f (x) = m(x + x) + c –
-6 (m x + c) = mx + mx + c – mx – c
= mx, whatever the value of c.
Figure 5. A family of lines with the same slope

The variable y is said to be proportional, or directly proportional, to the variable x, written y  x

if there exists a constant k such that y = kx. The number k = y/x is called the constant of
proportionality. Proportionality is a linear relationship.

1
[Link]

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 6 of 48
 3.4 Quadratic Functions

The quadratic function has the general form f(x) = mx2 + c. The simplest case for m = 1 and c = 0
is shown in Figure 6(a), where its graph contains the points (-3, 9), (-2, 4), (1-, 1), (0, 0), (1, 1),
(2, 4), and (3, 9). The graph is drawn by interpolation between these points.
f(x) = x2 – 5 f(x) = x2 /4
12 6
11 5
10 4
9 3
8 2
7 1
6
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
5 -1
4 -2
3 -3
2 -4
1 -5
-6
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6

(a) the graph of f(x) = x2 (b) changing m squashes or stretches the graph
and changing c moves it up and down
Figure 6. Quadratic functions

The effect of changing c is to move the graph up and down, and the effect of changing m is to
make the U-shaped graph flatter or narrower, as shown in Figure 6(b). How could the graph be
moved from side to side? E.g. how can the graph in Figure 6(a) be moved up to the right?

12
Figure 7 shows how this can be done. Here
11
it is desired that the lowest point in the
10 graph is (3, 1). To move the graph up one
9 simply changes c, in this case c = 1. To
8
move the graph to the right, one effectively
subtracts 3 from x before squaring it Thus
7
the function is f(x) = (x – 3)2 + 1.
6
5 This graph of this function includes the
4
points (0, 10), (1, 5), (2, 2), (3, 1), (4, 2),
(5, 5), and (6, 10) as shown.
3

2 This function can be rewritten as

1 f(x) = (x – 3)2 + 1 = x2 – 6x + 9 + 1
= x2 – 6x + 10.
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
Figure 7. Shifting the graph up and to the right

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 7 of 48
 SAQ 4

Show that the function f(x) = m(x – k)2 + c has the shape of a quadratic function, and has its
lowest point ( k, c). What is its lowest point?
■

3.5 Cubic Functions, Quartic Functions and Polynomials

The simplest cubic function has the form f(x) = x3 and has the graph shown in Figure 8(a). In this
case the graph has a point of inflection at x = 0, where (left to right) it becomes horizontal before
turning up again. More generally cubic functions cubic functions have the form f(x) = ax3 + bx2 +
cx + d. The points at which a function becomes (momentarily) horizontal are called extreme
points. The curve shown in Figure 7(b) has two extreme points: at x = 0 there is a local maximum,
and at x = 2 there is a local minimum. In general if b2 = 3ac there is one critical point, a point of
inflection (in Figure 8(a), a = 1, b = 0, c = 0, and d = 0, so b2 = 02 = 3  1  0 = 3ac), if b2 > 3ac
then there are two critical points (in Figure 8(b) a = 1, b = 3, c = 0 and d= 0, so b2 = 32 > 3  1 
0 = 3ac), and if b2 < 3ac there are no critical points (we give no example of this).

60 60
50 50
40 40
30 30
20 20
10 10

-6 -5 -4 -3 -2 0 1 2 3 4 5 6 -6 -5 -4 -3 -2 0 1 2 3 4 5 6
-10 -10
-20 -20
-30 -30
-40 -40
-50 -50
-60 -60

(a) f(x) = x3 (b) f(x) = x3 - 3x2

Figure 8. Cubic functions

SAQ 5.

Give an example of a cubic function that has no critical points. Draw its graph to illustrate this.
■

As you may have guessed, a quartic function involves terms of the order of x4. The simplest
quartic function is f(x) = x4. The following snippet is taken from Wikipedia 2 .

2
[Link] accessed 28th May 2007.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 8 of 48
Having seen the functions f(x) = mx + c, f(x) = x2, f(x) = x3, and f(x) = x4, and some combinations
of them, the idea of a polynomial that combines them in the most general way is quite natural:

f(x) = a0 + a1 x + a2 x2 + a3 x3 + a4x4 + a5 x5 + a6 x6 + …

Thus, for example, f(x) = a0 + a1 x is a straight line function, where a1 is the gradient and a0 is the
intercept on the y-axis (a0 is the constant c and a1 is the gradient m in f(x) = mx + c).

In the term xn the number n is called an exponent. The degree, of a polynomial is the highest
exponent. Thus the degree of a quadratic is 2, the degree of a cubic is 3, and the degree of a
quartic is 4.

Polynomials play an important role in mathematics. Although for even modest degrees, xn, can
get very large very quickly. Ehen the signs on the ai alternate many interesting things can happen.

Even for a small number like n = 10, xn is very large. For example xn = 1011 when x =10, which is
a hundred billion. Senator Everett Dirksen is most often remembered for the quip: “A billion
here, a billion there, pretty soon, you're talking real money”. He made similar remarks but
probably not that exact one. Polynomials of high degree are rarely encountered in models of real
systems, either because they don’t exist or because we do not have the instruments to isolate
them.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 9 of 48
3.6 Scale Invariance

A function f(x) = axk is scale invariant if multiplying x by a constant, c, causes only a

proportionate scaling of the function itself, f(cx)  f(x):

f(cx) = a(cx)k = ck f(x)

In other words, scaling by a constant, c, simply multiplies the original power-law relation by the
constant ck. Thus scaling the function’s argument by c changes the constant of proportionality, ck,
but does not change the shape of the function. This is illustrated below for the function f(x) = x2.
As can be seen, the shape of the function is independent of the scale.

x -5 -4 -3 -2 -1 0 1 2 3 4 5
x2 , c = 1 25 16 9 4 1 0 1 4 9 16 25
(cx)2, c = 2 100 64 36 14 4 0 4 16 36 64 100
(cx)2, c =10 2500 1600 900 400 100 0 100 400 900 1600 2500

Table 2. The functions f(cx) for c = 1, 2, and 10

25 100 2500

20 80 2000

15 60 1500

10 40 1000

5 20 500

0 0 0
-5 -4 -3 -2 -1 0 1 2 3 4 5 -5 -4 -3 -2 -1 0 1 2 3 4 5 -5 -4 -3 -2 -1 0 1 2 3 4 5

(a) c = 1 (b) c = 2 (c) c= 10

Figure 9. Scale invariance: the graphs of the functions f(cx) for c = 1, 2, and 10

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 10 of 48
3.7 Measuring the slopes of functions

Figure 10 shows the function f (x) = x2. At three points are drawn the tangents to the curve. These
are straight lines that just touch the curve at the point. Each of the tangent lines has a slope or
gradient, given by the increase in f (x) for an increase in x.
In mathematics differences are often denoted by the Greek symbol , pronounced ‘delta’. Thus
x, read as “delta-x” is a horizontal distance on the Cartesian grid. Similarly  f (x), pronounced
‘delta-f-of-x’ is a vertical distance on the Cartesian grid. So, for example, the line going through
the point (5, 25) increases an amount  f (x) = 20 for an increase x = 2. The gradient at the point
(5, 25) is calculated as  f (x) /  x = 20/2 = 10.

35

33

31

29

27

25  f ( x) =20
23

21

19

17

15
13 x = 2
11
 f ( x) = 12 9

5
3
x = -2  f ( x) = 4
1
-6 -5 -4 -3 -2 -1 0 3 4 5 6
x = 2
Figure 10. Calculating the gradient at points on a curve.

SAQ 6. The table below shows the gradients for f (x) = x2. Complete it for x = 2, 3, 4 by placing a
ruler against the curve at those points, drawing the best tangent you can, and measuring the rate
of change of f (x) with x. Draw the graph of x plotted against the gradient at x. What is the
function of this graph?

x -5 -4 -3 -2 -1 0 1 2 3 4 5
x -2 -2 -2 -2 -2 2 2 2
 f (x) 20 16 12 8 4 0 4 20
Gradient at x
-10 -8 -6 -4 -2 0 2 10
=  f (x)/  x
■

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 11 of 48
In answer to SAQ 6, let the gradient at x be denoted by the symbol g(x). The gradients of the
function f(x) = x2 at x = 2 is g(x) = 4, at x = 3 is g(x) = 6, and at x = 4 it is g(x) = 8. This completes
the table in SAQ 6. The graph for this table is plotted in this figure. It is a straight line of the form
g(x) = m x + c. When x = 0, g(x) = 0, so c= 0. The gradient of this graph can be calculated from
the triangle shown, which has g(x)/ x = 8/4 = 2. Therefore the function is g(x) = 2x.
gradient
12

10

8 g(x) = 8

x = 4
2

-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
x

-2

-4

-6

-8

-10

-12

Figure 11. The graph of the gradient, g(x), of the function f(x) = x2.

The gradients of functions play a very important role in science because they give a way of
representing rates of change. Figure 12 shows an inclined plane used to repeat Galileo’s
experiment in which he rolled a ball down an inclined plane.

water
clock

shalf, thalf
t slength, tlength
thalf
shalf s

Figure 12. An experimental setup to repeat Galileo’s inclined plane experiments

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 12 of 48
Until the fifteenth and sixteenth centuries, science was dominated by the ideas of Aristotle,
including the idea that heavy bodies fall faster than light bodies. It is said that Galileo (1564-
1642) sought to show this was not true by experimental means, dropping objects of different
masses from the Leaning Tower of Pisa 3 and observing them hit the ground at the same time. He
further experimented with motion using an inclined plane, similar to that in Figure 12.

In the sixteenth century there were no mechanical clocks able to measure time in fractions of a
second, so Galileo contrived an ingenious experiment to investigate the nature of motion. The use
of an inclined plane slowed down the motion of the ball compared to dropping it. Galileo used a
water clock to measure time intervals. Water clocks work by allowing water to flow at a constant
rate out of a tube at the bottom of a large vessel, to be collected in a measuring cylinder. The flow
of water is started when the ball is released, and ended when the ball has travelled the prescribed
distance. Using this setup Galileo showed that the distance travelled was proportional to the time
squared, d  t2, or d = k t2 where k is a constant.

If we plot Galileo’s curve d= k t2, the gradient of the curve at any point is the rate of change of
position with time, i.e. the velocity at time t of the sphere rolling down the inclined plane, d/t.
But from our previous discussion it can be shown that the gradient of the function d= kt2, at a
point t is 2kt, so it can be calculated that the velocity of the sphere at time t is v = 2kt.

distance
d = kt2

d d

velocity at t is d /t = 2kt

t
time
t

Figure 13. The velocity of a sphere rolling down an inclined plane at time t is 2kt, with k a constant/

In a subsequent session we will return to these ideas. For the moment the important point is that
this mathematical formalism can be applied to the physics of motion. Furthermore the velocity at
t, v(t) = 2kt is predictive. In principle the constant k can be determined by calibration, and after
this the precise velocity of the sphere rolling down the inclined can be predicted for any practical
time interval.

This kind of prediction is a scientific triumph for physics, and for four centuries it has dominated
the view of what it means to make a scientific prediction. However, such predictions are usually
impossible in complex systems, and an important part of the new science is trying to understand
what it means to make a prediction about the behaviour of systems whose dynamics are not
determined in such a simple way.

3
There is no evidence that he did this experiment from the Leaning Tower.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 13 of 48
3.8 The Exponential and Logarithmic Functions

In the expression ba is the number b multiplied by itself a time. The number b is called the base
and the number a is called the exponent. The polynomials just considered are of the form xn,
where the exponent is fixed and the base varies.

In this section we will consider functions in which the base is fixed and the exponent varies. For
example, Figure 14(a) shows exponential functions with fixed bases ½, and 2, and Figure 14(b)
shows exponential functions with fixed bases 1/3 and 3.

f(x) = 1/2x f(x) = 2x f(x) = 1/3x f(x) = 3x

24 24

20 20

16 16

12 12

8 8

4 4

x x
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6

(a) f(x) = 1/2x and f(x) = 2x (a) f(x) = 1/3x and f(x) = 3x

Figure 14. Exponential functions

Large numbers are often expressed as powers of ten, where the term ‘power’ is another word for
the exponent. For example ten to the power of two is written as 102, which is one followed by two
zeros, or 100. As noted previously, the exponents follow the simple rule 10m  10n = 10m+n when
two numbers are multiplied. For example 102  103 = 100  1000 = 100000 = 105.

SAQ 7

(i) What is the base of the number 106?

(ii) What is the exponent of 102?
(iii) Is it true that ab = ba ?
(iv) Which is larger, 510 or 105 ? (Hint: 510 = 5  5  5  5  5  5  5  5  5  5 = 9765625)

As can be seen by comparing 510 or 105, varying the exponent rather than the base results in
functions that ‘grow’ much faster. For example, a googol is the number 1 followed by one
hundred zeros, and it can be written 10100. It is a huge number. Even so it is ten times smaller than
10101, and a hundred times smaller than 10102, and … a billion times smaller than 10109.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 14 of 48
 The exponential function, f(x) = ex, , where e  2.71828, is distinguished by the fact that its rate of
growth at x is exactly equal to the value ex of the function at x. The number e is called Euler’s
Number, after Leonid Euler. The exponential function is very important in mathematics and its
properties will be further investigated in subsequent lessons.
Consider the function fb(x) = bx for a fixed base number, b. We are interested in the inverse
function of f , say gb with gb(bx) = x, so that gb ° fb(x) = x. The function gb is called the logarithm
and is also written as x = logb(fb(x)).
fb
x bx exponential function

x logb
bx logarithmic function

This may appear a very mysterious thing to do, so first let’s try to see what the logarithmic
function might look like.
For simplicity, let the base be 10. Let log10 :  →  with 10 = . To draw a graph of the function
it is necessary to get pairs of  - values. Rather than start with values of  and calculating values
of  it is easier to work the other way round, starting with values of  and calculating the values
of  mapped to them. In other words, it is easiest to complete the following table.

 = 10
 = log10 () -6 -5 -4 -3 -2 -1 0 1 2 3

When  = -6,  = 10-6 = 1/1,000,000 which is very close to zero. The same is true for the other
values of  associated with  = -5, -4, -3, and -2. For  = -1,  = 10-1 = 1/10 = 0.1. For  = 0,
 = 100 = 1. For  = 1,  = 101 = 10. For  = 2,  = 102 = 100. For  = 3,  = 103 = 1000. For
values of  greater than 3, the associated value of  exceeds a thousand, which is quite large. So
the completed table is:
 = 10 .000,001 .00001 .0001 .001 .01 .1 1.0 10 100 1000
 = log10 () -6 -5 -4 -3 -2 -1 0 1 2 3

The graph of this function is drawn in Figure 15. As can be seen the graph is almost vertical
between  = 0 and  = 0.1, and it is almost horizontal after  = 10.
4
3
2
1
0
-1
-2
-3
-4
-5
-6
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

Figure 15. The graph of the function  = log10 

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 15 of 48
Logarithms with base 10 are called common logarithms. Of the many other possible choice of
base, the most common are the number 2, which has applications in information theory, and
Euler’s number, e  2.71828. Logarithms to the base e are called natural logarithms and are often
written ln(x). The logarithmic functions to these bases are shown in Figure 16.

Figure 16. A comparison of logarithms to different base(Source: Wesstein, 1999) 4 .

Exponentials and logarithms have a number important related properties, as shown in the table:

bx  bx’ = bx+x’ logb( x  x’ ) = log(x) + log(x’)

b b =b
x x’ x - x’
logb( x  x’ ) = log(x) – log(x’)
(b ) = b
x n nx
logb( xn) = n logb(x)

Conversions between logarithmic bases are easy. For example, to convert y = log10 x to y’ = loge
x, we have 10y = x and ey’ = x, so 10y = ey’. Taking logs of both sides gives log10 10y = log10 ey’.
Thus y log10 10 = y’ log10 e, or y  1 = y’ log10 e, whereby y/log10e = y’ or y’= log10x/log10e, or
loge x = log10x/log10e. In general:
logb x
log x =
logb

3.9 Logarithms, polynomials, and straight lines

It may seem astonishing, but the logarithmic functions in Figures 15 and 16 allow power laws to
be represented by straight lines. Consider the function y = x2. Take logarithms of both sides, to get
log y = log x2. This gives log y = 2 log x. More generally, consider y = xn. Take logarithms of each
side, giving log y = log xn, so that log y = n log x. Plotting the logarithm of y against the logarithm
of x gives a straight line with gradient n.

4
Image taken from Weisstein, Eric W. "Logarithm." From MathWorld--A Wolfram Web Resource. 1999,
[Link] (viewed 3 June 2007)

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 16 of 48
Most computers have a built-in calculator. In Windows it appears as shown in Figure 17. To get,
say, the logarithm of 9, you enter the number 9, press the ‘log’ key (circled in Figure 17) and you
get the answer 0.95424250943932487459005580651023, which is about 0.95 when rounded.

Figure 17. Using a computer calculator to find a logarithm.

SAQ 8

The ‘log’ key on the calculator gives logarithms base 10, while the ‘ln’ key gives logarithms to
base e. Use the calculator to show that log10 1234 = loge 1234 / loge 10.
■

SAQ 9. Complete the following table and plot the graphs y = x2 and log y = log y2 on the grids.

x 1 2 3 4 5 6 7 8 9 10
x2 1 4 9 16 25 36 49 64 81 100
log x 0 0.301 0.477 0.602 0.670 0.788
log x2 0 0.602 0.954 1.204 1.340 1.556

10 2.0
0
1.8
90
1.6
60
1.4
70
2
1.2
x log x2
60
1.0

50 0.8

40 0.6

30 0.4

0.2
20

0.0
0 1 2 3 4 5 6 7 8 9 0 0.2 0.4 0.6 0.8 1.0
x log x
■

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 17 of 48
3.10 Graphing on logarithmic paper
SAQ 10.
Instead of calculating the logarithms, log-log graph paper allows you to enter the data directly.
Plot the data below on the log-log graph paper. It has two vertical cycles. Since x only goes from
1 to 10 it requires only one cycle. However, x2 takes values between 0 and 100, and requires two
cycles. The upper cycle is ten times the lower cycle. So, for example, the number 16 lies between
1 ( 10) and 2 ( 20) in the position shown. The position for 25 is also shown.

x 1 2 3 4 5 6 7 8 9 10
x2 1 4 9 16 25 36 49 64 81 100

upper cycle

25

16

x2

lower cycle

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 18 of 48
3.11 Zipf’s Law

In his 1949 book Human Behaviour and the Principle of Least-Effort, George Kingsley Zipf gave
an argument that languages should be structured according to a principle of least effort to allow
the most efficient communication, and illustrated his argument by showing that languages obey
the power law frequency(word) = constant  rank(word)–1. The frequency is the number of times
a word appears in a corpus (text collection), and the words are ranked according to frequency, so
that the most common word has rank 1, the next most common rank 2, and so on.

The Brown Corpus of Standard American English was compiled by Henry Kucera and W. Nelson
Francis at Brown University. In 1967 they published their classic work Computational Analysis
of Present-Day American English which provided basic statistics on what is known today simply
as the Brown Corpus. The Brown Corpus was a carefully compiled selection of current American
English, totalling about a million words drawn from a wide variety of sources. One interesting
result is that even for quite large samples, graphing words in order of decreasing frequency of
occurrence shows a hyperbola: the frequency of the nth most frequent word is roughly
proportional to 1/n. 5

Table 1 lists a selection of words with their ranks and frequencies taken from the Brown Corpus,
which comprises 1,015,945 words with 47,218 unique words. The first ranked word is “the” with
a frequency of 69970. Suppose that the frequency and rank are related as f(n) = k/ na. Take
logarithms of both sides to obtain log(f(n)) = log(k/na) = log( k) – log(na) = log(k) – a log(n). This
is a straight line with gradient -a and intercept log(k). To illustrate this, Table 1 lists some of the
words from the Brown corpus, together with their ranks and frequencies, and logarithms of the
ranks and frequencies.

rank freq word log rank log freq rank freq word log rank log freq

1. 69970 the 0.00 4.85 80. 1237 over 1.90 3.09

2. 36410 of 0.30 4.56 90. 1016 before 1.95 3.01
3. 28854 and 0.48 4.46 100. 895 down 2.00 2.95
4. 26154 to 0.60 4.42 200. 426 took 2.30 2.63
5. 23363 a 0.70 4.37 300. 308 matter 2.48 2.49
6. 21345 in 0.78 4.33 400. 203 court 2.60 2.31
7. 10594 that 0.85 4.03 500. 191 finally 2.70 2.28
8. 10102 is 0.90 4.00 600. 163 doing 2.78 2.21
9. 9815 was 0.95 3.99 700. 143 industrial 2.85 2.16
10. 9542 he 1.00 3.98 800. 128 forms 2.90 2.11
20. 5180 I 1.30 3.71 900. 115 slowly 2.95 2.06
30. 3619 they 1.48 3.56 1000. 104 current 3.00 2.02
40. 2670 their 1.60 3.43 2000. 55 components 3.30 1.74
50. 2199 if 1.70 3.34 3000. 36 sympathy 3.48 1.56
60. 1789 them 1.78 3.25 4000. 25 terror 3.60 1.40
70. 1400 may 1.85 3.15 5000. 19 ladder 3.70 1.28

Table 1. Words and their frequencies in the Brown Corpus.

5
adapted from [Link] (accessed 9th June 2007)

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 19 of 48
SAQ 11.

(i) Plot the logarithm of the rank against the logarithm of the frequency for the data in Table 1.
What is the equation of the line?

5.0

4.0

log
frequency

3.0

2.0

1.0
0.0 1.0 2.0 3.0 4.0
log rank

(ii) Do the data support Zipf’s Law that f(n) = k/n a ? If so what are the values of k and a?
■

SAQ 12. The ranks and populations of twelve American cities are given below. Complete the
table and plot the log of the rank against the log of the population on the grid below. What
relationship, if any, was there between the rank and the population in 1990?

City Rank Population 1990 Log rank Log Population

New York 1 7,322,564 0.00 6.86
Detroit 7 1,027,974 0.85 6.01
Baltimore 13 736,014 1.11 5.87
Washington D.C. 19 606,900 1.28 5.78
New Orleans 25 496,938 1.40 5.70
Kansas City 31 434,829 1.50 5.64
Virginia Beach 37 393,089 1.57 5.59
Toledo 49 332,089
Arlington Texas 61 261,721
Baton Rouge 73 219,531
Hialeah 85 188,008
Bakersfield 97 174,820

Table 2. Relationship between cities ranked by population

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 20 of 48
■

Wentian Li writes “Probably few people pay attention to a comment by Miller in his preface to
Zipf’s book, that randomly generated texts, which are perhaps the least interesting sequences and
unrelated to any other scaling behaviors, also exhibit Zipf’s law. What he said was that Zipf’s law
is not exclusive for English or any other natural languages. Miller did not give a proof of his
statement, and it is the purpose of this short paper 6 to provide a very simple proof that random
texts do indeed exhibit Zipf’s-law-like word frequency distribution.”

3.12 Power Laws 7

A power law is any polynomial relationship that exhibits the property of scale invariance. The
most common power laws relate two variables and have the form
y(x) = axk + o(xk)
where a and k are constants, and o(xk) is an asymptotically small 8 function of x. Here, k is
typically called the scaling exponent, denoting the fact that a power-law function (or, more
generally, a kth order homogeneous polynomial) satisfies the criteria where c is a constant. That is,
scaling the function's argument changes the constant of proportionality as a function of the scale
change, but preserves the shape of the function itself. This relationship becomes more clear if we
take the logarithm of both sides (or, graphically, plotting on a log-log graph)
log( f(x)) = k log x + log a

Note that this has the form of a linear relationship with slope k, and scaling the argument induces
a linear shift (up or down) of the function, and leaves both the form and slope k unchanged.

6
[Link]
7
this section is adapted from [Link]
8
asymptotically small here means that as k gets large, o(xk) becomes small compared to axk.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 21 of 48
Power-law relations characterize a staggering number of natural patterns, and it is primarily in
this context that the term power law is used rather than polynomial function. For instance,
inverse-square laws, such as gravitation and the Coulomb force are power laws, as are many
common mathematical formulae such as the quadratic law of area of the circle. Also, many
probability distributions have tails that asymptotically follow power-law relations, a topic that
connects tightly with the theory of large deviations (also called extreme value theory), which
considers the frequency of extremely rare events like stock market crashes, and large natural
disasters.
Scientific interest in power law relations, whether functions or distributions, comes primarily
from the ease with which certain general classes of mechanisms can generate them. That is, the
observation of a power-law relation in data often points to specific kinds of mechanisms that
underly the natural phenomenon in question, and can often indicate a deep connection with other,
seemingly unrelated systems (for instance, see both the reference by Simon and the subsection on
universality below). The ubiquity of power-law relations in physics is partly due to dimensional
constraints, while in complex systems, power laws are often thought to be signatures of hierarchy
and robustness. A few notable examples of power laws are the Gutenberg-Richter law for
earthquake sizes, or structural self-similarity of fractals, and scaling laws in biological systems.
Research on the origins of power-law relations, and efforts to observe and validate them in the
real world, is extremely active in many fields of modern science, including physics, computer
science, linguistics, geophysics, sociology, economics and more.

3.13 The Pareto Distribution

The Pareto distribution, named after the Italian economist Vilfredo Pareto, is a power law
probability distribution that coincides with social, scientific, geophysical, and many other types of
observable phenomena. Outside economics it is at times referred to as the Bradford distribution.

Figure 18. Pareto Graph, top tail of estimated world distribution of wealth 9
Pareto originally used this distribution to describe the allocation of wealth among individuals
since it seemed to show rather well the way that a larger portion of the wealth of any society is
owned by a smaller percentage of the people in that society. This idea is sometimes expressed
more simply as the Parteo Principle or the “80-20 rule” which says that 20% of the population
owns 80% of the wealth.
It can be seen from the probability density function (PDF) graph on the right, that the
“probability” or fraction of the population f(x) that owns a small amount of wealth per person (x)

9
Source: The World Distribution of Household Wealth, James B. Davies, Susanna Sandstrom, Anthony Shorrocks, and
Edward N. Wolff, 5 December 2006. [Link]
5-12-2006/[Link]

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 22 of 48
is rather high, and then decreases steadily as wealth increases. This distribution is not limited to
describing wealth or income distribution, but to many situations in which an equilibrium is found
in the distribution of the "small" to the "large".

Probability density function Cumulative distribution function

(a) Pareto probability density functions for various (b) Pareto cumulative didstribution functions for
k with xm = 1. The horizontal axis is the x various k with xm = 1. The horizontal axis is the x
parameter. Pr( X > x ) = ( x/xm) – k parameter.

Figure 19. The Pareto Distribtion

The following examples are sometimes seen as approximately Pareto-distributed:

 Frequencies of words in longer texts (a few words are used often, lots of words are used
infrequently)
 The sizes of human settlements (few cities, many hamlets/villages)
 File size distribution of Internet traffic which uses the TCP protocol (many smaller files,
few larger ones)
 Clusters of Bose-Einstein condensate near absolute zero
 The values of oil reserves in oil fields (a few large fields, many small fields)
 The length distribution in jobs assigned supercomputers (a few large ones, many small
ones)
 The standardized price returns on individual stocks
 Sizes of sand particles
 Sizes of meteorites
 Numbers of species per genus (There is subjectivity involved: The tendency to divide a
genus into two or more increases with the number of species in it)
 Areas burnt in forest fires

If X is a random variable with a Pareto distribution, then the probability that X is greater than
some number x is given by
Pr( X > x ) = ( x/xm) – k
for all x ≥ xm, where xm is the (necessarily positive) minimum possible value of X, and k is a
positive parameter. The family of Pareto distributions is parameterized by two quantities, xm and

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 23 of 48
k. When this distribution is used to model the distribution of wealth, then the parameter k is called
the Pareto index.
Figure 19(a) shows the Pareto distribution for xm = 1 and k = 1, 2, 3. The functions are therefore
Pr( X > x, xm = 1, k = 1) = 1/x, Pr( X > x, xm = 1, k = 2) = 1/x2, and Pr( X > x, xm = 1, k = 1) = 1/x3.
Thus in these Pareto distributions are well-known power laws.
Since you have not yet encountered probability or integration on this course the details of the
following are beyond the scope of this session. They are included for completeness and need not
be studied in depth. The probability density function (Fig. 19(b)) is
f(x; k, xm) = k ( xmk / xk+1) for x  xm.
This means the probability of a particular x given k and xm is
f(x; k, xm) = (k/x)  (xmk / xk) = (k/x)  (xm/ x)k = (k/x)  (x / xm)–k = (k/x)  Pr( X > x ).

Pareto distributions are continuous probability distributions. Zipf’s law also sometimes called the
zeta distribution, may be thought of as a discrete counterpart of the Pareto distribution.

3.14 Examples

Power laws are ubiquitous, occurring in many different kinds of system, as illustrated here.

3.14.1 Earthquakes 10
In seismology the Gutenberg-Richter law expresses the relationship between the magnitude and
total number of earthquakes in any given region and time period:
log N = A – bM
or
N = 10 A – bM
Where N is the number of events in a given range, M is a magnitude range, and Aand b are
constants. The relationship was first proposed by Charles Francis Richter and Beno Gutenberg.
The relationship is surprisingly robust and does not vary significantly from region to region or
over time.
The constant b is typically equal to 1.0. This means that for every magnitude 4.0 event there will
be 10 magnitude 3.0 quakes and 100 magnitude 2.0 quakes. A notable exception is during
earthquake swarms when the b-value can become as high as 2.5 indicating a large proportion of
small quakes to large ones. A b significantly different from 1.0 may suggest a problem with the
data set; e.g. it is incomplete or contains errors in calculating magnitude. The "roll off" of the b-
value is an indicator of the completeness of the data set at the low magnitude end.
The A-value is of less scientific interest and simply indicates the total seismicity rate of the
region. Modern attempts to understand the law involve theories of self-organized criticality or
self-similarity

10
This section is taken from [Link]

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 24 of 48
3.14.2 Allometric laws11
An allometric law describes the relationship between two attributes of living organisms, and is
usually expressed as a power law
y  xa
or in a logarithmic form
log y ~ a log x
where ~ means ‘approximately equal to’, a is the scaling exponent of the law. Methods for
estimating this exponent from data tend to involve a particular kind of ‘principle component
analysis’. For example, Figure 20 shows an allometric law of cruising speed versus body mass.

Figure 20 The allometric law of cruising speed versus body mass

Some examples of allometric laws:

 Keleiver’s Law: the proportionality between metabolic rate q0 and body mass M raised to
the power 3/4
q0 ~ M 3/4
 the proportionality between breathing and heart beating times t and body mass M raised
to the power 1/4
t ~ M 1/4
 mass transfer contact area A and body mass M:
A ~ M 7/8
 the proportionality between the optimal cruising speed Vopt of flying bodies (insects,
birds, airplanes) and body mass M in kg raised to the power 1/6
Vopt ~ 30M 1/6 ms-1

11
taken from [Link]

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 25 of 48
3.14.3 Power-law distributions in financial markets
The following is based on a paper in Nature in 2003 12 . Define pt as the price of a given stock and
the stock price ‘return’ rt as the change of the logarithm of stock price in a given time interval t;
rt  ln pt – ln pt–t . The probability that a return has an absolute value larger than x is found
empirically to be (see Figure 21)
–r
P(| rt | > x ) ~ x (1)
with r  3.  is the Greek symbol zeta.  means ‘approximately equal to’. | rt | means the absolute
value of rt , i.e. rt without its sign, so that | rt | is always positive.

Figure 21.

Cumulative distributions of the normalized

15-min absolute returns of the 1,000 largest
companies in the ‘Trades and Quotes’
database for the 2-yr period 1994–1995.

Empirical studies also show that the distribution of trading volume Vt obeys a similar power law:
– v
P(Vt > x) ~ x (2)
with  v  1.5 ,while the number of trades Nt obeys:
– N
P(Nt > x ) ~ x
with zN  3.4.

The ‘inverse cubic law’ of equation (1) is rather ‘universal’, holding over as many as 80 standard
deviations for some stock markets, with t ranging from one minute to one month, across
different sizes of stocks, different time periods, and also for different stock market indices.
Moreover, the most extreme events—including the 1929 and 1987 market crashes—conform to
equation (1), demonstrating that crashes do not appear to be outliers of the distribution. Gabaix et
al tested the universality of equations (2) and (3) by analysing the 35 million transactions of the
30 largest stocks on the Paris Bourse over the 5-yr period 1994–1999. This analysis shows that
the power laws (2) and (3) obtained for US stocks also hold for a distinctly different market,
consistent with the possibility that equations (2) and (3) are as universal as equation (1).

12
‘A theory of power-law distributions in financial market fluctuations’, Xavier Gabaix, Parameswaran Gopikrishnan,
Vasiliki Plerou, H. Eugene Stanley, Nature, 423, 267-270, 15 MAY 2003.
[Link]

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 26 of 48
3.15 Of Mice and Elephants: A Matter of Scale 13

Below is a reproduction of an article from the New York Times. It is included because it
illustrates power laws, and includes comments from distinguished scientists who have studied
them. The article suggests that power laws are fundamental and can be of practical use.

By GEORGE JOHNSON
January 12, 1999

Scientists, intent on categorizing everything around them, sometimes divide themselves into the
lumpers and the splitters. The lumpers, many of whom flock to the unifying field of theoretical
physics, search for hidden laws uniting the most seemingly diverse phenomena: Blur your vision
a little and lightning bolts and static cling are really the same thing.
The splitters, often drawn to the biological sciences, are more taken with diversity, reveling in the
34,000 variations on the theme spider, or the 550 species of conifer trees.

But there are exceptions to the rule. When two biologists and a physicist, all three of the lumper
persuasion, recently joined forces at the Santa Fe Institute, an interdisciplinary research center in
northern New Mexico, the result was an advance in a problem that has bothered scientists for
decades: the origin of biological scaling. How is one to explain the subtle ways in which various
characteristics of living creatures -- their life spans, their pulse rates, how fast they burn energy --
change according to their body size?
As animals get bigger, from tiny shrew to huge blue whale, pulse rates slow down and life spans
stretch out longer, conspiring so that the number of heartbeats during an average stay on Earth
tends to be roughly the same, around a billion. A mouse just uses them up more quickly than an
elephant.

13
Taken from [Link]

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 27 of 48
Mysteriously, these and a large variety of other phenomena change
with body size according to a precise mathematical principle called
quarter-power scaling. A cat, 100 times more massive than a mouse,
lives about 100 to the one-quarter power, or about three times, longer.
(To calculate this number take the square root of 100, which is 10 and
then take the square root of 10, which is 3.2.) Heartbeat scales to mass
to the minus one-quarter power. The cat's heart thus beats a third as
fast as a mouse's.
The Santa Fe Institute collaborators -- Geoffrey West, a physicist at
Los Alamos National Laboratory, and two biologists at the University
of New Mexico, Jim Brown and Brian Enquist -- have drawn on their
different kinds of expertise to propose a model for what causes certain
kinds of quarter-power scaling, which they have extended to the plant
kingdom as well.
In their theory, scaling emerges from the geometrical and statistical
properties of the internal networks animals and plants use to distribute
nutrients. But almost as interesting as the details of this model, is the
collaboration itself. It is rare enough for scientists of such different
persuasions to come together, rarer still that the result is hailed as an
important development.
"Scaling is interesting because, aside from natural selection, it is one
of the few laws we really have in biology," said John Gittleman, an
evolutionary biologist at the University of Virginia. "What is so
elegant is that the work makes very clear predictions about causal Source: Dr. Geoffrey West, Los
mechanisms. That's what had been missing in the field." Alamos National Laboratory
Brown said: "None of us could have done it by himself. It is one of
the most exciting things I've been involved in."
It might seem that because a cat is a hundred times more massive than a mouse, its metabolic rate,
the intensity with which it burns energy, would be a hundred times greater -- what
mathematicians call a linear relationship. After all, the cat has a hundred times more cells to feed.
But if this were so, the animal would quickly be consumed by a fit of spontaneous feline
combustion, or at least a very bad fever. The reason: the surface area a creature uses to dissipate
the heat of the metabolic fires does not grow as fast as its body mass. To see this, consider (like a
good lumper) a mouse as an approximation of a small sphere. As the sphere grows larger, to cat
size, the surface area increases along two dimensions but the volume increases along three
dimensions. The size of the biological radiator cannot possibly keep up with the size of the
metabolic engine.
If this was the only factor involved, metabolic rate would scale to body mass to the two-thirds
power, more slowly than in a simple one-to-one relationship. The cat's metabolic rate would be
not 100 times greater than the mouse's but 100 to the power of two-thirds, or about 21.5 times
greater.
But biologists, beginning with Max Kleiber in the early 1930s, found that the situation was much
more complex. For an amazing range of creatures, spanning in size from bacteria to blue whales,
metabolic rate scales with body mass not to the two-thirds power but slightly faster -- to the three-
quarter power.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 28 of 48
Evolution seems to have found a way to overcome in part the limitations imposed by pure
geometric scaling, the fact that surface area grows more slowly than size. For decades no one
could plausibly say why.
Kleiber's law means that a cat's metabolic rate is not a hundred or 21.5 times greater than a
mouse's, but about 31.6 (100 to the three-quarter power). This relationship seems to hold across
the animal kingdom, from shrew to blue whale, and it has since been extended all the way down
to single-celled organisms, and possibly within the cells themselves to the internal structures
called mitochondria that turn nutrients into energy.
Long before meeting Brown and Enquist, West was interested in how scaling manifests itself in
the world of subatomic particles. The strong nuclear force, which binds quarks into neutrons,
protons and other particles, is weaker, paradoxically, when the quarks are closer together, but
stronger as they are pulled farther apart -- the opposite of what happens with gravity or
electromagnetism.
Scaling also shows up in Heisenberg's Uncertainty Principle: the more finely you measure the
position of a particle, viewing it on a smaller and smaller scale, the more uncertain its momentum
becomes.
"Everything around
us is scale
dependent," West
said. "It's woven
into the fabric of the
universe."
The lesson he took
away from this was
that you cannot just
naively scale things
up. He liked to
illustrate the idea
with Superman. In
two panels labeled
"A Scientific
Explanation of
Clark Kent's
Amazing Strength,"
from Superman's
first comic book
appearance in 1938,
the artists invoked a
scaling law: "The
lowly ant can
support weights
hundreds of times its own. The grasshopper leaps what to man would be the space of several city
blocks." The implication was that on the planet Krypton, Superman's home, strength scaled to
body mass in a simple linear manner: If an ant could carry a twig, a Superman or Superwoman
could carry a giant ponderosa pine.
But in the rest of the universe, the scaling is actually much slower. Body mass increases along
three dimensions, but the strength of legs and arms, which is proportional to their cross-sectional
area, increases along just two dimensions. If a man is a million times more massive than an ant,

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 29 of 48
he will be only 1,000,000 to the two-thirds power stronger: about 10,000 times, allowing him to
lift objects weighing up to a hundred pounds, not thousands.
Things behave differently at different scales, but there are orderly ways -- scaling laws -- that
connect one realm to another. "I found this enormously exciting," West said. "That's what got me
thinking about scaling in biology."
At some point he ran across Kleiber's law. "It is truly amazing because life is easily the most
complex of complex systems," West said. "But in spite of this, it has this absurdly simple scaling
law. Something universal is going on."
Enquist became hooked on scaling as a student at Colorado College in Colorado Springs in 1988.
When he was looking for a graduate school to study ecology, he chose the University of New
Mexico in Albuquerque partly because a professor there, Brown, specialized in how scaling
occurred in ecosystems.
There are obviously very few large species, like elephants and whales, and a countless number of
small species. But who would have expected, as Enquist learned in one of Brown's classes, that if
one drew a graph with the size of animals on one axis and the number of species on the other
axis, the slope of the resulting line would reveal another quarter-power scaling law? Population
density, the average number of offspring, the time until reproduction -- all are dependent on body
size scaled to quarter-powers.
"As an ecologist you are used to dealing with complexity -- you're essentially embedded in it,"
Enquist said. "But all these quarter-power scaling laws hinted that something very general and
simple was going on."
The examples Brown had given all involved mammals. "Has anyone found similar laws with
plants?" Enquist asked. Brown said, "I have no idea. Why don't you find out?"
After sifting through piles of data compiled over the years in agricultural and forestry reports,
Enquist found that the same kinds of quarter-power scaling happened in the plant world. He even
uncovered an equivalent to Kleiber's law.
It was surprising enough that these laws held among all kinds of animals. That they seemed to
apply to plants as well was astonishing. What was the common mechanism involved? "I asked
Jim whether or not we could figure it out," Enquist recalled. "He kind of rubbed his head and
said, 'Do you know how long this is going to take?"'
They assumed that Kleiber's law, and maybe the other scaling relationships, arose because of the
mathematical nature of the networks both animals and trees used to transport nutrients to all their
cells and carry away the wastes. A silhouette of the human circulatory system and of the roots and
branches of a tree look remarkably similar.
But they knew that precisely modeling the systems would require some very difficult
mathematics and physics. And they wanted to talk to someone who was used to trafficking in the
idea of general laws.
"Physicists tend to look for universals and invariants whereas biologists often get preoccupied
with all the variations in nature," Brown said. He knew that the Santa Fe Institute had been
established to encourage broad-ranging collaborations. He asked Mike Simmons, then an institute
administrator, whether he knew of a physicist interested in tackling biological scaling laws.
West liked to joke that if Galileo had been a biologist, he would have written volumes cataloging
how objects of different shapes fall from the leaning tower of Pisa at slightly different velocities.
He would not have seen through the distracting details to the underlying truth: if you ignore air
resistance, all objects fall at the same rate regardless of their weight.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 30 of 48
But at their first meeting in Santa Fe, he was impressed that Brown and Enquist were interested in
big, all-embracing theories. And they were impressed that West seemed like a biologist at heart.
He wanted to know how life worked.
It took them a while to learn each other's languages, but before long they were meeting every
week at the Santa Fe Institute. West would show the biologists how to translate the qualitative
ideas of biology into precise equations. And Brown and Enquist would make sure West was true
to the biology. Sometimes he would show up with a neat model, a physicist's dream. No, Brown
and Enquist would tell him, real organisms do not work that way.
"When collaborating across that wide a gulf of disciplines, you're never going to learn everything
the collaborator knows," Brown said. "You have to develop an implicit trust in the quality of their
science. On the other hand, you learn enough to be sure there are not miscommunications."
They started by assuming that the nutrient supply networks in both animals and plants worked
according to three basic principles: the networks branched to reach every part of the organism and
the ends of the branches (the capillaries and their botanical equivalent) were all about the same
size. After all, whatever the species, the sizes of cells being fed were all roughly equivalent.
Finally they assumed that evolution would have tuned the systems to work in the most efficient
possible manner.
What emerged closely approximated a so-called fractal network, in which each tiny part is a
replica of the whole. Magnify the network of blood vessels in a hand and the image resembles
one of an entire circulatory system. And to be as efficient as possible, the network also had to be
"area-preserving."
If a branch split into three daughter branches, their cross-sectional areas had to add up to that of
the parent branch. This would insure that blood or sap would continue to move at the same speed
throughout the organism.
The scientists were delighted to see that the model gave rise to three-quarter-power scaling
between metabolic rate and body mass. But the system worked only for plants. "We worked
through the model and made clear predictions about mammals," Brown said, "every single one of
which was wrong."
In making the model as simple as possible, the scientists had hoped they could ignore the fact that
blood is pumped by the heart in pulses and treat mammals as though they were trees. After
studying hydrodynamics, the nature of liquid flow, they realized they needed a way to slow the
pulsing blood as the vessels got tinier and tinier.
These finer parts of the network would not be area-preserving but area-increasing: the cross
sections of the daughter branches would add up to a sum greater than the parent branch, spreading
the blood over a larger area.
After adding these and other complications, they found that the model also predicted three-
quarter-power scaling in mammals. Other quarter-power scaling laws also emerged naturally
from the equations. Evolution, it seemed, has overcome the natural limitations of simple
geometric scaling by developing these very efficient fractal-like webs.
Sometimes it all seemed too good to be true. One Friday night, West was at home playing with
the equations when he realized to his chagrin that the model predicted that all mammals must
have about the same blood pressure. That could not be right, he thought. After a restless weekend,
he called Brown, who told him that indeed this was so.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 31 of 48
The model was revealed, about two years after the collaboration began, on April 4, 1997, in an
article in Science 14 . A follow-up last fall in Nature 15 extended the ideas further into the plant
world.
More recently the three collaborators have been puzzling over the fact that a version of Kleiber's
law also seems to apply to single cells and even to the energy-burning mitochondria inside cells.
They assume this is because the mitochondria inside the cytoplasm and even the respiratory
components inside the mitochondria are arranged in fractal-like networks.
For all the excitement the model has caused, there are still skeptics. A paper published last year in
American Naturalist by two scientists in Poland, Dr. Jan Kozlowski and Dr. January Weiner,
suggests the possibility that quarter-power scaling across species could be nothing more than a
statistical illusion. And biologists persist in confronting the collaborators with single species in
which quarter-power scaling laws do not seem to hold.
West is not too bothered by these seeming exceptions. The history of physics is replete with cases
where an elegant model came up against some recalcitrant data, and the model eventually won.
He is now working with other collaborators to see whether river systems, which look remarkably
like circulatory systems, and even the hierarchical structure of corporations obey the same kind of
scaling laws.
The overarching lesson, West says, is that as organisms grow in size they become more efficient.
“That is why nature has evolved large animals,” he said. “It's a much better way of utilizing
energy. This might also explain the drive for corporations to merge. Small may be beautiful but it
is more efficient to be big.”

3.16 Power Laws and Prediction

The ubiquity of power laws makes them important for analysing and interpreting data. But some
people ask the “so what” question. For example, it seems that settlements obey a power law
distribution. Apart from telling us that there are many small settlements and few really large
settlements, what else do such power laws tell us?
The current great interest in power laws is partly because people think they can be used for
explaining or predicting things. In the case of physics, and Galileo’s inclined plane, the formula
v = k t2 is predictive. It allows the velocity at precise points of time in the future to be determined.
Many physical systems seem to exhibit power laws properties that make them predictable in this
traditional sense.
Generally complex systems are not predictable in this way. It is not possible to make point
predictions of their precise state at some precise future time. In this context the analysis of data
from the perspective of power laws seems to be a duty of the scientist. If the data admit power
law properties we should know this, and not to know it would be negligent. There then emerges a
‘static’ versus ‘dynamic’ dichotomy. To look at one power law at a snapshot in time, as with the
US cities of the 1990s, might not tell us much. But, presumably, the populations of US cities have
changed since then? Perhaps the slope of the line has changed? Perhaps the rankings have
changed? Perhaps both? What might this tell us about the underlying dynamic processes that
generate these distributions? The work of Brown, Enquist and West discussed in the previous
section suggests that they can tell us a great deal.

14
‘A General Model for the Origin of Allometric Scaling Laws in Biology, Geoffrey B. West, James H.
Brown, Brian J. Enquist. Science, 276(4), 122-126, April 1997. [Link]
15
Brian J. Enquist, James H. Brown, Geoffrey B. West, 'Allometric scaling of plant energetics and
population density, Nature, 395, 163-165, 10 September 1998.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 32 of 48
Pumain writes 16 “The simplest alternative to the static point of view is the approach which treats
urban hierarchy as the product of stochastic process distributing population growth between
cities. This involves transferring a statistical model for the dynamic description of an urban
system. In his book: Les inégalitiés économiques (1931), Gibrat demonstrated that when cities are
growing at the same average rate but with fluxtuations or growth inequalities, the distribution of
city sizes will consistently take a lognormal form 17 . He explained that whereas an additive
process of growth would lead to a normal distribution, multiplicative growth, which he calls “the
law of proportional effect”, will result in a lognormal distribution of sizes.(page 198) …
According to this process, the “attractivity” of cities does not have to be assumed to increase with
their size (according to an “increasing returns” hypothesis for instance) to explain a highly
uneven size distribution, yielding a small number of very large cities. A more satisfactory
“explanation” for the shape of the distribution of city sizes is thus obtained when we view it as
resulting from a dynamical process for the distribution of urban growth rather than simply as the
expression of a static equilibrium. Compared with these static interpretations, Gibrat’s model is in
fact the only one to show why this form of organisation is unchanging over time, and to
demonstrate a gradual concentration of the population in increasingly large cities at the top of
increasingly skewed distributions. The model thus provides an “explanation” for the gradual
differentiation in size between cities that are involved in the same evolution (page 199). … Many
new models of distributed growth that can generate power laws have been developed recently.
The novelty is that they are not mathematical models of aggregated growth, they are an attempt to
generate global distributions from agent-based models or multi-agent systems that define rules of
interaction at the micro-level.”

Moore’s Law 18 is an example of an empirical power law that is widely cited and used. The most
popular formulation is of the doubling of the number of transistors on integrated circuits every 18

16
Pumain, D., ‘Alternative explanations of hierarchical differentiation in urban systems’, 169-222 in
Hierarchy in Natural and Social Sciences, D. Pumain (ed), Springer (Dordrecht, Netherlands), 2006.
17
A lognormal distribution is a distribution whose logarithm is normal with bell shaped curve. The details
are beyond the scope of this less and will be dealt with in a later lesson.
18
[Link] , [Link]

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 33 of 48
months. At the end of the 1970s, Moore's Law became known as the limit for the number of
transistors on the most complex chips. However, it is also common to cite Moore's Law to refer to
the rapidly continuing advance in computing power per unit cost, because increase in transistor
count is also a rough measure of computer processing power. On this basis, the power of
computers per unit cost - or more colloquially, "bangs per buck" - doubles every 24 months (or,
equivalently, increases 32-fold in 10 years).
Although Moore's Law was initially made in the form of an observation and forecast, the more
widely it became accepted, the more it served as a goal for an entire industry. This drove both
marketing and engineering departments of semiconductor manufacturers to focus enormous
energy aiming for the specified increase in processing power that it was presumed one or more of
their competitors would soon actually attain. In this regard, it can be viewed as a self-fulfilling
prophecy.

It is possible that power laws and log-periodicity could be used in earthquake prediction 19 . “We
must add a third and last touch to the picture, which uses the concept of discrete scale invariance,
its associated complex exponents and log-periodicity, as discussed above. In the presence of the
frozen nature of the disorder together with stress amplification effects, we showed that the critical
behavior of rupture is described by complex exponents, in other words, the measurable physical
quantities can exhibit a power law behavior (real part of the exponents) with superimposed log-
periodic oscillations (due to the imaginary part of the exponents). Physically, this stems from a
spontaneous organization on a fractal fault system with “discrete scale invariance”. The practical
upshot is that the log-periodic undulations may help in “synchronizing” a better fit to the data. In
the above numerical model, most of the large earthquakes whose period is of the order of a
century can be predicted in this way 4 years in advance with a precision better than a year. For the
real earth, we do not know yet as several difficulties hinder a practical implementation, such as
the definition of the relevant space-time domain. A few encouraging results have been obtained
but much remains to test these ideas systematically, especially using the methodology presented
above to detect the regional domain of critical maturation before a large earthquake [Sornette and
Sammis, 1995]. While encouraging and suggestive, extreme caution should be exercised before
even proposing that this method is useful for predictive purpose but the theory is beautiful in its
self-consistency and, even if probably inaccurate in details, it may provide a useful guideline for
the future.”

In the context of flood prediction, David Dawdy 20 writes “However, it was soon realized that a
physical understanding of statistical scaling of floods, strictly applied, was valid only for nested
basins, that is, basins with subbasins nested within, and time scales of individual rainfall-runoff
events (Gupta et al. 1996). Although the USGS regions contained similar basins, they were not
nested basins, and the time scale was annual not event based. The observational evidence for
power laws in quantile regression analysis of annual floods in nested basins for regional or spatial
analysis of annual flood frequencies is a recent development, because only two papers have been
published on it: one for the Walnut Gulch basin, Ariz. (Goodrich et al. 1997) and another for the
Goodwin Creek basin, Miss. (Ogden and Dawdy 2003). The USDA has operated these two basins
as experimental facilities for many years, and they are quite unique as natural laboratories in
terms of high density of rainfall and streamflow data. The Walnut Gulch basin is in the semiarid
southwest United States, and the Goodwin Creek is in the humid southeast United States. Ogden

19
[Link]
20
[Link]

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 34 of 48
and Dawdy (2003) also conducted the first event-based analysis of spatial flood statistics for the
Goodwin Creek basin and observed a presence of power laws. Surprisingly, they found that the
power law parameters vary from one event to the next. Following on this work, Furey and Gupta
(2005) have tested the hypothesis that the observed variability in the power law parameters can be
attributed to temporal variability in effective rainfall intensity and duration among events”.

Power laws and scale invariant distributions play a significant role in the science of complex
systems. As this and the previous section show, finding any particular power law may not
advance knowledge much, but understanding the dynamics that underlie power laws may add
greatly to our understanding of complex systems. Already we have seen that power laws can
drive scientific research and discovery.

3.17 Conclusions

In this session you have covered the following ideas

 Representing functions graphically
 Linearity and linear relationships
 Quadratic Functions, Cubic Functions, Quartic Functions and Polynomials
 Scale invariance
 Measuring the slopes of functions
 The Exponential and Logarithmic Functions
 Logarithms, polynomials, and straight lines
 Zipf’s Law, Power Laws, and Pareto Disribution
You have seen them used in a number of fields including
 dynamics of a ball on an inclined plane
 word rank-frequency distributions
 city population rank-frequency distributions
 earthquakes
 allometric laws
 financial markets
 Moore’s Law
 flooding

The very fact that power laws are so common in such a wide variety of systems, and they are so
widely studied in complex systems, suggests that anyone working in the complex systems field
should understand the basic underlying mathematics.

This is not to say that power laws are the answer to all questions in complex systems science, or
that all work involving power laws is useful or insightful.

As always in science, individuals have to make up their own minds, and we have to argue our
perspectives with others in the hope of reaching synthesis, agreement, and new science. Now you
have the necessary mathematical tools to you can – and should - join in this process.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 35 of 48
3.18 Multiple choice questions

You have now completed all the formal material for this session.

Answer each of the following multiple choice questions by selecting what you think is the right
answer. If you cannot answer the question reply with ‘don’t know.

Q1.

Which of the following is incorrect

(a) xm  xn = xm+n
(b) x – k = 1/xk
(c) xm  xn = xm – n
(d) xm  xn = xm/ n
(e) (nx)k = nk xk
(f) x = (xk) –k
(g) x2  x3 = x5
(h) don’t know

Q2.

Who invented the Cartesian grid?

(a) Galileo
(b) Aristotle
(c) Bourbaki
(d) Descartes
(e) Euclid
(f) Cartan
(g) Julius Caesar
(h) don’t know

Q3.

A straight line graph y = mx + c cuts the y-axis at the point

(a) (7, 3)
(b) (x, y)
(c) (m, c)
(d) (c, 0)
(e) (m, 0)
(f) (0, c)
(g) (0, m)
(h) don’t know

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 36 of 48
Q4.

A straight line graph y = mx + c cuts the x-axis at the point

(a) (0, -c/m)

(b) (x, y)
(c) (m, c)
(d) (c, 0)
(e) (-c/m, 0)
(f) (0, c)
(g) (0, m)
(h) don’t know

Q5.

A function is said to be linear if which of the following hold

(a) f(a+b) = f(a) + f(b)

(b) f(a+b) = f(a)  f(b)
(c) f(a+b) = f(a)  f(b)
(d) f( a) = f(( b)
(e) f( a) =  f()
(f) f( a) = f(a)
(g) f( x ) = 1/x
(h) don’t know

Q6

Let f(x) = ax3 + bx2 + cx + d. Which of the following is true

(a) f has a point of reflection if ax3 < bx2.

(b) f has no critical points if b2 < 3ac.
(c) f has one critical point if b2 = 3ac.
(d) f has two critical points if b2 > 3ac
(e) f has four critical points if b4 > 3ac
(f) f has a point of inflection if b2 = 3ac.
(g) f has three crtical points if d > 0.
(h) don’t know

Q7 Which of the following is the largest

(a) 1234567
(b) 109
(c) 910
(d) 209
(e) 1020
(f) 220
(g) 10003
(h) don’t know

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 37 of 48
Q8.

A function is scale invariant if

(a) f(cx) = cx
(b) f(cx) = cx
(c) f(cx) = c + x
(d) f(cx)  x
(e) f(cx) = f(c) f(x)
(f) f(cx) = f(c) + f(x)
(g) f(cx)  f(x)
(h) don’t know

Q9.
Which of the following are true

(a) Galileo showed that d = kt2

(b) Galileo’s formula is predictive
(c) Galileo dropped objects from the Leaning tower of Pisa
(d) Galileo used a water clock in his experiments
(e) Galileo used a pendulum clock in his experiments
(f) Galileo discovered a power law
(g) Heavy objects fall fastest, except in a vacuum
(h) don’t know

Q10

Which of the following are true

(a) The base of 106 is 6

(b) The base of 106 is 10
(c) The base of 26 is 6
(d) ab = ba
(e) 510 > 105
(f) the exponent of 102 is 2
(g) the exponent of 102 is 10.
(h) don’t know

Q11

Which of the following are true

(a) e = 2.71828 is called Euler’s number.

(b) if ba = x then logb x = a
(c) Logartithms with base 10 are called ‘common logarithms’
(d) log( x  x’ ) = log( x ) + log( x’ )
(e) log x = logb x / logb 
(f) log( x - x’ ) = log( x ) - log( x’ )
(g) log( x  x’ ) = log( x ) - log( x’ )
(h) don’t know

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 38 of 48
Q12

Consider the function y = x2. Which of the following are true

(a) log y = log x2

(b) log y = 2 log x
(c) plotting log y against log x gives a straight line with gradient n
(d) log y = log x2 only if y > x
(e) the graph of the function is a straight line with gradient 2
(f) the gradient at any point x is 2x
(g) plotting log y against log x gives a straight line with gradient 2
(h) Don’t know

Q13

Which of the following is true concerning the Brown Corpus

(a) Zipf’s Law states that word frequency is directly proportional to word rank.
(b) Zipf’s Law states that word rank is directly proportional to word frequncy.
(c) Zipf’s Law states that city size is directly proportional to rank.
(d) Zipf’s Law states that city rank is directly proportional to size.
(e) Zipf’s Law states that the logarithm of word size is inversely proportional to word length
(f) Zipf’s Law states that word frequency is proportional to the reciprocal of word frequency.
(g) The Brown Corpus involves about a million words drawn from a wide variety of sources.
(h) Don’t know.

Q14.

Which estimates agree with the SAQ 11 formula, population = 5,625,000/rank0.75?

(a) The 200th ranked city is estimated to have population of 106,000

(b) The 200th ranked city is estimated to have population of 160,000
(c) The 200th ranked city is estimated to have population of 17,000
(d) The 400th ranked city is estimated to have population of 62,000
(e) The 400th ranked city is estimated to have population of 108,000
(f) The 400th ranked city is estimated to have population of 150,000
(g) The 400th ranked city is estimated to have population of 17,000
(h) Don’t know.

Q15

Which of the following are true

(a) f(x) = axk is a power law

(b) Galileo discovered a power law empirically
(c) There is a power law between city populations and their ranks
(d) For a power law f(x) = axk, log f(x) = k lig x + log a
(e) f(x) = 1/xk is a power law
(f) f(x) = log x is a power law
(g) power laws are scale invariant
(h) Don’t know

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 39 of 48
Q16

Which of the following are true

(a) The Pareto principle states that 20% of the population owns 80% of the wealth
(b) The Pareto distribution is given by Pr( X > x) = (x/xm)-k, where k and xm constants.
(c) Pr( X > x) = 1/x is a Pareto distribution
(d) The height of the population follows a Pareto distribution
(e) The age of the population follows a Pareto distribution
(f) The size of meteorites follows a Pareto distribution
(g) The speed of cars on long straight roads follows a Pareto distribution
(h) Don’t know.

Q17

Which of the following are true

(a) The total number of earthquakes in any region in a given period is N = 10A - bM
(b) Keleiver Law gives the body mass – metabolic relationship as q0 ~ M 3/4
(c) The optimal cruising speed in terms of body mass, M, is Vopt ~ 30M 1/6 ms-1
(d) Allometric laws concern scale invariant distributions of metals
(e) Heart beat time, t, and body mass, M, are related as t ~ M1/4
(f) In financial markets, profits, p, vary with share price, s, as p = s8/3
(g) In financial markets P(|ri| > x ) ~ x -  r, where ri is return, and r  3.
(h) Don’t’ know

Q18

Which of the following are true?

(a) The average elephant weighs 220,000 times as a mouse but requires 10,000 as much energy.
(b) As animals get bigger, the pulse slows down and they live longer
(c) Most animals tend to have a billion heart beats in their lifetimes
(d) Elephants have less heart beat than mice during their lives.
(e) A cat has 1000 times the mass of a mouse and lives 1000  33 times as long.
(f) The grasshopper leaps to what would several city blocks for a human.
(g) Scaling laws do not occur in plants,
(h) Don’t know

Q19

Which of the following are true

(a) The human circulatory systems and the roots and branches of trees look similar
(b) Life is the most complex system, but has simple scaling laws.
(c) If man is a million times more massive than an ant, he can lift only 1,000,0002/3 as much
(d) Kleiber found in the 1930 that metabolic rates scales with body mass to three-quarter power.
(e) Monkeys can carry 100 times their body weight.
(f) An ant can lift 100 times its own weight while a man can lift 0.6 times his own weight.
(g) All mammals have approximately the same blood pressure
(h) Don’t know.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 40 of 48
Q20

Which of the following are true

(a) Gibrat demonstrated that city sizes will consistently take lognormal form.
(b) Moore’s Law suggests computer power will double every eighteen months
(c) Moore’s law could be a self-fulfilling prophesy.
(d) Power laws may be useful in predicting earthquakes
(e) Power laws can predict random processes like the outcome at roulette.
(f) Power laws can be used to investigate flooding
(g) Power laws are an important area in complex systems science
(h) Don’t know.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 41 of 48
3.19 Answers to the SAQs

SAQ 1
(a) (0,1)
(b) (-2, 0)
(c) the gradient is y/x = 2/4 = ½
■

SAQ 2

(a) The function in SAQ 1 function illustrates a wide class of functions that have the general form
f (x) = mx + c, where m and c are constants. On the grids below draw the graphs of the functions
f (x) = mx + c for m = 1, c = 0 and for m = – 2, c = 2.

6 6
5 5

4 4

3 3

2 2

1 1

-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
-1 -1

-2 -2

-3 -3

-4 -4

-5 -5
-6 -6

f (x) = x ,(for m = 1, c = 0) f (x) = -2x + 2, (for m = -2, c = 2)

(b) what are the gradients of these two graphs? 1 and -2.

(c) What are the x- and y-intercepts of these two graphs?

(0,0) if the intercept for the first. (1,0) and (0,2) are the intercepts for the second.

(d) What is the relationship between the y-intercept and the constant c?

The graph cuts the y-axis at the point (0, c).

SAQ 3.

The SAQ is answered in the text that follows it.

■

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 42 of 48
SAQ 4

f(x) = m(x – k)2 + c is the function f(x) = mx2 + c shifted to the right by a distance of k. It has a
minimum at the point (k, c).
■

SAQ 5

A cubic function has no critical points if b2 < 4ac. So, for example, let b = 1 and let a = 1 and c=
1, with f(x) = x3 + x2 + x, so that 1 < 4.

x -5 -4 -3 -2 -1 0 1 2 3 4
x2 25 16 9 4 1 0 1 4 9 16
x3 -125 -64 -27 -16 -1 0 1 16 27 64
x3+ x2 + x -105 -54 -21 -14 -1 0 3 22 39 84

2.0

1.8

1.6

1.4

1.2

-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
1.0

0.8

0.6

0.4

0.2

SAQ 6

The answer to this SAQ is in the text that follows it.

■

SAQ 7

(i) The base of the number of 106 is ten.

(ii) The exponent of 102 is two.
(iii) Is it true that ab = ba ? No! for example 23 = 8  9 = 32.
(iv) Which is larger, 510 or 105 ?
510 = 5  5  5  5  5  5  5  5  5  5 = 9,765,625 > 100,000
■

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 43 of 48
SAQ 8

Use the calculator to show that log10 1234 = loge 1234 / loge 10.

log10 1234  3.091

ln 1234  7.118
ln 10  2.303

loge 1234 / loge 10 = ln 1234 / ln 10  7.118 / 2.303  3.091  log10 1234

SAQ 9. Complete the following table and plot the graphs y = x2 and log y = log y2 on the grids.

x 1 2 3 4 5 6 7 8 9 10
x2 1 4 9 16 25 36 49 64 81 100
log x 0 0.301 0.477 0.602 0.670 0.788 0.845 0.903 0.954 1.0
log x2 0 0.602 0.954 1.204 1.340 1.556 0.169 1.806 1.908 2.0

100

90

80

70

60
x2 log x2
50

40

30

20

10

0
0 2 4 6 8 0 0 0.2 0.4 0.6 0.8 1.0

x log x
■

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 44 of 48
SAQ 10

x 1 2 3 4 5 6 7 8 9 10
x2 1 4 9 16 25 36 49 64 81 100

upper cycle

25

16

x2

lower cycle

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 45 of 48
SAQ 11

rank freq word log rank log freq rank freq word log rank log freq

1. 69970 the 0.00 4.85 80. 1237 over 1.90 3.09

2. 36410 of 0.30 4.56 90. 1016 before 1.95 3.01
3. 28854 and 0.48 4.46 100. 895 down 2.00 2.95
4. 26154 to 0.60 4.42 200. 426 took 2.30 2.63
5. 23363 a 0.70 4.37 300. 308 matter 2.48 2.49
6. 21345 in 0.78 4.33 400. 203 court 2.60 2.31
7. 10594 that 0.85 4.03 500. 191 finally 2.70 2.28
8. 10102 is 0.90 4.00 600. 163 doing 2.78 2.21
9. 9815 was 0.95 3.99 700. 143 industrial 2.85 2.16
10. 9542 he 1.00 3.98 800. 128 forms 2.90 2.11
20. 5180 I 1.30 3.71 900. 115 slowly 2.95 2.06
30. 3619 they 1.48 3.56 1000. 104 current 3.00 2.02
40. 2670 their 1.60 3.43 2000. 55 components 3.30 1.74
50. 2199 if 1.70 3.34 3000. 36 sympathy 3.48 1.56
60. 1789 them 1.78 3.25 4000. 25 terror 3.60 1.40
70. 1400 may 1.85 3.15 5000. 19 ladder 3.70 1.28

Table 1. Words and their frequencies in the Brown Corpus.

SAQ 10.

(i) Plot the logarithm of the rank against the logarithm of the frequency for the data in Table 1.
What is the equation of the line?

5.0

4.0

log
frequency

3.0

2.0

1.0
0.0 1.0 2.0 3.0 4.0
log rank

The slope of my line is almost -1, so the equation is log(rank) = -log(frequency) + k. My point at
which the line cuts the y-axis is ( 0, 4.9), so k = 4.9.

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 46 of 48
By trial and error with the calculator I found that 4.9  log 80,000.

log(rank) = - log(frequency) + 4.9 can be rewritten as

log(rank)  -log(frequency) + log 80,000, which can be rewritten as

log(rank)  log(80,000/frequency ) wherby

rank = 80,000 / frequency, or

frequency = 80,000 / rank.

This rather overestimates the frequency for rank 1 at 80,000 when it is actually 69970.

For the hundredth word the prediction is 800, when in reality it is 895.

For the thousandth word the prediction is 80, when in reality it is 104.

(ii) Do the data support Zipf’s Law that f(n) = k/n a ? If so what are the values of k and a?

Although the values are not prefect, I would say this result supports Zipf’s Law.

In the function f(n) = k/n a, k = 80,000 and a = 1.

SAQ 12. The ranks and populations of twelve American cities are given below. Complete the
table and plot the log of the rank against the log of the population on the grid below. What
relationship, if any, was there between the rank and the population in 1990?

City Rank Population 1990 Log rank Log Population

New York 1 7,322,564 0.00 6.86
Detroit 7 1,027,974 0.85 6.01
Baltimore 13 736,014 1.11 5.87
Washington D.C. 19 606,900 1.28 5.78
New Orleans 25 496,938 1.40 5.70
Kansas City 31 434,829 1.50 5.64
Virginia Beach 37 393,089 1.57 5.59
Toledo 49 332,089 1.69 5.52
Arlington Texas 61 261,721 1.79 5.42
Baton Rouge 73 219,531 1.86 5.34
Hialeah 85 188,008 1.93 5.27
Bakersfield 97 174,820 1.99 5.24

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 47 of 48
7.00

6.00

5.00
1.00 2.00
The gradient of this line is close to -0.75. The intercept is at (0, 6.75), so

log(size) = -0.75 log(rank) + 6.75, or by trial and error with the calculator
log(size) = -0.75 log(rank) + log(5,625,000), or
log(size) = log( 5,625,000/rank0.75), so that
size = 5,625,000/rank0. 75 is the population size- rank relation.

To test this let rank = 1. Then size = 5,625,000/(10. 75) = 5,625,000/1 = 5,625,000, which of the
same order of magnitude as the tabulated value of 7,322,564.

Now let rank = 97. Then size = 5,625,000/(970. 75). What is 970. 75?

Let x = 970. 75. Then log x = log 970. 75 = 0.75 log 97 = 0. 75  1.987 = 1.490. So log x = 1.490.
By trial and error with the calculator I found that log 30.9 = 1.519, so x  30.9.

Then size == 5,625,000/ 30.9 = 182,038, which is close to the tabulated population value of
174,820 for the 97th city.
■

CSS - ASSYST Course on Mathematics for the Science of Complex Systems. Lesson 3 Page 48 of 48