Fractals: theory and applications

Didier Gonze

Unit de Chronobiologie Thorique

Service de Chimie Physique - CP 231
Universit Libre de Bruxelles
Belgium
Overview

Theory of fractals
Definition and properties
Dimension

Examples of fractals
Cantor set & devil's staircase
Mandelbrot set
Julia set
Sierpinsky triangle

Appications in biology
Physiology
Biological time series
Kinetics
Definition

A fractal is generally "a rough or fragmented geometric shape

that can be subdivided in parts, each of which is (at least
approximately) a reduced-size copy of the whole" (B.
Mandlebrot). This property is called self-similarity.

The term was coined by Benot Mandelbrot in 1975 and was

derived from the Latin fractus meaning "broken" or "fractured".

Benot Mandelbrot
(1924-2010)

Source: wikipedia
Britain coast

How long is the coast of Bretagne?

Britain coast

Lewis Fry Richardson (1881-1953)

While studying the causes of war between two countries,
Richardson decided to search for a relation between the
probability of two countries going to war and the length of their
common border. While collecting data, he realised that there was
considerable variation in the various gazetted lengths of
international borders. For example, that between Spain and
Portugal was variously quoted as 987 or 1214 km while that
between The Netherlands and Belgium as 380 or 449 km.

As part of his research, Richardson

investigated how the measured length of a
border changes as the unit of measurement is
changed. He published empirical statistics
which led to a conjectured relationship. This
research was quoted by mathematician Benot
Mandelbrot in his 1967 paper "How Long Is
the Coast of Britain"?

http://fr.wikipedia.org/wiki/Lewis_Fry_Richardson
Britain coast

This hypothetic map illustrates the notion of self-similarity. Note however

that in reality, coasts are not purely self-similar!
Properties

A fractal often has the following features:

It has a fine structure at arbitrarily small scales.
It is self-similar (at least approximately).
It is too irregular to be easily described in traditional Euclidean geometric
language.
It has a dimension which is non-integer and greater than its topological
dimension (i.e. the dimension of the space required to "draw" the fractal).
It has a simple and recursive definition.

Because they appear similar at all levels of magnification, fractals are often considered
to be infinitely complex (in informal terms). Natural objects that approximate fractals to a
degree include clouds, mountain ranges, lightning bolts, coastlines, and snow flakes.
However, not all self-similar objects are fractals for example, the real line (a straight
Euclidean line) is formally self-similar but fails to have other fractal characteristics.
Examples of natural fractals

Brocoli Lung

Mountain landscape
Fern
Geographic map
Examples of mathematical fractals

Mandelbrot set Julia set

Koch snowflake Dragon curve Sierpinsky

triangle
Cantor set

The Cantor set is obtained by deleting recursively the 1/3 middle

part of a set of line segments:
S0
S1
S2
S3
S4
S5
S6
...
S

Properties of the Cantor set C (obtained after an infinite number of

iterations):
C has a structure as arbitrarily small scale (like worlds within worlds).
C is self-similar: it contains smaller copies of itself at all scales.
The dimension of C is not an integer (see later).

Source: Strogatz
Cantor set

The Cantor set is obtained by deleting recursively the 1/3 middle

part of a set of line segments:
S0
S1
S2
S3
S4
S5
S6
...
S
How long in the Cantor set?
Each set Sn completely covers all the sets that come after it in the construction.
Hence the Cantor set C=S is covered by each of the set Sn. So the total length of
the Cantor set must be less than the total length of Sn, for any n. Let Ln denote the
length of Sn. Then from the construction, we see that L0=1, L1=2/3,
L2=2/3*2/3=(2/3)2, and in general Ln=(2/3)n. Since Ln -> as n -> , the Cantor set
as a length = limn->((2/3)n)=0.
This suggest that the Cantor set id small. However, the Cantor set contains an
infinite (uncountable) number of points.... Source: Strogatz
Dimension

Any point on a 1D curve can be Euclidian

represented by one number, the geometry
distance d from the start point.

Any point on a 2D surface can

be represented by two numbers.
One possible method is to grid
the surface and to measure two
distance along the grid lines

Any point in 3D can be

represented by three numbers.
Typically these numbers are the
coordinated of the point using an
orthogonal corrdinate system.
Dimension

Any point on a 1D curve can be Euclidian

represented by one number, the geometry
distance d from the start point.

reference
point

0
distance
Koch curve

The Koch curve is obtained as

follows: start with a line segment
S0. To generate S1, delete the
middle 1/3 part of S0 and
replace it with two other 2-sides
of an equilateral triangle.
Subsequent stages are
generated recursively by the
same rule. The limit K=S is the
von Koch curve.

Source: Strogatz
Koch curve
What is the length of the Koch cuve?
Following the same procedure as for the
Cantor set, we find:
L0=1
L1=4/3
L2=(4/3)2
Ln=(4/3)n
L=limn->(4/3)n =
The length is thus infinite.
In fact the same infinite limit is obtained for
any value of L0. Thus, the distance
between any pair of points on the Koch
curve is infinite.
This suggests that K is more than 1D, but
would we say that it is 2D? It has no
"area"...
Source: Strogatz
Fractal dimension
The simplest fractals (such as the Cantor set and the Koch curve) are self-similar:
they are made of scale-down copies of themselves. The dimension of such fractals
can be defined by extending an elementary observation about "classical" self-similar
objects. A mathematical description of dimension is based on how the "size" of an
object behaves as the linear dimension increases.
Example:

number of squares = m = 1 m = 4 m = 9
scale factor = r = 1 r = 2 r = 3

If we shrink the square by a factor of 2 in each direction, it takes 4 of the small

squares to equal the whole. If we scale the originl square by a factor of 3, we need
9 small squares. In general, if we reduce the side dimensions of the square by a
factor r, it takes r2 smaller squares to equal the original.
If we play the same game with cubes, we notice that we need 8 cubes of scale 2, 27
cube of scale 3, ... i.e. r3 cubes of scale r. Source: Strogatz
Fractal dimension
In summary:
In 1D (line segment), we need r segments of scale r to equal the original segment.
In 2D (squares) we need r2 squares of scale r to equal the original square.
In 3D (cube) we need r3 cubes of scale r to equal the original cube.

The exponents 1,2,3 are not accident. They reflect the dimentionality d of these
"classical" objects. This relationship between the dimension d, the scaling factor r
and the number m of rescaled copies required to cover the original object is thus:

Rearranging the above equation gives an expression for dimension depending

on how the size changes as a function of linear scaling:

Source: Strogatz
Fractal dimension

Application to the Cantor set

S0
S1
S2
S3
S4
S5
S6
...
S

Each set is composed of m = 2 copies of itself.

Each copy is scaled by a factor r = 3.
Thus the dimension of the Cantor set is

Source: Strogatz
Fractal dimension

Application to the Koch curve set

Each set is composed of m = 4 copies
of itself (from one segment, we get 4
segments).
Each copy is scaled by a factor r = 3
(each segment has a length=1/3 of the
length of its parent).
Thus the dimension of the Koch curve
is:

Source: Strogatz
Fractal dimension

The box dimension

To deal with fractals which are not self-similar, we need to further generalize the
notion of dimension. Various definition have been proposed. All the definitions
share the idea of "measurement at a scale " and how the measurement varies
as ->0.
One kind of measurement involved covering the set with boxes of size .

The minimum number of

segments of size needed The minimum number of squares of size
to cover the original needed to cover the grey aera A is:
segment of length L is:
N() = A / 2
N() = L / Source: Strogatz
Fractal dimension

The box dimension

In general, for "classical" objects, we have a power-low relationship:

N() ~ 1 / d
We interpret d as a dimension, usually called the box dimension. An equivalent
definition is:

Application to the Cantor set

Recall that the Cantor set is covered by each of the set Sn used in its
construction. Each Sn consists of 2n intervals of length (1/3)n, so if we pick
=(1/3)n, we need 2n of these intervals to cover the Cantor set. Hence, N=2n
when =(1/3)n. Since ->0 as n->, we find:

Note that this is consistent with the similarity

measurement introduced previously
Source: Strogatz
Fractal dimension

Application to a fractal that is not self-similar

Let's consider the following fractal: A square region is divided into 9 equal
squares. One of them is selected at random and discarded. Then the process is
repeated on each of the 8 remaining squares and so on.

What is the box dimension of this fractal?

Pick the unit of length to equal the side of the oringinal square. Then S1 is covered
(with no wastage) by N=8 squares of side =1/3. Similarly S2 is covered by N=82
squares of side =(1/3)2. In general N=8n when =(1/3)n. hence:

Source: Strogatz
Fractal dimension

Hausdorff dimension
When computing the box dimension it is not always easy to find a minimal
cover. There is an equivalent way to compute the bow dimension that avoids
this problem. We cover the set with a square mesh of boxes of side , count the
number of occupied boxes N(), and compute d as before.
Even with this improvement the box dimension is rarely used in practice. Its
computation requires too much storage and computer time. the box dimension
also suffers from some mathematical drawbacks.
One alternative dimension, called the Hausdorff dimension, is also based on
covering sets with boxes but uses boxes of varying sizes. It is however hard to
compute numerically.

List of fractals objects sorted by their Hausdorff dimension:

http://en.wikipedia.org/wiki/List_of_fractals_by_Hausdorff_dimension
Source: Strogatz
Multifractals

Multifractals are a generalization of a fractals which are not

characterized by a single dimension, but rather by a continuous
spectrum of dimensions.
Fractal vs Euclidian geometry

Euclidian geometry Fractal geometry

traditional modern invention

based on a characteristic size or scale no specific size or scale

suits description of man made object appropriate for geometry in nature

described by a usually simple formula described by an algorithm

example: example
x2+y2+z2 = r2 zn+1=zn2+z0
described a sphere describes the Mandelbrot set
Examples of fractals
Cantor set: variants

Cantor comb

Cantor collar

Cantor quare
Cantor set: variants

Cantor cube
(Cantor dust)

Cantor tartan
Devil's staircase

As Pacman eats the dots, he gets heavier. Imagine that his weight after eats
all the dots is 1. Lets graph his weight with time. Obviously, it will not increase
gradually, but will instead go up like this:

This is called the Devils

Staircase. In this picture, we
colored the area below it to
make it more visual.

The Devil's stair case is also

called the Cantor function.

Source:
http://library.thinkquest.org/26242/full/fm/fm7.html
Devil's staircase
Suppose that we pick a point at random from the Cantor set. The Cantor function
(or Devil's stair case) actually gives the probability P(x) that this point lies to the
left of x, where 0<x<1 is some fixed point.

P(x)

Source: Strogatz; Figure from http://en.wikipedia.org/wiki/Cantor_function

von Koch Island
Menger sponge
Mandelbrot set

The Mandelbrot set is the set of points z0 in the complex

plane for which the iteration zn+1 = zn2 + z0 remains bounded.
Examples
z0 = 1 z1 = z02+z0 = 1+1 = 2
z2= z12+z0 = 4+1 = 5
z3= z22+z0 = 25+1 = 26
z4= z32+z0 = 25+1 = 677
... => diverge => does not belong to the set => c
z0 = i z1 = z02+z0 = i2 + i = -1+i
z2 = z12+z0 = (-1+i)2 + i = - i
z3 = z12+z0 = (-i)2 + i = -1+i
... => oscillates between 2 values => bounded
=>belongs to the set => g
Source: wikipedia
Mandelbrot set

The Mandelbrot set is the set of points z0 in the complex

plane for which the iteration zn+1 = zn2 + z0 remains bounded.

Algorithm
foreach z0 in the complex plane A more colourful picture can be
for i=1:N generated by coloring points not in
z=z^2+z0 the set according to how quickly
end
if |z|>th
the sequence diverges to infinity
plot z in while
else
plot z in black
end
end
Mandelbrot set

The Mandelbrot set is the set of points z0 in the complex

plane for which the iteration zn+1 = zn2 + z0 remains bounded.

Image: wikipedia
Mandelbrot set

The Mandelbrot set is the set of points z0 in the complex

plane for which the iteration zn+1 = zn2 + z0 remains bounded.

Image: wikipedia
Mandelbrot set

Properties of the Mandelbrot set

The Mandelbrot set has a fractal boundary.
The Mandelbrot set is connected (Hubbard & Douady).
The area of the set obtained by pixel counting is 1.50659177 +/- 0.00000008
(Sloane's; Munafo; Lesmoir-Gordon et al. 2000) and by statistical sampling is
1.506486 +/- 0.000004 with 95% confidence (Mitchell 2001).
The region of the Mandelbrot set centered around -0.75+i is sometimes
known as the sea horse valley because the spiral shapes appearing in it
resemble sea horse tails. Similarly, the portion of the Mandelbrot set centered
around 0.3+0i with size approximately 0.1+0.1i is known as elephant valley.

Source: Wolfram
Mandelbrot set

A striking correspondence
between the Mandelbrot set
and the logistic map

Source: wikipedia
Julia set

A Julia set is a kind of mirror image of the Mandelbrot. Take the

same basic family of functions: f(x)=x2+c. But instead of varying c
(as done to obtain the Mandelbrot set), keep c fixed, and vary x.
The Julia set for c is the set of x-values for which iterating f does
not diverge. There's an infinite number of Julia sets - one for every
possible c.

c = -0.4+0.6i c = 0.285+0.01i c = 0.8+0.156i

Figures: wikipedia
Julia set

Link with the

Mandelbrot set

Figure from:
http://mail.colonial.net/~abeckwith/00
6B0D39-70E903AC-006B0D39
Weierstrass function

Weistrass function
Weierstrass function

Weistrass function
Applications in biology
Natural fractals

Both, geometrically and dynamically, biological systems

are intricate. Concerning geometry, the natural world
can not be described in terms of the familiar geometry
of lines, triangles, squares, and circles. Instead,
mountains clouds, and coastlines are fractals structures
that always look the same as portions of them are
enlarged.

Quotation from: Agn et al (2000) Cell Biol Int

Fractals in physiology
Fractals in physiology
Some of the most visually striking examples of fractal forms are found in physiology:
The respiratory, circulatory, and nervous systems are remarkable instances of
fractal architecture, branches subdividing and subdividing and subdividing again.
Although no clear genetic, enzymatic, or biophysical mechanism yet have been
shown to be responsible for this fractal structure, few doubt this. Careful analysis of
the lungs reveal fractal scaling, and it has been noted that this fractal structure
makes the lungs more fault-tolerant during growth.

Lungs Heart Vessels

Source: http://classes.yale.edu/fractals/panorama/Biology/Physiology/Physiology.html
Fractals in physiology
Dog For comparison with human lungs,
here are some other mammalian
lungs. Note the considerable
differences in branching geometry.

Pig

Manatee
(lamantin)

Camel

http://classes.yale.edu/fractals/panorama/biology/
physiology/animallungs/animallungs.html
Fractals in biological time series

The fractal concept can be applied

not just to irregular geometric forms
that lack a characteristic (single)
scale of length, but also to certain
complex processes that lack a
single scale of time. Fractal
processes generate irregular
fluctuations across multiple time
scales, analogous to scale-invariant
objects that have a branching or
wrinkly structure across multiple
length scales. The irregularity seen
on different scales is not readily
distinguishable, suggesting
statistical self-similarity.

Goldberger et al (2002) Fractal

dynamics in physiology: Alterations
with disease and aging, PNAS
99:2466-72
Fractals in biological time series

patient with severe

congestive heart Note that the healthy
failure record (B) is far from a
homeostatic constant
state (it is characterized
by non-stationarity and
healthy patient patchiness). These
features are related to
fractal and nonlinear
properties. Their
patient with severe breakdown in disease
congestive heart failure may be associated with
the emergence of
excessive regularity (A)
subject with a cardiac and (C), or uncorrelated
arrhythmia, atrial fibrillation, randomness (D).
which produces an erratic
heart rate.

Goldberger et al (2002) Fractal dynamics in physiology:

Alterations with disease and aging, PNAS 99:2466-72
Fractals in the nervous system

Werner G (2010) Fractals in the nervous system: conceptual implications for

theoretical neuroscience, Frontiers in Physiology
Fractal (enzyme) kinetics

Savageau MA. (1995) Michaelis-Menten mechanism reconsidered: implications of fractal

kinetics. J Theor Biol 7:115-24.
Fractal (enzyme) kinetics

Savageau MA. (1995) J T heor Biol 7:115-24.

Savageau MA. (1998) Biosystems 47:9-36.
Fractal in biological systems

Fractal geometry in physiology

?
Fractal time series Fractals kinetics
Some more fractals...
Fractals-generated landscapes
Examples of artistic fractals
Fractal crop circles

In the mid 90s some of the found

crop circles were based on
fractals. For instance the so-called
'Julia set' at Stonehenge in 1996
and the 'Triple Julia set' at Windmill
Hill the same year. Or the Koch-
Fractals of extra-terrestrial origin ;-) fractals in 1997.
Source:
http://www.cropcircleconnector.com/
Bert/3dfractals.html
Matlab code

http://homepages.ulb.ac.be/~dgonze/FRACTALS/fractals.html

Cantor
Koch
Koch flake
Sierpinsky
Menger
Menger 3D
Ford
Julia
Mandelbrot
Tree
Fern
Landscape
Further reading