# A Study of the Roughness of Gothic Rose Windows

- 2.1k Downloads
- 1 Citations

## Abstract

The rose window is one of the most representative elements of Gothic art and architecture. In this work we analyze fifteen rose windows from fifteen Gothic cathedrals using fractal geometry. Specifically, we examine the texture and roughness of these rose windows focusing on three factors, their designs, glass areas and solid areas. In this investigation we generate parameters which provide a measure of roughness of the rose windows in order to find out if they show a general non-random fractal pattern. The paper concludes that statistically, there is a characteristic fractal pattern in the solid and glass areas of the rose windows of the Gothic style, but not necessarily in their overall design.

## Keywords

Rose windows Gothic architecture Fractal geometry Fractal parameter## Historical Background: Light in the Gothic Cathedral

The Gothic style signaled a veritable revolution in architecture as a result of its technical innovations which challenged the conventional concepts of construction at the time and also changed the manner in which large indoor spaces were conceived. Louis VI, King of France, aspired to rule all of the vast territories which had formed the Carolingian Empire three centuries before. To achieve this aspiration he was assisted in this task by Abbot Suger of Saint-Denis, an advisor with both intelligence and diplomatic skills. In gratitude for the services rendered, Louis VI granted the abbey of Saint-Denis great privileges so that they would have a competitive advantage and be able to hold their prosperous annual fair, which gathered both the faithful and merchants providing major benefits for the abbey. As a result of this, Saint-Denis became the richest Benedictine monastery in France after Cluny. Influenced by the Neoplatonic thoughts which spread amongst scholars of the early twelfth century in the Paris region, Suger was fascinated by light as a means to connecting with God. With this concept as an intellectual guide, Suger inferred that the House of God, the Christian church, had to become a temple of light and even more: “a city bathed in the light of God” (Berger 1906; Panofsky 1970). In order to achieve this ideal, it was necessary to modify and improve the construction system of the great Romanesque churches. That is, it was imperative to remove some walls and tear others from top to bottom in order to place large windows in them which would capture the sunlight. The name of the architect who found the solution to Suger’s problem is unknown. Perhaps it was the master builder who directed the construction of the Romanesque structure in the abbey of Saint-Denis. Regardless, shortly before finishing this work, Suger ordered the following inscription to be places, in Latin verses on the main door of the abbey church. “Portarum quisquis attolere quaeris honorem, Aurum nec sumptus, operis mirare laborem. Nobile claret opus, sed opus quod nobile claret, Clarificet mentes, ut eant per lumina vera, Ad verum lumen, ubi Christus jaunua vera. Quale sit intus in his determinat aurea porta”^{1} (Grosse 2004; Panofsky 1970).

## Mathematical Background: Fractal Parameter

As a summary, and by way of conceptual and intuitive explanation in order to clarify the technique which we will introduce next, we argue that the roughness of an object, whether or not it is a fractal object, is geometrically expressed as “its space-filling ability”. This space-filling ability is measured by so-called “fractal parameters”. As explained later in this section, fractal parameters are generated through extrapolated calculations of the theoretical geometric measures of roughness in fractal objects; and these theoretical measures are the different geometric dimensions of fractals. Broadly speaking, roughness is the spatial infiltration behavior of an object across several scales, and a fractal parameter is a value that provides a measure of this infiltration, stating how much that object seems to fill the space as we use finer and finer scales.

This geometric concept is applied in several scientific fields. For example, in medicine, when considering neuronal networks and their pathologies; in electronics, when considering the physical behavior of circuits on smaller manufacturing scales; in chemistry, when obtaining different properties of substances depending on their roughness (Kiselev et al. 2003). This concept is also used in architecture, when seeking evidence of the objective influence between buildings designed by renowned architects, or in order to obtain mathematical evidence with regard to whether or not certain important architects respected the compositional continuity of the artificial or natural environment in which their design was intended to be constructed (Bechhoefer and Bovill 1994; Batty and Longley 1997; Hammer 2006; Sala 2006; Joye 2007; Rian and Park 2007; Bovill 2008; Ostwald 2001; Ostwald et al. 2008; Ostwald and Vaughan 2009; Vaughan and Ostwald 2009, 2011).

Finally, with all the data generated for the drawings (three parameters for each rose window), we make a robust statistical study of the obtained values. Therefore, the geometric results thus achieved are not subjective and they can be expressed in architectural and geometric language, as we will state in the conclusions of this paper.

### Fractal Dimension Theory

This sub-section includes a technical description and references required to fully understand the work which is presented in this paper. In this subsection we summarize the dimensions *T*(*M*), \(D\left( M \right)\), \(\bar{F} \left( M \right)\), *F*(*M*), *H*(*M*), *S*(*M*) of the fractal objects *M*; and some of their properties. This is the theoretical basis from which the ideas for generating the fractal parameters *P* _{ s }(*M*) arise. First, we will establish a few basic notions. More detailed mathematical references and demonstrations of these are available (Falconer 1990, 1997; Edgar 1998).

*M*,

*N*) such that:

*M*is homothetic to

*N*with a homothety

*h*

_{ r }of ratio

*r*) where

*g*

_{ i }(

*N*) is a displacement of

*N*—in (1) the union is disjoint. We say that

*M*has an homothetic structure and its homothetic dimension is:

*M*be a bounded non-empty object of the Euclidean space

*A*

^{ n }, such that \(M = \mathop \coprod \nolimits_{i = 1 }^{i = m} S_{i} \left( M \right)\), where

*S*

_{ i }is a contractive similarity, i.e. \(S_{i} : A^{n} \to A^{n}\) such that ∀(

*x*,

*y*) ∊

*A*

^{ n }×

*A*

^{ n }⇒ d(

*S*

_{ i }(

*x*),

*S*

_{ i }(

*y*)) =

*k*

_{ i }

*d*(

*x*,

*y*) with 0 <

*k*

_{ i }< 1. Then

*M*is called a self-similar object, and its self-similarity dimension is the value

*S*(

*M*) such that \(\sum\nolimits_{i = 1}^{i = m} {k_{i}^{S(M)} = 1}.\) If a self-similar object

*M*is a homothetic object, then

*S*(

*M*) =

*H*(

*M*). Besides, the objects

*M*of space

*A*

^{ n }have their topological dimension

*T*(

*M*), where

*T*(

*M*) = 1, 2 or 3 if

*M*is a line, a surface or a three-dimensional body, respectively.

Even though *M* is self-similar, if *S*(*M*) = *T*(*M*) we say that *M* is a non-fractal self-similar object. But when *M* is self-similar and also *S*(*M*) ≠ *T*(*M*), then we say that the self-similar object *M* is fractal. However, objects *M* in general do not have a homothetic structure nor a self-similar structure; therefore, they do not have a homothetic dimension *H*(*M*) nor a self-similarity dimension *S*(*M*). In spite of that, there is a generalization of the self-similarity dimension *S*(*M*) which is called Hausdorff-Besicovich dimension, noted as *D*(*M*). If *M*, either with or without homothetic or self-similar structure, verifies that *D*(*M*) ≠ *T*(*M*), then *M* is a fractal object.

The definition of *D*(*M*) uses geometric-mathematical concepts which fall outside the purpose of this paper. Intuitively we can say that an object is a fractal when in an infinite number of its points it does not have tangent space, or to put it more colloquially, it has an infinite number of points where it seems to be fractured. In any case, since the calculation of *D*(*M*) falls out from the scope of this paper, we consider another value \(\bar{F}\left( M \right)\) which is an upper bound of \(D\left( M \right)\). This bound, \(\bar{F}\left( M \right) \ge D\left( M \right)\), is called Minkowski-Bouligand dimension of *M*, also called upper fractal dimension of *M*. The object *M* may or may not be a fractal object, but, regardless of this condition, \(\bar{F}\left( M \right)\) is a measure of its irregularity, its space-filling-ability or roughness.

*s*

_{ m }is the number of

*δ*-mesh cubes of

*A*

^{ n }, that intersect

*M*, with \(\delta = {\raise0.7ex\hbox{$1$} \!\mathord{\left/ {\vphantom {1 {2^{m} }}}\right.\kern-0pt} \!\lower0.7ex\hbox{${2^{m} }$}}\).

For this reason, \(\bar{F}\left( M \right)\) is also called the upper box-counting fractal dimension. If \(\bar{F}\left( M \right)\) is equal to \(\mathop {\lim \inf }\nolimits_{m \to \infty } \frac{{ln \left( {s_{m} } \right)}}{{ln \left( {2^{m} } \right)}}\), then \(\mathop {\lim }\nolimits_{m \to \infty } \frac{{\ln \left( {s_{m} } \right)}}{{\ln \left( {2^{m} } \right)}} = F\left( M \right)\) exists, and \(\bar{F}\left( M \right) = F\left( M \right)\). This limit *F*(*M*), if it exists, is called fractal dimension of *M* or box-counting fractal dimension of *M*.

In Falconer (1990) we find that: (1) \(T\left( M \right) \le D\left( M \right) \le \bar{F}\left( M \right) \le n\). (2) If *M* is a self-similar object, then \(D\left( M \right) = F\left( M \right) = S\left( M \right)\).

Limit (3) is a theoretical limit of a geometric object *M*; however, in real cases such as urban plots, the theoretical limit (3) is always substituted by a similar finite calculation. This similar calculation generates a parameter which we will call fractal parameter *P*(*M*), and which also offers a roughness measure of *M*.

### Fractal Parameter and Method

After setting out the \(\bar{F}\left( M \right)\) theory in the former subsection, now we will explain how to generate the fractal parameters *P*(*M*).

Architectural structures *M* are not fractal objects, however we can consider their unevenness, which determines their space-filling ability (that is, their level of roughness), and we can generate a parameter for those non-fractal objects. Since the rose window’s design *M* is a non-fractal real object, the parameter thus generated cannot be the theoretical value *D*(*M*) or \(\bar{F}\left( M \right)\). The architectural composition *M*, despite showing repetitions in some scales, does not really have a homothetic structure nor a self-similar structure. Therefore, we will extrapolate the theoretical calculations of \(\bar{F}\left( M \right)\) and *S*(*M*) in order to generate parameters which provide measures of roughness. We will call these parameters *P* _{ s }(*M*) and *P* _{ r }(*M*).

Parameter *P* _{ s }(*M*) will provide a measure of roughness without taking into account the self-similarity aspect [this measure will come directly from the theoretical process described in (3) for \(\bar{F}\left( M \right)\)]; and parameter *P* _{ r }(*M*) will provide a measure of roughness taking into account the self-similarity aspect. Since these parameters come from theoretical extrapolations which are not applicable to the real structure *M*, they cannot be separated from the generation process. Consequently, any study which is made with such parameters must fulfill two conditions: firstly, the parameter-generation process must be clearly defined, and secondly, the results of such study are not the parameters themselves but the general conclusions drawn from the parameters, regardless of their specific values.

### Summary of the Fractal Parameter’s Generation Process

*P*

_{ s }(

*M*) through calculations with self-created software is summarized as follows:

- 1.
Given the architectural design

*M*, first we generate its design in AutoCad vector format, using black color and line width \(\sim 0.00 \,mm\). From this AutoCad format we obtain the pdf vector format. - 2.
From the pdf vector format we generate a black-and-white digital bitmap file, sized 1024 ×

*v*pixels, showing the architectural design with its size adjusted to full width and height. - 3.
*P*_{ s }(*M*)—Using self-created software, we calculate the fractal parameter*P*_{ s }(*M*) based on the slope on the last point of a continuous graph*ln*–*ln*. In the following sections we explain which continuous graph we are talking about and which calculations are made.

The reason we have created special software is twofold: firstly, we will have total control of the calculations and so we will ensure they are correct. Secondly, commercial software like *Benoit 1.31 de TruSotf Int’I Inc* does not use the slope on the last point of the continuous graph *ln*–*ln*.

*P*

_{ r }(

*M*):

- 3.
*P*_{ r }(*M*)—using our self-created software, we calculate the fractal parameter*P*_{ r }(*M*) based on the slope of the regression line corresponding to the discrete set of points in the graph*ln*–*ln*; and for this calculation we use square meshes, the finest mesh having 4 × 4 pixel squares, and the coarsest mesh having 32 × 32 pixel squares.

We would like to point out that, even though the different calculation methods may be subject to certain variation (Ostwald 2013) and the value of the fractal parameter (of non-fractal objects) depends on the method being used and also on the level of graphic detail of the drawing (Ostwald and Ediz 2015), we have been able to draw conclusions from our method because they are based on the global statistic correlations between all the fractal parameters obtained here.

### Practical Example Applied to Rosette Troyes’ Cathedral

Since *N* is a pixelated digital image file, the calculation process to generate *P*(*M*) will have a finite number of steps. The finest mesh used to generate *P*(*M*) is a 4 × 4 pixel square mesh, because 4 = 2^{ n } gives the finest mesh which is similar to the theoretical meshes for the theoretical calculation of \(\bar{F}\left( M \right)\) (Falconer 1997). This is true because the 4 × 4 pixel squares have inside points and border points. Therefore, in order to calculate *P*(*M*) we will use four meshes the squares of which are 4, 8, 16 and 32 pixels in length, respectively. The reason to use four meshes is that, as we will see later, *P*(*M*) is generated with the slope of a function of a continuous graph *ln*–*ln*. Using the classical interpolation methods, 4 points of a function are enough to find a good approximation of that slope. And we should not use more than 4 interpolation points because of the well-known Runge Phenomenon in numerical calculation. Then, our software generates a square mesh, which we have called *g* _{5}, consisting of 32 × *h* _{5} square boxes with an edge dimension *a* _{5} = 1024 × 2^{−5} = 32 pixels. Then we apply that mesh on the image of *N* and we calculate *ln*(*s* _{5}), where *s* _{5} is the number of boxes of *g* _{5} which have black pixels. Then we repeat the process with the other square meshes *g* _{6}, *g* _{7} and *g* _{8} having \(64 \times h_{6}\), 128 × *h* _{7}, 256 × *h* _{8} square boxes, respectively. The edge dimensions are *a* _{6} = 1024 × 2^{−6} = 16, *a* _{7} = 1024 × 2^{−7} = 8 and *a* _{8} = 1024 × 2^{−8} = 4, respectively. Then we calculate *ln*(*s* _{6}), *ln*(*s* _{7}), and *ln*(*s* _{8}), where *s* _{6}, *s* _{7} and *s* _{8} are the number of boxes with black pixels in each mesh *g* _{6}, *g* _{7} and *g* _{8}, respectively. For example, Fig. 4 shows the data corresponding to: \(\left( {h_{5} , h_{6} , h_{7} , h_{8} } \right) = \left( {32, 64, 128, 256} \right), \left( {s_{5} , s_{6} , s_{7} , s_{8} } \right) = \left( {984, 3182, 9259, 24426} \right)\). As a result of the above mentioned process we obtain the coordinates of four points \(\left( {\ln \left( {2^{5} } \right), \ln \left( {s_{5} } \right)} \right)\), \(\left( {\ln \left( {2^{6} } \right), \ln \left( {s_{6} } \right)} \right)\), \(\left( {\ln \left( {2^{7} } \right), \ln \left( {s_{7} } \right)} \right)\), \(\left( {\ln \left( {2^{8} } \right), \ln \left( {s_{8} } \right)} \right)\), in a graph *ln*–*ln* shown in the center of Fig. 4.

## Parameter *P* _{ s }(*M*)

*ln*–

*ln*on the fourth point \(\left( {\ln \left( {2^{8} } \right), \ln \left( {s_{8} } \right)} \right)\). Such slope is an extrapolation of the process used to calculate the theoretical limit of the upper fractal dimension \(\bar{F}\left( M \right)\). To confirm that the preceding claim is true, you can consider 4 and use l’

*Hopital*’

*s*rule. In order to calculate that slope, the software implements the classical four-point formula 5, where

*h*=

*ln*2 and

*y*

_{ i }= ln(

*s*

_{5+i }). The final result

*y*′

_{3}given by our software is the fractal parameter

*P*

_{ s }(

*M*). In the example, the fractal parameter is \(\frac{1}{6\ln 2}\left( { - 2\ln \left( {984} \right) + 9\ln \left( {3182} \right) - 18\ln \left( {9259} \right) + 11\ln \left( {24426} \right)} \right) \approx 1.33\).

## Parameter *P* _{ r }(*M*)

*P*

_{ s }(

*M*) is generated by means of the slope in the fourth point of the continuous graph

*ln–ln*. However, in the theoretical case of fractal self-similar objects such a graph is a straight line. Therefore, if we generate a fractal parameter under the hypothesis of self-similarity, then we can use the slope of the regression line corresponding to the discrete set of the four points belonging the graph

*ln*–

*ln*. So, the calculation is the quotient of the covariances \(\frac{{\sigma_{xy} }}{{\sigma_{xx} }}\) where:

This fractal parameter will be called *P* _{ r }(*M*). In the case of the rose window displayed in Fig. 4, we have *P* _{ r }(*M*) ≈ 1.55.

## Calculation Results

*P*

_{ s }(

*M*) and for

*P*

_{ r }(

*M*): the 15 designs of the rosettes (Fig. 5); the 15 configurations of solid areas (Fig. 6) and the 15 configurations of the glass areas (Fig. 7).

Calculated results

F.P. design | F.P. solid areas | F.P. glass areas | ||||
---|---|---|---|---|---|---|

| | | | | | |

| 1.53 | 1.72 | 1.79 | 1.74 | 1.80 | 1.70 |

| 1.37 | 1.56 | 1.65 | 1.62 | 1.84 | 1.92 |

| 1.37 | 1.69 | 1.89 | 1.81 | 1.89 | 1.69 |

| 1.67 | 1.82 | 1.89 | 1.84 | 1.84 | 1.70 |

| 1.35 | 1.63 | 1.54 | 1.63 | 1.87 | 1.93 |

| 1.21 | 1.52 | 1.78 | 1.70 | 1.95 | 1.83 |

| 1.25 | 1.56 | 1.90 | 1.80 | 1.88 | 1.73 |

| 1.29 | 1.75 | 1.82 | 1.89 | 1.66 | 1.62 |

| 1.01 | 1.53 | 1.73 | 1.65 | 1.91 | 1.88 |

| 1.50 | 1.68 | 1.69 | 1.72 | 1.85 | 1.91 |

| 1.12 | 1.42 | 1.79 | 1.74 | 1.98 | 1.86 |

| 1.41 | 1.65 | 1.81 | 1.71 | 1.96 | 1.82 |

| 1.59 | 1.67 | 1.89 | 1.73 | 1.92 | 1.82 |

| 1.64 | 1.65 | 1.76 | 1.65 | 1.81 | 1.84 |

| 1.33 | 1.55 | 1.61 | 1.60 | 1.81 | 1.95 |

## Discussion

### The Fractal Pattern of the Rose Windows Design

*P*

_{ s }(

*M*) and the standard deviation \(\sigma_{{D_{s} }} \approx 0.187\). Therefore, the Pearson’s coefficient of variation of

*P*

_{ s }(

*M*) is \(CV_{{D_{s} }} = 14\,\%\). In general, when the Pearson’s coefficient of variation is under 25 % it is considered that there is little scattering around the mean, or that the mean is representative. As a result, the mean is representative. In order to determine the probability of the mean \(m_{{D_{s} }}\) being a non-random result, we have applied Pearson’s Chi squared test with 19 degrees of freedom in the following two-way table, Table 2, where \(I_{1} = \left[ {1.01, 1.05} \right],\, I_{2} = \left[ {1.06, 1.10} \right], \ldots , I_{19} = \left[ {1.91, 1.95} \right], I_{20} = \left[ {1.96, 2} \right]\); and the result is

*χ*

^{2}≈ 13.684.

Two-way table of s parameters for the architectural designs

| | | | | | | | | | | | | | | | | | | | Total | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

| 1 | 0 | 1 | 0 | 2 | 1 | 2 | 2 | 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 15 |

| 14 | 15 | 14 | 15 | 13 | 14 | 13 | 13 | 14 | 14 | 14 | 14 | 14 | 14 | 15 | 15 | 15 | 15 | 15 | 15 | 285 |

Total | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 300 |

In conclusion this table is a non-random table with probability \(P_{{D_{s} }} = 0.198\, \ll 0.998\). This probability is very low, consequently the mean \(m_{{D_{s} }}\) is a random result. But, despite the size of Table 2, which has 19 degrees of freedom, some of the 40 expected frequencies are less than 5. Therefore, one might think that we have taken the wrong decision using Pearson’s Chi squared test. In order to dispel any doubts, we apply Fisher-Irwin’s exact test to the 1855967520 possible matrices. By calculation, we find that the value *P* _{ cutoff } of conditional probability for this table’s matrix is \(P_{cutoff} = \frac{{15!285!\left( {15!} \right)^{20} }}{{300!\left( {0!} \right)^{8} \left( {1!} \right)^{9} \left( {2!} \right)^{3} \left( {15!} \right)^{8} \left( {14!} \right)^{9} \left( {13!} \right)^{3} }} \approx 5.788725 \times 10^{ - 9}\). The number *n* _{ cutoff } of matrices having conditional probability *P* _{ value,i } ≤ *P* _{ cutoff } is *n* _{ cutoff } = 1820641656. Finally, the two-side *P* _{ value } of the test is \(P_{value} = \mathop \sum \nolimits_{i = 1}^{{n_{cutoff} }} P_{value,i} \approx 0.741525 = P_{{D_{s} }}\). In conclusion, using the exact test, this table is a non-random table with probability \(P_{{D_{s} }} = 0.741525 \gg 0.05\). This probability is very high, consequently the mean \(m_{{D_{s} }}\) is a random result. From this we cannot conclude that the rose windows show a fractal pattern. If the mean for *s* or *r* parameters is not representative or is a random result, the geometry of the configuration does not have a geometric pattern; the existence of the pattern must be independent of the type of parameter used, giving a numerical value for the roughness of the geometric configuration.

### The Fractal Pattern of the Solid Areas in the Rose Windows

_{S}(M). Therefore, the Pearson’s coefficient of variation is \(CV_{{S_{s} }} = 6\;\%\). As a result, the mean is very representative. If then we apply Pearson’s Chi square test with 19 degrees of freedom to the data in Table 3 (the two-way table of

*s*parameters for the solid areas)—as we did with Table 2—, the result is

*χ*

^{2}≈ 44.561. The table is a non-random table with probability \(P_{{S_{s} }} \ge 0.998\), so the mean \(m_{{S_{s} }}\) is a non-random result with a very high probability. We can also use Fisher-Irwin’s exact test, the result of which is \(P_{cutoff} = \frac{{15!285!\left( {15!} \right)^{20} }}{{300!\left( {0!} \right)^{13} \left( {1!} \right)^{3} \left( {2!} \right)^{2} \left( {4!} \right)^{2} \left( {15!} \right)^{13} \left( {14!} \right)^{3} \left( {13!} \right)^{2} \left( {11!} \right)^{2} }} \approx 9.018023 \times 10^{ - 12}\),\(n_{cutoff} = 250220824\) and \(P_{value} = \mathop \sum \nolimits_{i = 1}^{{n_{cutoff} }} P_{value,i} \approx 0.000742 = P_{{S_{s} }} \ll 0.01\), so the mean \(m_{{S_{s} }}\) is a non-random result with a very high probability.

Two-way table of s parameters for the solid areas

| | | | | | | | | | | | | | | | | | | | Total | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

| 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 2 | 1 | 1 | 4 | 2 | 4 | 0 | 0 | 15 |

| 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 14 | 15 | 13 | 14 | 14 | 11 | 13 | 11 | 15 | 15 | 285 |

Total | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 300 |

*P*

_{ r }(

*M*) for the solid areas are, \(m_{{S_{r} }} \approx 1.722\) (Mean), \(\sigma_{{S_{r} }} \approx 0.082\) (Standard Deviation) and \(CV_{{S_{r} }} = 5\;\%\) (Pearson’s Coefficient of Variation). The mean is very representative and if we apply the Chi square test to the data in Table 4 (the two-way table of

*r*parameters for the solid areas), the result is

*χ*

^{2}≈ 52.5982 and \(P_{{S_{r} }} \ge 0.998\), so the mean \(m_{{S_{r} }}\) is a non-random result. The Fisher-Irwin’s exact test for Table 4 generates the result of

*P*

_{ cutoff }≈ 2.834236 × 10

^{−12},

*n*

_{ cutoff }= 132157864 and \(P_{value} \approx 0.000133 = P_{{S_{r} }} \ll 0.01\), so the mean \(m_{{S_{r} }}\) is a non-random result. Thus, the means of the parameters

*s*and

*r*are a non-random results and they are very representative. Therefore, the solid areas of the rose windows follow a fractal pattern.

Two-way table of r parameters for the solid areas

| | | | | | | | | | | | | | | | | | | | Total | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

| 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 4 | 1 | 5 | 1 | 2 | 1 | 0 | 0 | 15 |

| 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 14 | 11 | 14 | 10 | 14 | 13 | 14 | 15 | 15 | 285 |

Total | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 300 |

### The Fractal Pattern of the Glass Areas in the Rose Windows

_{S}(M). Therefore, the Pearson’s coefficient of variation is \(CV_{{L_{s} }} = 4\;\%\). As a result, the mean is very representative. If we apply Pearson’s Chi square test with 19 degrees of freedom to the data in Table 5 (the two-way table of

*s*parameters for the glass areas) the result is \(\chi^{2} \approx 52.982\). The table is a non-random table with probability \(P_{{L_{s} }} \ge 0.998\), so the mean \(m_{{L_{s} }}\) is a non-random result with a very high probability. Using Fisher-Irwin’s exact test, the result of which is

*P*

_{ cutoff }≈ 1.910485 × 10

^{−12},

*n*

_{ cutoff }= 112801120 and \(P_{value} \approx 0.000048 = P_{{L_{s} }} \ll 0.01\), the mean \(m_{{L_{s} }}\) is a non-random result with a very high probability.

Two-way table of s parameters for the glass areas

| | | | | | | | | | | | | | | | | | | | Total | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

| 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 1 | 5 | 3 | 3 | 2 | 15 |

| 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 14 | 15 | 15 | 14 | 15 | 14 | 10 | 12 | 12 | 13 | 285 |

Total | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 300 |

*P*

_{ r }(

*M*) for the glass areas are \(m_{{L_{r} }} \approx 1.820\) (mean), \(\sigma_{{L_{r} }} \approx 0.094\) (standard deviation) and \(CV_{{L_{r} }} = 5\%\) (Pearson’s coefficient of variation). The mean is very representative and if we apply the Chi square test to the data in Table 6 (the two-way table of

*r*parameters for the glass areas), the result is

*χ*

^{2}≈ 45.614 and \(P_{{L_{r} }} \ge 0.998\), so the mean \(m_{{L_{r} }}\) is a non-random result. Again we apply Fisher-Irwin’s exact test, the result of which is

*P*

_{ cutoff }≈ 2.605207 × 10

^{−12}, \(n_{cutoff} = 125181064\) and \(P_{value} \approx 0.000115 = P_{{L_{r} }} \ll 0.01\), so the mean \(m_{{L_{r} }}\) a non-random result. The means of the parameters

*s*and

*r*are a non-random results and they are very representative. Therefore, the glass areas of the rose windows follow a fractal pattern.

Two-way table of r parameters for the glass areas

| | | | | | | | | | | | | | | | | | | | Total | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

| 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 3 | 1 | 0 | 4 | 2 | 4 | 0 | 15 |

| 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 14 | 15 | 14 | 12 | 14 | 15 | 11 | 13 | 11 | 15 | 285 |

Total | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 15 | 300 |

## Conclusion

*P*

_{ r }and

*P*

_{ s }simultaneously. Therefore, using the fractal geometric parameterization, it has been proven that the rose windows designs do not follow any characteristic roughness pattern (Table 7). With all these results, we can conclude that each of these rose windows was designed according to the particular stylistic approach of the corresponding architect, the construction budget and, finally, the architectural composition of the main elevation. However, if we analyze the techniques which glassmakers and stonemasons applied to the geometry of the solid and glass surfaces, we find more interesting results. After analyzing the solid and glass areas of the rose windows we do find a characteristic roughness pattern because there is a fractal pattern in both types of parameters (Table 7). This means that the rose windows were designed with the same roughness model for solid areas and glass areas. This results allows us to conclude that there is a characteristic fractal pattern not only in the Gothic structures (floor plan, elevation and cross-section) (Samper and Herrera 2014), but also in the rose windows (solid areas and glass areas), which are one of the most representative elements of the Gothic style.

The summary table of results

| | | \(P \le 0.01 \,\left( {Fisher} \right)\) | \(P \ge 0.998 \,\left( {\chi^{2} } \right)\) | Fractal pattern | ||||||
---|---|---|---|---|---|---|---|---|---|---|---|

| | | | | | | | | | ||

Design | – | 1.369 | – | 0.187 | – | 14 | – | 0.741525 | – | 0.072 | No |

Solid | 1.722 | 1.768 | 0.082 | 0.106 | 5 % | 6 | 0.000133 | 0.000742 | 0.998 | 0.998 | Yes |

Glass | 1.820 | 1.865 | 0.094 | 0.077 | 5 % | 4 | 0.000115 | 0.000048 | 0.998 | 0.998 | Yes |

## Footnotes

- 1.
“All you who seek to honor these doors, marvel not at the gold and expense but at the craftsmanship of the work. The noble work is bright, but, being nobly bright, the work should brighten the minds, allowing them to travel through the lights to the true light, where Christ is the true door.”

## References

- Bechhoefer, W. and C. Bovill, 1994. Fractal Analysis of Traditional Housing in Amasya, Turkey. In:
*Proceedings of the fourth Conference of the International Association for the Study of Traditional Environments.*Berkeley: International Association for the Study of the Traditional Environment.Google Scholar - Berger, É. 1906.
*Les Aventures de la reine Aliénor. Historie et Légende. Académie des Inscriptions et Belles*-*lettres*. Paris: Alphonse Picard et fils, Éditeurs.Google Scholar - Bovill, C. 2008. The Doric Order as a Fractal.
*Nexus Network Journal***10**(2): 283–290.CrossRefGoogle Scholar - Edgar, G. 1998.
*Integral, Probability, and Fractal Measures*. New York: Springer.CrossRefzbMATHGoogle Scholar - Falconer, K. 1990.
*Fractal Geometry, Mathematical Foundations and Applications*. Chichester: John Willey.zbMATHGoogle Scholar - Falconer, K. 1997.
*Techniques in Fractal Geometry*. Chichester: John Willey.zbMATHGoogle Scholar - Grosse, R. 2004.
*Suger, Personnage Complexe par Jean Dufour dans Suger en Question: regards Croisés sur Saint*-*Denis*. Oldenbourg: Oldenbourg Editor.Google Scholar - Hammer, J. 2006. From Fractal Geometry to Fractured Architecture: The Federation Square of Melbourne.
*Mathematical Intelligencer***28**(4): 44–48.MathSciNetCrossRefGoogle Scholar - Joye, Y. 2007. Fractal Architecture Could be Good for you.
*Nexus Network Journal***9**(2): 311–320.MathSciNetCrossRefzbMATHGoogle Scholar - King, T. 1858.
*An Original Architectural Plate from: The study*-*book of Mediaeval Architecture and Art. Working Drawings of the Principal Monuments of the Middles ages*. London: Bell and Daldy.Google Scholar - Kiselev, V., K. Hahn, and D. Auer, 2003. Is the Brain Cortex a fractal?
*Neuroimage***20**(3): 1765–1774.CrossRefGoogle Scholar - Ostwald, M., J. Vaughan, and S. Chalup, 2008. A Computational Analysis of Fractal Dimensions in the Architecture of Eileen Gray. In:
*Biological Processes and Computation,*ACADIA 08: 256–263.Google Scholar - Ostwald, M. J. 2001. Fractal Architecture: Late 20th Century Connections between Architecture and Fractal Geometry.
*Nexus Network Journal***3**(1): 73–84.MathSciNetCrossRefzbMATHGoogle Scholar - Ostwald, M. J. and J. Vaughan, 2009. Calculating Visual Complexity in Peter Eisenman’s Architecture: a Computational Fractal Analysis of Five Houses (1968-1976). In:
*Proceedings of the Fourteenth Conference on Computer Aided Architectural Design Research in Asia*. CAADRIA 2009: 75–84Google Scholar - Ostwald, M. J. and J. Vaughan, 2012. Significant Lines: Measuring and Representing Architecture for Computational Analysis. In:
*46th Annual Conference of the Architectural Science Association*, ANZAScA 2012: 50–60.Google Scholar - Ostwald, M. J. and J. Vaughan, 2013. Representing Architecture for Fractal Analysis: a Framework for Identifying Significant Lines.
*Architectural Science Review***56**(3): 242–251.CrossRefGoogle Scholar - Ostwald, M. J. 2013. The Fractal Analysis of Architecture: Calibrating the Box-Counting Method Using Scaling Coefficient and Grid Disposition Variables.
*Environment and Planning B: Planning and Design***40**(4): 644–663.CrossRefGoogle Scholar - Ostwald, M. J. and Ö. Ediz, 2015. Measuring Form, Ornament and Materiality in Sinan’s Kiliç Ali Pasa Mosque: an Analysis Using Fractal Dimensions.
*Nexus Network Journal***17**(1): 1:5–22.Google Scholar - Panofsky, E. 1970.
*Architecture Gothique et Pensée Scolastique*. Paris: Les éditions de Minuit.Google Scholar - Rian, I. and J. Park, 2007. Fractal Geometry as the Synthesis of Hindu Cosmology in Kandariya Mahadev Temple, Khajuraho.
*Building and Environment***42**(12): 4093–4107.CrossRefGoogle Scholar - Sala, N. 2006. Fractal Geometry and Architecture: Some Interesting Connections.
*WIT Transactions on The Built Environment***86:**163–173.Google Scholar - Samper, A. and B. Herrera, 2014. The Fractal Pattern of the French Gothic Cathedrals.
*Nexus Network Journal***16**(2): 251–271.CrossRefzbMATHGoogle Scholar - Vaughan, J. and M. J. Ostwald, 2009. A Quantitative Comparison Between the Formal Complexity of Le Corbusier’s Pre-Modern (1905-1912) and Early Modern (1922-1928) Architecture.
*Design Principles and Practices: An international journal***3**(4): 359–372.Google Scholar - Vaughan, J. and M. J. Ostwald 2011. The Relationship Between the Fractal Dimension of Plans and Elevations in the Architecture of Frank Lloyd Wright: Comparing the Prairie Style, Textile Block and Usonian Periods.
*Architecture Science Research: ArS***4**: 21–44.Google Scholar - Vaughan, J. and M. J. Ostwald, 2014. Measuring the Significance of Façade Transparency in Australian Regionalist Architecture: A computational Analysis of 10 Designs by Glenn Murcutt.
*Architectural Science Review***57**(4): 249–259.CrossRefGoogle Scholar