Chapter 9

Fractal Geometry of Architecture

Fractal Dimension as a Connection Between Fractal
Geometry and Architecture

Wolfgang E. Lorenz

Abstract In Fractals smaller parts and the whole are linked together. Fractals are
self-similar, as those parts are, at least approximately, scaled-down copies of the
rough whole. In architecture, such a concept has also been known for a long time.
Not only architects of the twentieth century called for an overall idea that is mirrored
in every single detail, but also Gothic cathedrals and Indian temples offer self-
similarity. This study mainly focuses upon the question whether this concept of
self-similarity makes architecture with fractal properties more diverse and interest-
ing than Euclidean Modern architecture. The first part gives an introduction and
explains Fractal properties in various natural and architectural objects, presenting
the underlying structure by computer programmed renderings. In this connection,
differences between the fractal, architectural concept and true, mathematical Frac-
tals are worked out to become aware of limits. This is the basis for dealing with
the problem whether fractal-like architecture, particularly facades, can be measured
so that different designs can be compared with each other under the aspect of
fractal properties. Finally the usability of the Box-Counting Method, an easy-to-use
measurement method of Fractal Dimension is analyzed with regard to architecture.

9.1 Fractal Concepts in Nature and Architecture

9.1.1 From the Language of Fractals to Classification

For a long period of time, nature has been an inspiration for architects, which implies
copying natural forms, translating them into floral ornamentation or using underly-
ing structures found in nature for static optimization and many other possibilities

W.E. Lorenz ()

Digital Architecture and Planning (IEMAR) E259.1, Institute of Architectural Sciences,
Vienna University of Technology, Treitlstrasse 3/1, 1040 Vienna, Austria
e-mail: [email protected]; www.iemar.tuwien.ac.at

P. Gruber et al. (eds.) Biomimetics – Materials, Structures and Processes, Biological 179
and Medical Physics, Biomedical Engineering, DOI 10.1007/978-3-642-11934-7 9,
© Springer-Verlag Berlin Heidelberg 2011
180 W.E. Lorenz

of translation. Fractal Geometry has provided scientists with an improved approach

to analyzing and generating natural forms. In 1975, the computer-scientist Benoit
Mandelbrot introduced the term Fractal to describe irregular, non-smooth curves,
and to distinguish self-similar non-smooth structures from smooth Euclidean ones.
When he writes that clouds are not spheres, mountains are not cones or bark is
not smooth, this shows very clearly that Euclidean geometry lacks the capability
to describe natural objects [1]. With his concept of Fractal Geometry, Mandelbrot
has increased and broadened our insight in nature. Over two thousand years,
our environment had primarily been described in terms of classical Euclidean
Geometry – the geometry of simple shapes – and people had focused on a simplified
view of nature. Fractal Geometry, however, offers methods to describe and produce
nature-like objects directly, using the underlying structure rather than describing
them with simple forms or reducing the overall form by dividing it into more simple,
smooth components. The language of Fractals enables us to describe the twisted,
rough and irregular surfaces of our environment by a few simple rules without
reducing their complexity. The major question in this article is whether the language
of Fractals may also be applied to architecture.
The language of Fractals can be illustrated with the help of the Barnsley
fern, a computer generated fern named after the American mathematician Michael
Barnsley (Fig. 9.1). It is supposed to resemble the Black Spleenwort, Asplenium
adiantum-nigrum. At first sight, the shape of the natural fern can only be described
precisely by defining each detail in an extremely time-consuming process. This type
of description could be compared with a non-compressed computer image, saving
information about every single dot in large files. Describing a facade in that way
would mean providing information on size and position of every single architectural
element, including roof, doors, windows, window-strips and columns down to even
small-sized ornaments. Barnsley used the iterated function system (IFS) to generate
an image of the fern with only four relatively simple transformation rules. Four
different configurations of translating, reducing and rotating an initial rectangle or
converting it into a single line – and the right fine-tuning – are the basis for a fern-
like image. In that process the overall information is reduced to a few underlying
construction rules – the algorithm is then the language that is used to describe
Fractals. Even though the complex output may not be an exact copy of its natural

Fig. 9.1 Barnsley fern

9 Fractal Geometry of Architecture 181

counterpart, the character is the same and, by fine-tuning, the image can become a
very close approximation. How can we draw an analogy between that approach and
methods found in architecture? Some architects used basic ideas and basic motifs
as a designing tool. Horizontality for instance can be regarded as the basic idea of
Robie House by Frank Lloyd Wright – not only for a first impression but also for a
deeper understanding.
Focusing on details of the fern we become aware of its fractal characteristics. No
matter which part is analyzed, it looks like the whole or, putting it differently, the
same characteristics can be discovered on each level of scale – the basic settings of
the configuration are present in each part. This phenomenon is called self-similarity,
and one of the main properties of Fractals is as follows: smaller parts of an image are
hierarchically linked to the whole. Analyzing nature in terms of Fractals shows that
nature, from mountains to coastlines and down to plants, is based on self-similarity –
it is self-similarity that makes nature so fascinating from the large to the small.
In case of Robie house horizontality is evident in the overall view as well as in
the horizontally stretched roofs, window-strips and even in details such as long-
stretched bricks. The basic idea of horizontality is the common denominator for
all the individual components that form a complex whole. As the example of the
fern illustrates, Fractal Geometry can be used to simulate complex natural objects
with the help of simple algorithms, even if, in many cases, no clear rule of the
object’s development can be identified at first sight. That is, the right configuration
of translating, reducing and rotating has to be found first. But as soon as the right
configuration has been detected, complex natural objects can be described with the
help of Fractal Geometry. This may also be true for complex artificial objects.
In architecture Mandelbrot distinguished buildings of the Beaux-Arts, which
are close to Fractal Geometry, from buildings by Mies van der Rohe, which he
calls scale-bound throwback to Euclid [1]. It seems that Euclidean Geometry of
Modern architecture – throughout this study the term of Modern Architecture is
used for those styles using the design vocabulary of simplification of form and the
elimination of ornament – is mostly reduced to a few elementary objects. Even if
every single architectural element of a smooth architectural design has to be defined
by its size and position, the resulting data set remains manageable. But, would
this mean that the Fractal concept is not valid for classical Modern architecture?
At first sight, no connection between different elements can be found that might
reduce the data set to a set of a few construction rules or to basic ideas. But on
closer observation there are such rules, for instance rules developed out of systems
of proportion that are applied to all elements from the overall facade to the very
small detail. Fractal-like architecture is characterized by the presence of various
details of several different sizes that are linked together. Le Corbusier demonstrates
how this connection can even be based on a certain angle only, e.g. at the facade
of a villa he designed in 1916 [2]. There a specific angle defines the diagonal of
the overall facade whose numerous parallels together with their perpendicular lines
determine the dimension of elements of the second order such as doors, windows
down to certain details. Le Corbusier created a similar example with House Ozenfant
in 1923. In those examples all elements such as windows and individual parts but
182 W.E. Lorenz

also the whole are based on a system of proportion. The only difference between
these examples and architecture that is regarded as more fractal-like is the limited
range of scales.
If architecture is analyzed from the point of view of Fractal Geometry, an index
of coherence can be introduced. On the very low end, there is an absolutely smooth
plane then, an empty rectangle, which is not fractal but belongs to Euclidean
Geometry. The index of coherence is increased if fractal properties such as self-
similarity and others, which will be described in the following section, are present.
Facades belonging to this section are then called fractal-like, because they are not
Fractals in a mathematical sense, but offer fractal properties within a limited range
of scales. In case of such facades, single architectural components of different sizes
are combined, leading to a consistent overall composition. At the upper end of the
index of coherence, we can find architectural examples with many unlinked details,
hence confusing patterns. “Unlinked details” means that architectural components
of different sizes are not interrelated, for instance by a formal basic idea. That
means while approaching such a building the observer is confronted with constantly
changing and often confusing new impressions.
The systems of proportion at the facade of House Ozenfant are only applied
within a restricted range of scales and, in addition to that, the facade looks rather
smooth. Those are reasons that put the facade of House Ozenfant on the lower end
of the index of coherence, while Robie House can be placed in the middle because
of its self-similar characteristics.

9.2 Fractals: A Definition from a Mathematical

and an Architectural Point of View

Fractals can be explained by their properties. These include roughness, self-

similarity, development through iterations, infinite complexity, dependence on
starting conditions, they are common to nature and their Hausdorff dimension
exceeds their topological dimension. A simplified definition of the Hausdorff
dimension can be illustrated by covering a set of points of finite expansion in a three
dimensional space with a minimal number of balls N.R/ of radius .R/. Decreasing
the radius will enlarge the number of spheres. The Hausdorff dimension .dH / is then
defined by
log.N /
dH D  lim :
R!0 log.R/

9.2.1 Roughness and Length Measurement

Roughness can be described very well if we look at the coastline of Norway,

where Fjords with their sub-bays, inlets, cliffs and rocks lead to a very fractional
9 Fractal Geometry of Architecture 183

border in contrast to a circle offering a smooth border. This also becomes evident
in connection with length measurement. Benoit Mandelbrot introduced Fractal
geometry with the question: “How long is the coastline of Britain” [1]. This is not
a trivial question. For length measurement, different maps of various scales can be
used, which represent different distances between the observer and the coastline.
A large scale may correspond to the view out of an aeroplane hundreds of meters
above ground, while a small scale may correspond to the impression an observer
gets when walking along the coastline on foot. If we keep in mind the example of
the Fjords, it is obvious that each time a smaller part of the coastline, represented
on a map of smaller scale, is analyzed, new sub-bays will become visible, which
could not be identified in the previous large-scale map. The length measurement
of the coastline on a map using a larger scale presents a rough image of the real
coastline as it only includes the larger bays. If the same section of the coastline
is measured on a small-scale map offering a more detailed version of reality, the
larger bays mentioned before are indented by smaller bays so that the measured
length increases. In other words, length depends on the details presented on the map
used, or on the measuring devices in the real world. In contrast to the Fjords, if
the borderline of a circle is analyzed in the same way, the length measurement will
follow a different behaviour. If we use smaller and smaller measuring devices, length
will tend to a limiting value very quickly (approximation through polygon with “N”
edges). This is because no additional details will be presented while zooming in.
The difference between Fractals and Euclidean objects (e.g. squares and circles)
then lies in the fact that length measurement fails for the first ones.
What does length increase look like in a mathematical fractal such as the Koch
curve, a famous representative of a classical Fractal? The construction rule of the
Koch curve starts with a straight line of length one, which is called the initiator.
This line is then divided into three equal parts (Fig. 9.2). The middle part is replaced
by an equilateral triangle, whose base line is removed. This adjustment of four
smaller lines is called the generator of the Koch curve. As the resulting structure
consists of four scaled-down copies (four lines) of 1/3rd of the initiator (one line
of the initial size one), the overall size increases by 4/3rd. In the next step, each of
the four new lines is again replaced by the generator, resulting in 16 scaled-down
copies. The scaling factor in relation to the initial line is now 1/9th. That means
the length increase is 4/3rd by 4/3rd or in other words 16/9th. From this it was not
only followed that the length increases by 4/3rd from one iteration – that is a step of
replacement – to the next, but also that the offsets of the curve increase – a higher
degree of roughness can be identified. For the third iteration length increase – in
relation to the initiator – is 4/3rd by 4/3rd by 4/3rd or 4/3 above 3. Expressed in a
more general way this leads to the following equation

L.i / D .N.1/  s.1/ /i

with .i / indicating the number of iterations, .L.i/ / the length of the curve after .i /
iterations, .N.1/ / the number of single lines of the generator and .s.1/ / the reduction
factor of the generator, being one third for the Koch curve. From one iteration to the
184 W.E. Lorenz

Fig. 9.2 Koch curve

next, the curve gets longer and longer, hence more and more twisted, which means
it is getting rougher from iteration to iteration. Because mathematical Fractals are
results of infinite iterations, the final length is infinite.

9.2.2 Scale Range and Distance

As the above example of coastlines shows, scale range and distance determine the
degree of roughness and detail the viewer is able to see. These factors also influence
a viewer’s perception of a building. From far away, the viewer will only perceive
very large components, the silhouette and significant edges. This view corresponds
to an elevation of large scale that only includes a few details. Approaching the
building, the observer’s attention begins to focus upon the sequence of base, middle
or roof part. Then windows and doors or the rhythm of columns are the most
prominent parts. Zooming to one of the next levels it turns out that (e.g.) windows
consist of smaller details such as window frames and window handles, but even
walls offer smaller parts such as face bricks or tiles. Thinking about grain of wood,
the cascade of details will come down to the material itself. For Salingaros, the size
of the smallest detail has to correspond to the smallest perceivable scale [3]. This
clearly shows that facades are in general not smooth Euclidean two-dimensional
planes, which is per se valid for Modern architecture as well as organic architecture
or any other style. Elevations as two-dimensional translations of facades are rather
something between lines, defined by edges, and the plane, defining the surface.
Before architecture can be analyzed and buildings compared with each other from
the point of view of coherence between elements of different sizes, or more precisely
with regard to self-similarity, it is important to define the range of scale first – it has
to be standardized. The range correlates with the distance of the observer while
approaching the building and the smallest possible detail that can be perceived from
certain distances.

9.2.3 Self-Similarity: An Important Attribute of Fractals

Self-similarity is an important property of Fractals. In mathematical terms, two

objects are similar if their corresponding angles are identical and their correspond-
ing sides are in proportion regardless of their size. An object is called self-similar if
one or more of its parts look like the whole. The parts can be exactly similar or
approximately similar. The Koch curve is exactly self-similar because each part
is an exact, scaled-down copy of the whole. The final Koch curve is the result
9 Fractal Geometry of Architecture 185

of applying similarity transformations (scaling, translation, rotation) to the initial

object, which modifies proportion by the same factor. The resulting scaled-down
copy may be rotated or transformed while the shape remains the same. If the pieces
of the object are scaled down by different amounts in different directions, the fractal
is then called self-affine instead of self-similar. When simulating nature the factor of
chance has to be added. The new structures are called statistically self-similar, that
is whenever small copies, looking like the whole, have variations [4]. Parts show
the same statistical properties at many different levels of scale. This is important for
describing natural objects, but also architecture. When we, for instance, examine
certain parts of coastlines, they do not just represent scaled-down, transformed and
rotated copies of the whole, but show a similar character and degree of irregularity.
Cliffy coastlines offer the same strong irregularity from the very large to the very
small level of scale, similar to gently twisted coastlines offering the same softness
all over as well.
Geometric shapes with fractal properties had already been known long before
the term Fractals was introduced and long before all the facts about Fractals
were combined to form a theory. The Koch curve for instance – a continuous
curve without tangents – was already presented in 1904. In connection with length
increase, it was indicated that the number of components (single lines) of smaller
scales increases from one iteration to next. In architecture, fractal properties had also
been used consciously but also unconsciously, which is true for Gothic cathedrals
as well as for Robie house by Frank Lloyd Wright [16, 17]. Carl Bovill already
drew an analogy between writings of Frank Lloyd Wright and the fractal concept
[5]. According to Frank Lloyd Wright, nature should be a source of inspiration.
Carl Bovill demonstrates that for Frank Lloyd Wright the underlying organizing
structure of nature – which is finally the fractal concept – is of particular interest.
The fundamental idea of bringing the characteristics of a horizontal environment,
the Prairie, into the building by Frank Lloyd Wright, Robie House, is implemented
on many levels of scale, from the wide overhanging roof and stretched storeys
over window strips and horizontal parts of walls to the design of the horizontal
and straight joints [15]. The number of components that are evident on a specific
level of scale increases from the overall view over the level of scale that includes
windows down to ranges of scale on which details of bricks and material become
visible. Analogously, a cauliflower consists of a smaller number of similar scaled-
down components that again consist of many scaled-down components. In this sense
examples such as Robie House can be called fractal-like architecture. Architecture
is called fractal-like if the whole and all other formal elements are derived from
one basic idea and, by that, a simple, specific form characterizes the expression of
the building. Parts are then reflections of the whole, the formal elements are held
together in scale and character [5]. Eero Saarinen also pointed out that a building
should follow the strong, simple concept of the whole – each part has to be an
active component of a certain overall theme. This is valid irrespective of the point
of time when decisions to follow such an overall theme are taken: Right from the
start with regard to ground plan or construction-system or at a later stage, when
186 W.E. Lorenz

detailed elements such as colours inside the house or even door-handles are in the
architect’s focus of interest [6].

9.2.4 Architectural Examples

There are different approaches to self-similarity in architecture: Self-similarity in

architecture may arise from a basic proportional system, from a basic form or
from an overall idea. To give another example, Rietveld-Schröder house in Utrecht
(Netherlands, 1924) offers coherence from the large elements to the small ones [17].
The architect Gerrit Rietveld translated the use of form of the Stijl, with basic forms
and basic colours, into architecture. Those basic forms are taken from Euclidean
Geometry and include straight lines, planes and slabs that are detached from each
other and seem to glide past – the floating space playing an important role [7]. The
building is characterised by large openings, horizontal and vertical elements and
an intersection between inside and outside elements. The cascade of architectural
components starts with basic spatial and constructive structure, formed by few main
elements of large white slabs (windows and doors are cut out). Then, on a smaller
level of scale smaller slabs for balconies and canopies follow, while linear elements
such as rainwater pipes add vertical and horizontal accents. Inside, an analogy can
be drawn between the basic elements and window-frames, sliding and revolving
panels and their linear hanging. Coming even closer, the basic elements are repeated
in the design of the furniture, from chair to modular cupboard. The translation of
basic elements into smaller levels of scale connects the furniture to the architecture
around. This example illustrates that even though the number of iterations is limited,
the basic idea of separating components by means of form and colour and the
composition of planes and linear accents is nevertheless evident within all levels
of scale.
There is one more example we should deal with: in 1956, Bruce Goff constructed
the Joe Price studio in Bartlesville, Oklahoma. The basic idea of construction was
focused on 60ı angles and their multiplication or subdivision to hexagons and
triangles [8]. Once again self-similarity occurs on the basis of a simple component, a
certain angle. Variations of the basic element can be found from the large elements,
including the roof, with the basic shapes forming a twisted outline, down to the
small size of decorative elements. Inside in turn, the ceiling, the walls, the hexagonal
sitting hole, which is a hollow in the floor, and certain details are based on triangular
shapes. The individual components of different sizes are variations of the basic
theme and therefore offer coherence within all levels of scale. Such variation arises
from the material used and from purpose such as deflecting sound in the music room.
The work of Antoni Gaudı́ offers fractal-like architecture as well [16]. Although
buildings by Antoni Gaudı́ appear very complex, they are nevertheless coherent.
This is not a surprise because Gaudı́ pointed out the importance of dealing with
details to produce a complete work of art. As an example the use of curved
construction stones and nature-like, organic forms can be found all over Sagrada
9 Fractal Geometry of Architecture 187

Familia from outside to inside from large elements to small details. Gaudı́ was
interested in the forces of nature that act behind the surfaces rather than in the
shape, hence the surface itself [9]. Inspired by nature, Antonio Gaudı́ introduced
catenary shapes as idealized forms of arches [10]. He developed catenarian-models
and cable-models with sandbags illustrating the reverted interrelation of forces for
columns and pillars. The result was then an upside down model of the structure
in the building. Gaudı́ found out that the use of parabolic arches and inclined
buttresses could withstand the forces involved. On smaller levels of scale, hyperbolic
paraboloids can be detected in the vaults but also at the base of columns [10]. Antoni
Gaudı́ used cone, cylinder, simple hyperboloids, hyperbolic paraboloids to design
non-smooth architectural components. The analysis of a small part of the building
will make the observer think about the whole, and the building is coherent in form
and character.
Self-similarity alone, however, does not define a Fractal. Considering that a
line between two points can also be divided into smaller parts that are scaled-
down copies of the whole, the line nevertheless does not increase in length from
one iteration to the next, and there are no additional details offered either when
zooming in. Or, zooming in on a circle will show a more and more straightened part
of the circumference – apart from irregularities arising from the drawing itself –
but does not offer additional details. The latter two examples belong to Euclidean
Geometry because no further details arise when zooming in.

9.2.5 Developed Through Iteration

Generalized for a first approach, Gothic cathedrals can be described by verticality

and light-flooded interior rooms, which was made possible by pointed arch, flying
buttress and ribbed vault. They are an expression of unity between single compo-
nents, inside and outside, and the whole: In numerous functional and decorative
elements such as windows, portals, baldachins, pinnacles, attics, Gothic gables and
tabernacles, references to the whole can be identified. Examples such as the Gothic
windows of the Southwest tower of the cathedral of Cologne or of the Angel Choir
of the Cathedral in Lincoln definitely provide excellent examples to describe how
Fractals can be constructed. The overall shape, a pointed arch, defines the initiator,
which is then replaced by the generator, a pointed arch of the same size vaulting two
smaller ones of half the size. In the next step, each pointed arch is again replaced by
the generator and so forth. Figure 9.3a shows a series with the initiator that is the
starting object, the generator that is the replacing rule and some iterations. Although
the replacing rule for the Gothic window may be applied infinite times in theory, in
the real world there are some restrictions arising from the material used and from the
usability as a window. The built version tries to find a harmonious balance between
a slight big opening and construction rules and therefore already ends after a few
levels of replacement or iterations, respectively. Furthermore, the algorithm may be
adopted with regard to the reduction of the diameter of the shafts.
188 W.E. Lorenz

Such replacement procedures can also be identified as underlying construction

elements of Indian temples, such as the Sikbam of a Jaina temple on Mount Girnaz
[11]. The second image in Fig. 9.3b shows an output of a construction rule that
is derived from this example. The basic curved shape of the whole temple, rising
out of a square, has undergone a transformation in size and position, bringing up
four additional similar shapes at the sides of the main part. Each of these four new
elements is then transformed in the same way, resulting in a complex whole, where
all parts are formally linked with the whole. Figure 9.3b shows the initiator and the
fourth iteration, each of three different configurations of such an insertion rule. The
basic transformation rule, scaling down the initial shape and positioning it at each
side, is adjusted in the way that different reduction factors and movements are used.
It is also conceivable that the reduction factor varies within the same iteration by the
factor of chance. Using different basic shapes (initiator) and adjustments, the basic
rule may then lead to high rising as well as vaulted structures (Fig. 9.4a, b).
The output of such computer-generated self-similar structures cannot be used
for architectural design without changing or adapting it with regard to the limits
imposed by the material used, construction, function and environmental influences.
Such computer outputs are only useful to show how self-similar patterns in
connection with architecture can be generated in a simplified way. If facade
generating programs are developed to compute facades with visual depth, the
question of fabrication and usability has to be considered. Such facade generating
programs might start with a simple box, dividing it in .n/ by .m/ boxes, which are
removed or added in certain ways. Single boxes vary in their expansion to the front,
so that variation in depth is achieved. Then the new boxes are manipulated in the
same way, using the same parameters or slightly adopted ones with, e.g. random

Fig. 9.3 Various examples of geometric output – (a) Gothic window algorithm, (b) temple
algorithm

Fig. 9.4 Different insertion rules – (a) high rising, (b) vaulted, (c, d) window element
9 Fractal Geometry of Architecture 189

factors for expansion. Figures 9.4c, d are based on such a rule and offer self-similar
patterns after a few iterations. Windows are either included in such algorithms by
cutting out middle parts or such outputs are simply arranged around existing cuts. In
the first case, one large cut in the facade is surrounded by smaller ones, surrounded
by even smaller cuts and so on.

9.2.6 Differences Between Architectural and Mathematical

Fractals

The main difference between nature or fractal-like architecture and mathematical

Fractals is the limited number of iterations. Although mathematical Fractals are
theoretical constructions offering infinitely small parts, self-similarity of nature and
architecture only exists for a limited range of scale. It could be said that nature
and architecture only uses a limited number of iterations. This will be illustrated by
comparing the Sierpinski triangle, a mathematical fractal, with Castel del Monte,
situated in the Apulia region, by the Holy Roman Emperor Frederick II. The
Sierpinski triangle can be simulated by starting with an equilateral triangle as the
initiator. This initiator is then replaced by a generator consisting of three equilateral
triangles, each scaled-down copies of the initiator by the factor of 0.5. All corners
of the initiator act as fixed points for one of those three new triangles when they are
scaled down. In other words an equilateral triangle is cut out in the middle of the
initiator where the corners of this cut out triangle are situated on the middle points of
the sides of the initiator. For the next iterations, each triangle is subjected to the same
procedure, leading to nine triangles. In theory, this is repeated infinite times leading
to a real Fractal, where even the smallest part is a scaled-down copy of the whole.
In case of Castel del Monte, which offers the beginning of a replacing algorithm,
the difference between mathematics and architecture becomes evident. The basic
octagonal shape is complemented by further octagons at its corners. For the next
iteration, each of these octagons would be complemented by further octagons at
their corners and so on. But the “real” Castel del Monte only offers the first iteration.
Nevertheless, the circumference of Castel del Monte increases from the initiator
to the first iteration and offers a rough outline. Together with the rough stonewall
this distinguishes Castel del Monte from smooth surfaces with a lower index of
coherence.

9.2.7 Fractals as a Design Aid

The competition contribution for Cardiff opera house in Wales by Greg Lynn in
1994 mandates a new concept for waterfront urban space that is nonetheless in
conformity with the history of the site and Cardiff’s waterfront [12]. The analysis
190 W.E. Lorenz

of the coastline for self-similar structures from the large down to the small level
of scale (water’s edge being captured by land) together with the oval form of the
basin was translated into a rule with the starting body being replaced by three parts
of different sizes and orientation. Some iterations of this Fractal basic shape in
combination with the basin defined the character of the opera house next to it – the
opera house is then a continuation of Cardiff’s waterfront. The output was adapted
with regard to function, construction and form. Volumes were rearranged because
of requirements of the foyer, auditorium and acoustic properties, stages, studios,
offices and other purposes of the opera house.
In general, fractals can also be used as a basic design for breaking open an
otherwise straight line. For example, the characteristics of a natural coastline can
be simulated by a fractal, if the right configuration or the right insertion rules are
found. The twisting of the Fractal then leads to a coastline section, which is based on
a more environmentally appropriate scale. What can be learnt from such attempts –
forming buildings and border-lines? In general, using such Fractal-based designs
means that the architect looks for a rule coming close to the one that is inherent in
the environment on a larger scale to continue the cascade of similar character down
to the size of harbors or buildings and even to human level connecting smaller scales
to the whole. Man-made interaction will then continue the natural characteristics of
the environment and in turn will not restrict the range of scale. The basic Fractal
nevertheless has to be adapted with regard to usability concerning function and
costs, but also with regard to the fact that the result may again consist of straight
parts though on an even smaller scale.

9.2.8 Fractals Are Common to Nature

Dealing with Fractal Geometry also means focusing on nature: Clouds, bark and
trees are not smooth but rough, and snowflakes or the distribution of stars offer
self-similarity. Different fractal methods, which take advantage of self-similarity,
underline the connection between nature and fractals [20]. The right algorithms
produce models of plants, mountains, crystals and entire natural scenes. Self-
similarity may in turn be the reason why nature seems so fascinating from the
large elements to the small ones and why forms of nature seem better balanced to
us than Euclidean smooth shapes. If so, this explains why Gothic cathedrals, rural
houses and organic buildings, which rather contain self-similarity, are so fascinating
to many observers. Frequently those buildings offer details, which are prominent
on different levels of scale and are consistent with each other and the whole.
Consequently, while we are approaching the building, new details of smaller size
get into the focus of our attention reminding us of the whole. Because of variation
of components, which differentiates self-similarity from self-sameness, they also
remain diversified. Assuming that those parts are exact, scaled-down copies of the
whole, this would mean the observer can judge from looking at the whole what the
detail will exactly be like. Because of variation he can envisage what comes next but
9 Fractal Geometry of Architecture 191

may be slightly surprised and confirmed at the same time. Mandelbrot believes that
fractal art is more acceptable because it imitates nature to make the observer guess
its rules and is therefore more familiar to us [1].

9.2.9 The Factor Chance

Nature-like images produced by strict self-similar construction rules are often

too clinically “perfect” to imitate their originals to their last consequence. Local
influences such as temperature, wind, waves or nutritive substances, which deform
the objects, are missing. Effects of such influences can be imitated by the factor of
chance. This can be illustrated by means of the tree algorithm (Fig. 9.5a). Starting
from a stem of certain diameter and length, three or any other number of scaled
down copies of it are moved to the end of the stem and rotated in different directions.
Each of these branches is again copied, scaled down, moved to the end and rotated
in different directions. This rule is applied to a couple of iterations. If the rotation
is chosen randomly between certain limits as well as the length of each branch, the
resulting image resembles nature even more closely. The basic shape is a smooth
Euclidean cylinder, but the object itself becomes rougher from one iteration to the
next, while smaller pieces are added. On the one hand, the resulting tree can be
described by defining length, position and direction of each single branch, which
produces large data quantities. On the other hand, the resulting tree can also be
described just by its insertion rule.
With the factor of chance, even the Koch curve can be turned into a natural-
looking coastline. Figure 9.5b shows three such Koch curves added to one Koch
Island. In the basic rule, the middle vertex of the generators of each Koch curve
looks to the outside and so does every replacing step. To get a more nature-like
model, the middle vertex is also allowed to look to the inside. The choice whether
it is pointed to the outside or inside is chosen randomly. After some iterations, the
resulting curve is not as clinical as the origin without the factor of chance, but never-
theless offers the same characteristics and length increase. Both examples, the tree

Fig. 9.5 (a) Simulation of a tree, (b) Koch Island with factor of chance, (c) simulation of a
mountain
192 W.E. Lorenz

algorithm and the coastline, indicate the importance of variation for natural-looking
images. This has been taken into account for certain Fractal methods producing
images of plants, trees, mountains and even planets. The rule used for such a method
can be very simple, while the output will look extremely complex. This can be illus-
trated with a program imitating mountains, which can be implemented into a CAD
package quite easily. The starting image to generate such a mountain is a triangle.
Then the midpoints of each of the three sides of the triangle are marked. These points
are then moved up or down by chance and by a certain factor. The higher the factor,
the rougher the resulting mountain is. In the next step, each point is connected to
a new triangle with its neighbouring points. These triangles are then subjected to
the same rule as before (Fig. 9.5c). Parts are not identical with the whole but the
overall character remains the same from one zoom level to the next – analyzing
small parts will offer similar structures as the whole. From this follows that, the other
way round, underlying construction rules of natural objects are difficult to identify
because of variation, which have been here simulated by the factor of chance.

9.3 From Simulation to Measurement

Fractal methods generating architectural structures, as they have been introduced

with the Gothic window, the Indian temple and the facade designer, are only a
first approach to Fractal architecture. All those attempts have in common that
their algorithm is very simple, just demonstrating basic rules of existing buildings,
simulating them with the help of a computer and producing outputs of different
shapes. This first attempt can be made more sophisticated with the factor chance to
develop symmetry breaking designs such as those that may result from environmen-
tal restrictions or adjustments because of internal function. Including parameters
that simulate the influence of daylight and shadow, compactness, functional fitness
or resources directly will increase computing time. With the help of a computer, a
large number of different alternatives can be generated from which the architect can
finally make his choice. This choice can then be modified and developed further as in
the case of the competition contribution for Cardiff opera in England by Greg Lynn.

9.3.1 Curdling

Several examples of fractal-like architecture from Gothic buildings to Robie house

and simplified architectural simulations have not only shown that, similar to nature,
architecture possesses fractal properties up to a certain degree, but also that those
properties are difficult to describe because of modification by certain influences.
Consequently, for comparing works of architecture with each other – with regard
to the degree they are fractal-like – a consistent measurement method has to be
developed. If self-similarity is present in architecture, then this is also expressed
by a similar distribution of architectural elements from one level of scale to the
9 Fractal Geometry of Architecture 193

Fig. 9.6 Curdling

next – each level of scale has its elements of specific size that again contains smaller
details of similar distribution. Such connections between different levels of scale
can be demonstrated with the help of the so-called curdling, introduced by Benoit
Mandelbrot to demonstrate the process that produces a disconnected set of points
with nevertheless clustered characteristics [5]. The process is called curdling since
originally uniform mass distribution clogs together forming many small regions
of high density [4]. There, structures are generated by the factor chance where
zooming in and analyzing parts will offer similar characteristics like the whole.
Finally, such connection between many levels of scale can be given by a set of
values, characterizing the cluster. The measurement method of these values will
now be introduced with curdling.
“Curdling” was the name Mandelbrot coined for a procedure that produces a
random fractal dust in two dimensions [1]. With curdling, the starting object, a
simple square, is divided into a grid. In our example, the grid consists of three by
three cells. Tossing a coin determines whether a cell is deleted or again divided
into a three by three grid. The coin can also be replaced by a probability factor.
High probabilities lead to a higher chance for cells to remain and consequently to a
higher ‘density’. Figure 9.6 shows some iterations for different probabilities.
Considering the probability of one out of nine, this means that mathematically
one cell will remain after the first iteration. That also means that eight out of nine
cells are deleted. Because of the element of chance, in practice none may be chosen
as well, which stops the algorithm, but also more than one is possible. What happens
to the number of cells when using the probability two out of nine? Mathematically,
after the first iteration only two of nine cells remain and seven are deleted. For the
second iteration, both remaining cells are again divided into a three by three grid.
Again in each grid, two cells remain and the others are deleted. This increases the
total number of remaining cells up to two by two, hence four. In the next iteration,
each of these four cells is again divided into a three by three grid. Again two cells
of each grid remain, increasing the total number of remaining cells to two by two
by two, hence two above three. Simplified, this leads to the equation

N.i / D .N.1/ /i

giving the connection between increasing numbers of remaining cells and iteration.
There .N.i / / is the total number of remaining cells after .i / iterations and .N.1/ / the
number of remaining cells of the generator. For the probability three out of nine,
194 W.E. Lorenz

the number of remaining boxes after three iterations is then three above three, hence
N.3/ D 27. For the same iteration but for the probability eight out of nine it is
N.3/ D 512.
Simulations use a random generator of a certain probability that determines for
each cell whether it remains or not. Therefore, different simulations of the same
probability may lead to different numbers of remaining cells. The two examples on
the right in Fig. 9.6 show two different results for the probability of two out of nine,
the middle two for the probability six out of nine and the two on the left for the
probability eight out of nine. These examples indicate that the remaining cells after
.i / iterations vary for one and the same probability. For simulations, the simple
connection between the number of remaining cells and iteration of the previous
equation is then no more valid.
Considering that the grid represents a building site, the same algorithm may
simulate distribution of buildings. This time the grid does not have to be regular
and the algorithm will be stopped at a proper size, that of buildings. Then cells
of the last iteration are moved from the border defining streets in between. Once
again this is just a formal simulation using one basic rule only, but it will produce
choices of different distributions as discussion bases. More practicable models can
be developed, enlarging the basic rule by additional parameters accounting for
different influences.

9.3.2 Fractal Dimension

For indicating consistency between different iterations, we have to look for a com-
parison between the increase in number of remaining cells and the reduction factor.
With the Koch curve, the number of single lines .N.i // and the reduction factor
.s.i / / are both increasing/decreasing by the same index: number of parts after the
first iteration .N.1/ / D 4, the reduction factor of these parts compared with the size
of the initiator .s.1/ / D .1=3/; number of parts after the second iteration .N.2/ / D
42 ; .s.2/ / D .1=3/2I .N.3/ / D 43 ; .s.3/ / D .1=3/3 . From this it was derived that
there exists a connection between the number of single lines after .i / iterations
.N.i / / and the reduction factor .s.i / /. This connection is shown in the equation
 Ds
1
N.i / D
S.i /

introducing index .Ds /, which is the self-similar dimension. Modified, this leads to

log.N.i / /
DS D  ;
log S1.i /

which equals .Ds / 1.26 for the Koch curve with four lines of one third each after
the first iteration .i / D 1.
9 Fractal Geometry of Architecture 195

With the theoretical output of curdling, self-similar dimension can be calculated

in the same way. Using the example from above – tiling each cell into three by three
smaller cells – with a probability of two thirds, this means that mathematically six
cells will remain .N.1/ / after the first iteration. The reduction factor of the grid-size
is one third .s.1/ D 1=3/. Inserting these values into the last equation, the self-similar
dimension .Ds / equals 1.631. With the probability of one ninth, mathematically
only one cell remains after the first iteration while the reduction factor is again one
third. This leads to the value zero, equal to the topological dimension given for a
dot. At the other extreme with the probability of nine ninth all nine cells remain,
resulting in the value two, which equals the topological dimension for a plane.
What is of interest next is how the value (we will call it the index of coherence),
defining the connection between the number of single elements and the reduction
factor, behaves throughout the range of scales. With curdling, this means to compare
the increase in the number of remaining cells with the decrease of the reduction
factor. Using once more the example from above with a grid of three by three cells,
mathematically the number of remaining cells after the first iteration is six for the
probability of two thirds and the reduction factor is one third. Then the second
iteration, using equation N.i / D .N.1/ /i , increases the number of cells to 36 with
the reduction factor decreasing to one ninth. Between two iterations, the differences
of both, number of cells as well as reduction factor, can be examined. Consequently,
the equation for calculating Ds is rearranged, leading to the equation
 DB  DB
1 1
N.i /  N.i 1/ D  :
S.i / S.i 1/

The difference of cells is compared with the difference of the inverse reduction
factors above the value of index, this time called .DB / with regard to analyzing
boxes (we called them cells). This equation is rearranged to

log.N.i / /  log.N.i 1/ /

DB D    :
log S1.i/  log S.i1/
1

Inserting the mathematically calculated values of our example with a probability

of 2/3rd – that is for the number of remaining cells N.1/ D 6 with the reduction
factor s.1/ D 1=3 for the first iteration and N.2/ D 36 with s.2/ D 1=9 for the second
iteration – .DB / again equals 1.631. Analyzing the third and fourth example from
Fig. 9.6, both using a random calculator with the probability of 2/3rd, the results
change. While the reduction factor remains the same, the number of remaining
cells changes. That is N.1/ D 7 for the first example after the first iteration and
N.2/ D 47 after the second iteration. The calculated value DB between first and
second iteration then equals 1.73. DB of the second example equals 1.75 with
N.1/ D 7 and N.2/ D 48.
So both values are slightly higher than the result inserting the mathematically
calculated values. Using the last equation, a set of index values .DB / can be given
for a certain number of iterations. Each iteration is compared with the next one,
196 W.E. Lorenz

which leads to a certain .DB / from the first to the second iteration, to another from
the second to the third and so forth. Because of the consistency between iterations,
self-similar structures will then offer similar values throughout the range of scales.

9.3.3 Perception and Distance

Comparing buildings with each other means that their appearance has to be
standardized first, including influences of color, shadow but also depth of details –
equals the number of iterations or levels of scale. For a first approach, just black
and white elevations will be analyzed, which means removing all influences other
than depth of details. It is the set of index values .DB / for a certain range of scale
that is of interest. Because two-dimensional plans will be analyzed, a first step of
standardization is to define the translation of the original building into the elevation.
There the smallest detail presented in the elevation depends on the distance of the
viewing point from the building in reality and on the human eye. This derives from
the fact that the smallest possible detail depends on the reading field, which is inside
a cone of 0ı 1’ [13]. With the aid of trigonometric function, the relationship between
detail and distance can be given by
 
Minimum Size of Detail > Distance to Detail  tangents 0ı 10 :

This means that for a given distance of 10 m, the size of the smallest perceivable
part is approximately 3 mm. Since the smallest perceivable detail depends on
the distance between the observer and the building, consequently the distance
defines what should be presented in the elevation. From the smallest detail on,
all architectural elements of larger size have to be included and translated to the
elevation as significant edges. Only then coherence between these levels of scale
or sets of different sizes of architectural elements can be analyzed, but also only
then buildings can be compared with each other by their elevations, standardized by
the observer’s distance. The distance to the building, in turn, should be chosen in
the way that its whole extent can be perceived, mainly its vertical extent. Maertens
gives an indication, where a distance equal to the relevant height of a building is
appropriate to view details of the object [13]. Then the uppermost part is within an
angle of 45ı above horizon. A distance of double height equals 27ı and there the
whole building can be viewed for itself. Finally at a distance of the observer of three
times the height, the building will become one with its environment. From these
angles, the distance may then be derived.

9.4 Fractal Dimension and Architecture

Fractal curves are the result of insertion rules after infinite iterations. The Peano
curve, a representative of a classical fractal, is an endless, twisted curve between
two points that does not exceed a certain space. As it can be separated by removing
9 Fractal Geometry of Architecture 197

just one point from the set, it is said to have a topological dimension of one. But
on closer observation, it passes through the two-dimensional plane completely – it
offers area-filling property. The Koch curve again does not fill the two-dimensional
plane but is also endlessly twisted. Since length is infinite, a point on such curves
cannot be defined by only giving the distance. But also a mountain, being rough,
does not fill the three-dimensional space completely. Such structures can then be
characterised in a better way by their Fractal Dimension. The Fractal Dimension
expresses how fast a fractal curve tends to infinity from one iteration to the next or
how completely a fractal appears to fill space.
Facades are rough surfaces that consist of cuts, different architectural elements of
different sizes and details. Hence, they are no flat smooth two-dimensional planes.
Likewise, their expressions on paper, elevations are more than a one-dimensional
line that defines the silhouette, but they also do not fill the plane where they are
situated in completely. They are rough structures and Fractal Dimension is then the
expression of the degree of this roughness, which means how much texture an object
has [5]. With fractal-like architecture, self-similarity cannot simply be identified by
rescaling parts transforming them into the whole again. Then Fractal Dimension is
an adequate possibility to describe such structures, where coherence of roughness
can be analyzed by calculating a set of index values .DB / for a certain range of scale.

9.4.1 Fractal Dimension and Approaching a Building

Expressed in a different way, Fractal Dimension is the degree of mixture of order

and surprise. For objects that are visualized on paper, Fractal Dimension can be
measured by the Box-Counting Method. This measurement method was first applied
to architecture by Carl Bovill [5, 19]. First, a grid is put over the object to be
measured. The grid-size is defined by the number of boxes across the bottom row
of the grid, that is the numbers of boxes in x-direction (1/s). Its inverse value then
defines the scale of the grid .s/. Then those boxes, which cover relevant parts of
the elevation, are counted. Relevant parts are the outline, the roof, windows, doors,
walls, but also certain details. The depth of details of the analyzed elevation depends
on the scale of plan. For the first grid-size .1=s1 /, the number of boxes that contain
relevant parts is defined as .N1 /. For measurement, the grid-size is then reduced to
.1=s2 /, and the number of boxes that contain relevant parts is counted again .N2 /.
Finally, the Box-Counting Dimension between two scales of grid-size is calculated
by the relationship between the difference of the logarithms of the number of boxes
that contain relevant parts and the difference of the logarithms of grid-size as given
in the equation
log.N.S 2//  log.N.S 1/ /
DB.12/ D     :
log S12  log S11
198 W.E. Lorenz

The Box-Counting Dimension DB is another special application of Mandelbrot’s

Fractal Dimension [4, 14]. The Box-Counting Method compares the roughness –
represented by lines – between different grid-sizes and thus allows measuring the
complexity of a structure across certain sizes of details. Equivalence exists between
the scale of the elevation, the scale of the grid and coming closer to a building. In
the first case, a large scale of the elevation only gives an impression of the building.
In this case, larger grid-sizes are used for measurement. Then reducing the scale of
elevation, which means including smaller details, will allow us to identify more and
more details. This asks for smaller grid-sizes. The same is true when approaching
the building in reality. On the level of scale of far distance, smaller details are faded
out, because they cannot be perceived. Consequently, they have to be excluded
from the elevation for measuring the Box-Counting Dimension. Then from a shorter
distance, bigger architectural elements such as windows and doors are perceived that
could not be distinguished before, followed by window-frames and door-handles.
For measuring the Box-Counting Dimension for this distance, the elevation has
to include these components. If the building follows the Fractal concept, the kind
of roughness nevertheless remains the same for all steps. In the logic of Fractal
Geometry, they are linked together by the depth of similar roughness.

9.4.2 Results of Measurement

The behaviour of the relationship between grid-size and number of boxes that
contain relevant parts is analyzed in a double logarithmic graph, where the slope of
the replacing line defines the Box-Counting Dimension for a certain range of grid-
sizes. The result for many different measurements of Robie House by Frank Lloyd
Wright remains between 1.6 and 1.65, taking different elevations with different
detail-richness into account: from an overview to plans of smaller scale, including
details such as stained glass and brick. That means different elevations were used for
measurement to include different distances. For the first distance only main edges
such as outline, windows and doors were included. Then approaching the building,
hence using smaller boxes, sections of the elevation were analysed, including
stained glass and bricks. For all measurements, it is valid that the result depends
on what is included in the elevation and how it is presented.
The double logarithmic graph illustrates that certain measuring points of Robie
House are very close to their replacing lines. From this follows that, although Box-
Counting Dimensions between two single grid-sizes may vary when approaching
the building, the set of measurement points in the double-logarithmic graph is
nevertheless stable over a large range. The slope of the replacing line then gives quite
a significant value .DB / for this range. Comparing these results with the Koch curve,
whose Self-similar Dimension is known, it can be indicated that measurements
for the Koch curve are even more stable. Nevertheless, while comparing different
buildings with each other certain influences have to be dealt with [14, 18]. Some
derive from the measurement method itself – that is dependence on starting position
9 Fractal Geometry of Architecture 199

or overall grid-size – and others from range dependence of architectural elements.

Architectural elements only emerge locally, which means they have a specific range
of distance of the observer in which they are significantly present. In general, to
minimize local influences a whole range of grid-sizes is analyzed rather than only
two single levels. The standard deviation of the graph then gives the degree of
coherence. For Robie House, it turned out that the measurement points in the double
logarithmic graph are quite stable for a broad range hence indicating coherence for
a broad range.

9.5 Conclusions and Outlook

Basically the fractal concept of architecture means that details of different sizes
are kept together by a central rule or idea, respectively – avoiding monotony by
using variation. In architecture, this concept is the reason why Gothic cathedrals
and examples of the so-called organic architecture are so interesting and diversified.
Modern architecture may also offer fractal properties but not for a broad range
of scale. For measuring the presence and coherence of architectural elements
across many levels of scale, the Box-Counting Method turned out to provide a
first verifiable measurement method. The double logarithmic graph – grid-size vs.
number of boxes covered – gives a first indication for similar density across certain
scales. Although the resulting graph does not tell us anything about the quality of a
building or about its form, it provides a first impression of the coherence between
levels of scale as it is true for Robie House by Frank Lloyd Wright.
Future focus lies on further Box-Counting measurements mainly of different
architectural styles for comparison of fractal-like works of architecture with rep-
resentatives of Modern architecture. Those buildings where the single measurement
points in the double logarithmic graph are very close to the replacing line will
be analyzed more closely with regard to a possibly underlying Fractal concept.
An interesting aspect will then, however, be in how far such a concept has some
influence on architectural quality and on the acceptance of the building by observers.

References

1. B.B. Mandelbrot, Die fraktale Geometrie der Natur (Birkhäuser, Basel [u.a.], 1991)
2. Le Corbusier, 1922, Ausblick auf eine Architektur (Vieweg & Sohn, Braunschweig/
Wiesbaden, 1993)
3. N.A. Salingaros, A theory of Architecture (Umbau-Verlag, Solingen, 2006)
4. H.-O. Peitgen, H. Jürgens, D. Saupe, Chaos and Fractals – New Frontiers of Science (Springer,
New York, NY [u.a.], 1992)
5. C. Bovill, Fractal Geometry in architecture and design (Birkhäuser, Boston, Mass. [u.a.], 1996)
6. H. Borcherdt, Architekten, Begegnungen 1956–1986 (Georg Müller Verlag, München, 1988)
7. M. Küper, I. van Zijl, Gerrit Th. Rietveld, The complete works (Centraal Museum,
Utrecht, 1992)
200 W.E. Lorenz

8. C. Jencks, The Architecture of the Jumping Universe (Academy Editions, London, 1996)
9. R. Zerbst, Antoni Gaudı́, 1852–1926 (ein Leben in der Architektur, Taschen, Köln, 1993)
10. M. Burry, Gaudı́ unseen – Die Vollendung der Sagrada Famı́lia (Jovis Verlag, Berlin, 2007)
11. P. Portoghesi, Nature and Architecture (Skira editore, Milan, 2000)
12. G. Lynn, Animate Form (Princeton Architectural Press, New York, NY, 1999)
13. H. Maertens, Der optische Maßstab (Verlag von Ernst Wasmuth, Berlin, 1884)
14. K. Foroutan-pour, P. Dutilleul, D.L. Smith, in Applied Mathematics and Computation,
Advances in the implementation of the box-counting method of Fractal Dimension estimation,
vol. 105, Issue 2–3 (Elsevier Science Inc., New York, NY, 1999), pp. 195–210
15. D. Hoffmann, Frank Lloyd Wright’s Robie House (Dover Publ., New York, NY, 1984)
16. W.E. Lorenz, Master Thesis, Fractals and Fractal Architecture (Vienna University of Techno-
logy, 2003)
17. W.E. Lorenz,in First International Conference on Fractal Foundations for the 21st century
architecture and environmental design, Fractal Geometry as an approach to quality in
architecture (ffractarq, Madrid, 2004), CD-ROM
18. M.J. Ostwald, J. Vaughan, C. Tucker, in Nexus VII: Architecture and Mathematics, Character-
istic Visual Complexity: Fractal Dimensions in the Architecture of Frank Lloyd Wright and Le
Corbusier (Kim Williams Books, Turin, 2008) pp. 217–231
19. M.J. Ostwald, C. Tucker, in Techniques and Technologies Transfer and Transformation: IVth
International Conference of the Association of Architecture Schools of Australasia 2007,
Measuring architecture: Questioning the application of non-linear mathematics in the analysis
of historic buildings (University of Technology and AASA, Sydney, 2007), pp. 183–189
20. H.-O. Peitgen, D. Saupe, The science of Fractal images (Springer, New York, NY, 1988)

Picture credits

All pictures by Wolfgang E. Lorenz.