NATURE’S GEOMETRY

SOUMITRO BANERJEE∗

1. Introduction continued to be idealized ones that one can

think up in one’s head but never finds in
Geometry essentially deals with two types nature.
of entities: objects and spaces. Points, cir- Nature abounds in irregular objects.
cles, lines, curves, cylinders, tetrahedrons “Mountains are not cones”, as Benoit Man-
— these are examples of the objects one delbrot, the founder of the new geometry
finds in geometry books. These objects put it, “clouds are not spheres, lightnings
live in spaces that hold them. A point are not straight lines”. Towards the end of
(a zero-dimensional geometrical object) can the twentieth century we seem to be break-
rest on a one-dimensional line or a two- ing out of the compartment of Euclidean
dimensional plane. A curve (a one- dimen- objects. Geometers are now considering
sional geometrical object) can rest on a 2-D these irregular objects as valid subjects of
surface or in a 3-D volume. We thus have study. And that is what fractal geometry is
objects embedded in spaces of dimension all about.
greater than or equal to that of the objects.
For much of human history, geometers 2. Euclidean objects versus
majorly studied idealized objects with reg-
natural objects
ular shapes: triangles, circles, spheres and
rectangular parallelopipeds. This was the What distinguishes natural objects from
legacy of Pythagoras and Euclid. The em- Euclidean objects? The first thing that
bedding space was always thought to be comes to one’s mind is the irregularity of
Euclidean: a flat piece of paper in case of the shape of natural objects. But that needs
2-D space, a volume with ‘flat’ characteris- to be specified in mathematical terms.
tics in case of 3-D space etc. If we take a very small length of a curve,
Towards the end of the 19th cen- say y = f (x), it approximates a straight line.
tury, there was a revolution in geometry. The closer we look at it, the more it loses
Lobachevsky, Riemann and others showed its structure. That is how we can define the
that space can also be curved, and then first derivative as a limiting value of the rate
Einstein found a profound application of of change. Same is the character of Eu-
the new geometry in his theory of gravi- clidean surfaces—they smoothen out into
tation. At least in the domain of spaces flat planes as you take a close look.1
the narrow confines of Euclidean space was Here lies the main distinction between
broken. natural objects and the idealized objects.
But objects still remained Euclidean. We Natural objects never flatten out—at what-
still talked about angles contained in trian- 1 The derivative is
gles in Euclidean space and non-Euclidean
dy ∆y
space. The objects of study in geometry = lim
dx ∆x→0 ∆x
∗ Dr. Banerjee is in the faculty of the Department of This limit would exist only if the curve smoothens out
Electrical Engineering, I.I.T. Kharagpur - 721302. into the tangent as ∆x → 0

Breakthrough, Vol.13, No.4, January 2009 1

 From the Breakthrough archives
ever level of magnification you may look at just by looking around yourself.
them. Think of any natural surface — the These objects, evidently, need a new
surface of your skin or that of a tree trunk. mathematical tool for characterization.
They contain complexities within complexi- There comes the question of dimension.
ties which come in view at higher levels of
magnification. These are continuous but 3. Dimension of geometrical
nondifferentiable surfaces. objects
This applies to some data sets also. No
It is strange that in school and college level
economist ever tries to differentiate the
mathematics curricula we never learn what
curve for share prices or exchange rates. No
‘dimension’ means in geometry. Our con-
electrical engineer would think of differen-
ception, naturally, is mostly derived from
tiating the curve representing the load on a
common sense: if we have a line-like object
power station. One cannot take the deriva-
we say it is one dimensional, if we have a
tive of the curve for oscillation during earth-
surface-like object we say it is two dimen-
quakes. The reason is, these curves as ge-
sional and so on.
ometrical objects, are continuous but not
We are accustomed to thinking of dimen-
differentiable anywhere. They reveal more
sions as integers, and have considerable
and more rich forms as you zoom closer.
difficulty in visualizing anything otherwise.
The next major difference is illustrated by The ancients had the same sort of difficulty
Mandelbrot’s famous question, “how long is when the only numbers they knew were
the coastline of England?”. A little thought natural numbers. Three horses, ten men,
makes it evident that the measured length twentifive bananas — such numbers came
of the coastline depends on the yardstick naturally to them. But finally they had to
of measurement. If the yardstick is long, shake off the narrow confines of integers
much of the detail of coastline geometry and conceive fractions — in order to rep-
would be missed. As smaller and smaller resent length, area, weight etc.
yardsticks are used, the creeks and bends It now appears that we again have to
come in view and the measured length shake off the notion of integral dimensions
increases. And in the limit—when the and conceive fractional dimensions when
yardstick length shrinks close to zero—the we face the task of representing natural ob-
length of the coastline becomes infinitely jects.
large. It must first be understood that the di-
Yet the area of England is finite. Thus the mension of any object and that of the em-
coastline is a curve of infinite length enclos- bedding space are two different quantities.
ing a finite area. The dimension of the embedding space is
Same is the case of all 3-D natural ob- given by the degree of freedom. In 1-D
jects enclosed by natural surfaces. Take space one can move only left or right, in 2-
the structure of our lungs as an example. D space one can move left-right as well as
The task of the lungs is to absorb oxygen front-back, in 3-D space one can also move
from air. In order to have a high absorption up-down. The embedding dimension, natu-
rate, the surface area needs to be very high. rally, has to be an integer.
Yet, the volume must be small—it has to be The dimension of an object can not be de-
accomodated within the rib cage. Thus it fined in a similar manner. It has to be de-
has the same geometrical characteristics as fined according to the way it fills space. To
the coastline. You can multiply examples probe the question, let us take a simple Eu-

2 Breakthrough, Vol.13, No.4, January 2009

 From the Breakthrough archives

Figure 1: To find how a triangle fills space, it is covered by a grid and the grid size is
successively reduced. The boxes required to cover the object is shown in shade. In the
limit we find that the triangle fills space in the same way as the square. But the boundary
of the carbon particle seen in the electron microscope photograph does it in a different
way.

clidean object: the square. We know that it To generalize, we can write

is two dimensional. But how do we obtain  2
the number 2 from the structure of this ob- 1
N () = K
ject? 
In order to see how it fills space, we where K is a constant.
cover it with a grid as shown in Fig.1. If We can extract the dimension (2 in this
the square has 1 cm sides and the dis- case) from it as follows:
tance between two consecutive grid lines is
1
1/10 cm, then 100 grid elements would be ln N () = ln K + 2 ln
necessary to cover the square. If we now 
reduce the grid length by half, 400 grid el- ln N () ln K
2= −
ements would be required. We see that the ln 1 ln 1
number of grid elements required to cover The second term would vanish as  → 0.
the object increases as the square of the re- Thus the dimension D of the object is given
ciprocal of grid length. by
 2 ln N ()
1 D = lim
N () = →0 ln 1


If instead of the known Euclidean objects
where  is the grid length and N () is the we take some natural object like a carbon
number of grid elements required to cover particle or a map of the Andamans, and
the object (a function of ). subject it to the above procedure, we would
If the object taken is a right angled trian- find that the dimension turns out to be a
gle, the count of covering boxes finally con- fraction. This is a general characteristics
verges to of all natural objects as distinct from ideal-
 2
1 1 ized objects. The fractional dimension thus
N () =
2  provides a method of characterizing natural
objects. In fact, such objects are defined by
And for a circle we have
this property.
 2
π 1 Geometrical objects with fractional di-
N () = mensions are called Fractals.
4 

Breakthrough, Vol.13, No.4, January 2009 3

 From the Breakthrough archives
4. What use is fractal square
2.0
dimension? trian
gle
How do we quantify the fractal character of fractal object
an object? What experimental procedure

Dimension
would enable us to judge if two dissimilar
fractal objects are closely related or geomet- straight line
1.0
rically equivalent? parabo
Looking at any object, we always have a la
intuitive feeling about how densely the ob-
ject occupies space, how crooked, twisted,
broken it is. Looking at a curve plotted
from a set of data, we do feel how wavy or
“noisy” it is. But these are subjective feel- Epsilon
ings. We need a concrete objective method-
ology to assess this quality. Measurement Figure 2: Measurement of the fractal di-
of the fractal dimension provides the means mension of a natural object.
to achieve this end.
The method can easily be guessed from
the definition of dimension given in the last When two surfaces make electrical con-
section. We divide the embedding space tact (as in a switch), all the points never
into a number of equal “boxes”. In case of touch each other. The amount of actual
2-D space, we divide the sheet of paper into electrical contact is determined by the frac-
small squares as in a graph paper. In case tal character of the surfaces, and that is
of 3-D space we divide it into cubes. quantified by the dimension of the two sur-
Then we count how many of these ele- faces.
mental boxes are required to cover the ob- Some silt particles float in river water and
ject. This is our N . The side of the box is some settle quickly. The precipitation obvi-
. Subsequently, reduce the size of the box ously depends on the size and specific grav-
in steps and repeat the procedure of count- ity of the particles. But these two parame-
ing. Then plot D versus . The plot approxi- ters being equal, the more crooked the sur-
mates a horizontal straight line as  tends to face, the more it will get carried by flowing
zero. When a reasonable approximation is water. This property is again quantified by
obtained, the point where it cuts the y-axis the fractal dimension.
gives the dimension of the object (Fig.2). When metals, semiconductors or alloys
But what use in this new piece of infor- crystallize, the crystal grains have bound-
mation? It actually quantifies the surface aries that are fractals. Scientists have
characteristics. And if any of the properties found a number of properties of such mate-
of the body is determined by the character- rials that are related to the fractal character
istics of the surface, fractal dimension can of the grains.
supply vital information. Any surface is fractal. However much
For example, the fractal dimension of the you may polish and smoothen a surface,
surface of the carbon particles in auto- at some level of magnification irregularities
mobile exhaust is related to the afficacy must show up. And the light absorption
with which the particle will attach with the property of the surface is determined not
breathing ducts and cause harm. only by the property of the material but also

4 Breakthrough, Vol.13, No.4, January 2009

 From the Breakthrough archives
by the character of the surface. This, again, entists try to express the changes or “dy-
can be quantified by fractal dimension. namics” by equations so that the status at
When rock structures form in various ge- any point can be computed from its history.
ological processes, the characteristic sig- Generally the dynamical equations may
nature of the formative process is left in be quite complicated. But simplified ver-
the geometry of the rocks. And these sions often reveal properties observed in
are nothing but fractals. You may look complicated systems as well. Let us take a
at large mountain ranges, large boulders, simple dynamical system expressed by the
small pebbles or miniscule grains that be- equation:
come visible when you venture inside the Zt+1 = (Zt )2 + C
rocks by cutting them — you see structures
inside structures — that wait to tell you where Zt is the state of the system at the
much about their character. t-th instant and Zt+1 is the state at the next
The afficacy of solid catalysts in help- instant. The Z’s and C are complex num-
ing or retarding a reaction depends on how bers which have a “real part” and an “imag-
much surface area is exposed to the reac- inary part” expressed as√a + ib; and i is
tants. Same is the case for the electrodes the “imaginary” number −1. Such com-
in electrochemical reactions. None of these plex numbers can be plotted as points on
surfaces are smooth, and it is easy to see a “complex plane” with the real part along
how their behaviour would be related to the the x-axis and the imaginary part along the
fractal dimension of the surfaces in ques- y-axis. Dynamics of the complex number
tion. Z can then be viewed as the changing po-
The veins in the plant leaves, similarly, sition of a point on a sheet of paper — the
are fractals. The arteries and veins in your complex plane.
body are fractals. Tumours and cancer cells If we choose a value of C and a starting
are fractals. The economists’ data set for point Z0 , we can calculate subsequent val-
inflation rates are fractals. And in all these ues of Z and observe the dynamics. We find
cases, the fractal dimension gives a direct that for some values of Z0 the system re-
measurement of the character of the object. mains bounded, for some other values Z in-
creases without bounds, i.e., runs towards
5. Mandelbrot and Julia sets infinity. If we now plot those values of Z0 for
which the system remains within bounds,
Can we generate geometrical objects with we get a set of points making up a picture.
such characters mathematically? Benoit This turns out to be a fractal—the Julia Set
Mandelbrot considered this problem first (Fig.3).
and came up with a solution. He found that We can also vary C while taking the start-
such structures can not be generated by the ing point Z0 same in all computations. We
known mathematical procedures of writing find that for some values of the parameter
equations and functions. For it, one has to C the system remains within bounds and
follow the procedure of repeating the same for some other values it doesn’t. We can
operation again and again. again plot on a sheet of paper the locations
Suppose there is a system whose status is of those values of C which make the sys-
changing continuously. In scientists’ per- tem bounded. We again get a fractal—the
lance it would be called a dynamical sys- Mandelbrot set (Fig.4).
tem, and it is easy to appreciate that all nat- The Julia set and the Mandelbrot set have
ural systems are dynamical systems. Sci- become sort of a universal representation of

Breakthrough, Vol.13, No.4, January 2009 5

 From the Breakthrough archives

Figure 3: Examples of Julia Sets

sizing their symmetry, beauty, complexity

and all that, obtained from the simple equa-
tion Zt+1 = Zt2 + C. And in most cases the
basic idea is lost: they are derived from a
simplified representation of a natural pro-
cess, a dynamical system.
Nature is full of dynamical systems. And
for each one there would be certain param-
eters and certain initial conditions. The les-
son that we learn from Mandelbrot’s work is
that for every dynamical system there exist
fractals in the space formed by the system
parameters and in the space formed by the
initial conditions. Most dynamical systems
in nature, in fact, work on fractal geometry.
Let us take an example from engineer-
Figure 4: The Mandelbrot set. In this ge- ing. A ship is stable in the upright position
ometrical object, however small part of the and tends to come back to this position if
figure you may want to observe, it reveals slightly disturbed from it. But with exces-
rich internal structures. The same is true sive amount of tilt, it may capsize. Thus,
for Julia sets and all other fractal objects. depending on the initial condition of the
tilt, its state may be bounded (upright po-
sition) or can escape without bounds (cap-
fractal geometry. Most books, articles and size). A real ship in an ocean would con-
popular expositions on fractal geometry be- tinuously be bombarded by waves, and its
gin with them and end with them, empha- dynamics would depend on the intensity of

6 Breakthrough, Vol.13, No.4, January 2009

 From the Breakthrough archives
the waves (let it be denoted by F ). Now we
write down the equation of motion of the
system2 , and find out which initial condi-
tions keep the system bounded. We can
now plot these initial conditions in a plane
with the tilt angle in the x-axis and the rate
of change of tilt angle in the y-axis. We get
a picture. If we now change the value of
F in steps and draw the picture again, we
find that it changes dramatically with in-
creasing wave intensity. The boundary of
the zone for stable initial conditions can not
be properly discerned and the initial condi-
tions from which the system collapses get Figure 5: The set of initial conditions for
mixed up with it. The picture becomes a which a ship remains stable becomes frac-
fractal (Fig.5). While rocking in the wave, tal under certain conditions. This is the
if the state of the ship ever touches any of reason for many cases of ship capsize.
the white dots, it will capsize. Thus frac-
tal structures are of vital importance to en-
gineers. And it is generated by the same and the second number for the distance we
methodology as Julia sets. go parallel to the y-axis in order to reach
the point. Hence a sheet of paper, in the
6. Iterative Function Systems view of mathematics, is a space formed by
two real numbers. Any picture drawn on a
There is an entirely different approach to paper is nothing but a set of points in that
the problem of modelling real-life objects space, that is, a collection of pairs of two
with mathematical methods. Developed by real numbers.
Professor Michael Bernsley of Georgia Insti- How can you reach a point starting from
tute of Technology, USA, this method uses another point? Here comes the concept of
elegant mathematical reasoning to generate functions. We know the functions of a real
images of natural objects on the computer number, like f (x) = a.x + b. Here, x is a real
screen. number and and f (x) is also a real number.
Any black and white two-dimensional pic- So both the starting point and the destina-
ture is nothing but a set of black dots in a tion are elements of the real line (Fig.6). The
white background. It can be viewed as a set function (or mapping) carries a point on the
of points in a 2-D space. Now, any point in real line to another.
the 2-D space can be located with the help
of two real numbers: the first number giv-
ing the distance that we go along the x-axis
2 Neglecting unnecessary details, the equation may

be written as
ẍ + β ẋ + x − x2 = F sin ωt Figure 6: A function takes one point on the
real line to another.
where β represents the frictional damping and waves
of intensity F strike the ship with a frequency ω. For
the computations that generated the pictures, we took There can be such functions in the 2-D
β = 0.1 and ω = 0.85 space also — carrying a point on a sheet of

Breakthrough, Vol.13, No.4, January 2009 7

 From the Breakthrough archives
paper to another. Such a simple function of
the form
x2 = ax1 + by1 + e

y2 = cx1 + dy1 + f

is called affine transformation.

We can apply such transformations on all
the points of a figure. Thus, while we can
get one point from another by application of
the transformation, we can also get a whole
geometrical figure from another using the
same transformation. And by application of
the transformation again and again we get
a “sequence” of figures.
The question is, will such a sequence lead
us anywhere? We know that a convergent
sequence of numbers (like 1, 12 , 14 , ...) always
has a specific number as the limit point,
and for any given number we can always
define a suitable converging sequence to
give that number. If the sequence of fig-
ures is made convergent, it will also have
a particular figure as its limit. This is in
fact guaranteed by a theorem called “Ba-
Figure 7: Successive application of the
nach fixed point theorem”.
function system transforms a square into a
In order to get such a meaningful se-
fern. All information about this immensely
quence we only need to ensure that the se-
complicated picture is contained in just 24
quence of figures is convergent. That is en-
numbers. For any figure, it is possible to
sured if two points obtained from the trans-
define similar iterated function systems.
formation are closer to each other than the
original pair of points. Such transforma-
tions are called contraction mappings. want. The collection of contraction map-
Thus we can get any figure by suitably pings that generate a particular picture is
defining a contraction mapping. For com- called the “iterative function system”. The
plicated constructions, we only have to de- picture of the fern shown in Fig.7 was gen-
fine a number of such transformations, erated from a square, by iterating a func-
each giving a figure. The resultant figure tion system comprising four affine transfor-
would simply be the combination (or union) mations defined by a,b,c,d,e and f as fol-
of these sub-figures. lows.
In this method it is immaterial from
which figure we start. We can as well start fn a b c d e f
from a square. On repeated application 1 0 0 0 0.16 0 0
of the transformations (iterations) the fig- 2 0.85 0.04 -0.04 0.85 0 1.6
ure will gradually change shape before your 3 0.2 -0.26 0.23 0.22 0 1.6
eyes and “converge” on to the figure you 4 -0.15 0.28 0.26 0.24 0 0.44

8 Breakthrough, Vol.13, No.4, January 2009

 From the Breakthrough archives
It has been shown that all geometrical fig- jects scientists have discovered that very
ures that can be drawn on a piece of pa- simple rules, applied in an iterative way,
per can be generated from such transfor- can generate extremely complicated struc-
mations. We only have to find the trans- tures. Does it hold a clue to the immense
formations for any particular figure. The complexity of natural objects? It has al-
mathematics to do this has also been de- ways intrigued scientists how the informa-
veloped. We can thus squeeze the informa- tion stored in a single molecule—the DNA—
tion contained in pictures and photographs can generate an unimaginably complex en-
in the from of just a few numbers. tity like a human body. Now we know it is
This has immense technological conse- possible, mathematically.
quence. For example, when spacecrafts far
in space send photographs, the picture is
divided into grids and the grey level in each
grid has to be coded and transmitted to
earth. For good pictures this needs long
transmission time as a huge amount of in-
formation has to be sent. Now the fractal
image compression offers the possibility of
compressing the image into a few numbers
enabling very fast transmission. Programs
for coding the image into such numbers has
been developed and may soon become an
international protocol for satellite transmis-
sion.

7. Last Words
Much of the popular literature on fractals
present it as a mathematical game of gen-
erating fancy pictures in a computer. In
contrast, this article presents it as a neces-
sary tool to model the real geometry of na-
ture. In fact in recent years the discovery of
fractal geometry has caused a sea change in
the geometers’ approach. We are no longer
thinking up “perfect” mathematical shapes
and specifying apriori what nature must be
like. We are now taking real lessons from
nature.
This has led to new ways of measur-
ing the complexity of natural objects. And
we find that the fractal dimension quan-
tifies diverse phenomena in nature, which
have real scientific and technological signif-
icance.
In an attempt to model the natural ob-

Breakthrough, Vol.13, No.4, January 2009 9