Architectural Morphology - J.P Steadman PDF
Architectural Morphology - J.P Steadman PDF
Architectural
Morphology
J PSteadman
A rchit ectura l M orpho log y
J P Steadman
Contents
Introduction
2 T he 'dimensionless' representation
of rectangular plans
3 The symmetries of rectangular plans
4 Generating and counting rectangular
ar rangements: dissection and additive
methods
5 Generating and counting rectangular
arrangements : tilings and colourings
on grids
6 Graphs of plans and their arrangement
7 Embedded, colou red, and weighted graphs
of plans
8 Properties of rectangular arrangements,
and their classification
9 Floor plan morphology in design
10 Plan morphology and bui lding science
11 Plan morphology and architectura l history
12 Afterwords: prospects for an architectural
morphology
Appendix :
Diagrams of rect angular dissect ions up ton =7
with tabu lations of their properties
References
I ndex
p Pion Limited, 207 Brondesbury Park, London NW2 SJN "If something like a theory of architecture will ever be develo ped, then one of its
first chapters will deal with the theory of cell configurations ..."
H Rittel (I 970)
© 1983 Pion Limited
This book provides an introduction to an area of architectural research
All rights' reserved. No part of this book may be reproduced in any form which has been emerging over the last ten years, and which has gone under
by photostat microfilm or any other means without written permission the name, variously, of 'architectural morphology' or 'configurational
from the publishers. studies' in architecture. It is concerned centrally with the limits which
ISBN O 85086 086 5 geometry places on the possible forms and shapes which buildings and
their plans may take. The use of the term 'morphology' alludes then to
Goethe's original notion, of a general science of possible forms, covering
not just forms in nature, but forms in art, and especially the forms of
architecture.
This research work has been published up to now in scattered papers,
many of them so technical as to be inaccessible to the general reader.
The material is here brought together for the first time, put into order,
and, together with the necessary mathematical foundations, set out in a
self-contained treatment for the nonspecialist student of architecture. The
last three chapters of the book explore in a more speculative way the
broader implications of the work for design, for building science, and for
architectural history.
The hope is that the book may be of interest to practising designers and
architectural students. But it is directed most specifically to theoreticians:
to building scientists, understood in the most general sense of that
term ; to historians of architecture who have interests in the description
and explanation of the basic underlying forms of buildings, and their
methods of composition; and to those working on computer aids for the
representation and manipulation of building form in design.
Much of the work described here is that of my own immediate colleagues
and students in three university departments: the Martin Centre at the
Cambridge School of Architecture, the School of Architecture and Urban
Planning at the University of California, Los Angeles, and the Centre for
Configurational Studies at the Open University. Of these colleagues I should
like to name specifically, at Cambridge, Leonardo Combes and Cecil Bloch;
at UCLA, Bill Mitchell, Robin Liggett, and George Stiny ; and at the Open
University, Lionel March, Chris Earl, Ray Matela, and Ramesh Krishnamurti.
I owe a great deal to the many discussions which I have had with these
individuals and others in these three places; and I hope I have dbne justice
to their work where I have presented it, in necessarily highly summarised
form, here.
A glance at the references will show how a very large proportion of the
entries refer to a single source- the journal Environment and Planning B.
Quite apart from his own protean contributions, Lionel March in his
editorial capacity has promoted this field of research to an international
Printed in Great Britain b y Page Bros (No rwich) Limited
audience with such enthusiasm and vigour that everyone working on these
Preface and acknowledgements
subjects must be greatly in his debt. If this book can introduce the field For Leonardo:
to readers unfamiliar with Environment and Planning B, and lead them on another exhibit for Dubuffet's museum?
into the fuller and more technical discussions of the journal itself, then it
will have served much of its purpose.
Outside these immediate circles, I must mention the intellectual stimulus
and challenge continuously provided by the morphological work of Bill
Hillier and of his colleagues and students at University College London.
Some of the arguments recounted here have been raised and debated in an
interuniversity seminar series held jointly between the Open University,
Cambridge, and University College groups.
I would like to thank Chuck Eastman, as editor of this series, and also
Chris Earl, for reading the manuscript, for making many helpful suggestions,
and for saving me from at least some of my errors. I am especially
grateful to Cecil Bloch for his kindness in allowing me to reproduce part
of his catalogue of rectangular dissections as an appendix. The manuscript
was typed by Sue Ayers and the diagrams drawn by Sarah Couch.
Thanks are due to the following publishers, for their kind permission to
reproduce copyright material:
Cambridge University Press, Cambridge, for figures 2.3 and 2.4;
Petrocelli/Charter, New York, for figure 2.8 ;
Artemis Verlag, Zurich, for figure 8.12(a);
Macmillan, New York, for figures 8.12(c) and (d);
MIT Press, copyright Massachusetts Institute of Technology, Cambridge, MA,
for figure 9. I ;
Oxford University Press, London, for figures 9 .2 and I I .9(a) ;
Granada, St Albans, for figure 10.7 ;
Faber and Faber, London, for figure 11.1 ;
Manchester University Press, Manchester, for figures 11.S(a) and 11. 7;
Architectural Press, London, for figure 11 .11 (a).
Contents
Introduction
References 269
Index 277
1
Introduction
"It seems very unaccountable that the generality of our late architects dwell so
much upon [the) ornamental, and so slightly pass over the geometrical, which is the
most essential part of architecture."
Sir Christopher Wren (1750)
(I) Two of the best and most succinct discussions of these distinct meanings of
' functionalism' are to be found in Summerson (1949, page 149), and in Goodman and
Goodman (1947, pages 8-9). As the Goodmans say, 'Form follows function'" ... in
the original statement of Louis Sullivan ... meant that the form is not given by the
function but is appropriate to the function; in his words 'a store must look like a
store, a bank must look like a bank'. This is an aesthetic principle; for it includes
certain ideas, the genres of buildings, whose unity is formal over and above the utility;
they are given by the sensibility of the culture ... But in the more radical interpretation
of the Bauhaus the formula means that the form is given by the function: there is no
addition to the arrangement of the utility, but it is presented just as it works. As such,
this is not an aesthetic principle at all ...". (See also Steadman, 1979, chapter 13.)
2 Chapter I Introduction 3
involved an aesthetic belief that buildings should 'explain themselves', None of these limitations force the designer's hand. Rather they serve to
should present in their design a rational argument about their function and determine the extent of the field of possibilities within which his choice
means of construction. By contrast the functionalism of, say, Hannes must be exercised. As he restricts himself to a geometrical discipline, to a
Meyer, or more recently of some ad herents of the 'design methods' set of formal elements, to some dimensional constraints, so his choice is
movement, tended towards a much more radical doctrine: that functional further narrowed.
considerations could, if subjected to sufficiently precise analysis, be made Other authors have adopted the terminology of ,linguistics in this
to define the form of a building in a necessary and automatic way- what contex t, and have referred to the syntax (Frank Lloyd Wright spoke of
has been called 'functional determinism'. 'grammars') of architectural form. We might pursue this analogy with
It is popularly assumed that any architectural research of a mathematical language a little farther. The academic study of architecture can be seen
nature must have functionalist aims in this second sense: that it seeks to as being divisible into three areas or disciplines. There is the professional
devise ways in which the design of a building can be formulated as a training of designers, in architectural schools. The analogy here is with
mathematical 'problem', and mathematically 'solved' . learning to speak a language. There is the study of architectural history,
This book aims to provide an introduction to an area of research which , from a critical and aesthetic point of view. This must correspond to the
though mathematical in nature, and although computers are certainly used, study of literary history and literary criticism. And , third, there is what I
has quite different aims and is based on a very different co nception of the would suggest is the architectural counterpart to the discipline of linguistics:
nature of architectural design. It takes the view that design is, always has that is, an architectural science, devoted to a general investigation of the
been, and always will be concerned at its central core with the manipulation cultural and technological systems within which all architects work, and all
of f orm, with composition , understood as the putting together of two- buildings are produced.
dimensional and three-dimensional components, either spaces or material This 'architectural linguistics' is itself divisible into two parts. There is
elements, in arrangements or configurations. that part which deals with the syntax of possible architectural forms and
The architect's choice of forms is made according to his artistic purpose, arrangements. And there is that part which deals with the semantics, the
and is directed towards the satisfaction o f (though not by any mea ns systems of meaning, which the syntactic forms and stru ctures come to
uniquely determined by) his client's tastes, desires, and utilitarian support. (For example, see Norberg-Schulz, 1974; Bonta, 1979; Broadbent
requirements, as well as being limited by techni cal and structural possibility. et al, I 980.) The present book is confined entirely to the subject of
But whatever is expressed or signified by an architectural work, whatever architectural syn tax.
practical functions it might serve, and however it is constructed , this It is in many ways a sequel to The Geometry of Environment which
choi ce of form in design is constrained above all by limits on what is Lionel March and I published just twelve years ago (March and Steadman,
geometrically and topologically possible. It is the purpose of this whole 197 1). (In fact the material presented in the two books overlaps to an
book to demonstrate the detailed nature of some of these limits. As an extent, especially that on symmetry, and graph theory.) One of our
example, and speaking generally, if it is required that a number of rooms slightly hidden motives in writing The Geometry of Environment was a
be laid out on a single floor level such that specified pairs of rooms are or belief that learning to understand geometrical limitations and geometrical
are not adjacent (without consideration of their shape or size) then there possibility formed a valuable, and at that time neglected, part of the
exists only a finite number of possible such arrange ments (perhaps no ne at general education of the designer<2>.
all). Such limitations are of a topological nature. Not that this knowledge would be applied necessarily in specific
Again there exist choices for the overall geometrical discipline according mathematical techniques or methods, but that it could constitute part of
to which plans may be laid out, for instance with their walls aligned o n a the broad intellectual makeup of the designer, part of the mental apparatus
rectangular or some other form of grid; for the geometrical elements from which he brings to bear in design. There was admittedly a short presentation
which the plan is to be made up - for example, rectangles or o ther shapes in The Geometry of Environment of some computerised 'design methods'
corresponding perhaps to rooms; and for conditions on the assemb ly o f
those elements- for example, that they should pack closely without gaps. <2> It is perhaps worth recalling that we felt it necessary, in the introduction to The
that they should not overlap, and so o n. Limits on the variety of possible Geometry of Enviro11me11t , to apologise for our "total disregard in this book for any
arrangements of such elements under such conditions can then be expressed but the most simple functional requirements". Thus we could have been more
justifiably accused on that count of forma list rather than functionalist tendencies. For
in terms of well-defined rules for their composition. The introduction of coverage of functional aspects we referred readers, among other works, to Alexander's
dimensional constraints will reduce this variety of arrangement still further. ( 1964) Notes on the Synthesis of Form, which as George Stiny has said , should more
properly have been called Notes on the Analysis of Function.
4 Chapter 1
I. Introduction 5
intended for generating supposedly 'optimal' architectural layouts on the ' infinitude of possibilities' which paralyses the imagination. Thus Stravinsky
basis of a circulation-minimising criterion. Such methods had been (I 970, page 85) says:
formulated very much in a 'functional determinist' spirit, and were "What delivers me from the anguish into which an unrestricted freedom
criticised in the book on those grounds. plunges me is the fact that I am always able to turn immediately to
Some similar methods are described briefly here at the beginning of the concrete things that are here in question. [He is referring to such
chapter 9; but the purpose is most emphatically not to endorse the view acoustic and physiological facts as the nature of the octave, its chromatic
of architectural design which they embody. They are introduced because intervals, and their perception, etc.) Let me have something finite,
they make use of some of the mathematical ideas and techniques which definite- matter that can lend itself to my operation only insofar as it is
are described here; and more importantly because they led the way commensurate with my possibilities. And such matter presents itself to
towards the development of methods for enumerating all permutations of me together with its limitations. I must in turn impose mine on it. So
plan arrangement within given constraints, and hence of exploring the here we are, whether we like it or not, in the realm of necessity. And yet
'outer limits' of geometrical possibility in design, as already described. which of us has ever heard talk of art as other than a realm of freedom?
Some designers are strangely unwilling to acknowledge the fact of the This sort of heresy is uniformly widespread because it is imagined that art
existence of such geometrical limitations at all- witness the reaction of is outside the bounds of ordinary activity."
Mitchell's acquaintance- and adopt evasive tactics in discussion. When one In architecture too there are individuals whose creativity might be
makes the argument that, say the number of rectangular plans with so credited at least in part to their knowledge of geometry as a subject;
many rooms is (in a certain sense, defined in chapter 2 in terms of either explicit, as in the case of Wren, or of a more intuitive and informal
topological equivalence) strictly finite, then the response tends to be kind, such as that of Wright, who is well known to have made much of his
"Ah well, but you haven't included triangular plans ... " or "What about education from an early age in the constructive appreciation of abstract
circular plans? ... or 'free-form' plans? ... ". By moving the ground of the pattern and form.
discussion, by trying to step outside the system, the architect attempts
A final word should be said about the mathematical treatment. For
vainly to escape this geometrical prison, as he sees it, in an effort to
many people, I know, the mere sight of a mathematical symbol is sufficient
reassert his creative freedom.
to make the eyes glaze over and the brain go numb. This, however, is an
But this supposed freedom is illusory ; and anyone who denies the
existence of geometrical limitations in design is only doomed to stumble irrational phobia, and like all phobias it is treated by urging the sufferer to
face and handle the feared object, to discover for himself that it is by no
against them blindly. The fact is that for many good geometrical and
means as frightening as he had always believed.
practical reasons, as will be clear from later chapters, architects do choose
If you are a mathematophobe, you should be reassured that the prior
to confine themselves to a rectangular discipline in design. And where
they do so, then certain limitations on geometrical possibility must
knowledge of mathematics required is no more than basic ari thmetic and a
few simple algebraic concepts. The mathematics is very informally
necessarily apply. They have perfect freedom , certainly, to reject the
presented, most equations are spelled out in words, and frequent examples
discipline of the right-angle if they so wish, and adopt say a triangular or a
are given , illustrated with explanatory diagrams. Do not be put off by
curvilinear geometry. But of course these disciplines in their turn impose
algebraic formulations (of which there are few). You should expect to
their own limits on possibility- in some respects more severe, as we shall
read mathematics slowly in any case: and more is learned by doing rather
see, than those of rectangular geometry.
than just reading- so exercises are included at the end of several of the
It might be objected that architects have been managing to compose
chapters. All the treatment here is self-contained ; but in some of the
forms in design very satisfactorily for a couple of thousand years or so,
topics covered, background textbooks have been cited for those who want
without any conscious knowledge of such limitations. In terms of our
to prepare more widely around the subject.
earlier analogy, a person can certainly achieve fluency in a language-
For the mathematically equipped reader, on the other hand, I should
become a poet indeed- without any formal knowledge of the science of
apologise for the absence of formal definitions and proofs, and for any
linguistics. Perhaps there is no call for the practical training of designers
possible tedium which the repetition of ex planations or illustrations might
to include any formal geometrical education in these subjects.
cause. References are given to the specialised literature in which formal
On the other hand it is possible to point to some undeniably creative
treatment of the mathematical issues may be found.
practitioners of other arts, who have argued that a mastery of the formal
'rules of composition' serves precisely to liberate, rather than to enslave,
the artist; that it is exactly the dizzying (and deceptive) notion of an
2 The 'dimensionless' representation of rectangular plans 7
The 'dimensionless' representation of rectangular plans All the faces are made from essentially the same drawing, from the same
set of curved lines joined in an unvarying arrangement; but squeezed or
stretched, so to speak, as the grid dimensions are altered. It is as though
"The splitting into something discrete and something continuous seems to me a the drawing were made on a sheet of rubber.
basic issue in all morphology, and the morphology of ornaments and crystals In other illustrations, Diirer encloses the head in an oblique grid of
establishes a paragon by the clearcut way in which this distinction is carried out." equal parallelograms (figure 2.2). The grid is again transformed , this time
Hermann Wey! (1952) by altering the angle between the coordinates- so subjecting the drawing
to a shear transformation- to generate a different kind of series of related
There are some famous drawings by Albrecht Diirer ( 1528), in which he profiles.
uses a particular geometrical method to describe the proportions of the The same idea is developed by D'Arcy Wentworth Thompson (1961) in
human head and face. The method consists in drawing a square or one of the best-known chapters "On the theory of transformations, or the
rectangle to enclose the head, and dividing this rectangle up into a grid of comparison of related forms" , in his book on biological morphology On
lines which mark the positions of the various features- brow, eyes, nose, Growth and Form. Thompson's purpose here is to compare the shapes of
chin, and so on. Diirer shows how a series of different faces may be animals which belong to the same zoological class. To do this he makes
produced by altering the relative spacing of the lines of the grid (figure 2.1 ).
m
m m
n n n
Geryon Para lorn is
I, l\s
~ttw
I
,I
I I,
m
m
m I ,
r- I
n n Lupa Corystes
"
Figure 2.1. Different faces produced by assigning varying dimensions to the intervals
of the superimposed grid (from Dilrer, 1528).
Scyramathia Chorinus
Figure 2.2. Different faces produced by changing the angle between the coordinates of Figure 2.3. Carapaces of various crabs, related together in shape by D' Arey Thompson's
the superimposed grid, in a shear transformation (from Dnrer, 1528). 'method of coordinates' (from Thompson , 1961 , figure 142).
8 Chapter 2 The 'dimensionless' representation of rectangular plans 9
use of several types of 'deformation• of systems of rectangular coordinates: curvature, every area its magnitude; but on the other hand every point
by simple enlargement or red uction, by stretching along one or other of and every line continues to exist, and keeps its relative order and position
the axes, by shearing, by various logarithmic transformations of the throughout all distortions and transformations". Speaking rather loosely,
dimensions of the grid, and through different forms of curvature of the we might say that D'Arcy Thompson's, and Di.irer's, method separates out
grid about one or more cent res (figure 2.3). the essential configuration or Gestalt of the shapes of faces, bodies, or
By these means, Thompson illustrates how species which are closely other figures, from the particular relative sizes which the separate parts of
related in evolutionary terms can often be shown to possess forms which the figure may assume.
can be produced one from another by simple transformations (figure 2.4). Let us see how a similar approach might be taken to describing the
By use of similar techniq ues, that is, through transformations which can be shapes of architectural plans (compare Eastman. 1970: March, 1972). We
referred to mechanical or geometrical principles of growth, he demonstrates will start with plans organised according to a rectangular geometry, for the
alternatively how the same animal or plant changes in shape as it develops. sake of simplicity- although the coordinate method is by no means
Thompson reproduces in the chapter some of the drawings by Di.irer confined to the description of rectangular forms, as we have seen from
already mentioned. The same method of coordinates can be extended to these biological applications- and we can extend the argument to take in
three dimensions, and Thompson discusses how the solid forms of the bodies other geometries in due course.
of fishes can be related together by means of geometric transformations in The following figure illustrates in very diagrammatic fo rm a small plan
which say a 'round' fish is 'rolled out' into a flat fish , 'as a baker rolls a of four rooms:
piece of dough'.
In every case the related shapes can be said to show a 'topological
similitude'. As Thompson ( 1961, page 32 1) expresses it, "There is
something, an esse ntial and indispensable something, which is common to
them all , something which is the subj ect of all our transformations, and
remains in variant (as the mathematicians say) under them all. In these
Let us assume at this stage that the thicknesses of walls can be ignored,
transformations of o urs every point may change its place, every line its
so that the walls are represented by single lines. Also we ignore any
door or window openings. We impose on this plan a coordinate system or
grid. There is a sufficient number of grid lines in either direction to mark
the positions of all walls, an d no more (that is, there are no empty grid
Jines). Thus we have three grid lines in the 'east- west' direction (as the
plan is oriented north to the top on the page) and four lines in the
'north- south' direction. These lines are of course unequally spaced ; we
can show the dimensions of their spacing alo ng two orthogonal (x and y)
Argyropelccus olfersi Scarus sp. Po lyprion axes as in :
/1'-s
Y, a
C d
Y, b
'-
x, x, x, /
/ '- This representation of the shape is unique-given a unit dimension for the
a a
grating- and we can choose for convenience to regard it as the standard or
b C d canonical version. We have removed any relative differences in dimension
C d
between the grating intervals, and so it may be termed the 'dimensionless
b representation' of the plan.
It is possible to get back to the original dimensioned plan from the
'/ 'dimensionless' version, clearly, by means of the inverse transformation, in
which the grid intervals are reassigned their correct sizes. What is more,
it could be stretched in either direction: any other differently dimensioned version of the plan may be completely
described by means of two types of information: the 'dimensionless
/ '-. / [',.
representation' to describe the basic shape, together with an appropriate
a a
I---- - set of dimensions, giving the required spacing for the grating in the x and
C d y directions.
C d b
b The following figure illustrates some examples of dimensionless
'-
/ representations of the same plan, which are not minimal in the sense
described earlier:
'/
or the relative spacing of the grating lines might be changed- in the same a
way as Di:trer's first series of portrait heads- to any dimensions we choose: C d
/ [',. /I'-
a
a
C d C d That is to say, there exists one or more empty grating lines: these can be
b
b removed without losing any part of the shape represented. It is important
'-
'/ / to avoid this condition in the standard versions of shapes, since otherwise
they will not be unique; there are many such nonminimal versions for
It will be seen that in this way an infinite number of particular plans every shape.
might be produced, all of which would, however, share the same essential This method of representing rectangular forms has effected a distinction,
four-room 'shape'. The actual sizes of all the rooms might change, but then, between the description of configuration, and the description of the
not their overall disposition in relation to each other. See how this relative sizes of the parts. It is the same kind of separation, between
follows from the fact that the walls cannot change their disposition: a 'something discrete' and 'something continuous', to which Hermann Weu_
wall which is to the 'north' of another, or to the 'east' of another, must (19 52) refers in the quotation which heads this chapter.. This splitting he
always remain so. As a consequence, in the example, room a is adjacent says is "a basic issue in all morphology". In the study of crystal forms
to rooms b and c in the original plan, and will remain so under all and the patterns of repetitive ornament with which Weyl is specifically
transformations of the kind we are considering. The same will be true or concerned, the reference is to the distinction between the underlying
all other such ' topological' relations of adjacency between the rooms- symmetry lattice or grid by which the pattern is organised ( the 'something
since these are dependent in turn on the relative positions of the walls. discrete') and its metric properties, the particular sizes which the grid units
Now take that specific transformation in which the grating intervals are and the angles between the grid lines take (the 'something continuous').
all made equal in both directions: The distinction is a most important one, as Wey! emphasises, because
although the metric properties may vary continuously-so giving rise to an
Y, a
infinity of possibilities of scale or size- it can be shown that there are
-- C d- only a strictly limited number of distinct types or configurations of
Y, b symmetry lattice. For regular two-dimensional patterns which fill the
·-
X I X I x plane, the so-called wallpaper groups, this number is 17. And for the
12 Chapter 2 The 'dimensionless' representation of rectangular plans 13
equivalent three-dimensional case, the space-filling symmetric lattices so, since it requires only a slight extension of the idea to describe plans
which correspond to the structures of crystals, the number is 230. (On with nonrectangular boundaries.
symmetry in the arts and sciences generally, see Wey!, 1952; Shubnikov If, for example, one of the component rectangles at the corner of the
and Koptsik, 1974; Rosen, 1975. Plane-filling or space-filling symmetry arrangement is taken to be, not a room, but simply a part of the exterior
of infinite extent is nevertheless perhaps of limited interest for architecture.) space surrounding the plan, then in this way L-shaped plans can be
It is in this sense that it is legitimate to speak of there being only a represented as in the left-hand figure below and U-shaped plans by means
certain restricted range of possibilities in, for example, the design of of a similar 'dummy room' on one of the edges as in the right-hand figure :
repetitive wallpaper, mosaic, or fabric patterns. The limitation is on the
possible types of symmetry ; whereas the particular details of the repeated
motifs, their size, colour, and so on are capable of indefinite variation.
We shall see in the next chapter how our method for describing
architectural plans, and indeed any comparable kinds of rectangular
designs, allows us to make some equivalent statements about the range of
possibilities for such plans. There are only certain limited numbers of Ring-shaped or courtyard plans can be treated by the expedient of taking
arrangements for the 'dimensionless representations', depending on how, one of the rectangles in the centre of the arrangement to be an open space
precisely, we distinguish those arrangements. And so we can say that the
range of 'possible designs' in this sense is finite, and all possibilities can be Y, I I
Y, 0
listed. None of this alters the fact , of course, that each one of these y• I
particular dimensio ned plans, since those dimensions are capable in principle I
Y, I
of continuous variation . Y,
Mea nwhile, let us look briefly at some other applications of the method Y, I
of coordinates, to show how different kinds of representation of buildings
Y.
I
are possible with its use. Figure 2.5(a) shows the ground-floor plan of an I
Y,
English vernacular house typical of the Eastern counties, and taken to I I
Y, I LOO
New England by colo nists in the seventeenth century. [It is of the type
classified by R W Brunskill ( 197 I) as the 'central fireplaces family'.) As a
Y,
o l - ,. ~ . ~
I
~ ~
~
matter o f fact this is the same plan as in the previous examples; but now (a)
it is shown in greater detail, with walls drawn at their true thickness and
window and door openings and the central fireplaces shown.
The principle o f the me thod of description is exactly the same, only in
this case we require additional lines in the grating to capture the extra
y • ...m.;;4---l---!-f.S --l--+
detail. The grating lines correspond in effect to all those positions to Y , -1$.~b,,,l,,,,MH
which a draughtsman would have to move the edges of his T-square and y6 ~- ..: •,!:•.•~ ::::...~ ......... ::_~,~,:.:.,;_ -+ ~-+-..._f:':'-':-.t--- - ---1',;:,t-
set-square when drawing the plan. Figure 2.5(b) shows the dimensionless Ys li . I.:'...
Y,
version of the plan, together with two lists of numbers (or vectors) to Y, x dimensions:
indicate the dimensions of the grating intervals. Since walls are now given Y,.__,-;,,.,+~;;;;
i _., {25, 50, 5~55, 2~ 110, 75, 75, 40, 15,
~zu J,•~· .z::· 20,40, 20, 50, 60, 70,25}
thickness, it is necessary to have some convention for distinguishing those Y1
y dimensions:
cells in the grating which represent solid material from those which {25, 20, 45, 25, 95, 15, 30, 30, 55, 20, 25 }
represent voids (here shown as shaded and unshaded cells, respectively). (b)
It might be objec ted that as a means of representing architectural plans
the rectangular grating is rather limited ; not only are we restricted to a Figure 2.5. (a) Ground-floor plan of an English seventeenth-century house of the
'central fireplaces' type (after Brunskill, I 971 , page I 03) with superimposed grating to
rectangular geometri c discipline overall, but the perimeter of the building mark the positions of details. (b) The same plan in dimensionless form . Shaded cells
represented must itself be a simple rectangle. However, this is not strictly represent solids, and unshaded cells voids. The two dimensioning vectors give the
dimensions of the grating intervals in x and y directions, in arbitrary units.
14 Chapter 2 The 'dimensionless' representation of rectangular plans 15
or light-well instead of a room: large numbers of grid lines and grid cells unoccupied, and the representation
becomes highly uneconomical [figure 2.6(c)]. (This can be an important
consideration from the computing point of view.) The dimensioned
grating by contrast allows precise dimensional accuracy in the representation,
while using the minimum of lines and cells in the descriptive grid. These
points are discussed more fully in relation to the t}Uestion of computer
representation in Eastman ( 1970) and Mitchell ( 1977, chapter 6).
Similarly we are not obliged to regard the rectangles in the interior as Several authors, of whom the first was Frew ( 1973), have suggested that
separate rooms. They might be zones or areas of any kind, not necessarily a different class of rectangular designs might be used to represent
separated by walls; for example an L-shaped room could be represented architectural plans (see also Frew et al, 1972; Mitchell and Dillon, 1972;
by a pair of adjacent rectangles: Matela, 1974). These are the so-called 'animals' or 'polyominoes', the
latter being the name given by Golomb ( 1966) who was largely responsible
for creating popular interest in their study. A polyomino is analogous to
a domino, but may possess more than two faces. It consists of a number of
square cells joined together along their edges. There are only two types of
In principle, very complex room or plan shapes might be depicted, given 'tromino' with three cells, a straight one and a bent one:
that they could be broken down into rectangular pieces.
In the case of the plan we must imagine a bounding rectangle set
around the outside, touching it on all four sides, with the interior
component rectangles distinguished as to whether they represent exterior
open spaces, or rooms ( or parts of rooms) in the interior:
EP
With four cells there are several possibilities:
many shapes. Some can have holes in the middle, whereas others are
rectangular overall:
'
/ '
/
the version of the ground floor of the central fireplaces house type in that is, composed of equilateral triangles, regular hexagons, and other shapes
figure 2.7. Alternatively it would be possible to treat the form of the made from the combination of these elements? The plans of some of the
whole building 'sculpturally', as though it were a solid mass, without any later houses of Frank Lloyd Wright provide good examples.
~ccount taken of internal organisation. Some examples of high-rise office (This question is not as straightforward as it perhaps appears. You
buildings treated in this way are illustrated in figure 2.8. The whole might immediately consider the idea of three dimensioning vectors, giving
building is approximated in each case as a single polycube. dimensions in the directions of three sets of grating.lines at 60° to each
other. However, the position of any point can be fixed by any two such
coordinates; and indeed the size of an equilateral triangle or of a regular
Exercises hexagon is fixed by giving the dimension of just one side. The point is
2.1 Draw dimensionless representations for some of the plans given in that these shapes cannot be squeezed and stretched in the same way as
figure 11.5. Treat the walls as having no thickness, ignore all openings in can rectangles, without their internal angles changing. Alternatively, if the
the walls and other minor details, and approximate the rooms as simple angles are kept fixed, such shapes may be transformed in size only by an
rectangles or combinations of rectangles. (You will discover that in some overall reduction or enlargement. This fact places severe constraints on
cases this involves some interpretation, and that there is no single definitive the possible ways of packing together such shapes, and must be recognised
way of representing the plan.) as one of the reasons for plans with triangular and hexagonal geometry
being rare in architecture, compared with those of rectangular geometry.)
2.2 Take one of the plans given in figure 11.5, and draw a dimensionless
representation to include wall thicknesses, openings, etc. List the dimensions 2.6 As a more open-ended exercise, you might like to look through some
of the plan in the two dimensioning vectors for the x and y axes. (Use books illustrating a variety of architectural plans, and see to what extent
the scale given in the figure; or else make some arbitrary assumptions as these can be represented by the methods indicated in this chapter. You
to dimensions.) might look for plans which can be approximated as polyominoes, or
packings of polyominoes; or for buildings whose whole forms could be
2. 3 Which of the plans illustrated share the same dimensionless
represented as polycubes. You could notice the frequency of occurrence
representation? (Ignore the particular orientation of each plan on the page.)
of plans with different types of geometry, and try to see what problems
are presented by their dimensionless representation. (Circular plans are
not so infrequent, but often consist in effect of just one principal room.
It is not unusual to find circular, elliptical, or octagonal rooms in plans
whose geometry is largely rectangular, but which are formed by 'carving'
these shapes out of the thickness of the walls of what are basically
rectangular units.)
Of the books listed in the references, Durand (1801 ), Fletcher and
Fletcher (1896), and Pevsner (1976) might be suitable for this exercise.
"The basic concepts of ... symmetry are those of rotations, such as the symmetry of
a flower; of inversion, as in the difference between the right hand and the left hand; □
We can now start the process of dissection proper, and divide this original
and the combination of these with each other and with direct movements in space."
JD Bernal (1937) rectangle into two rectangular parts. Before we do so, however, it will be
useful to take a slight detour via the subject of symmetry; since symmetry
We have seen in the last chapter how the treatment of shape and dimension properties are of some importance in distinguishing between, and counting,
can be separated in the description of rectangular designs such as plans. the possibilities of arrangement with larger numbers of rectangular parts.
Let us now set all consideration of dimensional properties on one side, We need concern ourselves with two types of planar, that is, two-
and concentrate for the moment wholly on the question of shape. Let us dimensional, symmetry: rotational symmetry and reflected symmetry.
further narrow our focus, and examine those designs which, like the first A figure is said to possess rotational symmetry if, when it is turned in the
diagrammatic plan of the 'central fireplaces' house, consist of an arrangement plane by a certain angle about some point (the centre of rotation), it
of rectangles (corresponding to rooms) set within a boundary which is remains unchanged. This is true for any planar figure for a turn through
itself a simple rectangle. The thicknesses of walls, and all openings in 360° , which merely returns it to its original position. If, however, we
those walls, are ignored. take a figure such as the square 'pinwheel' made up of five rectangles:
We can imagine that these rectangular arrangements will thus represent,
in a somewhat approximate way, a certain particular class of architectural
plans. Later on we will consider more critically what degree of
approximation this involves: how closely, that is, such representations
correspond to the typical plans of real buildings. And we will also examine
other classes of arrangement, both rectangular like the polyominoes, and
nonrectangular.
Meanwhile, for the sake of simplicity in the discussion which follows, we can appreciate that a rotation of 90° (or a quarter-turn) will leave the
the simple component rectangles of the arrangement will be referred to as arrangement unaltered. The same is true of a rotation through 180° (a
'rooms', and the arrangement of the whole as the 'plan'. But it should be half-turn), through 270° (a three-quarter turn), through 360°, or indeed
remembered that, as we have already seen , this is not necessarily their sole through any number of right angles or multiples of 90°. Such a figure is
interpretation in architectural terms, and the arrangements would serve said to possess cyclic symmetry C4 , where the subscript 4 indicates that
equally to depict L-shaped, U-shaped, and many other shapes of rooms or the figure is symmetrical with respect to (that is, is left unchanged by) a
plans. rotation of (360/ 4) or 90°.
Such designs or plans can be thought of either as being made up of sets Another example of this type of cyclic symmetry is afforded by the
of rectangular pieces, like tiles or jigsaw pieces, set together by a process of arrangement of four rectangles:
addition to make larger rectangular mosaics. Or else they may be imagined
as being produced through a process of division , by cutting a large rectangle
into smaller rectangular pieces. For the latter reason they have been
termed 'rectangular dissections', and will be referred to as such here;
although, as we shall see, there have been different approaches taken to
the problem of enumerating such designs, which have adopted, variously,
both dissection, and addition or 'tiling' techniques. In this case the figure is unchanged by a rotation through 180°, or any
We consider then the dimensionless representations of these 'rectangular multiple of 180°. It thus has the symmetry C 2 •
dissections', depicted on gratings in which the spacing is equal in both The second type of symmetry we have to consider is symmetry under
directions. Taking the very simplest case first, a single rectangle, that is, reflection. We must imagine a mirror set perpendicular to the plane of
a plan composed of one rectangular room only, the dimensionless the paper. The mirror intersects the plane in a reflection line. If the
figure remains unchanged by reflection in this mirror, then it is said to
22 Chapter 3 The symmetries of rectangular plans 23
possess reflection symmetry. Consider the following arrangement: The one cannot be produced from the other by rotation, only by reflection.
.. Taken together in the positions illustrated, the pair of arrangements shows
lE
It shows mirror symmetry about the central reflection line (shown dotted)
reflected symmetry. They are 'handed' like a pair of gloves or shoes
(although as mentioned neither figure has reflected symmetry on its own).
To come back to our dimensionless representation of a one-room plan-
the simple square. It might seem pedantic to insist on an enumeration
of the symmetry properties of this more or less trivial case. However, by
starting at this simplest arrangement we can be clearer about some of the
more complicated considerations of symmetry, which will emerge later.
as indicated. This is a very rudimentary example of what is perhaps the
The square possesses rotational symmetry of the dihedral type D4 ;
most commonly found type of symmetry in organic nature, at least in the
dihedral, because the square also has reflected symmetry, about the two
forms of animals: mirror symmetry about a single central axis, otherwise
perpendicular axes:
bilateral symmetry, as in the division of the human body into matching
left and right halves.
Now it is possible for a figure to show both rotational and reflected ·n···
symmetry. The arrangement made up from two component rectangles: ..··LJ··..
(We may notice that, in addition, it has reflected symmetry about the two
diagonal lines passing through opposite corners, though this is of limited
interest for our present purposes.)
All these same symmetry properties are displayed by the square
is an example with rotational symmetry through 180°, as well as mirror arrangement of four squares and also the square arrangement of nine
symmetry in two perpendicular reflection lines ( dotted) as shown. Where squares in the following figure:
the rotated figure has bilateral symmetry about a line passing through the
centre of rotation, as here, the rotational symmetry is referred to as
dihedral. This is a case of D2 dihedral symmetry, where the subscript 2
indicates the rotational symmetry through 180° as before(3>. The previous
example of bilateral symmetry was a case of what is denoted by strict
convention as D 1 • (It has rotational symmetry only through 360°, that is,
by returning the figure to its starting position.)
EE
We can say that, of all kinds of dissections of a rectangle into rectangles,
Notice that, by contrast, our earlier example of symmetry C2 had no these 'square patterns of squares' possess the highest possible degree of
reflected symmetry. There is a distinct 'direction of spin' in its rotational symmetry (the number of their symmetries is at a maximum).
symmetry. We may imagine two otherwise equivalent arrangements, but At the opposite extreme, there will be arrangements with no symmetries
with opposite directions of spin, clockwise and anticlockwise: by reflection or rotation at all. One of the simplest of such cases is the
dissection composed of four rectangles<4>:
Suppose that the grating on which this figure is represented in dimensionless rotations and reflections of the kind described will leave it unchanged.
form is set on a pair of x and y axes, like the examples in chapter 2. There is only in effect one possible orientation on the coordinate system.
How many ways- that is, in how many orientations- might it be set in In between these extremes, of eight possible versions, and only one
relation- to those axes? We can turn it through 90°, 180°, or 270°, and in possible version, there are two intermediate situations which may arise. In
each case, since the figure has no symmetry, produce an arrangement the case of dissections with symmetry D2 or C4 there are, under reflection
which, considered in relation to the axes, is differently oriented: and/or rotation, only two possible versions in either case:
, /[' /~
---
Or we can reflect the arrangement in a mirror line parallel to one or the
B
only four distinct isomorphs:
'
/ '/
and again rotate the resulting mirror image through three successive
quarter-turns:
the dissection with symmetry D1 in the left-hand diagram below, are The first case can be oriented in two ways (if unlabelled, it has
labelled a and b, the rotational symmetry through 180° is Jost, as is one of symmetry D2 ) . Notice that the room labelled b in the first arrangement is
the symmetries by reflection, so that only symmetry Di remains. The adjacent both to room a and to room c, but that a and c are separate.
labelled dissections in the other two cases lose all their symmetries as a The second arrangement can be oriented in four ways (the arrangement-
consequence of the labelling: if the labelling is ignored- is bilaterally symmetric in itself, but is only
symmetrical under rotation through a full turn o( 360°, so that it is
strictly a case of Di symmetry). In this case any one of the rooms a, b,
and c is adjacent to both of the others. Notice also in this dissection that
there is an internal T-shaped junction between the walls of the rooms, of a
kind which does not occur in the first arrangement.
We are now ready to go back to the process of dissection, and to The process of dissection can be carried on, to make divisions of the
consider the case where the original single rectangle is cut into two original rectangle into four parts. At this point we begin to need some
rectangular parts. It will be clear that there is, in the terms of the present systematic procedure for determining all possibilities, other than simple
discussion, only one way in which this can be done; its dimensionless trial and error. We will look at several such procedures in the next chapter.
representation being: For the moment suffice it to say that there turn out to be seven distinct
alternatives, all of which are shown below:
rn
We have just seen how this particular arrangement has dihedral symmetry D2
and can be set on the coordinate system in two distinct orientations,
that is,
B
These correspond in effect to the situations in which the dissection, the
(a) (b) (c) (d) (e) (f) (g)
Several of these we have met before: arrangement (d) is without
reflection or rotation symmetries, arrangement (g) has all the symmetries
of the square, whereas arrangement (f) is our earlier example of C2
symmetry. You might care to identify the symmetries of the remaining
cut, is made either 'east-west' or 'north-south' (taking north to be at the four dissections, as an exercise.
top of the page). However, the results of the two operations are only A new feature which emerges, in these dissections into four parts, is
rotations of the same arrangement, and so we treat them as the same basic exemplified in the single case of arrangement (g). Here the four rooms
configuration. meet at a single point, giving rise to a cross-shaped or four-way junction
When the original rectangle is divided into three rectangular parts, then between walls. This type, and the three-way or T-junction:
matters become slightly more interesting. A little experimenting will show
us that there are now two quite distinct ways in which this can be done;
two different configurations which are not simply isomorphs. There is the
arrangement where the rectangle is cut into three parallel slices and also
the arrangement when the three parts are so to speak in a 'triangular'
configuration: are the only types of junction which can occur in the interior of any
dissection (see Biggs, 1969; Combes, 1976).
The four-way type of junction will, however, present us with a
problem at a later stage. At this point we can just take note of the
special relationship which exists between dissections (f) and (g).
28 Chapter 3 The symmetries of rectangular plans 29
:ffi~m
y) a
0JffiBd➔
d-
y C
-' ~ b Y,
-
= C
)'I e b e
x. x, x, x, x, ' /
·and suppose that the dimensions of the arrangement are described by the In the case on the right the dissection occupies a grating of size 3 x 2,
dimensioning vectors (x 1 , x 2 ) and (y 1 , Yz, Y3). Now suppose that the and there are two separate internal walls which are aligned, the wall
dimension y 2 is made equal to zero. What will happen? The arrangement between rooms a and b, and the wall between rooms d and e. In the case
will be transformed into the dissection (g); that is, the two three-way on the left, which is obviously closely related, the two corresponding walls
junctions will 'coalesce' into a single four-way junction: are not aligned, and the representation occupies a grating of size 3 x 3.
If we imagine dimensions assigned to the dissection on the left, and the
+
dimension y 2 set to zero, then the result will be to align the walls in
+ ➔ +
question and thus produce precisely the arrangement on the right.
➔ However, the converse is not true: there is no way in which dimensioning
of the second arrangement can ever result in the two walls moving out of
alignment.
If we performed a similar operation on most other dissections- set~ing ~ne For this reason it seems preferable to adopt the 'nonaligned' case as
of the dimensions in a dimensioning vector to zero- the effect, as 1s easily our basic configuration, and to regard the aligned case as a particular
imagined, would be the disappearance of one or more rectang!es from the dimensioned instance of the nonaligned representation. Of course, the
arrangement. In this case, however, no component rectangle 1s lost. There symmetries of the nonaligned version are different, but all eventualities
are four to start with and four remain. can be taken care of by appropriate rotations and/or reflections. There
For reasons associated with this fact it has proved convenient in some are many dissections into six or more rectangles where equivalent problems
circumstances when counting or making a catalogue, to regard arrangements arise, and which can be taken account of in a similar way by avoiding all
related in the way which (f) and (g) are, as one and the same dissection. such alignments across the representation.
Arrangement (g) is thought of as a particular dimensioned in~tan_..:e of (f), I have been using the word 'wall' in a general and obvious sense up to
and, in general, four-way junctions are treated as 'degenerate pairs of now. But it will be convenient to introduce some more formal terminology
three-way junctions, where the 'dimension by which they are separated has at this stage, to distinguish between different kinds of walls. I will refer
shrunk to zero. On this view we would therefore count the total number to a section of wall which runs between two junctions as a wall segment.
of distinct dissections into four rectangular components as six, not 1-even. This can either be an external wall segment, forming part of the perimeter
Dissections in which all junctions are three-way have been termed trivalent. of the plan, or an internal wall segment, in the interior. In the following
With five component rectangles, the number of distinct arrangements arrangement (left) there are six internal wall segments and ten external
including those with four-way junctions is twenty-three. There are two wall segments [notice that for this purpose the four outer corners of the
with four-way junctions: plan are counted as ('two-way' ) junctions]:
consist of one wall segment each; and two more internal walls running Generating and counting rectangular arrangements:
'north-south' consisting of two wall segments each. Clearly any plan dissection and additive methods
whose overall shape is rectangular like this one must have just four
external walls.
" ... we want to cover the field by a systematic research into possibilities. The
possibilities of walls and vaults, and of the relations be~ween the walls and the cell,
and between one cell and another, want investigating, as Lord Kelvin investigated
Exercises the geometry of crystalline structures and the 'packing of cells'."
3.1 Construct polyominoes with the symmetries D1, D2, D4, C2, and C4,
WR Lethaby (1922)
respectively. Construct a polyomino with no symmetries (that is,
'symmetrical' only under the 'identity transformation'). What ~re th~ The numbers of distinct possible dissections of a rectangle into
smallest polyominoes (that is, with the least number of cells) d1splaymg
successive numbers of rectangular parts are as follows:
each of these various symmetries? (Notice that many polyominoes possess
component rectangles: 1 2 3 4 5 6
reflected symmetry across diagonal axes.) numbers of dissections: 1 1 2 7 23 119
3.2 The illustration shows rectangular dissections with four-way junctions, By 'distinct' is meant that the canonical versions of the dissections are
and/or alignments of internal walls of the kind discussed in relation to the distinct. Isomorphs by rotation and/or reflection are not counted.
first figure on page 29. Draw the 'nonaligned' versions of these dissections Dissections with four-way junctions are included. Plainly, the number of
(this means that there may be only one section of continuous straight wall dissections is growing at an accelerating rate, as is typical of many
on each grating line), and replace four-way junctions by pairs of three-way combinatorial systems of this kind. Any attempt to find and list all
junctions, as discussed in relation to the second figure on page 28. possibilities exhaustively must involve devising some organised logical
procedure, or algorithm , so as to ensure that none are missed and none are
counted twice. Once such a procedure is formulated and has been shown
to work, it can be implemented as a computer program, and the machine
used to do the work of counting, where the numbers are large.
Several such algorithms have been written for generating rectangular
dissections; and since their results are known and the dissections can be
tabulated, the detailed technical working of these various methods might
perhaps be thought to have limited interest for the general reader, or for
the reader who is concerned principally with applications to architecture
and design. Moreover the invention of a more efficient, that is, faster and
more economical method, will for all practical purposes render previous
approaches obsolete.
Nevertheless I believe it is worthwhile looking briefly at the broad
principles of the various methods which have been used, even those which
are now effectively superseded; since this will draw attention to various
properties of dissections, various means whereby they can be represented,
and various techniques with which they can be manipulated and transformed
one into another. For those readers more deeply interested in the
technicalities and in the computing aspects, references are given to the
specialised literature.
The first attempt historically to devise an algorithm for generating
rectangular dissections was I believe made by myself (Steadman, 1973),
and was conceived very much as a 'cutting' or, precisely, a dissection
method. In the original single rectangle, it is possible to make a cut in
either the 'east-west' or the 'north-south' sense, so as to divide it into two
32 Chapter 4 Generating and counting rectangular arrangeme nts 33
rectangular parts (figure 4.1). The result of both operation s, as we have it into two parts by either an 'east-we st' or a 'north-so uth' cut, so as to
seen, is to give the same basic dissection , but rotated through 90°. Each yield a dissection into n rectangles.
can be converted to its appropria te dimensionless representation. With larger num bers of rectangles in the arrangement than three, there
The next step is to take the dissection with two rectangles·, and further may possibly be several ways of making each of these cuts. Take the
subdivide one of its compone nt rectangles, so as to give a dissectio n into example of the four-recta ngle arrangem ent illustrate d in figure 4.3. We
three rectangles. Supposing we start with the two-rectangle dissectio n in a make a 'north-so uth' cut, as it is oriented, to the. rectangle labelled a.
given orientatio n, as in figure 4.2, then we can in principle make the one There are two possible positions for this cut, resulting in two distinct
cut in four different ways. We can cut either of the two rectangles; and dissections of five compone nts (which are not just rotations or reflections
in each case we can make the cut either 'east - west' or ' north-so uth'.
This exhausts all possibilities. The results are shown at the bottom of the
figure . Again they are converted to their dimensionless represent ation.
Of the two distinct possible dissections one is produced twice in the same
orientatio n, and the other in two different orientatio ns. (By applying the
process to the two-recta ngle dissection in its alternative orientatio n, we
would produce the three-rec tangle arrangements in all their other possible
orientatio ns.)
So the procedur e carries on, taking each dissection with n - I rectangles in
tum, taking each of those n - I compone nt rectangles in tum, and cutting
be
Figure 4 .3. The rectangle labelled a in the arrangeme nt of four rectangles (top) can
cut in a 'north-sou th' sense in two possible positions (centre), resu lting in two non-
isomorphic dissections of four rectangles (below).
in the ' north -south' sense (centre), to produce the dissection of two rectangles in
two possible o rientations {below).
its
$B J
rn"'"' ++· rlt7
ITO
- ,-
ta- tjjJI
1
rn
I \ : \'
BJ BI ] BI ]
made Figure 4 .4 . Two distinct possible sequences of cuts by which the di ssection of two
Figure 4 .2. One cut is made to the dissection of two rectangles (top). It can be rectangles (top) is converted to the same dissection of fou r rectangles in the same
to either of the rectangles, and in an 'east -west' or a 'north-sou th' sense in each case o rientation {below).
(centre), to produce the two dissections of three rectangles (below).
Chapter 4 Generating and counting rectangular arrangements 35
34
of each other). The method must therefore ensure that all such possible will not be produced either. (I recognised this problem while working the
cuts are made in turn. method by hand, and added in the missing dissections.)
This algorithm is somewhat wasteful, in that it may generate one For the purposes of a computer method for generating dissections of a
dissection several times in either the same or differently handed or rotated higher order, it is obviously necessary to overcome this drawback. A
versions. It is possible that different sequences of cuts can produce the second algorithm, devised by myself and Mitchell (Mitchell et al, 1976)
same dissection in the same orientation, several times. We have already and implemented as a computer program by Sauda (1975), included
seen this in figure 4. 2. Another example, at the stage where three operations designed to generate the C4 swastika and related arrangements.
rectangles are converted to four, is shown in figure 4.4. After each There are three types of operation involved in this method, which works
operation it is thus necessary to compare and reject duplicates'. includin_g directly on the dimensionless representation:
isomorphs of the same dissection, so as to minimise the work uwolved in ( 1) The first of these operations is very similar to that employed in the
succeeding steps. hand-worked method as just described. In this case, however, the
The method has never been implemented as a computer program; operation treats only those component rectangles which are made up, in
however I used it to generate the dissections with six component any given dimensionless representation, from more than one cell in the
rectangl~s, by hand, and succeeded in coming reasonably close to what is grating. In the case of rectangles made up from two cells only, the
now believed to be the complete list of 119 dissections. division can be made in one way only, converting each cell into a separate
Unfortunately it is easy to show that this procedure cannot be rectangle [figure 4.S(a)). Where the original rectangle is made from
exhaustive. The 'pinwheel' dissection of five rectangles with symmetry C4: three cells in the grating, then the division may be made in two ways
[figure 4.S(b)]. Where the original rectangle is made from four cells, there
are five possibilities [figure 4.5(c)] . And so on. This type of subdivision
operation is applied to all the relevant component rectangles in the given
dissection of n - I rectangles in all possible ways in turn, so as to add a
new nth rectangle in each case.
(6 )Compare March and Steadman (1971, page 160). Curiously, and quite _incidentally,
n
EE
this C4 'swastika' is a figure which has cropped up more than once m architectural
Figure 4.5. Generation of rectangular dissections with operation (1) of a computer
history, as a way of arranging joists to support a square floor whose span 1s greater
algorithm due to Mitchell et al (1976 ). A rectangle made up from more than one cell
than the length of the available timbers. Serlio first mentioned the proble~, and the
in the grating of a dimensionless representation, is divided into two separate rectangles.
idea was taken up by J Wallis ( 1670) in his Mechanica, sive De Moru '. who illustrates
The figure shows possible subdivisions (a) where the original rectangle consists of two
exactly the arrangement of this figure, as well as many other geometnc schemes fo~
cells only, (b) where it consists of three cells, and (c) where it consists of four cells.
the structural design of floors. Both Serlio and Wallis are quoted by Wren (1750) m
Label n indicates the new nth rectangle added in each case to the original dissection of
his Parentalia in connection with a description of his flat roof design for the Sheldonian
n - 1 rectangles.
Theatre, Oxford.
Chapter 4 Generating and counting rectangular arrangements 37
36
(2) In the second type of operation a _new nth rectangle is added along the example, a further two possibilities [figure 4.7(b)] . The same procedures
whole boundary edge of an existing dissection with n - I rectangles are repeated in turn (where applicable) to all four boundary edges of the
(figure 4.6). Clearly there are four potentially distinct ways in which this dissection. It will be seen that the operation, in general terms, consists of
might be done, depending on the symmetrie s of the dissection in question, extending certain component rectangles so as to turn the n - I dissection
correspond ing to its four outside edges. If the size of the original of size / x m into an L-shape of size / x (m + I) or (/ + 1) x m; and filling
dissection, measured as the number of columns x rows in the grating, is the corner of this L with an nth rectangle. Figur~ 4.8 shows one way in
I x m , then this operation will produce a dissection either of size (/+I) x m which the pinwheel dissection for n = 5 may be produced with this
operation from a dissection with n = 4.
or size / x (m + 1).
(3) The third sort of operation is slightly more complex than the previous All three of the operations described are applied separately, in turn, to
two, and is intended to take care of the pinwheel type dissections (amongst all dissections with n - I rectangles. The method as a whole it is clear
others). It may be described thus. Take one of the outside boundary comprises a mixture of 'dissection' operations with what are' in effect '
edges of a given dissection with n - I rectangles. The operation is applied 'additive' procedures. The results will again include large numbers of
only where two or more rectangles border that edge. We work along from duplicates; that is, the same basic configurat ion may be generated several
times, as well as in various possible isomorphs.
one end of the edge in question to the other. Let us suppose that this is
left to right in the example shown in figure 4. 7(a). Take the first rectangle a The method employed in the computer program for removing these
which borders the edge, and extend this along its length by an increment duplicates is very straightfor ward. Each new dissection into n rectangles
of one unit in the grating. Fill the remainder of the new row with a new which is generated by some operation is subjected to three quarter-tur ns
nth rectangle. This operation produces a new dissection with n rectangles relative to the coordinate system. All four oriented arrangeme nts so
as on the left-hand side of figure 4. 7(a). produced are then subjected to reflection (figure 4.9). In this way there
Repeat the operation, by extending both the first rectangle a and the is produced a total of eight versions which, as we saw in the last chapter,
second rectangle b occurring along the edge, again by increments of one might comprise either one, two, four, or eight distinct isomorphs , depending
grating unit along their lengths. Fill the remainder of the new row with a
new nth rectangle (as shown on the right-hand side of figure 4. 7(a). Where
~I~
there are , rectangles bordering the edge of the dissection which is being
operated on, then the process is repeated, working 'left to right' along the
edge in this fashion, a total of ,-1 times; that is, up to and including
the penultima te rectangle. Since there are three rectangles bordering the
edge in the illustration given, the number of possible operations is two, as
J;) ~
shown.
A similar process is now carried out along the same edge, but working
back along in the opposite direction. This gives, from right to left in the
Ck J ff i ·
: I :, I : I
l•[l dE J
I
Cm [il l 5D BI ]
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
(a) (b)
Figure 4.7. Operation (3) of the Mitchell et al (I 976) algorithm. (a) The operation
~roceeds along the edge of a d_issection of n - I rectangles, in this example from left to
n~t (top). Rectangles bordenng the edge are extended by one unit in the grating-in
Figure 4.6. Operation (2) of the Mitchell et al ( 1976) algorithm. A new nth rectangle this case rectangle a, or rectangles a and b together (centre)-an d the space remaining
is added along the whole boundary edge of a dissection of n - I rectangles. This w!thin the resul_ting L-shape is filled by an nth rectangle (below). (b) As in (a), but
operation can be carried out in four potentially distinct ways, corresponding to the with the operation now proceeding from right to left along the edge of the dissection.
four sides of the original dissection.
Chapter 4 Generating and counting rectangular arrangements 39
38
on the symmetries of the basic configuration in question. By a matching program did not distinguish arrangements with wall alignments from their
of the eight against each other, any duplicate versions resulting from such equivalent 'nonaligned' versions (compare the first figure on page 29), and
symmetries can be detected and removed. The remainder are matched consequently gave totals which are lower than where these are separately
against all other dissections with n rectangles so far produced. If one of counted. (This question is taken up again in chapter 5, where table 5.2
the new arrangements duplicates some dissection already found, then they gives counts for the inclusion of and also for the exclusion of arrangements
are all rejected; if not, one version is chosen and retained. In this way with wall alignments. For nonaligned dissections, .and including those with
the final list contains only one version for each dissection. four-way junctions, the numbers are 735 for n = 7 and 5 527 for n = 8.)
The computer program embodying these procedures was used to generate The program was also relatively slow by computing standards; in particular
dissections for n = 7 and n = 8. The numbers of dissections produced the process of sorting out duplicates is inefficient, since it involves comparing
differ slightly from those found by other authors however, since the up to eight isomorphs of each new dissection against all of up to several
thousand other dissections (in the case of n_ = 8) in turn.
It has furthermore been shown by Earl (1977) that, quite apart from
the problems with alignment, this algorithm is not completely exhaustive
(a) (b)
Figure 4.10. Examples of dissections with sixteen rectangles (a) and fifteen rectangles (b)
of the type which Earl ( 1977) shows cannot be generated with the Mitchell et al ( I 976)
algorithm.
Figure 4.8. One way in which the C4 'swastika' dissection for n = 5 may be produced
from a dissection for n = 4, with operation 3 of the Mitchell et al (1976) algorithm.
Figure 4 .9. Rejection of isomorphs in the Mitchell, Steadman, and Liggett algorithm.
Each dissection generated is subjected to three quarter-turns relative to the coordinate
system; and all four oriented arrangements so produced are subject to reflection.
(The diagonal line marks the dissection here in such a way as to illustrate that
Figure 4.1 1. The two dissections of figure 4.10 in their nonaligned versions.
potentially it may have eight isomorphs.) All these eight versions are then matched
against each other, and any duplicates removed.
40 Chapter 4 Generating and counting rectangular arrangements 41
for larger values of n . This he demonstrates with the 'particularly elegant' (2) Add a new rectangle of depth one unit and filling the whole width of
counterexample of figure 4.1 0(a), which contains sixteen rectangular the dissection:
components. This dissection would not be produced by any of the three
operations described.
Notice that no two adjacent rectangles in the arrangement can be
coalesced into a single rectangle by the removal of one wall; so , conversely,
no pair of rectangles could have been produced by the dissection of a
single existing rectangle, as with operation (I). The arrangement along
any of the four outside edges of the dissection is not such as could have (This corresponds to the second operation in the previous algorithm.)
been produced by the addition of a single I x I or I x m rectangle along (3) Extend one or more rectangles at a corner of the dissection by one
the side of an existing dissection, as with operation (2). Nor is the edge unit, to form an L-shape, and fill the opposite corner with a new rectangle:
arrangement such as could have been produced by operation (3), since
this increases the dimension of the grating by one unit only; and the
corner rectangles here all have dimensions 2 x 2. Figure 4. I 0(b) shows a
further example of a dissection impossible to produce with the algorithm.
In this case the number of components is fifteen. Earl believes this to be
the smallest value of n at which the particular situation arises.
A new algorithm , devised by Earl ( 1977), avoids the difficulty, and is (This corresponds to the third operation in the previous algorithm.)
claimed by him to be completely exhaustive for all n. Earl's method is (4) Extend one or more rectangles at both corners of the dissection by
designed, unli ke those so far descri bed, to generate only trivalent one unit, to form a U-shape, and fill the central gap with a new rectangle:
nonaligned, or f u11dame11tal dissections. (Aligned arrangements and those
with four-way junctions may be generated by appropriate dimensioning
operations on nonaligned dissections as explained in chapter 3.) The
counterexamples of figu re 4. 10 would ap pear in their nonaligned forms as
in figure 4. 11 .
4
~ I ~1
Earl shows tha t it is possible to dispense with 'dissection' or subdivision
operations in the interior of an existing arrangement, of the kind represented (It is this operation which is required to generate Earl's coun terexamples
by the first operation in the previous algorithm. Instead it is sufficient to of figure 4. 11.)
carry out four kinds of procedure at the edge of any dissection. Imagine All those operations are to be carried out along the edge in question in all
that the edge in question is at the 'top' or 'north' of the dissection as it possible ways, very much as described for the previous algorithm. It is
appears on the coordinate system and on the page. Then : not, however, necessary as before to apply the rules to all fo ur edges of
(I ) Take a recta ngle which lies at the top edge, and divide it into two each dissection in turn. The reason will be clear from figure 4 . 12.
rectangles with a wall running in the vertical direction, not aligned with Suppose we start with the unit rectangle, and work only on its 'north'
any other existing verti cal wall in the remainder of the dissection: edge (drawn in thicker line). By applying rules I and 2, respectively, we
obtain the dissection for n = 2 in its two possible orientations. By then
applying to these the various rules which are relevant in each case, as
indicated on the figure, and still working always at the top edge only, we
I generate the two dissections for n = 3, one in its two isomorphs and the
~
other in its four isomorphs. [Only rules (1 ) to (3) are employed, since
there is no situation where rule (4) applies up to this point. )
Earl's method thus has the great merit that each dissection is produced
once and once only in each and all of its possible rotations and reflections.
(It follows that this operation involves an increase in the width of the (It is obviously necessary to maintain consistency about which edge is the
grating by one unit.) 'working' edge.) It is not possible for the same isomorph of the same
dissection to be generated via two distinct sequences of operations.
Chapter 4 Generating and counting rectangular arrangements 43
42
Another way of representing and working on dissections which can also be The enclosing brackets O serve to identify each wall, the prefixes I and - I
used for an exhaustive enumeration has been devised by Flemming (1978). are used to signify east-west and north-south walls, respectively, and the
Flemming makes use of what he calls wall representations of rectangular complete wall representation of the plan as a whole is enclosed in braces { }.
plans. He uses 'wall'· here in the formal sense introduced in the last A typical operation in Flemming's method for generating dissections is
chapter, to mean a maximal continuous straight run of wall. Thus in the illustrated in figure 4.13(b). A new fifth rectangle, labelled e, is introduced
dissection of figure 4. l 3(a) there are three walls oriented 'east-west' and into the dissection with four rectangles, by 'pushiug' d over towards the
four walls oriented 'north-south'. Flemming confines his attention to east and inserting a new north-south wall as shown. Besides this new wall,
trivalent dissections- he calls them 'T-plans', since every junction is T-shaped. there are three old walls which are affected by this operation: all four are
Notice how the external walls on the north and south sides of the dissection drawn in heavier lines in the figure. The new wall representation is as
in figure 4. l 3(a) are by convention shown extended beyond the outer follows. See how the change is registered in the lines corresponding to the
corners of the plan, so as to turn these into three-way junctions. three altered walls and the new wall (in heavy type).
Suppose the rooms in figure 4. I 3(a) are labelled a, b, c, and d as shown and N N
the four regions around the outside of the dissection are labelled N, E, S, - - ~
a -
b - -- ~ a b --
and W for the points of the compass. The extensions of the external walls W E W E
on the north and south sides serve to separate these four exterior regions. C d c e d
The wall representation now consists of a list of walls, first all east-west
-- s - - - -
s
- -
(a) (b)
oriented walls, and then all north-south walls. Each wall is represented
by listing in order the rooms or regions which lie on either side of it. An {(l, (N), (W, a, b, E)), ( 1, (a, b ), (c, e, d)), (1 , ( W, c, e, d, E), (S )),
east-west wall is represented by listing first the rooms to the north of it, (-1, (W), (a, c)), (-1, (c), (e)), (-1, (e), (d)), (- 1, (a), (b)), (-!, (b, d), (E))}.
in order from west to east; and then the rooms to the south of it, in the
same order. Similarly for a north-south wall, first the rooms to the west Figure 4.13. Wall representations of rectangular dissections, from Flemming (I 978).
A wall here is a maximal continuous straight run of wall. Thus there are three walls in
are listed, in order from north to south; and then the rooms to the east, the dissection illustrated in (a) oriented east -west , and four walls oriented north-
in the same order. Thus the northern boundary wall of the dissection in south. The two external east-west walls are extended beyond the corners of the plan,
figure 4. I 3(a) is represented as (N), (W, a, b , E), where the brackets () so that all junctions are three-way (the dissection is trivalent). In (b) a new rectangle e
divide the rooms or regions to north and south, respectively. The complete is inserted into the dissection of (a), through an operation in which rectangle d is
wall representation for this dissection is pushed to the east, and a new north-south wall is introduced. Besides this new wall
there are three existing walls affected by this operation- they are all four marked in
{<l, (N), (W, a, b, E)), (I , (a, b), (c, d)), (1, (W, c, d, E), (S)), heavier line. The new wall representation is shown beside the dissection. The changes
are registered in the four lines corresponding to the three altered walls and the new
(-1, (W), (a, c)), (-1, (c), (d)), (-1, (a), (b)), (-1, (b, d), (E))}. wall (marked with asterisks).
~~ II
(a) (b)
(c) (d)
Figure 4.12. Generation of dissections of up to three rectangles, with an algorithm Figure 4.14. Types of operation on wall representations of (trivalent) rectangular
dissections, from Flemming (I 978). In each case a new rectangle is introduced by
due to Earl ( I 977). The process starts with the single rectangle (top); and all rules
are applied in turn where relevant to the 'north' edge (shown in heavier line). The pushing aside some existing rectangle towards the east and inserting one new north-
numbers indicate the rule applied in each case. See how every dissection is produced south wall. Operation (a) is the one employed in the previous figure. Operation (d) is
of the general type required for making the C4 swastika and related arrangements.
in all its isomorphs. (There is no situation in which rule (4) applies up to this level.]
44 Chapter 4 Generating and counting rectangular arrangements 45
Every operation in the generating process is similar to this one just would serve equally to describe the two dissections:
illustrated in that a new rectangle is introduced by pushing aside some
existing r:ctangle and inserting one new wall. In some cases this requires
that other existing adjacent rectangles are 'pulled' or extended to fill part
of the space where the first is pushed aside.
--
a ~
W
- -
b - -
c
N
s
d
E
--
--BB--
- -
W
a
c
N
b
d
E
- -
Differences arise out of the ways in which the continuations of the four s
walls bounding the new rectangle are oriented, north-south or east- west. Of these, one would be without its labels a mirrored isomorph of figure
Four possibilities are illustrated in figure 4.14. In all cases existing 4. l 3(a), but differs by virtue of the labelling so that, for example, whereas
rectangles are pushed towards the east and new north-south walls rooms a and d are adjacent in figure 4. I 3(a), they are not in the above
introduced. The operation of figure 4. l 3(b) appears here as figure 4.14(a). figure. In the other, the two three-way junctions are coalesced into a
The operation shown in figure 4. l 4(d) is of the general type needed, a~ four-way junction. In general this whole question of overlaps or alignments
will be appreciated, to make 'pinwheel' and related dissections. Reflections across walls is taken care of by Flemming at the stage when dimensions
and rotations of these four basic situations give rise to a total of sixteen are reintroduced into the wall representation, as described in chapter 9.
possibilities as illustrated in figure 4.15. So an enumeration of (undimensioned) wall representations would yield
Flemming shows how each of these geometrical operations can be smaller numbers than that for rectangular dissections as previously defined.
expressed as a corresponding logical operation on the wall representatio~,
by maki ng substitutions in the sequences of letters or symbols representmg
rooms. To generate all trivalent dissections using Flemming's approach, it Exercises
would be necessary to start from the configuration with no rooms, just the 4. 1 Use Earl's procedure to carry the process illustrated in figure 4 . 12 one
external walls, as in stage further, to generate dissections with four component rectangles.
The results can be checked by referring to the first figure on page 27 of
N
chapter 3. There should be, in all, twenty-six arrangements counting all
w\ E isomorphs. [Notice that the one dissection containing a four-way junction,
s dissection (g) in the first figure on page 27 is not produced.] Rule (4) will
and to apply suitable sequences of operations as described, to yield be applicable in one instance only.
arrangements for successive values of n. It should be said though that You might wish to go on to enumerate dissections for n = 5. There
Flem ming's approach is a more general and abstract form of representation are twenty-one of these, excluding arrangements with four-way junctions
of a packing of rectangles into a rectangle than is a dimensionless (not counting isomorphs). You can check your results against figure 5.1.
representation on a grating. The wall representation, by its nature, does 4. 2 Make wall representations of the dissections shown in the figure. By
not always specify the 'overlapping' adjacencies of rooms across a wall. which of the operations, as illustrated in figure 4.15 , can the dissections
Thus the wall representation given above for the dissection of figure 4. l 3(a) with four rectangles be made into the dissections with five rectangles?
IT 6 7 8
--~- -
w
b
N
C d
E
--[EB-- N
b c
2 3 4 5 -~ a d ~ -
-- - -
s S
µ - - ~ --
N
- - b~e c- -
N
9 10 11 12 13 14 15 16 W a E
w e E d
figure 4.1 S. Reflections and rotations of the basic operations on wall representations C d -- - -
shown in figure 4.14 , giving rise to sixteen possibilities in all; from Flemmmg (1978). b s
-- s
- -
Generating and counting rectangular arrangements 47
5
N
)( ~ [EJ cm
Dissections for n = 5 take up gratings of size l x 5, 2 x 3, 2 x 4, and 3 x 3.
In this case there are three arrangements of size 2 x 3: two with four-way
N
N
)(
.,.,
[E EE
junctions and one with an alignment. The two with four-way junctions )( I I I I I I
may be generated from nonaligned dissections of size either 2 x 4 or 3 x 3, ....
)( I I II I
whereas the remaining aligned arrangement takes size 3 x 3 when nonaligned
as was shown in chapter 3:
M
)( [I]]
N
,1'. )( [I]
.... .,.,
y
y
'
-'
yI
a
b
C
d-
e
y,=Offid
-
b
C
e
N
"' "'
II
"' "'
'
x, x, . x, /
These results, together with the numbers of dissections of each size, are
set out in table 5.1.
Even considering these relatively few cases, it emerges that there exists,
for each value of n , a range of possible grating sizes, from the extremely
elongated to the square or 'squarish'. Of these the dissections most nearly
square are relatively more numerous, whereas there is for each value of n
only one dissection of size Ix n. It is obvious that this latter type is the
Chapter 5 Generating and counting rectangular arrangements 49
48
line is made
there can be which is chosen delibe rately to emphasise how every grating
most elonga ted shape possible for given n; and equall y that wall. By contra st the 'dense st' kinds of
only one such dissect ion for any value of n, tha t is, the arrang ement necessary by the presen ce of one
maxim um for the given grating dimens ions,
where all n rect angles are strung out in a row. dissect ion, where n is at a
require d every cell of the grating represe nts a compo nent
It is possible to predic t theoretically the range of grating sizes must be those where
genera te and measu re rectangle, and n = Im . The case fo r grating size 4 x 5 is:
for dissections with n rectang les, withou t the need to
largely to
all individual cases. The following demon stratio n is due very 4 x5
er in the first place fundam ental dissect ions, withou t
Bloch ( 1976). Consid
simple
four-way junctio ns or alignm ents. Fo r this type there is a
n I, m , and n (Biggs, I 969; Earl, 1977), such that
relationship betwee
(5 . 1) 1---+--1 --4--- 1--.J I = 4
n=l+ m-1 . m= 5
apply for the '---'--1 ---'--- 1--' n = Im = 20
You might like to confirm that this relatio nship does indeed
4 , and 2 x 3 for n = 5). dissection of
Obviou sly for given values of l and m there is only ever one
values of n in table 5. I (excluding 2 x 2 for n =
er the smalle st possib le
The reason is quite straigh tforward. Consid this type.
n , I, and m all have the for n for
dissect ion, of one cell in the grating, in which The following figure gives examples of the two extreme values
value 1. The relatio n obtain ed here is I = I + I - 1. A new
rectangle is various grating sizes:
one interna l wall. (It follow s that the
introd uced by the additio n of just
nt dissect ions- must always equal n - I ,
numbe r of interna l walls- in trivale I x4 4x4
r of rectang les.) See how each of Earl's
that is, one less than the numbe
1--
of l and m will be found. Figure 5.2(a) shows this curve illustrated for possible sizes of gratings needed. We take each size in turn and divide it
n = Im = 6. Any points below this curve cannot represent compatible up into, or fill it with, n rectangular pieces. Bloch (1978; 1979a) has
values of l and m for n = 6. himself employed just this technique; his algorithm is conceived in terms
All that we need now is to specify the upper limit on combinations of / of placing rectangular 'tiles' in such a way as to fill empty gratings of
and m. This we do by plotting the straight line corresponding to predetermined size<7>.
l+ m-1 = n, to represent the most sparse types of dissection-those The first step is to take the total number of cells Jn the grating, given
where the sum of/+ m is at its greatest for given n. The overall result is by l x m, and divide this number into n parts, each of which will correspond
to define a closed, roughly triangular area on the graph [figure 5.2(b)], to a tile. Thus a grating size 4 x 5 will contain twenty cells, which might
bounded by the two straight lines l = m and l + m - l = n, and by the be partitioned (in just one way out of many) into eight tiles as
curve Im = n. All points representing integral values of l and m falling on { 4, 3, 3, 3, 3, 2, 1, 1} .
or inside this boundary represent possible grating sizes for dissections with
n component rectangles. Figure 5.2(b) illustrates the situation for n = 6. The numbers in the partition are arranged for convenience in descending
Figure 5.3 shows the points plotted similarly for successive values of n order of size; each represents the area of a single tile; together they sum
between 1 and 9. See how in all cases they fall within a triangular area to the area of the whole grating.
delimited in the way described; and how in all cases there is one grating Each value in the partition must be further broken down into a pair of
size, at the extreme right-hand corner, where l = 1, corresponding to that factors to represent the actual dimensions (in grating intervals) of that
unique dissection where the n rectangles are set out in a single row. We particular tile. The result is what Bloch calls a 'factored representation' of
can confirm these results for n up to 5, by comparison with the sizes the partition. For the partition given above, we might have a factored
listed previously in table 5. I. Remember that it is possible that to one representation
point, or size l x m , particularly where the shape is squarish, there may {(2 X 2), (1 X 3), (1 X 3), (I X 3), (1 X 3), (1 X 2), (1 X 1), (I X 1)} .
correspond a large number of different dissections.
This demonstration of Bloch's allows a new approach to the problem of s
generating dissections exhaustively. For given n we now know all the 4- n= 1 n =2 n=3
3-
2- 0
f:- f:-
7 \.q 7 \.q 1- 0 0 0
0
6 o6x4 6
s s
s n=4 n=6
[~
4
I 4 04x6 4 3 0 0
2 0 0 0 0 0
3 3 0 0
2 2 0
Im= 6 Im= 6 5
n = 1 n = 8 0 0 0
n=9
4 0 0 0
0 0 3 0 0 0 0 0 0 0 0 0 0 0 0
0 2 3 4 s 6 7 8 0 I 2 3 4 s 6 7 8 2 0 0 0 0 0 0 0 0 0 0 0 0
m m 1 0 0
(a) (b)
0
Figure S.2. (a) Graph of possible grating sizes, corresponding to pairs of integral values 0 1 2 3 4 5 6 7 8 9 0 1 234567 8 9 0 1 23456789
for grating dimension s I and m; after Bloch (1976). Square gratings lie on the line m
I = m passing through the origin at 45° to the coordinates. Points placed symmetrically Figure S.3. Possible grating sizes / x m for values of n between 1 and 9; after Bloch
either side of this diagonal correspond to gratings of the same size, but with the values I
(1976).
and m exchanged; as with the sizes 6 x 4 and 4 x 6 illustrated. The lower limit on
pairs of values for I and m for given n , is given by the hyperbola Im = n, plotted here
7
for n = 6. (b) The upper limit on pairs of values for I and m for given n, is given by ( ) In the mathematical literature, problems of this type are strictly classified as
the straight line I+ m - I = n, plotted here for n = 6. All compatible pairs of values 'packing' problems. The term 'tiling' refers to filling the plane with regular patterns
for / and m for n =6 lie inside or on the boundaries of the area (shown shaded) composed of repeated copies of some few component shapes. However, 'tiling' very
defined by I = m, Im= 6, and l+m-1 = 6. aptly describes the nature of the process involved here.
51 Chapter 5 Generating and counting rectangular arrangements 53
For each specified grating size (that is, each pair of values for / and m) it identified and rejected. Those that remai~ at the end include dissections
is necessary to determine every possible partition- of which there may be with four-way junctions, as well as dissections with wall alignments .
many-and for each partition every possible factoring- of which there may Using this technique, Bloch has generated all dissections up to and
again be many. The result is a listing of all possible sets of n tiles, with which including n = 9; and he has used the computer to make a catalogue of
the grating might be filled. There are various conditions applying to drawings of each dissection up ton = 8 (see Bloch and Krishnamu rti, 1978;
factored representa tions- restriction s on admissible sets of tiles- which Bloch, 1979a; 1979b). This catalogue is reproduced here (page 250) as an
Bloch discusses. These include conditions on the largest size of tile, and appendix, but with results for n = 8 omitted for lack of space. The
on the maximum dimensions or combinatio ns of dimensions for large tiles. catalogue is organised according to a whole series of classificati ons which
I will not go into the details of Bloch's algorithm for generating are explained in chapter 8. It seems that Bloch's method is not capable
dissections, which is rather complex; but in general outline he proceeds of an enumerati on beyond n = 9, because of the fact that it produces
as follows. Every grating size is taken in turn, and in each case every increasingly larger numbers of the nonminima l and isomorphic arrangeme nts
factored representa tion is taken in turn. The set of tiles which the factored which have to be weeded out.
representa tion comprises is fitted into the empty grating, in all the The exact numbers of dissections in the various categories- with and
arrangeme nts which are possible. Three of the possibilities for the 4 x 5 without alignments and four-way junctions- for these higher values of n,
grating and factored representa tion given above are: are given at the end of this chapter. Meanwhile, we should look at one
last method for generating dissections, which exactly confirms Bloch's
results, and which is due to Krishnamu rti (Krishnam urti and Roe 1978·
Bloch and Krishnamu rti, 1978). ' '
I Rather than filling the empty grating with tiles, the essential concept
- here is that of assigning imaginary colours to the grating cells. Each
rectangle in the dissection is then represente d either by a single coloured
cell or by a group of adjacent cells all coloured the same. Different
The tiles are fitted by working systematic ally from one end of the rectangles are coloured differently ; and therefore as many colours are
grating to the other, in the direction of the longer grating dimension m: needed as there are rectangles, n, in the dissection.
Imagine a grating set with its long dimension m 'east-west ' on the page:
□ DODD □ □□
m
By taking in turn all feasible sequences and permutatio ns in which the ,__ I '/
tiles can be placed, and so proceeding until the grating is filled (or the r-...J
process fails), all possible tilings are generated. The process is such that
~
many dissections are finally produced which are nonminim al (compare the r- r--- ----
figure on page 11) or which are isomorphi c despite precaution s being
taken to avoid some such conditions arising. Such configurat ions must be
----
Chapter 5 Generating and counting rectangular arrangements 55
54
Let us call the cell in this sequence which is to be coloured at each step,
the qth cell. .
The colours are applied according to a set of rules whose combined
effect is to ensure that only groups of adjacent cells which together form q-m
rectangular shapes are coloured the same; that exactly n colours are used;
and that the resulting dissections are minimal both in 'east-west' and in
cell type 3
'north-south' directions. The rules are of two principal types: those
which colour the qth cell the same as one of its predecessors in the It is easy to see that this must be the (q - m)th cell, counting back in
sequence (that is, they extend an existing rectangle), and those which sequence a number of cells equal to the width of the grating. Again the
apply a new colour (that is, they start a new rectangle). Let us symbolise options are to colour the new qth cell the same as the (q- m)th, or to
the new colour by µ. assign a new colour µ.
Within this the rules may be further divided into four categories, ( 4) Finally, there are all the remaining cells in the grating, for which in
according to the cells in the grating to which they can apply. every case there will be three neighbouring cells already coloured. These
( 1) The simplest situation is that applying solely to the single cell at neighbours are the cells numbered q - l, q - m - 1, and q - m. Let us
bottom left in the grating, which is the first in the sequence, and which suppose that these have been coloured r, s, and t respectively, where r, s,
therefore must be assigned the first colour: and t might be the same or they might be different colours:
q - m-1 q- l rn q
cell type 1
□ q-m
q- m- 1r n q- m
cell type 4
(2) For those cells which are in the bottom row, each one can only have
Only certain combinations of colourings of this group of four cells
one neighbour already coloured, and that is the (q - 1)th cell on its
(including the qth cell) are permissible, within a rectangular dissection.
immediate left. There are two options for the qth cell here: either it is
Thus it is not allowable for three cells to be of one colour, and the fourth
coloured the same as cell q - 1, or it is coloured differently, with a new
a second colour, since this would create an L-shape in the final arrangement.
colour µ: Four situations are permitted for rule 4:
4(a) r = s = t. Only one rule applies: the qth cell must be coloured the
same as the other three:
q-1 q
cell type 2
(The four cells belong to the same rectangle.)
Either of these is compatible with the arrangement becoming a rectangular 4(b) s = t and r is different. Two rules can apply: the qth cell may be
dissection. coloured the same as r, or it may be assigned a new colourµ:
(3) When one or more rows have been completely coloured, a special
condition arises with each new cell in the column at the extreme left. In
this case the only neighbouring cell already coloured is that immediately
below:
Chapter 5
56
(The four cells belong either to three or to four rectangles.) Notice that
of all these operations it is only in this very last case, where the four cells
in question are all coloured differently, that a four-way junction is created.
Two general conditions on the applications of these rules ensure that
exactly n colours are used for colouring the complete grating. Rules of
the first general type, those which extend an existing rectangle, apply only
where lm - q ~ n - µ . The total number of cells in the grating is given by
lm , and since q is the number of the current cell, lm - q must give the
number of cells remaining to be coloured. The number of unused colours,
that is, the number of new rectangles remaining to be created is given by
n - µ . The condition therefore states, in effect, that where there are as
many cells remaining as there are rectangles still to be created, each one
must be coloured differently.
Meanwhile , only when n > µ , that is, the number of colours used so far
is less than n , are the rules of the second general type- those which assign
a new colour and thus start a new rectangle- applicable.
It remains to ensure that the dissections are minimal both in the 'north-
south' and in the 'east -west' senses. Recall that in a minimal arrangement
there must be at least one wall on every line in the grating. The first
point at which it is necessary to make a check for minimality (in the 'o .8
'north-south ' sense), is reached at the extreme right-hand cell in the .. e
.8 ::,
second row up (assuming l > 2): e"
g~
.:a 8
-.. .,"
-5 r: ::l
:ig8
I o bO
t: <) .s
1
1
1
1
2
2
3
3 4
I
1
1
1
2
2
3
3
5
4
~:fl
e ~,
1
E "._
- .. 0
0
58 Chapter 5 Generating and counting rectangular arrangements 59
Suppose, as in the example illustrated, that there is no_wall so far o n the (Although by no means all numbers of the appro priate length will
grating line between first and second rows. In these circumsta_nces, the correspond, obviously, to dissections.)
colouring rule applied must be of the second general type, which_creates Krishnamurt i and Roe choose to retain as their representativ e or
an east-west wall on the grating line. Similar tests must be applied to all canonical isomorph in each case, that arrangement which possesses the
cells further up the extreme right-hand column of the gratirlg. They must lowest number. In the above example, this would therefore be
also be carried out when the colouring process reaches the top-most row, arrangement (a) whose code number is 11234 2. Consider the stage at
starting from the second cell in that row, so as to ensure minimality ir1 the which this particular colouring is first generated. It is subj ected to rotations
'east-west' direction. and reflections (as in previous methods) to produce its possible isomorphs:
Krishnamurt i is able to generate separately the different classes of
dissection-w ith or without alignments or four-way junctions-b y the
judicious omission or inclusion of selected rules. There is only one type ~ ~
of rule, as already noted- that of 4(d) above, where four cell~ a~e all
coloured differently-w hich creates a four-way junction. Om1ss1on of this
. ~ ~
rule results in only trivalent dissections being produced. If alignments of These isomorphs are t hen 'recoloured' in the standard cell sequence, and
walls are to be avoided, then rules of the type which create a new rectangle their correspondin g code numbers derived [as in (b), (c), and (d) above] .
cannot be applied in t wo circumstances, both of them relating only to Since these new numbers are all greater than that of the original
cells o f the fourth category. These are eit her when s = t [refer to 4(b)] colouring (a), if follows that (a) must be the canonical isomorph. When t he
and there is a wall already on the relevant 'north- south' grid line (with same process comes to be applied to (b), (c), or (d), and these are reflected,
which the 'north-south ' wall crea ted by the new rectangle will be aligned rotated, and recoloured in a similar way, at least one of the resulting code
but not continuous). Or similarly where r = s [refer to 4(c)], and there numbers in every case will be lower than that of the original colouring-
is a wall already on the relevant 'east-west' grid line. . . . hence none of these can be canonical and they are all in turn rejected.
Figure 5.4 gives a sample illust ration of the w~ole algont~m_1~ action, The ingenious feature of this scheme is that there is no need to store
for a grating size 2 x 3 and n = 4. The 'colours h~r~ ar~ s1g111f1ed b~ the and compare colourings to detect isomorphs. It is possible to inspect a
numbers I to 4 . Where appropriate the tests for m1111mahty are applied legitimate colouring in isolation, and determine whether or not it is the
and their results indicated. See how the application of all relevant rules at canonical version and should be retained.
each step produces the five dissections for these values, in all the isomorphs With this algorithm Krishnamurt i was able to generate dissections up to
which this particular orientation of the gra ting allows. . and including n = l 0, the furthest any enumeration has been taken to
Krishnamurt i and Roe have an economical technique for detectmg and date. Table 5. 2 summarises the results obtained by Bloch and Krishnamurt i
removing isomorphs. Consid er the following four isomorphs of the (1978) for values of n from 5 to l 0 , with the different classes of dissections
dissection with fou r rectangles, labelled (a) to (d) as they were generated counted separately in each case. Krishnamurt i quotes a figure of three
in order by the colouring process:
Table 5.2. Numbers of dissections of differen t classes up to n = IO.
~ l1l3E] ~ l4l4E n Number of dissections
Exercises
5.1 List the possible grating sizes for dissections of order 12. (Draw a
graph as in figure 5.2(b), plot the relevant equations, and identify the
integer points.) In the first case the room labelled b is adjacent to both a and c; but a
5. 2 Use the general principles of Bloch's approach to generate all possible and c are separate. In the second dissection each of the three rooms is
tilings on 2 x 3 gratings. You will need first to partition the six grating adjacent to both the others. Imagine these, or indeed any other dissections,
cells into n tiles in all possible ways. Notice that tiles of area 5 are not subjected to the kinds of transforma tion described in chapter 2, where the
possible, since these could only take the shape 1 x 5 which will not fit on grating intervals in the x and y directions are assigned particular dimensions.
the grating. With the partition {5, 1} omitted, then, you should find that Evidently, the relationships of adjacency between rooms will not be
there are ten remaining partitions, including the extreme possibilities {6} changed und er such transforma tion, no matter how the re lative sizes of
and { 1, 1, 1, 1, 1, 1 }. Each of these has as it turns out only one factored the rooms may be al tered.
representa tion (the shapes of tiles are here given uniquely by their areas). Suppose tha t we represent each room in an arrangeme nt by a poin t ;
You will now need to work by trial and error, fitting each set of tiles into and that wherever two rooms are adjacent, we draw a line joining the
the grating in all possible ways. See how many of the arrangeme nts so respective points. (For these purposes we define rooms to be adjacent
produced are nonminima l- indeed the simple presence of a 2 x 2 tile, or where they share some length of wall in common. So two rooms which
of the single 2 x 3 tile ensures that the arrangeme nt must be nonminima l. touch only at a corner, for example, rooms diagonally opposite at a four-
You should end up with nine distinct dissections : five of order 4, three of way junction, are not regarded as adjacent.) Such figures constitute what
order 5, and one of order 6. are known mathematically as graphs; in this case we can refer to them as
adjacency graphs. The adjacency graphs of the two dissections for n = 3
5.3 Use Krishnamu rti's colouring technique to generate all dissections of are:
order 5 on 2 x 3 gratings, following the lines of figure 5.4. The first stages
will be identical to those in figure 5 .4; but the criterion Im - q < n - µ
b
(fewer cells remain than colours) will begin to apply sooner. Remember a c
0--0---0
to make the checks for minimality. The results should be as for the
dissections of order 5 in exercise 5.2 above- but they will occur in all the These adjacency graphs are rather similar to what architects call
different isomorphs which the orientation of the grating allows. 'functional ' or 'linkage' diagrams, and sometimes make use of in the early
stages of layout design. Another term is 'bubble diagram'- where the
5.4 Of the isomorphs of the three dissections produced in the previous rooms are the 'bubbles' -although this, as we shall see, is closer to what
exercise, which is the canonical isomorph in Krishnamu rti and Roe's strictly should be termed a 'plan graph'. In general , the meaning and
terms, in each case? (Derive the number for each isomorph as in the definition of such diagrams is imprecise, however, and they are used to
figure on page 58 and identify that with the lowest number.) combine topological with shape and dimensiona l informatio n in a way
62 Chapter 6 Graphs of plans and their arrangement 63
which is inconsistent and conceptually confused. Here I shall try to be to the theory exactly how the graph is drawn. The vertices may be placed
more explicit in defining precisely which features of a plan the graph does anywhere on the page, and the edges joining them may be straight or
and does not represent. curved, even crossing each other. What matters is how many vertices and
Relationships of adjacency between rooms in the plan of a building are edges there are, and which edges join which pairs of vertices. So there is a
clearly of the greatest functional significance. Where two rooms share a large, indeed an infinite, number of ways of drawing the same graph. The
sufficient length of wall in common then it is possible for them to be graph, despite its name, is an abstract structure, which is depicted for
made accessible one from the other via a door. The overall patterns of convenience in graphical form.
adjacency between corridors, halls, and other rooms determine the ways in In general, graphs are used to represent sets of relationships (denoted by
which routes can run, along which people can circulate through the the edges) between sets of entities or elements of some kind (the vertices).
building. Again, the adjacencies of rooms to the exterior constrain the In o~r case these relationships are spatial ones- relationships of adjacency-
possibilities for placing windows in those rooms; and the same is true for but m othe~ applications the relationships represented might be completely
doors giving direct access to the outside. nongeometnc. For example, graphs have been used in sociology to record
There are other aspects of the planning of buildings which may have to personal relationships between members of social groups; in anthropology
do with the adjacencies of rooms. For example, it may be desirable that to represent marital relations and kinship structures (family trees) ; and in
two rooms be adjacent because they share common services, for example, the theory of organisations to show hierarchies and the structure of
plumbing; or that two rooms not be adjacent, for reasons of sound command or responsibility in businesses or institutions.
insulation perhaps, or privacy. We shall look more closely at some of This said, however, graphs are especially well-suited to the description
these functional and planning matters in chapter 10. of spatial configurations, and many of the mathematical puzzles and
The adjacency graph of a plan, then, captures important topological scientific problems from which the subject of graph theory sprang, have
features of the relationship between rooms- topology being that branch of been concerned with the structure of two-dimensional or three-dimensional
mathematics which deals with the properties of spaces as they form arrangements in space. The subject is agreed to have begun with a paper
connected pieces and have boundaries, independent of their size and shape. by Euler ( 1736) (from which the quotation at the head of this chapter is
Continuity, connectedness, and adjacency are thus all properties of drawn), in which he described a problem to do with the plan of the city
topological interest. o~ Konigsberg and its bridges. Since then the connection of graph theory
We can make use of some of the techniques and findings of the theory with the.study of the general properties of maps has been a continuing
of graphs in representing and manipulating plan arrangements. In fact one; notably through the infamous and only recently proven ' four-colour
there is now a sizeable body of work on the application of graph theory co~jecture'. (The conjecture concerns the printing of maps, such that
to architecture. It is beyond the scope of this and the next chapter to adJacent regions are differently coloured. Will a maximum of four colours
review all that work in detail. So I will try here to give a general account suffice for any map? The answer is yes, it transpires.)
of the most relevant ideas and results, together with sufficient references Another early application was to chemistry, in notations for the
to lead the reader into the literature of graph theory in general, and arrangement of atoms in molecules. It is from this source that the term
applications to architecture in particular. [For general accounts of 'graph' itself- from 'chemicograph'- originates; whereas graph theory
architectural applications see March and Steadman (1971, chapters I 0 ?orrowed the chemical idea of valency in return. The valency of a vertex
and 11), Steadman (1973), Mitchell (1977, chapter 6), and Earl and March m ~ graph (sometimes called its 'degree' or 'order') is the number of edges
( 1979). For introductions to graph theory see Ore ( 1963) and Preparata wluch meet at that vertex. Yet a further application of graphs has been in
and Yeh (1973, chapter 2). For a more advanced and comprehensive text the description of networks, for example, traffic networks or electrical
with useful appendices of diagrams see Harary (1969). An entertaining and circui~s. Methods and results from several of these areas- specifically
discursive historical account of the subject is given in Biggs et al (1976). I electncal networks and the study of maps- have relevance, as will emerge,
For the present purpose we will need some basic definitions and to the morphology of architectural plans.
explanations of concepts and terminology. A point in a graph is termed a Graphs have been used therefore to represent a variety of spatial
vertex, and a line joining two vertices is an edge(B). It is of no importance structures; but we should still keep it in mind that the graph is not a
picture of some such structure, in the conventional sense. We use a graph
(S) There are many variations in terminology here. Harary ( 1969), one of the principal
to represent the adjacencies between rooms in a plan; and in this case we
authorities, uses the words 'point' and 'line' themselves. Since we are using graphs in
some cases here, however, to represent geometric arrangements of lines (that is, plans), could, for clarity in drawing the graph, place the vertices in the centres of
it seems less confusing to refer to 'vertices' and 'edges' in the corresponding graphs. the rooms, and draw the edges between them as lines crossing the
64 Chapter 6 Graphs of plans and their arrangement 65
respective partition walls; adjacencies to the exterior of all rooms around the periphery of the plan:
1x1
.., therefore a face. And the infinite region around the outside of the plan is
Let us now examine the numbers of vertices, edges, and faces in both
graphs in the above example. In the plan graph there are six vertices, nine
c d b c
edges, and five faces (including the e?(teJior face). In the adjacency graph
And it would be equally possible (though probably less useful) to draw a there are five vertices, nine edges, and~faces (again counting the
graph- just to make this point- in which each edge had the opposite exterior face). The number of edges is the same in both cases; we see
meaning: that 'rooms r; and r; are not adjacent'. A 'nonadjacenc y that this must be so, in the nature of the original definition of the graphs.
graph' of this kind, for the same plan, is: Each wall segment or edge in the plan graph separates two faces- either
two rooms, or a room and the exterior region- and each corresponding
edge in the adjacency graph represents the adjacency of those same two
faces. As they are drawn in the figure, each edge in one graph crosses the
corresponding edge in the other graph.
Furthermore , to every face in one graph there corresponds a vertex in the
There are yet other possible graphs of buildings, in which the edges other. For faces in the plan graph this property follows directly from the
(and vertices) have different meanings again. Suppose, for example, we way in which we produced the adjacency graph in the first place, by
represent the junctions between walls by vertices; and for each wall . assigning a vertex to every room (or region) in the plan. But the converse
segment running between two given junctions, we draw an edge. Keeping is also true, and there is a face in the adjacency graph for every vertex in
in mind the strictly symbolic character of the graph- each edge is not a the plan graph. Each vertex here represents a junction between wall
drawing of a wall segment, but rather has the meaning that 'a wall segments; and the corresponding face in the adjacency graph is bounded
segment runs between junctions V; and v;'- nevertheless the result is a by edges representing the adjacencies across all wall segments which meet
kind of diagrammatic version of the plan itself. We can call it the plan at that junction. Graphs related in this way are said to be duals ; and in
graph: the architectural context the combination of plan and adjacency graphs
has been termed the dual graph representation.
In this example we used one vertex to denote the whole of the exterior
face in the plan graph. Notice, however, that the dual relationship still
holds, if we divide the space outside the plan into several separate regions.
For a plan of overall rectangular shape, it may be convenient for example
(Notice that the four outer corners of the plan are not counted here as to distinguish four external regions, on the 'north', 'east', 'south', and
junctions.) . 'west' sides (as in Flemming's wall representations). In functional terms
A plan graph and its corresponding adjacency graph bear a spe_c1al this makes it possible to represent the orientation of rooms in relation to
relationship to each other. This relationship holds, however, only 1f we are sunlight or views, say ; or to show adjacency relationships to particular
careful to take account of the exterior region around the plan. We include areas around the plan such as the street front, a garden, other neighbouring
one more vertex o in the plan graph to represent the whole of this outside buildings, and so on.
space; and we now need edges in the adjacency graph to indicate the
Graphs of plans and their arrangement 67
66 Chapter 6
We can represent the exterior regions by adding four more infinite faces on this fact. They involve setting the reader, in effect, the impossible task
of drawing nonplanar graphs without crossings.
in the plan graph (figure 6.1 ). These faces are separated by four infinite
edges attached to vertices at the four outer corners of the plan. We now One such puzzle is the 'utilities' problem, otherwise the puzzle of the
houses and wells. In the latter version nine paths are to be drawn,
need appropriate edges in the adjacency graph to show both adjacencies of
connecting each of three houses to all of three wells, in such a way that
the rooms to these regions, and the adjacencies between the four regions
no paths intersect. The corresponding (nonplanar) graph is:
themselves. This type of adjacency graph for a rectangular plan has been
termed its augmented dual (Earl and March, 1979).
In some circumstances in the architectural context it is useful to draw
an adjacency graph in which only the interior adjacencies between rooms
in a plan are included, and adjacencies to the exterior are ignored (as was
the case in the figure at the beginning of the chapter). Here then there is
a vertex in the adjacency graph for every room as before, an edge for It is known as K 3, 3 , where the subscripts signify that three vertices are
every internal wall segment, and a face for every interior junction between each connected to all of three other vertices.
those wall segments. However, the full dual relationship between plan A second problem, discussed by Mobius (see Biggs et al, 1976), involves
graph and adjacency graph is not now preserved. Earl and March (1979) dividing an inherited estate between five brothers such that each holding is
call this type of adjacency graph the weak dual of the plan graph. adjacent to all four others; that is, a plan or map is required for which
It is irrelevant to the theory, as mentioned, if graphs are drawn with the adjacency graph has five vertices, with edges joining all pairs of vertices.
their edges crossing. It naturally tends to be clearer if they are shown in This graph, also nonplanar, and referred to as K 5 , is as follows:
such a way that edges do not cross; and graphs which it is possible to
draw without crossings-that is, to embed in the plane- are called planar
graphs. A graph which has been embedded in the plane is called a plane
graph. It follows that any plan graph of the kind we have been examining
must be plane, since the edges represent wall segments, and the vertices,
junctions between those wall segments. There is no way in which it would
make sense to speak of walls crossing each other, except at a junction.
There do, however, exist nonplanar graphs: no matter how their vertices In general, graphs in which edges join all pairs of vertices, like K 5 , are
are disposed on the plane of the page, or how circuitously the edges are called complete graphs. All of the complete graphs KP for values of p up
drawn, there will still necessarily be edges which cross somewhere. Certain to 5 are:
traditional puzzles from the literature of recreational mathematics depend
0>---0
w e Notice that the complete graphs for p less than 5 are all planar.
The two graphs K 5 and K 3, 3 have a special significance in the
characterisation of planarity. To explain this, some further definitions are
required. Two graphs are said to be homeomorphic ('of a similar form') if
one can be obtained from the other through a process of subdivision of
edges. This subdivision is effected by a series of progressive substitutions
Figure 6.1. A plan graph (thin line) and its augmented dual adjacency graph (heavy
line). Four exterior regions, n, e, s, w are separated by the four infinite edges attached
to the vertices at the corners of the plan. Adjacencies of rooms to these regions, and
of the regions to each other, are included in the adjacency graph.
68 Chapter 6 Graphs of plans and their arrangement 69
of one edge by a pair of edges plus a vertex: The architectural implication of the fact that Ks is nonplanar is that in
no plan is it possible to have five rooms all mutually adjacent. Four is the
greatest number of rooms (whatever their shape, not just rectangular) for
which this is possible. This follows since all complete graphs on more
than five vertices contain K 5 as a subgraph; hence they too are nonplanar.
So far I have illustrated adjacency and plan graphs exclusively in relation
to rectangular arrangements. It should be appreciated, however, that in
neither type of graph is the fact of rectangularity of the plan represented
The graphs illustrated are all homeomorphs of K 5 • (It is as though 'extra' as such. Figure 6. 2 shows four plans, for example, one of curvilinear form,
vertices were introduced along the existing edges.) one with a 45° and 90° geometry, one with 'hexagonal' (60° and 120°)
A graph is a subgraph of a second graph, if it has all of its vertices and geometry, and one of rectangular form. If the labelling of the vertices is
edges in that graph. (The relationship is thus something like that of a ignored for the moment, all four plan graphs can be said to be the same,
subset to a set.) As an example, in the following figure the two graphs on in that there is the same number of vertices in all cases, and these vertices
the right are subgraphs of the graph on the left: are always connected together in pairs in the same arrangement by the
same number of edges. The graphs can be matched or 'mapped' one to
another in exact correspondence; in technical terms they are said to be
isomorphic ('of the same form'). The dual adjacency graphs of all four
plans are also isomorphic (figure 6.3). (In the example of figure 6.2 the
plan graphs can all furthermore be said to be embedded in the plane 'in
the same way', in a sense which we will come to examine shortly.)
We are now equipped to state a theorem due to Kuratowski ( 1930) which If, however, the labelling of the vertices is taken account of, we see that
defines the conditiohs for planarity in graphs: "A graph is planar if and
all four plan graphs are not now precisely the same; we can match edges,
only if it does not contain any subgraph homeomorphic to K3,3 or Ks"· vertices, and labels only for the first, third and fourth graphs, but not any
So the presence of one or other (or both) of these types of subgraph is of these with the second graph. When speaking of isomorphism it is
sufficient to render a graph nonplanar. And conversely, if a test shows important to be specific therefore as to whether the graphs are taken as
that neither is present in a graph, then the graph is planar. labelled or unlabelled.
Now every plan graph has a planar dual: and no nonplanar graph can If an adjacency graph is to correspond to a plan taking the form of a
have a dual which is planar. We already know that any plan graph must rectangular dissection say, then there are certain limitations which this
be plane. It follows that any adjacency graph must in turn be planar, if it will impose in turn on the form of the graph itself. Later we will come to
is to correspond to a plan. Or, looked at in another way, if a series of look at the precise nature of these limitations.
adjacencies are specified between rooms or regions, which together form Let us suppose for the moment, however, that we will allow plans
an adjacency graph which is nonplanar (as in the problem of Mobius) then without 'geometrical' constraints of any kind; that is to say, we place no
those adjacencies cannot all be realised in any actual plan.
a
g h b
C
k
Figure 6.3. The adjacency graph which is the same for all the four plans of figure 6.2
Figure 6.2. Four plans of differing geometry, but with the same (unlabelled) plan graph. (adjacencies to a single exterior region are included).
70 Chapter 6 Graphs of plans and their arrangement 71
limits on the angles between walls in a plan, these walls can be curved or I do not want to suggest that very many (if any) architects actually do
straight, and the rooms, and the exterior perimeter of the plan, may ~ake design in this way. Most work directly on the geometry of the plan itself.
curvilinear or polygonal shapes of any kind. Such plans can be conveniently Nor is it true, of course, that adjacency of spaces is the only constraint on
drawn as packings of 'bubbles': the planning of buildings. But for the purposes of the present argument it
is convenient to assume this rather simplified or idealised logical sequence
of operations. (Amongst other reasons, this will throw light on issues
arising in several of the graph-based design methods, to be discussed in
chapter 9.)
The first point to notice is that this adjacency requirement graph is
rather unlikely to be isomorphic to the adjacency graph of the final plan
or plans. On the one hand, it is improbable that the architect will wish to
specify the entire adjacency graph of the plan in advance. His requirements
[which is not to say that real bubbles do not have very definite geometries
will generally be confined to those instances where rooms must be
(Almgren and Taylor, 1976)), and these bubbles, it must be imagined, can
contiguous for definite functional reasons; most often so that direct
be squeezed and stret ched as desired. access is possible via a door.
It is possible to imagine a process of design in which the architect But in any actual plan there will be adjacencies between rooms occurring
would start with some set of spaces, together with some requirements as more or less fortuitously, which are not specifically required on functional
to which of these spaces should ideally be contiguous in a plan. He could or aesthetic grounds, but arise simply out .of the packing of spaces into a
represent the spaces by vertices, and the requirements for adjac:~c~ compact whole. (There are economic reasons for plans to be compact
between rooms and to the exterior region or regions, by edges Jommg the certainly, including the reduction of constructional and heating costs; but
relevant vertices. Together these would form what we might call an these do not necessitate the particular adjacencies of specified pairs of
adjacency requirement graph. A hypothetical example for the ground . rooms.) Here then, for any actual plan in which all the requirements are
floor of a small house is illustrated in figure 6.4. (It is assumed that this satisfied, the adjacency requirement graph will be a subgraph of the
plan is four-sided and that the sides are oriented to points _o~ the compass adjacency graph. An example for our house plan of figure 6.4 is illustrated
n, e, s, and w. There are four rooms: a living room 1, a dmmg room d, a in figure 6.5.
kitchen k, and a hall h.) The designer would then want to know what On the other hand, there is the possibility, as we have noted, that an
range of possible plan graphs- and ultimately plans- would correspond to, architect might unknowingly stipulate a set of adjacency requirements
or satisfy, these stated requirements. which together turn out to constitute a nonplanar graph. In a graph-
n
theoretic design method following this line of approach, it would therefore
I d
0 0
n
d k /
0 0
I
'' /
/
h
0 0 h "
h k w e
0 0 0 0
h n
0 0
..
I
0 0
s '•·• ....
k s
0 0
/
/
/
''
Figure 6.4. Adjacency requirements (left) for a small house plan with_ four rooms:
'
a living room 1, a dining room d, a kitchen k, and a hall h. The plan 1s assumed to be Figure 6.5. A plan (thin line) satisfying the adjacency requirements of figure 6.4. The
four-sided: adjacencies are specified to exterior regions at n_orth n and sout~ s. These adjacency requirement graph (heavy solid line) is a spanning subgraph of the adjacency
requirements are then put together into the adjacency requirement graph (nght). graph (solid plus dotted lines). It contains all the vertices, but only some of the edges
of the adjacency graph.
72 Chapter 6 Graphs of plans and their arrangement 73
be wise to test any requirement graph for planarity before attempting to would mean that the architect specifies all of the rooms in a plan, but not
proceed further. Figure 6.6 shows how the addition of only one more all of their adjacencies.
requirement for adjacency to the graph of figure 6.4 serves to make it Even this would not necessarily be the case. It might for example be
nonplanar (given that edges ne, es, sw, and wn are included). The that the architect would want to introduce new circulation spaces such as
requirement graph now contains K 3 , 3 as a subgraph. halls, lobbies, or corridors, during the design process. These spaces might
Several techniques have been formulated as algorithms, and hence as not be strictly required in themselves, but are incorporated to meet certain
computer programs, by which to determine whether or not a given graph needs, say, for indirect access between other rooms. Thus in a graph-
9
is planar (Shirey, 1969) (although they are mostly rather complicated < >). theoretic representation, vertices would be added to the requirement graph
If the graph failed such a test, then some of the requirements would have and edges also added or rearranged (even removed), to produce the final
to be abandoned, that is, an edge or edges would have to be removed from adjacency graph (as illustrated in figure 6. 7).
the graph, before it could correspond to the adjacency graph of any plan. There are several other possible complications. In some instances the
(There is the further possibility, if additional constraints are placed architect might be prepared to allow a series of alternative adjacencies:
subsequently on the geometry of the plan or on the shapes and sizes of either room a must be next to room b, or else it must be next to room c.
rooms, that even the requirements in a planar adjacency requirement graph This simply means that there will be two, or more, equally acceptable
may not be capable of satisfaction simultaneously with these other requirement graphs. If it is stipulated that certain rooms must not be
constraints; in which case the architect must make a choice of which type
n n
of requirement to retain and which to relax.)
Let us assume, however, that there exists at least one feasible plan, and
that its adjacency graph contains the adjacency requirement graph as a
subgraph; this could be one of several different kinds of subgraph. If
h
a subgraph contains all of the vertices, but only some of the edges of a k
➔ k
graph, it is said to be a spanning subgraph. This is the relationship of the
two graphs in figures 6.4 and 6.5. In the architectural interpretation, this
..····•
.....•.... ··....
.·· ....
..······ ···•.... Figure 6.7. Example of a new vertex added to an adjacency requirement graph, to
.·· represent a circulation space h, with an accompanying rearrangement of edges.
wo:.
···...__ ...:oe
....
..··
..
...
·•
..
····...
···-.
....···········.... ./~
s h d
Figure 6.6. One more edge dn added to the adjacency requirement graph of figure 6.4
serves to make it non planar (given that the edges ne, es, sw, and wn are included). The
graph is redrawn in a way which makes clear that it contains K 3, 3 as a subgraph.
adjacent, then effectively a second 'nonadjacency requirement graph' is Several authors have referred to a slightly different type of graph which
needed. also has its uses in design methods and in morphological studies (Levin,
In general we can say of plan graphs that they must be connected. 1964; Hillier et al, 1978a). This is a graph in which the vertices signify
A connected graph is a graph in which all pairs of vertices are joined by the rooms of a plan, or exterior regions around a plan, as before, but in
continuous sequences of edges and vertices (paths); that is, the graph which each edge signifies the existence of a door or means of access
is in one piece and does not fall into two or more separate components. between the two rooms or regions in question. lt'may be called the
A disconnected plan graph could only signify the plans of two or more access graph. Figure 6.9 shows an imagined access graph for the plan
separate buildings [figure 6.8(a)]< 10>. illustrated in figure 6.5. Clearly the access graph must always be a
The adjacency graph of any plan must also be connected. [In fact even spanning subgraph of the adjacency graph of a plan and will generally be a
if the plan graph is not connected, its adjacency graph ~ill be,_ since all connected spanning subgraph. (Unless some secret rooms are completely
the separate 'plans' must be adjacent to the single exterior region, and the walled in, or there is for example an inaccessible light-well.)
corresponding edges will serve to conn~ct the entire gr~ph together-:- We could even define an 'access requirement graph' which would be a
figure 6.8(b).] But this is not necessanly true of an adJacency reqmrement subgraph of the more general adjacency requirement graph- there being
graph. requirements for adjacency which might be stipulated for reasons other
Consider, for example, the requirement graph of the ground floor of a than for access, such as adjacencies of rooms to the exterior to allow for
block of flats, in which separate access is to be provided to each flat windows.
direct from the exterior (not from a common hall). We might suppose
that the architect would set a series of requirements for the adjacencies of
rooms within each flat, and that he would want to pack all the flats Exercises
together into the single block; but that he would be largely indifferent as 6.1 For the plans illustrated (or any other plans with rectangular
to which rooms of different flats happened to be adjacent across party boundaries that you care to choose), draw in each case
walls. (Although in this particular example we can imagine that an (a) the plan graph,
architect might actually want to specify that certain rooms of different (b) the weak dual adjacency graph,
flats should not be adjacent across the party walls- for example, that a (c) the augmented dual adjacency graph, and
bedroom in one flat not be next to the living room in another.) (d) the access graph.
n Check in each instance that the dual relationship exists between (a) and (c).
(You will need to introduce infinite edges at the corners of the plan graph,
to divide the exterior into four regions, and show edges in the adjacency
graph, for adjacencies between these regions.)
Figure 6.9. The access graph of a plan , in which vertices r~present rooms or exterior
regions, and edges signify the existence of doors or other direct means of access
between those rooms or regions. The access graph must be a spannmg subgraph of the
adjacency graph of a plan.
6. 2 Draw all graphs on four vertices, including those which are not
Although, for example, the plan at ground level of a triumphal arch or any
(JO) connected. (There should be eleven of them. You may find it convenient
comparable bridge-like structure would have a disconnected plan graph. to classify them by their numbers of edges. It is possible to have a graph
Chapter 6 Graphs of plans and their arrangement 77
76
with no edges-just the four isolated vertices. At a maximum there can be 6. 6 Draw all distinct possible spanning subgraphs of the graph illustrated.
six edges, connecting all pairs of vertices, in the complete graph.) By 'distinct' is meant subgraphs which are not isomorphic. How many
pos~ibili~ie~ are there if, in determining isomorphism, (a) the labelling of
6. 3 Which of the graphs on six vertices illustrated is planar, and which vertices 1s ignored and (b) the labelling of vertices is taken account of?
non planar?
aK7f
c~ d
6. 7_ Th~ figure shows the plan of the city of Konigsberg (now Kaliningrad),
which hes on the river Pregel. There are seven bridges crossing the branches
6. 4 Which of the graphs illustrated are isomorphic? (A necessary- but of the river as shown. The problem which Euler ( 1736) addressed was
not sufficient-condition for two graphs to be isomorphic, is that for this: is it possible to take a walk around the city, crossing each bridge
every vertex in one graph, there exists a vertex with equal valency in the once only, and return to one's starting point? Euler showed that the
other graph, to which it can be matched. So a way to start checking for answer is no. Can you show why? Start by drawing a graph in which the
isomorphism is to label vertices with their valencies.) four distinct parts of the city are represented by vertices, and the bridges
by edges. (This is not strictly a graph proper, since there is more than
one edge joining the same vertices in some cases. It is a multigraph - see
chapter 7.) (Hint: consider the valencies of the vertices.)
6. 5 Which of the graphs illustrated are homeomorphic? 6. 8 Draw the weak dual adjacency graphs of all rectangular dissections for
n = 4. Imagine a set of four rooms with distinct functions labelled o: f3
'Y, and 8. Choose some set of required adjacencies between' those roo~s.'
Draw the adjacency requirement graph. In how many of the dissections
can these requirements be satisfied? (That is, in how many cases is the
requirement graph a spanning subgraph of the adjacency graph of the
dissection?)
For each dissection label the rectangles arbitrarily a, b, c, and d. In
every dissection where the same adjacency requirements as before can be
satisfied, ho w many distinct possible permutations of position of the room
functions are there? (In how many different ways, that is, can o:, {3, 'Y,
and 8 be assigned to a, b, c, and d such that the adjacency requirements
are met?)
78 Chapter 6 7
By this process you will be able to determine all possible plans of Embedded, coloured, and weighted graphs of plans
rectangular dissection type, which meet the given requirements-without
regard, of course, to any possible dimensional requirements which might
then be imposed. "Since much of design, particularly architectural and engineering design, is concerned
with objects or arrangements in real Euclidean two-dimensional or three-dimensional
6. 9 Carry out the same exercise as 6.8, but this time draw the augmented space, the representation of space and of things in space, will necessarily be a central
dual adjacency graphs of the dissections, and introduce some set of topic in a science of design."
required adjacencies of the four rooms to the four exterior regions on the Herbert Simon ( 1969)
north, east, south, and west.
You could also elaborate the exercise further by specifying that certain To come back to our imaginary architect of the last chapter, and his
adjacencies, between pairs of rooms or between rooms and certain exterior graph-theoretic method in design. He is working with an adjacency
regions, should not occur. In this case it would be convenient to draw requirement graph (not just requirements for access). Let us suppose that
both a graph of adjacency requirements, and a graph of adjacencies not he has tested this graph for planarity and found that it is planar. The
required- a 'nonadjacency requirement graph'. next problem that arises, is that there may be several distinct ways of
embedding this graph in the plane.
For example, take the following very simple requirement graph on four
vertices, which is drawn six times, but in only two different embeddings:
In the first three drawings the vertex d is outside the triangular face abc, and
in the other three it is inside. The one class of embedding is distinguished
from the other by virtue of this difference; it is as though the edge de
were folded over on the 'joint' represented by the vertex c.
Imagine what this would imply for the plan graph. We will draw the
faces (rooms) as curvilinear 'bubbles' as described in the last chapter. We
assume that all the four vertices in the requirement graph represent rooms
and none of them represents the exterior region. We assume also that no
more rooms will be added to the plan beyond these four. There are then
two distinct plan graphs, corresponding to the two embeddings of the
requirement graph- although exactly the same adjacency requirements are
satisfied in each case:
Room d is required only to touch room c, but it is not required to to~ch exterior region should also be shown:
a or b. There are various alternatives here. It is possible that there might
be a hole- perhaps a courtyard- in the ~Ian (shown sh~d_ed in the_above
figure). If the adjacency graph for this courtyard plan 1s drawn 1t ought
strictly to include a new vertex to represent the courtyard, however, plus
three new edges to represent the adjacencies of a, b, and d to the courtyard:
In the first embedding all four rooms are adjacent to the exterior, whereas
in the second only three are, and the room d is enclosed in the interior of
the plan.
We can look at this subject of embedding in a more general and
systematic way. Let us consider adjacency requirement graphs for the
On the other hand the architect might not want such a courtyard, and
moment which are connected, planar, and in which the n vertices represent
could extend room ct to fill the hole and so make it adjacent to a and b.
rooms in a plan (not the exterior region). We can ask "What are all the
Here then there would be two new adjacencies ad and bd not expressly
possibilities for the embeddings of such graphs, for successive values of n?"
required, but introduced because of the constraints of close-packing:
That is to say, we can take an exhaustive approach to the enumeration of
this class of graphs, rather in the same way that it is possible to take an
exhaustive approach to enumerating possible dissections, or possible
polyominoes. The general problem of counting planar graphs remains
unsolved. (Essentially the difficulty lies not in devising some generating
process, but in recognising and removing isomorphs.) The more particular
problem of counting planar embeddings has , however, been solved. For
small n it is possible to enumerate all cases by inspection. Figure 7 .1 is
adapted from tables published by Korf (1977), and by March and Earl
There are two further, slightly weird ancl unrealistic interpretations (1977), and illustrates possible embeddings of all connected planar graphs
which are possible for the embedded requirement graph: up to n = 5. (The two embeddings of the graph of the figure at the
beginning of the chapter appear in the fourth row of possibilities for
graphs on four vertices.)
The reader is invited to draw the plan graphs corresponding to some of
these cases, so as to gain a feeling for the dual relationships involved.
These plan graphs are technically speaking maps in graph-theoretic terms.
A map is a connected plane (that is, embedded planar) graph together
with all its faces. Obviously in its interpretation as a real, cartographic
map, the faces are countries or regions, and the edges their boundaries;
The first is that room cl is almost completely engulfed inside room c; thus where in architectural terms they are, respectively, rooms and walls.
d touches a and b only at a single point (a five-way junction) which does If it were adjacency requirement graphs which were being considered,
not count as adjacency. The second possibility is that room c is ring- then it would be quite permissible to add further edges and vertices, to
shaped and completely surrounds cl. In either case the plan ha~ no hole, produce adjacency graphs proper. This could be done in each case in very
all the required adjacencies are satisfied, and no other adJacenc1es occur. many different ways. Let us for the moment imagine, however, that these
However these are rather improbable kinds of plan in architectural terms. graphs in figure 7.1 represent as they stand the weak duals of plan graphs.
In the' full adjacency graphs of the plans corresponding to the two
embeddings of the requirement graph, the adjacencies of all rooms to the
Chapter 7 Embedded, coloured, and weighted graphs of plans 83
82
Notice some general features. A face in the embedded adjacency graph graph for n =5 a five-way junction, and so on:
implies a junction in the plan at which as many wall segments come
-$-
together as there are edges surrounding the face. Thus in the simplest
cases the 'triangular' adjacency graph for n = 3 implies that the three
rooms in the plan meet at the centre in a three-way junction, the
'quadrangular' graph for n = 4 implies a four-way junction, the 'pentagonal'
n = 5 (continued)
n=l 11=5
tr, op
A
0
n=2
I tr,op
A~
n=3
bop
AAAA
L ;,op
A~
II =4
Ac&kp
~ AA
~PA ~A
AP A
~LP &~PA
~,op
k~ tr indicates a tree
op indicates an outerplanar embedding
~,op Llp
Many of the graphs in figure 7. 1 represent plans in which either there form of representatio n of possible adjacency relations in architectural
are 'holes', or else interior rooms are engulfed inside others, as explained. plans, than the comparatively limited class of 'graphs' proper.
It should also be noted that there are certain adjacency relationships With this limitatio n in mind, and coming back to the (true) graphs of
which can arise in plans, indeed which occur quite frequently in real figure 7.1: see how the different embeddings are created in each case by the
buildings, but which are not represented by any of these adjacency graphs. ways in which parts of an adjacency graph can be 'hinged' or 'folded' over
It is possible that a room might be adjacent to itself, as for example, when to lie in its different faces (including the exterior face). The possibilities in
it completely surro unds a second room in a ring broken by a partition wall: each case here are determined by the degree of connectedness of the graph.
Wi th the use of the fo llowing example o nce again ; this graph is said to
be ]-connected:
I~
Qaloop d
\~
The room is then adjacent to itself across this partition. In the adjacency
graph the relationship would have to be represented by a loop.
A second possibility is that two rooms might be adjacent across two
l
a b
The removal of one vertex (vertex c in this case) together with all edges
(or more) distinct sections of common wall. Such a situation is where
incident wi th tha t vertex, results in the graph fall ing into two disconnected
two rooms surro und a third in a ring: components (the edge ab and th e isolated vertex d):
d d
Om". .,..,..
0
the cut vertex is at d, and the ve rtex e together with the edge de can be
embedded in any of the triangular faces abd, acd, bed. If the vertices
which multiple edges and loops are allowed, pseudograph s. The problem
were not labelled these three alternatives would be isomorphs, because of
is, for enumerating possible adjacency graphs, that in principle any number
the symmetry of the situation. Suppose the whole of the same graph
of loops or multiple edges may be added to each vertex or pair of already
is reembedded such that b, c, and d lie on the outer boundary, then e
joined vertices, without limit. There is a strong case nevertheless for
can be embedded either in the exterior face, or else in abd or acct
regarding pseudographs, or at least multigraphs, as a more appropriate
86 Chapter 7 Embedded, coloured, and weighted graphs of plans 87
(the last two again being isomorphic if the graph is unlabelled): An exactly symmetrical embedding is achieved by folding over ct, with ad
and cd, so that they lie inside abc:
e
d d
d
b c
~A
b c b C
(The graph is that represented in the top row of graphs on five vertices on
page 83 in figure 7.1, where all these possible embeddings are accounted for.)
If the graph were unlabelled these last two would be the same embedding.
In this case the graph is 2-connected, as it requires the removal of two
Notice that with I -connectedness we are not just referring to cases like vertices (with their incident edges), at a minimum, to disconnect the graph:
those so far illustrated where a single vertex is attached to a larger graph
a
by a single edge. For example, in the following figure:
i:x:i ➔ i x f
b e b c
·<1> C
In this case we have the embedding indicated in the figure in which a, b, Extending the idea of connectedness one step further: a graph in which
c and ct all lie on the exterior face. Or we might fold b together with ab three vertices and their incident edges must be removed to disconnect it,
a~d be over on the pair of hinges a and c, so as to lie inside the face acd. will be termed 3-connected. One of the very simplest of 3-connected
88 Chapter 7 Embedded, coloured, and weighted graphs of plans 89
graphs is The second shows a 2-connected graph, in which the two 'hinge' vertices
b b are common to two 3-connected subgraphs (both of them wheels on five
·$,
vertices).
d
➔ co
$ d
oc
A maximal n-connected subgraph of a graph is callee! an 11-component.
The first graph in the above figure contains two 2-components, and the
second graph two 3-components. A graph might well consist of some
combination of I-component, 2-components, and 3-components.
For reasons which are obvious from the way it is drawn here, it is called a To find all embeddings of a given planar graph involves then essentially
wheel. This is the wheel on five vertices. See how the removal of any isolating each 3-component (if any exist) and taking each of its faces in
two vertices leaves the graph still connected. Only the removal of three turn as the exterior face (that is to say, exterior to that component
considered in isolation). It is further possible that one entire 3-component
vertices, with their incident edges, is sufficient to break it into two parts:
might be capable of being, or might be req uired to be embedded inside
two isolated vertices.
one face, or in one of several alternative faces of another embedded
With this, as with all other 3-connected graphs, no possibility exists for
3-component- the possibility depending on the patterns of connectivity
'folding over' or 'pivoting' parts of the graph in the embedding, as was the
between these components. All permutations of possibility here must be
case with the I-connected and 2-connected examples. The only alternatives
accounted for. Then any 2-components must be folded about their 'hinge'
available for distinct embeddings are those presented by the choice of
vertices into different faces, and (where applicable) mirrored in all those
different faces to form the exterior face. Given that the wheel of the last
faces, in all distinct possible ways. Finally, every I-component must be
figure is labelled, there are five embeddings, in which either the single
embedded in each of the faces on which its 'hinge' or cut vertex lies.
quadrilateral face, or else each of the triangular faces in turn, becomes the
In many cases this might be a· matter of 'hinges on hinges', or 'folds
exterior face :
within folds' , so to speak. Thus in the following (relatively simple)
example:
·EB,, A A A A
d b cc d d e e b
Once again these last four embeddings are distinguishable only by virtue of
the labelling. a I-component hinges on a 2-component: and we must consider all
If we imagine the graph to be made of a net of elastic strings, it is as permutations of embedding of the one with the other. Which of the
though we take each loop of string surrounding a single hole in that net resulting embeddings in the general case should be regarded as distinct
(a face) , stretch it out, and lay it down to become the outer edge so that depends, first , on whether the vertices are labelled or not, and, second, if
the remainder of the ne t falls within that loop. the graph is unlabelled, on the symmetries of the situation. Considering
It can then be seen that the general problem of determining the possible embeddings of entire graphs; there may exist pairs of embeddings where
embeddings of any given planar connected graph is liable to be a complicated one is a symmetrical reflection of the other, and which differ only by
business (compare Korf, 1977). A graph as a whole might be I-connected ; virtue of that reflected symmetry. Whether these are regarded as distinct
but it could contain subgraphs which in themselves are 2-connected or or not is a matter of convention.
3-connected. An example of a graph which is I-connected , but where the In general, there must be at least as many distinct planar embeddings of
cut vertex is common to two subgraphs which are themselves 2-connected, a labelled graph, as there are faces in it. In practical architectural terms
is the first of the choice of a face in a (weak dual) adjacency graph to become the
exterior face implies that all those rooms represented by the vertices
belonging to that face- and only those rooms- will lie on the outside
perimeter of the plan. They will form a continuous 'ring' around all other
rooms, which will thus be completely internal and cannot have side-
windows or direct access from the outside. With the graph of the figure
90 Chapter 7 Embedded, coloured, and weighted graphs of plans 91
>-<
adding edges (and possibly new vertices) to produce an adjacency graph
from which the dual plan graph can be derived, can become a very
laborious and elaborate one. This complexity is reflected in several of the
design methods which are described in chapter 9.
A cycle in a graph is a continuous sequence of vertices and edges,
containing at least three vertices, which does not double back on or Primary plans and their adjacency graphs
intersect itself, and which returns to the original vertex. A tree is thus However, there is a converse way of looking at the whole subject, which
necessarily I-connected , and when embedded has only one face, the avoids much of the messiness created by the problem of alternative
exterior face. All of its vertices lie on this face. Notice that it is embeddings. Imagine an adjacency graph to which no more edges can be
nevertheless possible for a tree to have several distinct embeddings, since added in any position without rendering the graph nonplanar. Such a
it may be that its various 'branches' are capable of being arranged in graph is said to be maximal planar.
different order around a common vertex, in a manner similar to that It is the dual of a plan graph in which the adjacencies between rooms
illustrated in the third figure on page 86. All trees on n vertices up to n = 5 (including adjacencies to the exterior region) are at a maximum. It is
appear in their appropriate positions in the enumeration of embeddings in only possible to reduce the total number of adjacencies from this level,
figure 7 .1 , and are labelled there as such. through operations which effectively involve the removal or repositioning
Besides trees, there are other graphs which can also be embedded such of edges in the adjacency graph.
that all vertices again lie on the exterior face. All of these (including March and Earl ( 1977) have suggested that the set of plans- by which
trees) are known as outerplanar graphs. An example of an outerplanar they mean distinct embeddings of plan graphs- in which rooms are
graph containing cycles is maximally adjacent in this way, be considered as primary . A II other plans
can then be obtained from the primary plans by processes of what they
term ornamentation. Through ornamentation, either adjacencies between
rooms or regions are lost, or else a room which is adjacent to another (or
the exterior) across one wall segment may become adjacent to the same
room or region across a second wall segment, that is, multiple edges are
introduced into the adjacency graph._
92 Chapter 7 Embedded, coloured, and weighted graphs of plans 93
In a maximal plane adjacency graph (that is, an embedded maximal as they are labelled), and two three-way junctions are 'coalesced' into a
planar adjacency graph) every face is a triangle. This means that in the single four-way junction.
corresponding primary plan every junction between wall segments is three- This four-way junction in the plan graph may then be 'expanded' again
way. Junctions in which more than three wall segments meet together into two three-way junctions by the reintroduction of a new wall which
imply the Joss of adjacencies, by comparison with the primary plan. Such connects the plan in a different way:
junctions can be created by 'ornamentation', as we shall see.
The simplest primary plan of any interest is that with two rooms:
~ 0
V 0
The adjacency graph includes a third vertex for the exterior region, and
together the adjacencies make a triangle. There is no possibility of a
The centre room (labelled b) is now adjacent to the exterior across two
walls, and this implies two edges joining the relevant pair of vertices in the
adjacency graph. The two successive ornamentation operations described
account for all possible plans (besides the primary plan) with three rooms
graph on three vertices becoming nonplanar of course. But this triangle- (excluding, that is, plans in which multiple adjacencies occur between
the complete graph on three vertices- represents the maximal number of rooms , or in which rooms are adjacent to themselves- all of which could
adjacencies possible. (It simultaneously represents the minimum number be created by further ornamentation).
of adjacencies in this case, since the adjacencies of the rooms to the When we come to four rooms, there are two primary plans:
exterior region must occur in the nature of the situation, and if the
adjacency between the two rooms were not to occur, then the plan would
be disconnected- which is not allowed.)
The (only possible) primary plan for three rooms is: b o
~ ~ 0
w a
The interesting thing is that these share the same embedding of the same
O 0 adjacency graph. (It only has this one embedding.) The difference in the
The rooms are all adjacent to each other and to the exterior region, and plans arises out of the choice of vertex to represent the exterior region. It
the adjacency graph is the complete graph on four vertices this time, may be a vertex with valency three or with valency four, in which case
which when embedded has four triangular faces (including the exterior either three rooms or four rooms are on the exterior of the plan as shown.
face). The following shows one possibility for ornamentation of this See how the addition of one more edge to this adjacency graph (in the
primary plan, with a consequent reduction in the number of adjacencies: only available position, joining the two vertices of valency three) would
turn it into K 5 ; thus the graph must be maximal planar:
0 0
~)➔ @
If one interior edge (or wall segment) in the plan graph is removed, as
indicated in the figure, the adjacency is lost between two rooms (a and c
94 Chapter 7 Embedded, coloured, and weighted graphs of plans 95
Figure 7 .2. Primary plans with up to six rooms (left) and their adjacency graphs Figure 7.2 (continued)
(right) (from March and Earl , 1977). Each adjacency graph is shown in all its possible may give rise to different plans; and conversely the same plan may be produced by
embeddings. Notice that the correspondence between plans and graphs is not one-to- suitable choices of vertex for the exterior region in different adjacency graphs. Plans
one , since the choice of vertex in the adjacency graph to represent the exterior region marked • have distinct isomorphs by reflection. Plans marked t have no internal rooms.
96 Chapter 7 Embedded, coloured, and weighted graphs of plans 97
The next figure shows the primary plan with four rooms adjacent to the Again, since these adjacency graphs are maximal planar, they correspond
exterior: as described to plans with the highest degree of adjacency, the greatest
potential connectivity between rooms which is topologically feasible.
There is no point in an architect or his client idly setting 'requirements'
for adjacency in excess of these limits, since they can never be satisfied in
➔ any actual plan. Seppanen and Moore ( 1970) remark on the fact that as
larger primary plans are considered, the average face (that is, room)
approaches to being six-sided. (It is adjacent to six others.) This does not
mean, of course, that every face must be hexagonal; rather that if the
number of edges in a face is averaged over all faces, then the resulting
value tends to six. This is a fact which will have significance for some
Ornamentation by the removal of one edge as shown has the result of empirical studies of actual building plans- and other spatial patterns-
making the four rooms meet at a single four-way junction. The adjacency which we will come to in later chapters.
graph correspondingly possesses one less edge than the adjacency graph of As a matter of fact even the primary plans can, in turn, be related to a
the primary plans, and has become the wheel on five vertices. still smaller number of yet more fundamental 'objects', as March and Earl
Figure 7 .2 is taken from March and Earl (1977, figures 2 and 3) and demonstrate. So far we have considered the embedding of graphs in the
shows all primary plans together with their adjacency graphs, up to six plane, since it is in this form that they can be interpreted as architectural
rooms. The figure is arranged to show on each row the various possible plans. It is quite possible, however, to embed graphs in other kinds of
embeddings of each adjacency graph, and opposite these the corresponding surfaces, and specifically of interest here, to embed them on the sphere.
possibilities for plans. The correspondence is not one-to-one, since the We can envisage this as being like tightening the graph, like a string bag,
choice of vertices in the adjacency graphs to represent the exterior region over the spherical surface.
may give rise to different plans; and conversely the same plan may be It has been shown by Whitney ( 1932) that every 3-connected planar
given by suitable choices of vertices for the exterior region in different graph has a unique embedding on the sphere. The relation of the
adjacency graphs. embedding of such a graph on the sphere, to its possible embeddings in
What is the significance of considering only maximal planar adjacency the plane, can be imagined in the following way. The sphere is placed
graphs in this context? First, there are rather few of them, compared above the plane, and then the vertices and edges are projected down, from
with planar graphs generally. Furthermore it can be shown that every the spherical surface onto the plane surface (figure 7 .3). Perhaps the
maximal planar graph with four or more vertices is 3-connected (Harary, easiest way to think of this, is to imagine the sphere to be transparent,
1969). This means that the task of enumerating possible embeddings of and that a point source of light is placed above the sphere, shining down
such graphs is a relatively straightforward one. Every such graph has at a
maximum only as many distinct embeddings as it has faces - each of these
faces in turn being taken as the exterior face, as explained - and if
unlabelled it may very well have fewer embeddings, as a result of symmetries.
These facts are illustrated in figure 7 .2, where the numbers of distinct
primary plans for n = 3, 4, S, and 6, and their corresponding embedded
adjacency graphs are evidently few in number, and growing with increasing
n at a rather gentle rate. Compare, for example, the single maximal planar
adjacency graph on five vertices in this table, with the fifty embedded
planar connected adjacency graphs (all of them spanning subgraphs of the
maximal planar graph) illustrated in figure 7. 1<11 >.
(ll) The graphs in figure 7.2 include adjacencies to the exterior region; whereas we
were taking those in figure 7.1 to be the weak duals, without the exterior represented.
Nevertheless we can make this direct comparison if we regard one vertex in each of the
graphs of figure 7.1 to represent the exterior, instead of a room. Figure 7.3. A graph embedded on a sphere, and its projection onto the plane.
98 Chapter 7 Embedded, coloured, and weighted graphs of plans 99
on the graph as it lies on the sphere and casting its shadow onto the plane. , only nine, all of them illustrated in figure 7.4. March and Earl refer to
Less fancifully, it is the kind of stereographic projection which is employed them in this architectural interpretation, as fundamental plans.
in map-making (from the curved surface of the earth onto the flat sheet of March and Earl themselves mention how many writers on architecture
the map), or which underlies perspective drawing. and the arts, from Diirer to Buckminster Fuller, have been fascinated by
The graph as it is embedded on the sphere, may be projected down in polyhedral forms and their properties, and have held them out as perfect
different ways through each of its faces in turn. Thus the various possible models of order and spatial organisation in design .• Any connection with
embeddings of a 3-connected graph in the plane represented by taking the actual design of buildings has often seemed more of a metaphorical
each face in turn as the exterior face, correspond to the possible ways of one, however, until this work of March and Earl in which a mathematical
projecting the single embedding on the sphere, through each of its faces. mapping, albeit a many-stage and in some cases complex one, is established
Similarly, the dual of an adjacency graph, that is the corresponding plan between the class of trivalent polyhedra, and all possible architectural plans.
graph itself, can be conceived of as a graph embedded on the sphere, and It is known that, starting from the smallest trivalent polyhedron, the
projected down through each of its faces (rooms or regions) so as to give tetrahedron, all others with larger numbers of faces can be generated by
all the possible primary plans. combinations of three and only three operations (Eberhard, 1891 ). These
The embedding of the graph of the plan on the sphere can in its turn operations can be likened to a process of shaving off parts of the solid
be imagined as the skeleton of edges of a solid polyhedral figure. (It is as with a carpenter's plane: planing off one vertex to produce a new
though each curved face of the graph on the sphere were pressed flat.) triangular face, planing off two vertices and an edge to produce a new
Every vertex of this solid must be trivalent, that is, it is the meeting point quadrilateral face, or planing off two edges and three vertices to give a
of three faces. In this way March and Earl establish a correspondence new pentagonal face:
between a class of solid forms- trivalent polyhedra- and primary plans.
Because to each polyhedron there corresponds, in general, several primary
plans, the number of polyhedra is smaller than the number of plans. The
total number of such polyhedra with between four and seven faces, is
®@@@~
Figure 7.4. All trivalent polyhedra with up to seven faces (from March and Earl, 1977).
subject as a whole. From a practical or design standpoint nevertheless,
where the purpose is to arrive at some particular plan conforming to some
particular set of adjacency constraints, it should be appreciated that the
complications which we looked at earlier, in the possibilities for adding
edges to and embedding an adjacency graph, have by no means disappeared
here, but have simply been transferred into the very extensive and
These can be put into one-to-one correspondence with what March and Earl call
fundamental architectural plans.
potentially complex possibilities for 'ornamentation' of the primary plans.
100 Chapter 7 Embedded, coloured, and weighted graphs of plans 101
Coloured graphs of plans The next figure illustrates four dissections for n = 4, all of which have
Up to this point we have been looking at plans considered as graphs, the same weak dual:
without any constraint placed on the shapes and sizes of rooms. In a
a b
graph proper- either an adjacency graph or a plan graph- it is only
relations of a topological nature which are expressed. However, in a
~ ~ ~lLl1[l K7
~~~~ H
graph-theoretic approach to design, our imaginary architect would want to
go on from a plan graph, to a geometric layout with specific dimensions.
There are ways in which graphs can be coloured, and values or weights
added to their edges and vertices, so as to impose on the basic topological In this case we can see that the relationship of the rooms to the exterior
model some further representation of these geometrical and dimensional regions on the four sides of the plan are all different. If appropriate
aspects. vertices and edges are added to complete augmented dual graphs, then
A first question is "What limitations are placed on plan graphs and these differences are made manifest, and the four graphs become distinct
their embeddings, if any, by the requirement that they be capable of (figure 7 .5). (It is often difficult to determine by inspection whether two
transformation into plans with rectangular geometry?" In particular, what graphs are isomorphic. One necessary-but not sufficient- condition is
class of graphs may correspond, for example, to plans of rectangular that the number of vertices with a given valency be the same in both groups.
geometry? March and Earl state that every primary architectural plan can In figure 7 .5 all vertices are labelled with their valencies. This labelling
be represented such that every room is a polyomino and the walls are all helps to show that the graphs are indeed different.)
set orthogonally. There is no set of adjacency relationships in a plan
which cannot be realised in an orthogonal geometry and which necessarily 4 3
/
require any curvilinear or 'free-form' geometry. On the other hand, in a /
/
packing of polyomino-shaped rooms, the junctions between walls must be
either three-way or four-way. So any ornamentation of a primary plan
which produced junctions of a higher order could not be realised in an
orthogonal arrangement. 4 4 4
Similarly, if a plan graph is to be realised as a rectangular dissection,
the junctions between walls must also either be three-way or four-way.
It follows that the corresponding faces in the adjacency graph must be
triangles or quadrilaterals. Furthermore, any rectangular room in the
/
/ "" /
interior of a plan of this type must be surrounded by a minimum of four 4 " s
other rooms. Thus any embedded adjacency graph with a subgraph as
shown here in which all the vertices represent rooms (not the exterior s s
/
region) is inadmissible: "" ', /
/
/ /
The valency of every vertex which represents an interior room must be at / /
/ /
least four. 4 5
Let us look at the relationships between some rectangular dissections
Figure 7.5. Four plans of rectangular dissection form, with their augmented dual
and their adjacency graphs. Taking first just the weak dual graph in which adjacency graphs, showing different patterns of adjacency of rooms to the four
only interior relationships of adjacency between rooms are represented, exterior regions in each case. (Vertices are marked with their valencies to help
we find that in general many dissections may share the same graph. illustrate these differences.)
102 Chapter 7 T Embedded, coloured, and weighted graphs of plans 103
However, it should not be imagined, even when these exterior adjacencies To convert a plan graph into a particular rectangular dissection it is thus
on the four sides are taken account of, that the augmented dual graphs of necessary to specify in which of the two perpendicular directions at least
different dissections will themselves necessarily be distinct. Figure 7 .6(a) some of the wall segments run. The orientation of certain wall segments
shows two dissections for n = 6 in which not only are the adjacencies will then determine logically the directions in which others must lie. In
across the exterior walls of a similar pattern, but the two augmented duals graph-theoretic terms this can be formulated as a problem of colouring the
as a whole are absolutely identical. edges of the plan graph, in either of two 'colours', to signify the orientation
Notice that this means that the plan graphs themselves are also isomorphic. of the corresponding wall segments either 'east-west' or 'north-south'.
In what way do the dissections differ then? The difference lies in the fact This colouring must obey definite rules, if the resulting plan is to take
that where some wall segments in one dissection run in an 'east-west' the form of a rectangular dissection (Grason, 1968; 1970a; 1970b; 1970c;
direction, the corresponding wall segments in the other dissection run in a Earl and March, 1979). The 'colours' in the figures which follow are
'north-south' direction. Suppose that the pattern of wall segments in one denoted by broken and dotted lines. There are three rules required to
of these dissections (its embedded plan graph) was made from elastic ensure that the rooms themselves will be rectangular:
strings, knotted together at the junctions. It would be possible to peg out (I) At any vertex of valency three, the three incident edges must not be
the other dissection from the same set of strings, without altering any of coloured the same:
the knots. Figure 7.6(b) shows the first dissection transformed into the
second, in this kind of way. 0
4 4
.()
...lo\•···
I 0
/ / I
/ / I
/ /
(In any three-way junction two wall segments must run in one orientation
and the third wall segment in the perpendicular orientation.)
(2) At any vertex of valency four, opposite edges must be coloured the
same, and adjacent edges must be coloured differently:
4 5 4
I
1
0 ·· 0
.. L.bI ..!. ..
I I
O · cO
I
/
/
/ (At a four-way junction each pair of opposite wall segments must be
4 4 aligned, and the two pairs must run in the two perpendicular orientations.)
(a)
(3) The edges surrounding each face (not counting the exterior face) must
be coloured in such a way as to be divisible into four groups. Within each
group all edges must be coloured the same, and adjacent groups must be
coloured differently:
➔ ➔ ➔
\ I
0•\••0··>·-0
·· .. J \ I I
I . ··. \ / I
? 9'' ....
(b) - 'i I I
O••··T ····-O
I
Figure 7.6. (a) Two distinct plans of rectangular dissection form, but whose
augmented dual adjacency graphs are identical (vertices are marked with their valencies
(The face is a room, of which the edges making up the four walls must be
to help illustrate this identity). (b) A sequence of small shifts of wall segments
(without changes to any junctions) transforms the first dissection into the second; the capable of being oriented so as to form a rectangle.)
two differ only by virtue of the orientation of those wall segments, in the east-west or Last, if the plan as a whole is to have a rectangular boundary, rule (3)
north-south directions. must apply in addition to the edges surrounding the exterior face.
104 Chapter 7 Embedded, coloured, and weighted graphs of plans 105
(Without this condition on the exterior face, the application of the Weighted graphs of plans
previous colouring rules would be sufficient to create plans composed of Finally, there are ways of labelling either an adjacency graph, or else a
rectangular rooms but not necessarily with simple rectangular external new type of 'electrical' graph, so as to represent the dimensions of a
boundaries.) rectangular plan (Grason, 1968; 1970a; 1970b; 1970c; March and
Instead of colouring the plan graph according to these rules, it would Steadman, 1971; Earl and March, 1979). Consider a valid colouring of
be possible to colour the embedded adjacency graph according to a an augmented dual graph, such that it represents a. plan of rectangular
complementary set of rules. The two procedures are effectively equivalent. dissection form, as in figure 7 .8(a). The colouring divides the edges of the
Colouring the plan graph or the adjacency graph in all possible ways graph into two groups, according to whether they correspond to 'north-
such that the above rules are satisfied, will produce all plans of rectangular south' or 'east-west' walls in the plan. The graph can be split into two
dissection form to which the graph corresponds. Figure 7.7 shows the two subgraphs or 'half-graphs', each of which contains the vertices representing
alternative colourings for the plan graph of figure 7.6(b), giving the two all the rooms, two vertices representing opposite exterior regions, and all
different dissections as illustrated. (Notice that the colouring of the plan edges of one colour Ifigure 7.8(b)]. (The edges representing the adjacencies
boundary, and hence of some of the interior edges, is determined here by of the exterior regions to each other are omitted.)
the way in which the four corner vertices are positioned. If the positions Take the 'half-graph' which joins the vertices e and w representing the
of the corners were not fixed , then other colourings, giving other dissections, exterior regions on east and west, as in figure 7.9. Each edge in the
would also be possible.) Some of the design methods to be described in 'half-graph' corresponds to a wall segment aligned north-south in the plan.
chapter 9 incorporate colouring procedures which work along these lines. Suppose that values or weights corresponding to the lengths of these wall
segments are attached to the edges in the 'half-graph'. Suppose furthermore
that each edge is assigned a direction (indicated by an arrowhead) as it
">······y·······<'
/
L. . yl. .) '\
crosses the corresponding wall segment from west to east. (This direction
is arbitrary: it could equally be east to west- so long as all edges are
directed in the same way.)
.,
I
I I
\ I n n
\ .. , ~ 0
I \
' .· '
, --<\T..o····· , .•· /
\) ... .. / ~\ ../
....
\
I
I
I
I \
\
\
\
..'......,o. I I I I I
...···· b····Bt·b::6-•!..
••• I / I o·•·r I I
....•••• I
;\
\
I
0
I I
'\,.·······y ........✓ "u·······y· .. ····-✓ w o::·:..........\..&t·. -',b,t.. 6
••• 1··•; 1 I
·····o e
•..
0
I
O
\
I /
I
/ I \ / I \
I D \ / I ··... I I i°'··.• \/ I .....·· I \/
Q •. O -- --·O :o,........... .......,o. I 0
l _.-' ' , I I ,. ,. ··.. I
.P·· .. ·O, P····-O, 0 I
)t~j····\
/O·····O, /0-····0... I I
I 'o··· \ I ·.. ,- \ \ I
•d<::::: • •: i;~:::·.:::•:::>•
f I f O l
I : / \ i / \
\,
I
I
Figure 7.8. (a) A plan of rectangular dissection form, with its augmented dual adjacency
Figure 7.7. A plan graph (top) corresponding to either of the rectangular dissection
graph. The edges of the graph are coloured according to whether they represent
plans illustrated in figure 7.6. The edges (wall segments) are coloured (dotted o r
adjacencies across east-west or north-south wall segments in the plan. The graph may
broken lines) to correspond to their orientation (east-west or north-south, respectively)
be divided into two 'half-graphs' (b) corresponding respectively to adjacencies across
Two alternative legitimate colourings exist (below) for the five edges in the centre of
east-west wall segments (broken lines) and adjacencies across north-south wall
the graph, yielding the two distinct dissections of figure 7 .6.
segments (dotted tines).
106 Chapter 7 Embedded, coloured, and weighted graphs of plans 107
A graph with directed edges, to which weights are attached in this way, current in electrical networks. With an electrical network, clearly, wires
is a network02 >. Consider each vertex in this network. The total weight are represented by edges, and junctions between wires by vertices.
of the edges leaving vertex w, must equal the overall dimension of the Kirchhoffs law states that the sum of currents entering a junction or
plan from north to south. The total weight of edges entering e must be vertex must equal the sum of currents leaving that junction.
the same. As for each vertex representing a room; the sum of weights on A different kind of network representation of a rectangular plan has
edges entering the vertex must equal the sum of weights leaving that properties which parallel not just this one but both of Kirchhoff's laws for
vertex, and both sums must equal the dimension of the room in question electrical circuits (Kirchhoff, 1845). Vertices and edges here represent
in the north-south direction. quite different entities and relationships from all the graphs of plans
Exactly the same must apply for the second half-graph, running from considered up to now. Take the plan illustrated in figure 7 .1 O(a).
vertex s to vertex n, in which the weights relate to the dimensions of the Consider walls running north-south. We take not each separate wall
plan in the east-west direction. segment this time, but each maximal continuous straight run of wall-each
Looked at the other way round: given an augmented dual adjacency wall in the formal sense used in previous chapters-and assign to it a vertex.
graph, coloured in such a way as to correspond to a rectangular dissection Thus in the example of figure 7.lO(a) we take the whole extent of each
type of plan, then assignments of weights representing dimensions to the of the external walls on the east and west sides, and represent them by
coloured edges of this graph must obey this rule governing the sum of vertices E and W. We add two more vertices, to represent the wall P
weights on edges of one colour at each vertex, for the dimensioning of the between rooms c and f and d and e, and the wall Q which separates
plan as a whole to be consistent and feasible. rooms a and d from room b.
This condition on the sum of weights at the vertices of a network An edge in this network represents one of the rooms a to f and the fact
corresponds to one of two laws stated by Kirchhoff, governing the flow of that it lies between two of these north-south walls. Thus we place an
edge connecting vertex W with vertex Q, to express the fact that room a
7 5 A lies between the corresponding walls; and so on. There are therefore six
121
oo edges in the complete network, for the six rooms. As before the edges are
5 I \
4 3 9 I \ all assigned an (arbitrarily chosen but consistent) direction, in this case as
4 -:((\ ]I--.._ they run from west to east .
6 I / l5
Weights are assigned to the edges, to represent the dimensions of the
~
_2
7 : rooms in the north-south direction [figure 7.IO(b)]. An edge is shown
5
-;(!\fJ ? entering W, and another leaving E, with weights equal to the north-south
4 8
? i~
/!'--
\ I
dimension of the whole plan. The sums of weights at each vertex now
obey Kirchhoff's first law, as they did previously in the 'half-graph'
41 0
0 I (despite the fact that this network, and the 'half-graph', are quite distinct,
I I and are defined in quite different ways.) The sum of north-south
\~
dimensions of rooms on one side of a wall, equals the sum of north-
~I
' ,,I
south dimensions of rooms on the other side of that wall.
However, in addition weights are also attached to the vertices of the
)K network [shown in italic in figure 7. JO(b)]. These represent in effect the
121
dimensions of rooms in the east-west direction, but in a particular way.
Figure 7.9. A dimensioned rectangular plan and its representation by two networks. The weights give the distance of each wall from the eastern boundary
Each network corresponds to adjacencies across wall segments with a common wall E. Thus the vertex E itself takes the weight 0, Q takes weight 5,
orientation. Take the network representing adjacencies across north-sou th wall segments
(edges shown in dotted lines): each edge is directed, as it crosses the wall segment P takes weight 8, and W weight 12 for the east -west dimension of the
from west to east. The values (weights) on the edges give the lengths of the wall whole plan. Each edge in the network, with the weights attached both to
segments. The sum of weights on edges entering a vertex must equal the sum of itself and to the two vertices which it joins, now obeys Kirchhoff's second
weights on edges leaving that vertex. The same applies in the second network law. That Jaw relates together the resistance R of a wire in an electrical
representing adjacencies across east-west wall segments (edges shown in broken lines). circuit, the difference in voltage ( V- V') between the two ends of the
<12> A graph with directed edges (but no weights) is called a digraph. So a network is wire, and the resulting current A flowing from the larger voltage to the
a weighted digraph. Edges in networks are sometimes referred to as arcs.
108 Chapter 7 Embedded, coloured, and weighted graphs of plans 109
smaller, that is, In the electrical ne twork , the weight attached to an edge is the current,
and the weights attached to the vertices, the voltage levels. To what do
V-V'
A= R these correspond, by analogy, in a rectangular plan? The current , in the
network of figure 7. I 0(b), corresponds to the dimension of a room in the
north-south direction; whereas the voltage difference corresponds to its
a
dimension in the east-west direction . It follows tl}at we must interpret
b the e lectrical resistance R as the ratio of these dimensions, that is, the
Q proportion or shape of the room.
C
d
~::r. .·:P'.....!!;J For e xample, the edge corresponding to room a in figure 7. I 0(a) has a
-
f
t
Q e W q.:::-;;· -i. ····::::o E weight 5 (the 'current'), the vertices W and Q which it joins have weights
·. j· . . . . . ········€!······ 12 and 5, respectively (the 'voltages'), and Kirchhoff 's law gives
t t t ···◊ · · ·
12 - 5
w p E p
5 = -R- ' or
(a)
7 5 These relationships are illustrat ed in figure 7. I 0(c).
The same Kirchhoff's laws would apply to a complementary or
5 5 conjugate network (Earl and March, 1979) construc ted for the same plan
4 . 3 9
5 ••••• ·:f·····::n in the o ther orientation, and representing by its vertices the walls running
6
4
16 12...•.. ••·1A:
:4\
....... 016~ east-west [figure 7. I0(d) ). The dimensions of t he whole plan are given
- 2 .....':v-: ... !>1
/'"'.. ·~. / -:-:-, ....•····..O-··-;, by either one of the two networks.
5
7 ••I • •• °;,/
~•..:;l. "\J •.
....... •·1i This analogy between electrical ne tworks and re ctangu lar arrangements
8 was first demonstrated in I 937 by the four Cambridge mathematicians
4 8
(b) Brooks, Smith, Stone, and Tutte who called themselves the ' Impo rtan t
/I'.. Membe rs', and was used by them to solve the old puzzle of 'sq uaring the
112
~\o square' (Brooks e t al, 1940). The puzzle consists of finding some set of
-4-N ~~ squares, a ll of different sizes, which will pack together without holes to
srfz 's\ -
r -- , form a la rger square (that is, a dissection of a square into unequal squares).
-4-Z ~
I
uo..._x
~ ------' 19
OY
Since then several a uthors have suggested applications to the dimensioning
of architectural plans (see March and Steadman, 197 1), a subject which
- -4-X
-4-Y 'r:""4
\', I
JI will be taken up again in chapt er 9.
~6 As a final word on architectural applications of graph theory, it sh ould
i ......~>-······-6 -4-S 121 be mentioned that graph re presentations ca n be extended , without any
(c) (d) great theoretical difficulties arising, to treat 'three-dimensional' adjacency
relationships in buildings. Tha t is to say , an edge in a graph might as well
Figure 7.10. (a) A plan of rectangular dissection form with its north-south walls represent the fact that two rooms on different floors are adjacent, in the
(marked in heavier line) labelled W, P, Q, and E. In the corresponding network , the sense that the ceiling o f one fom1s the floor of the other, as re presen t
vertices represent these walls, and the edges (dotted lines) represent rooms lying their adjacency on the same floor across a common wall. On the other
between the walls. The edges are directed as they run from west to east. (b) The same
hand the re are no very significant func tional reasons for requiring such
plan dimensioned. Weights on the edges in the network now give the dimensions of
rooms in the north-south direction. Weights on the vertices give (east-west) distances adjacencies, and they tend to arise out of the constraints of gravity and
of the corresponding north- south walls from the eastern boundary wall. The sum of close-packing (rooms on one floor must be supported on those of the
weights on edges entering a vertex equals the sum of weights on edges leaving that vertex floor below). Nor is there any need for an adjacency requirement graph
(the first Kirchhoff law). (c) The single room a with its corresponding edge in the of this type to be planar, of course. There are nevertheless limits on the
network. The proportions of the room (7: 5) are given by the ratio of the difference possible adjacencies (across all surfaces) of rooms, whether box-like or of
between the weights on the vertices (12 - 5 = 7) to the weight on the edge (5) (the
second Kirchhoff law). (d) A conjugate network can be constructed for the same plan in other three-dimensional forms, packed together in space, just as there are
its other orientation, in which the vertices represent east-west walls (S, X, Y, Z, N).
110 Chapter 7 Embedded, coloured, and weighted graphs of plans 111
limits on adjacencies in the two-dimensional packing of rooms in plans 7.5 What is the degree of connectedness of the graphs illustrated?
(see Earl, 1978; Krishnamurti, 1979).
The principle of the 'electrical network' can also be extended to
represent the dimensional relationships occurring in the packing of cuboids
into a cuboid. Teague ( 1970) suggested the relevance of this representation
to building geometry, and proposed its application in computer-aided design.
Exercises
7.1 (This is if you did not already take up the invitation in the text.)
Draw the plan graphs corresponding to some of the embedded adjacency
graphs in figure 7. I (assume all vertices represent rooms in each case, and 7. 6 Draw all trees with eight vertices. (There are twenty-three of them.)
the exterior region is not represented). Some of the plans will contain
holes, or interior rooms engulfed inside others, as in the second figure on 7. 7 Draw the projections onto the plane of the tetrahedron (as in
page 79 and the third figure on page 80. figure 7.3), the cube, and the octahedron. Draw their duals. Of which
polyhedra are these duals projections onto the plane?
7. 2 Draw all distinct planar embeddings of the graphs illustrated. How
are the numbers of embeddings altered, by the fact of whether or not the 7.8 To which of the embedded adjacency graphs illustrated in figure 7.2
vertices are labelled? do the primary plans also shown in the same figure correspond?
(Remember that this correspondence is not one-to-one, and is dependent
on the choice of vertices in the adjacency graph to represent the exterior
region.)
A
7. 9 Draw the distinct possible planar embeddings of the trivalent polyhedra
(the 'fundamental plans') illustrated in figure 7.4 by projecting them
through each of their faces in turn. (Some of the results will be duplicates.)
These embeddings should give you the 'primary plans' illustrated in
figure 7. 2.
7. 3 Redraw the graph illustrated in a planar embedding such that the face
abc becomes the exterior face. 7.10 For the dimensioned rectangular dissection illustrated, draw the
types of electrical network, representing the adjacencies and dimensions of
the rooms in the 'north-south' and 'east-west' directions, whose principles
are illustrated in figure 7. 9 ( where the vertices represent rooms and regions).
For the same dissection, draw the types of electrical network illustrated
in figures 7.I0(b) and (d) (where the vertices represent walls).
6 5 4
3
6 -
7. 4 Identify all cut vertices in the graphs illustrated. 7
4
3
I 8 4
8 Properties and classification of rectangular arrangements 113
Properties of rectangular arrangements, and their One of the first attempts at a systematic categorisation was that
classification made by Combes (I 976), who directed his attention to the number of
wall segments in an arrangement. He counted the number of external wall
segments w in the perimeter of a dissection and compared this with the
"Art makes us realize that there are fixed laws which govern and point to the use number of internal wall segments p <13>_ He plotted a graph in which w
of the constntctive elements of the composition and of the inherent interrelationships and p are shown on the vertical and horizontal axes, respectively (figure 8.1 ).
between them."
Piet Mondrian (1937)
carried. Many real architectural plans, clearly , have more than nine or ten
i 6
C:
rooms on a single floor. Does this mean that the approach outlined here 1
~
is only applicable to small buildings? Even within the practical limits of -s ~-- ~- - ~ - - ~ -- ~ ----+-
enumeration, the further question arises whether, for larger values of n , a i .
simple listing of dissections would not be so long and cumbersome as to ]
be of limited practical use. ~ 4 ~ - -~ - - ~ - -~--~--__.;...
&l
Certainly where a catalogue of dissections is intended to be referred to
directly, it will need to be organised into some systematic order, and
properly cross-referenced, if arrangements with specified properties are to 3 ~ - - ~- - ~ - - ~ - - ~ - -__.;...
be picked out at all easily. And even if a listing of dissections is to be
searched by computer, it will still make this search much more efficient
2 - + - - -- + - - - - , . - - ~ - - - 1 - - - --1-
and economical if it too can be organised according to some classificatory
hierarchy (so that not every entry in the list has to be examined on every
run).
We have already looked at a number of properties of dissections in
earlier chapters: their symmetries, their graphs, their grating sizes. Indeed
in several instances some knowledge of these attributes has been exploited
deliberately in the design of algorithms by which the dissections have been 2 3 4 s
enumerated in the first place. Once the enumeration is complete, however, Internal wall segments, p
we can turn the question around, and ask how are such attributes
Figure 8.1. Graph of numbers of external wall segments w against numbers of internal
distributed over the whole population of possible arrangements? How
wall segments p in rectangular dissections (after Combes, 1976). All individual
many dissections possess symmetries of a certain type, how many occupy dissections are illustrated, attached to the appropriate integer points in the graph, for
a certain size of grating, how many possess a certain graph, and so on? values of p ~ 5, w ~ 10.
Certainly the first and most obvious basis for classification must be the
order of the dissection, n. Beyond this the choice of properties of interest <13> Combes himself refers to an external wall segment as a 'wall', and an internal wall
is wide open. segment as a 'partition'.
114 Chapter 8 Properties and classification of rectangular arrangements 115
Thus a dissection with given numbers of internal and external wall segments From the application of Euler's formula then to this adjacency graph
would appear on the graph at the appropriate integer point. The dissection considered as a map: the number of vertices is n, the number of edges
for n = 1 has four external and no internal wall segments; the dissection for is P, and the number of faces is (i3 + i 4 + I) (the 1 is for the exterior face).
n = 2, six external and one internal wall segments; the two dissections So
for n = 3, eight external and two internal, and seven external and three (8.1)
internal wall segments, respectively; and so on.
Combes was able to demonstrate a number of general relationships In the example illustrated above, n = 7, p = I 0 , i 3 = 3, i 4 = I , which
existing between w, p, n, and the numbers of junctions, three-way and gives
four-way , occurring in the interior of a dissection. Combes himself arrived 7-10+3+1+1 = 2.
at these formulae empirically. They are, however, derivable from the
polyhedral formula of Euler (Gutierrez, 1979; see also Biggs et al, 1976, Now take the dissection itself, considered as a plane map, and the
chapter 5). Euler's expression relates together the numbers of vertices v , relation between the total number of wall segments which it contains and
edges e , and faces f of any polyhedron: the numbers and types of all the various junctions (not just those in the
interior). There are i 4 four-way junctions, each of which has four internal
v-e+f=2 . wall segments incident with it, that is, it has valency 4. Similarly there
We have seen in the last chapter how any polyhedral form can be are i 3 three-way junctions in the interior, each with valency 3. We have
'projected' down through one of its faces onto the plane, to become a not yet counted those three-way junctions which lie on the perimeter of
plane map. You might like to confirm that Euler's formula applies to any the arrangement. Their number is related to the number of external wall
dissection, or indeed any plane map, you care to choose. (You can choose segments w. It is easy to see that the number must be w- 4. It is four
to count the four outer corners of a dissection as vertices of valency 2, or less than w because of the four 'two-way junctions' at the outer corners.
not- the relation applies in both cases. It will be convenient to include (Thus in the above figure , w = 11 and the number of three-way junctions
these vertices here, however, since edges will then correspond one for one in the perimeter is 7 .) These four corner junctions themselves have
to wall segments. Remember to count the exterior face.) valency 2. Since every wall segment joins two junctions, it follows that
Now the embedded adjacency graph of a rectangular dissection can be by adding all (junctions x valencies) we will count each segment twice. Thus
considered as a plane map. Suppose we draw the graph of the 'internal' 2(p+ w) = 4i4 + 3i3 + 3(w- 4 ) + (4 x 2). (8.2)
adjacencies between rooms only, and omit the exterior region around the
plan and all adjacencies to that region (it is the weak dual): For the above example:
2(10+ I I)= 4+ (3 x 3) + 3(1 I -4) + 8 = 42.
From equation (8.1) we have a value for p:
p = n + i 3 + i4 - I , (8.3)
into a single four-way junction. In this process one (internal) wall segment below the relevant line, by a distance in units equal to that number of
disappears.] junctions.
These relationships are demonstrated by the example of dissections for In those dissections without interior junctions of any kind, that is,
n = 5, shown in figure 8.2. These are all trivalent, and so the value t where both i 3 and i4 are zero, equations (8.3) and (8.4) reduce to
(= p+w) , is always sixteen, that is, (3 x S)+l.
Notice how the dissections are arranged in columns, according to the
p = n- I , (8.6)
pairs of values w and p, from w = 12, p = 4 at one extreme, to w = 8, w=2n+2. (8.7)
p = 8 at the other. It is as though, scanning from left to right in the
In fact for each value of n there is only one dissection of this type, and
figure, wall segments were progressively removed from the perimeter and
with these values for p and w. It is that arrangement where then
transferred into the interior. In the two remaining dissections for n = 5
rectangles are all arranged in a simple straight line:
which possess one four-way junction each, t equals 15 in both cases:
OJ II
EHJB
To return to Combes's graph on which dissections are plotted, it
It is clear by inspection why the number of internal wall segments must
be one less than the number of rectangles, and why the number of
external wall segments must equal 2n + 2. These are the dissections in
follows that trivalent dissections will all lie on Jines at 45° to the axes,
which the ratio of w top is at its greatest. They have the greatest possible
corresponding to the graphs of t = 3n + I for successive values of n
number of sides of their component rectangles lying on the perimeter;
(figure 8.3). Any dissection containing four-way junctions will drop down
they are the most 'outward-looking' arrangements. The line corresponding
w = 12 w = 11 w = 10 w =9 w = ·8 to the equation w = 2n + 2 on which they lie, marks an upper boundary to
p=4 p =S p=6 p=1 p =8 the area in Combes's graph within which all dissections must fall.
llj § g;g~gJ § 14
13
12
11
8 E-',-,
I
I
,
~ f-t--<;>c--+--t-9--t----,-<;\,
1~
0 -i---,- - 0~ - - + --<>--•--+
~,._.__,,__,__,_o--,___,, -,,
cm rn rn ~
10 ..--l--- :t--'
9 ~,.__.._,_._,,
[o $ 8
7 -j-rf--f-..,,>)'-+-t--+---+--t--!-t-+
6 +-~ +-l
~ Lill rn
5
rn cm 4
3 -t--\---t----,,-f-j---+--!---1---'--t-+---+
2
~ [mm ~
I 2 3 4 5 6 7 8 9 10 11 12
p
Figure 8.3. The Combes (1976) graph of external wall segments w against internal
wall segments p in rectangular dissections. Trivalent dissections for successive values of
n lie on the lines p+w = 3n+ I at 4S to the axes, as indicated. The space within
0
:::! lhn:; I
At the other, lower extreme, the boundary is set, for values of p greater
than 4, by the horizontal line at w = 8. This is a consequence of the fact
that dissections in which the number of external wall segments is least are
...... ffim] ~
those where four rectangles form a 'frame' around the outside, and all the
remaining rectangles are contained in the interior: g ~ mm
- -- -
"'
§ ~ [!ffl
00
[fill] mm ~
Only where n is less than four, is it possible for there to be fewer external
wall segments (compare figure 8.1).
On Combes's diagram then it is possible to delimit a wedge-shaped area
.... mm !wJE - ~
[ID]] ~ ~
within which all dissections fall. Moving from the origin towards the
upper right of the graph, corresponds to a progressive increase in the value
"'
of n. For given n , all trivalent dissections, as we have seen, lie along the
straight line t = 3n + 1. Where those with four-way junctions are included,
then all dissections for given n are contained within a roughly triangular V)
t[[I]]
area, of which the upper edge is marked by t = 3n + 1. For example,
the area within which all dissections for n = 9 lie is shown in figure 8.4. ... 1§ 11111
Much of Combes's work has been devoted to determining the exact
shape of the lower boundary to this triangle in the general case. The
demonstration is highly complicated, and need not concern us here. It will "' ffi 1111 1
be sufficient to observe that a move downwards within the triangle H 11ll ti
....
represents an increase in the number of four-way junctions, from O to
some maximum quantity reached at the lower apex. In the example of ·-·+ HI 111111 M ...
II 11-
n = 9 in figure 8.4 there is only one dissection at this extreme lower ~ 0 I II II I
0
I I ·-· I ·-• --· -----
e
point. It is the square pattern of nine squares in which there are four
four-way junctions.
. ....
0
the triangle for a given n, which is more densely populated with dissections
than is the upper left. (Figure 8.4 gives just one example of a dissection
g c:,.
for n = 9 located on each point.) This classification does not by any
means distinguish individual arrangements therefore.
But it does serve to classify dissections along a certain number of
'dimensions': first of all by increasing n; and then by the extent to
which dissections are 'inward-looking' or 'outward-looking', that is, how
many sides of the component rectangles are in contact with the perimeter, ~ ~ ~ ~~ ~ - - - , - r
"',--,r,-r+ 0
M
to those in the interior. (See how the number of interior junctions
i 3 + i 4 increases from O to 12, reading from left to right in figure 8.4.)
120 Chapter 8 Properties and classification of rectangular arrangements 121
·Finally, there is a classification of the extent to which arrangements arrangements at a quite different scale, for example arrangements of
contain four-way junctions. constructional components. The various types of 'bond' in which bricks
To what degree are these properties of any architectural relevance? are laid (figure 8.5) are all designed so as to avoid, as far as possible, four-
Certainly the matter of the contact of the component rectangles with the way junctions between bricks (looking at the pattern in the vertical plane).
perimeter of the dissection seems to be important. Rooms on the exterior There are sound functional reasons for this, of course: the overlapping is
of a plan can have windows, or they can have doors giving directly onto intended to ensure the best possible adhesion between adjacent bricks, and
the outside. Conversely, there are some types of building in whose plans to discourage vertical cracks from opening up. Just as with rooms in
one or a few central spaces are completely enclosed by other rooms plans, trivalent arrangements create more adjacencies between bricks, other
clustering around, such as theatres, law courts, or the operating suites of things being equal, than do those with four-way junctions.
hospitals.
The numbers and types of junctions between walls in a plan might not Bloch's catalogue of dissections
seem at first sight to be of any architectural interest. Architects do not Combes's work provides a classification of dissections by some of their
generally concern themselves over the distinction between three walls or properties, but these properties are limited ones, and in many cases serve
four walls coming together at a point. (Except perhaps in the design of to group very large numbers of dissections together. More recently, a
prefabricated walling systems, where different types of connector piece are comprehensive catalogue of dissections has been produced by Bloch ( 1979a;
needed in each case.) But even here there are certain consequences for 1979b), in which all arrangements up ton= 8 are depicted individually,
plan arrangement which are not without practical significance. and statistical tabulations are given for n = 9. (Part of this catalogue, up
Imagine a plan with a given number of rooms, corresponding geometrically to n = 7, is reproduced here with his kind permission, as an appendix.)
to a rectangular dissection. The rooms are all 'overlapped' one with another The entries are organised by means of an hierarchical classification. In
in their arrangement. Now imagine pairs of adjacent three-way junctions addition to the characteristics which determine this hierarchy, Bloch has
being progressively replaced by four-way junctions. The plan will become also investigated the frequency of occurrence of certain other properties
i,tcreasingly 'squared-up'; and the number of internal wall segments, as we which are arguably of architectural importance.
have seen, will decrease as a result. The existence of such a wall segment To some extent Bloch's catalogue is organised according to the properties
provides the opportunity to place a door giving access between two rooms. by which the dissections were originally generated, by means of his 'tiling'
With fewer internal wall segments there is correspondingly less flexibility algorithm (compare chapter 5). Dissections are classified first by their
in the way this placing of doors, and hence the design of the circulation order n and then by their grating dimensions. The catalogue contains all
pattern as a whole, can be made. In chapter 11 we will look at some dissections including those with alignments and four-way junctions, but
historical evidence (slight though it might be) to suggest changes related to with the sole exception of dissections for which Im = n , that is, where
these kinds of properties, going on in the evolution of house planning. each rectangle occupies one cell in the grating. Figure 8.6 reproduces a
We have been thinking of dissections as representing plans, and the page from the catalogue (Bloch, 1979a). It shows all dissections of order 5
rectangles as rooms. But they might just as well represent architectural (with the exception of the single arrangement with grating size 1 x 5). By
convention the gratings are set always with the shorter dimension / across
the page. The possibilities for grating sizes for n = 5 are 2 x 3, 2 x 4,
and 3 x 3 (compare figure 5. 1 and table 5.1), of which the 2 x 3 gratings
correspond only to arrangements with four-way junctions or alignments.
The next basis for classification is by what Bloch calls fronts. A front
is a rectangle which is in contact with the perimeter of the dissection (as
English English garden wall monk opposed to one surrounded in the interior). The boundary of the front
might contain one, two, or three external wall segments (four only where
n = I, obviously). This the term does not distinguish; in all cases the
rectangle is a front. As we have seen in Combes's work, for n ~ 4 the
minimum number of fronts is 4 ; and clearly the maximum number is
always n. In figure 8.6 there are only therefore dissections with four or
Flemish Flemish garden wall stretcher five fronts.
Figure 8.S. Types of traditional pattern, or bond, in which bricks are laid.
122 Chapter 8 Properties and classification of rectangular arrangements 123
The fourth dimension of Bloch's classification relates to the valencies of of valency 3, and none of valency 4:
vertices in the (weak dual) adjacency graphs of dissections. Bloch refers to
the distribution of those valencies between vertices as the graph partition.
This can be set out in formal notation as a string of digits in which the
first digit gives the number of vertices with valency 1, the ne~t giv~s the
number of vertices with valency 2, and so on. For the two dissections
which appear at the top left of figure 8.6, for example, the string 0320
indicates that there are no vertices of valency 1, three of valency 2, two
As a final classification, Bloch's catalogue gives the symmetry properties
Order 5, Grating (2, 3) Order 5, Grating ( 3, 3J of each dissection. The format is largely self-explanatory in terms of the
5 fronts 4 fronts notation introduced in chapter 3. Those dissections with no symmetries
0320 0041
D1,v iden C4 K. D, iden are labelled for 'identity' (they are left unchanged only by the identity
rn rn
0401
~§till~
5 fronts
0401
transformation). Dissections with all the symmetries of the rectangle are
labelled K4 for the 'Klein four-group [see footnote (3), page 22).
Dissections on nonsquare gratings with bilateral symmetry Di are
distinguished as to whether the axis of symmetry is vertical (D 1 v) or
K.
c.
w
Order 5, Grating (2, 4)
~
1211
horizontal (D 1, h) (that is, up and down, or across the page). Dissections
on square gratings with D 1 symmetry are illustrated always with the axis
of symmetry up and down the page. There is the further possibility that
square dissections may possess D 1 symmetry about a diagonal axis (D1, dg)·
5 fronts D, The cyclic symmetries C2 , C4 are marked as such.
1310
D1,v
§ Statistical analyses of dissections and polyominoes
To come to the statistical analyses which Bloch ( 1979a) has carried out on
these 'populations' of dissections: there is not the space to give his detailed
1130 results here, but it is worth commenting on the architectural significance
D1,v iden of some of his major findings. Grating area lm, which Bloch tabulates for
~~
all fundamental dissections, is in itself perhaps not very meaningful in
architectural terms. It obviously increases generally with n. And for given
n the number of distinct dissections generally (though not universally)
0221 increases with increasing grating area, as can be seen from an inspection of
~DH~~@~
the catalogue. We have already noted in chapter Show those arrangements
which contain four-way junctions or alignments have smaller grating areas
than corresponding fundamental dissections, because of the consequent
loss of grating lines.
Figure 8.6. Sample page from Bloch's catalogue of_recta~gular dissections _[from Bloch Grating shape I: m on the other hand has some significance, in that for
(1979a); see also appendix). The page shows al~ dissections for n = 5 (""'.1th the given n the numbers of distinct dissections on long thin gratings are fewer
exception of the single arrangement for grating size l x 5). They are cl~ss1fied by . than for gratings whose shape is closer to square. Thus we can imagine
grating size; number of fronts; graph partition; and sym~etry properties. A front 1s that there is more flexibility in the initial planning of architectural
a rectangle in contact with the perimeter. The gra~h partition r~lates to the valencies
of vertices in the weak dual adjacency graph of a dissection. It 1s set out as a strmg of
arrangements on squarer gratings, and more possibilities for the conversion
digits (four in this instance), where the first digit giv~s the number of vert_ices with of such plans after they are built, by rearranging internal walls etc.
valency l , the second digit the number of vertices ~th valency 2, etc. Dihedral and The shapes and sizes of gratings are linked with the numbers of fronts,
cyclic symmetries are marked by the usual conventions (D1 , C2, C4 , etc), and v and h since in larger and squarer gratings the proportion of grating cells in the
indicate whether the axis of Di symmetry is vertical or horizontal. Arrangements interior to those on the perimeter is higher. There are therefore more
marked K 4 have all the symmetries of the rectangle. Those marked iden have no
opportunities here for arrangements with interior rectangles, and hence
symmetries.
Chapter 8 Properties and classification of rectangular arrangements 125
124
with fewer fronts. The ratio of fronts to interior rectangles in a dissection is dissections showing symmetry properties of any kind will become very
another measure of how 'outward-looking' or 'enclosed' the arrangement is. small indeed. It is a characteristic of many real architectural plans, of
The measure is not by any means the same as Combes's ratio of external course, especially of Renaissance or Classical style buildings such as those
to internal wall segments; but it points to something of the same qualities. planned according to the precepts of the Beaux Arts, even those with very
Bloch's graph partition brings out properties which are of the greatest many rooms, that they possess at least bilateral and perhaps other
interest architecturally, since the valency of a vertex in the (weak dual) symmetries. Bloch's result suggests that such arrangements are very rare
adjacency graph corresponds in the dissection to the number of other when considered in relation to the whole range of possible rectangular
rectangles to which one rectangle is adjacent. Some rooms, such as configurations.
corridors or halls, are typically adjacent to many others (the corresponding The most frequently occurring type of symmetry in dissections generally
vertices have high valencies), indeed this is exactly their purpose; whereas is D1, and after this C2 • Symmetries of order 4, that is, D4 and C4 , are
with private rooms such as offices or bedrooms it is perhaps functionally much rarer. Of the D 1 symmetries, the orthogonal (horizontal or vertical)
necessary for them to possess only a single adjacency (the corresponding are much more common than the diagonal. The occurrence of symmetries
vertex has valency l): adjacency to the circulation. of the different types is related in rather complicated ways to the oddness
Many plan types are characterised by the distribution of this pattern of or evenness in various combinations of l, m, and n, and to the manner
adjacencies between rooms. There are hierarchical plans in which one or a in which central 'cores' of one or a few rectangles may be formed,
very few major spaces are adjacent to many other minor ones. A station around which others can then be symmetrically disposed.
concourse with its surrounding offices, bars, restaurants, kiosks, etc is a Bloch derived these statistics by generating all dissections, as explained,
case in point. And there are 'democratic' plans, with a more even and then picking out those with each kind of symmetry. Krishnamurti and
distribution of adjacencies throughout. The cellular plans of American Roe (l 978) have, however, published a method, using 'colouring'
Indian pueblos provide an example here. techniques similar to those described in chapter 5, for generating only
Since Bloch considers only the weak dual adjacency graph, he does not dissections possessing certain specified symmetries. Figure 8.8, for
take account of adjacencies of rooms to exterior regions, whose number example, illustrates all dissections with C4 and D4 symmetries, on square
must equal the number of external wall segments w. The patterns of gratings where l = m = 3 and l = m = 4. (It should be remembered that
distribution of these adjacencies are naturally also of great functional in all this discussion we are talking of the symmetries of configurations on
significance in buildings, in characterising different plan types. We shall dimensionless gratings. Such symmetries could well be destroyed with the
return to these topics in chapters l O and 11. assignment of dimensions to the grating intervals.)
As regards Bloch's results on the occurrence of different types of Besides the classifiers used in the catalogue, Bloch has also compiled
symmetry in dissections, the most important result is that illustrated in statistics for other properties. These include the occurrence of two further
figure 8. 7. This shows the relative frequency of dissections possessing
any symmetry (other than the identity) in the population of all dissections,
for values of n between 5 and 9. The percentage of symmetrical dissections
drops rapidly, from over 50% for n = 5 (compare figure 8.6) to around
2% for n = 9. Clearly, for still larger values of n the proportion of all
5 6 7 8
n
Figure 8. 7. Percentage of all rectangular dissections possessing any symmetry ( other Figure 8.8. All dissections with C4 and D4 symmetries on square gratings, I =m =3
than the identity), for values of n between 5 and 9 (from Bloch, 1979a). and I= m = 4 (from Krishnamurti and Roe, 1978).
Chapter 8 Properties and classification of rectangular arrangements 127
126
X
P = n + i4 - 1 = 9 , and w = 2n - 2i4 +2 = 14 .
- Our earlier demonstration of the equations for dissections made use of
Euler's polyhedral formula , which must still apply. We also relied on
(These parts would themselves be rectangular dissections.) In the weak the fact of the boundary of the ·dissection being rectangular to derive
dual adjacency graph of the dissection it corresponds to a cut vertex, equation (8.2), which related wall segments to the numbers and types of
hence the term. In architectural parlance a cut is a 'through-room'. all junctions. Clearly for polyominoes we would have to proceed
Very often a corridor in an office block is a cut in this sense. somewhat differently; but it requires only a slight extension of the same
Another type of rectangle is the snip. This is a rectangle which lies approach to show that equations (8.3), (8.4), and (8.5) still hoJd< 15>.
along the whole length of one of the sides of a dissection, and may be Since polyominoes differ from dissections in that their boundaries can
'snipped' off, leaving a complete di~ection of order n - 1 behind: take up many shapes, it is possible to apply to polyominoes new measures
which aim to capture some of these shape characteristics. Matela and '
March compute a simple shape index w/p. When this ratio is high the
snip polyomino is long and thin (although it may be branching or bent round
on itself); when it is low, it is compact. (The same is true for the
measure applied to rectangular dissections.)
A second index devised by March and Matela relates the number of
external wall segments to the number of component squares. This is a
perimeter index and is expressed as w/(4n). In this way its value for any
It is the type of rectangle which is added in operation (2) of Steadman shape is confined to a range between O and I , since for arrangements
and Mitchell's generating algorithm, or operation (2) of Earl's algorithm, without internal four-way junctions, w = 2n + 2, and the index takes the
as described in chapter 4. In architectural terms a snip is something like
an end bay. Figure 8.9 illustrates both a snip and a cut in the 'central
fireplaces type' house of chapter 2.
We will come back to Bloch's catalogue. In the meantime we should
look at some equivalent work which has been done for polyominoes, by
March, Matela, and O'Hare. These authors have compiled a number of
statistics for various properties of polyominoes, in some cases for Figure 8.9. Examples of a snip and a cut in the 'central fireplaces' type house of
n = 6, 7, and 8, and in other cases for the whole population up to n = 9. chapter 2. A snip is a mom in a plan of rectangular dissection type which lies along
March and Matela (1974) have tabulated the numbers of internal and t~e en~1re edge of the dissection and may be removed (snipped off) to leave a complete
external wall segments in polyominoes for n = 6, 7, and 8, in much the dissecl!On of ~rder _n - 1. (It is something like an end bay.) A cut is a room extending
same way Combes did for dissections. Indeed it turns out that, in general, across the entire width or length of the dissection, which if removed would leave two
separate dissections (it is a through-room).
the Combes equations (8.3) and (8.4) governing wall segments apply
equally to polyominoes. Of course, there are no internal three-way
(ls) The exception is where the polyomino contains a hole or holes, and these
junctions in polyominoes. But there may be four-way junctions (both 'courtyards' are treated as exterior regions, and the wall segments which border them
internal and on the perimeter); and the total number t of wall segments, as external wall segments. This possibility first arises with n = 8, where the eight cells
for some given n , is dependent on the number i 4 of internal four-way can surround a unit square hole in a complete square ring. (It is possible for a
junctions, as March and Matela point out. In the following example: polyomino o'. order 7 to form a ring, but the two 'ends' of the figure touch only at a
corner; and 111 these circumstances the Combes formulae still apply).
128 Chapter 8 Properties and classification of rectangular arrangements 129
The taxicab distance is likely to be, for rectangular plans, a second-best graphs of polyominoes, and the polyominoes themselves. See how, for
approximation, since circulation routes in buildings tend to run orthogonally. example, three out of the five polyominoes for n = 4 share the same graph.
Interest in comparing the three distance measures first arose in the Matela and O'Hare have enumerated all such possible adjacency graphs for
polyominoes up to n = 9, and illustrate in their paper the possibilities up
architectural context, in connection with computer methods for assembling
to n = 8.
plan layouts which, as mentioned in chapter 2, tended to make use of
The numbers of weak dual adjacency graphs f?r values of n up to
regular square grid systems of description. Rooms or zones were then
n = 9, compared with the corresponding numbers of polyominoes is
located on this grid so as to minimise circulation distances, supposedly;
given by:
these were measured as straight-line or Euclidean distances.
As Tabor ( 1976) pointed out, it would have been much more realistic, order 2 3 4 5 6 7 8 9
in the absence of any predefined circulation routes, to use a taxicab polyominoes 2 5 12 35 108 369 1285
measure of distance. Even this might depart quite substantially from the adjacency graphs I 3 4 IO 19 49 112
real distances between rooms, as measured along actual circulation routes, The ratio of graphs to polyominoes decreases with increasing n; that is,
depending on the shape of the plan overall- whether it contained court- for larger values of n there are some graphs which are shared by very
yards or was branching in shape, etc- as Tabor demonstrated in a series of many polyominoes. If the number of possible adjacency graphs is
analyses. In this light, it is not surprising that by attempting to minimise compared with the total number of all connected graphs, for increasing
the sum of Euclidean distances between rooms, such methods tended to values of n, this ratio is also found to decrease, rapidly:
produce plans of deep, compact, and clustered form.
vertices 2 3 4 5 6 7 8 9
There are implications too, in this work of Matela and O'Hare's on
distance measures, for the use of regular square grids in representing the adjacency graphs I I I 3 4 10 19 49 112
all connected graphs I I 2 6 21 112 853 11 117 261080
land surface of towns or regions in simulation models.
ln the present context we might question the significance of the absolute As Matela and O'Hare point out, this implies that out of all the possible
values obtained for distance measures on polyominoes, as such, however; patterns of adjacencies between rooms which a designer might wish to
since the polyomino represents an architectural plan in undimensioned create, only a relatively few, for larger values of n, can be accommodated
form, and when the configuration is dimensioned, so all the distances in polyomino-type plans.
between cells will change accordingly. The same argument would hold for It should be said, however, that this result is first of all a consequence
similar measures applied to 'dimensionless' rectangular dissections. On the of the fact that many connected graphs for large n will be nonplanar, for
other hand the relative values of these distance measures would continue a start. And after that, it is in large part attributable to the particular
to give information about shape and arrangement. geometry of the polyomino itself. The valency of the vertices in permissible
As a final class of properties of polyominoes, Matela and O'Hare (1976a)
investigate some of the relations between polyominoes and their (weak
dual) adjacency graphs. We have already seen in chapter 6 how the
relation between rectangular dissections and their adjacency graphs is not
one-to-one; and the same is true for polyominoes. lt is, however, possible
to distinguish classes of polyominoes which share the same adjacency
graph. And the properties of the graphs can serve to distinguish some
further, characteristic features of these classes of arrangement.
If we take the weak dual graph representing only internal adjacencies
between cells, then the number of vertices equals n and the number of
edges equals p, as for dissections. Out of all graphs, these must at least
be connected and planar. Since the polyomino may have in its interior
only four-way junctions, it follows that the (interior) faces in the embedded Figure 8.10. Relationship of polyominoes to their adjacency graphs for n = 4; after
graph may be only four-sided. Matela and O'Hare (I 976a). The top row shows all connected graphs for n = 4. The
Figure 8.10 is adapted from Matela and O'Hare. It illustrates in successive middle row shows all of those graphs which can be the weak dual adjacency graphs of
rows in the diagram, for n = 4 in each case, reading from top to bottom: polyominoes. The bottom row shows the polyominoes themselves. See how three out
of the five polyominoes share the same graph.
all connected graphs, all graphs which may be the weak dual adjacency
132 Chapter 8 Properties and classification of rectangular arrangements 133
adjacency graphs must not exceed 4. And faces in the embedded graph But it would be plausible to imagine that it might perhaps be structured
must be four-sided. The graphs of rectangular dissections, for example, on five principal levels (figure 8.11) (compare Korf, 1977)< 16>.
are not so severely restricted; although here too we would again expect The first level would be a classification by the simple number of cells or
there to be many possible connected graphs, even connected planar graphs, rooms (and hence the number of 'functions' possible to accommodate, one
which cannot be mapped into dissections. to a room) in the plan. Thus the rooms are considered at this stage as an
Matela and O'Hare (1976a) have made some studies of the possible unordered set. At the second level, adjacency graphs with the appropriate
subgraphs of the adjacency graphs of polyominoes, which are discussed number of vertices (and perhaps with other specified properties, such as
here in chapter 10 in connection with the adaptability of plans. And particular graph partitions etc) would be considered. At this level, then,
finally they have investigated the occurrence of cut vertices and bridges in topology is introduced. At the third level the adjacency graphs are
the adjacency graphs of polyominoes. A cut vertex would correspond to embedded. At the fourth level shape would be introduced; that is, the
a cell in the polyomino which if removed would leave two or more classification would move to some particular class of geometrical designs,
disconnected parts (themselves smaller polyominoes). It is the exact such as polyominoes or rectangular dissections. At the lowest level,
analogue of Bloch's cut in a dissection. Some examples are:
level l : number of spaces ao
ob
eo g od
level 2: adjacencies - -- - - - -1- - - -
ao ob
a 0----0 d
b 0---0 C
A bridge in the graph, on the other hand , is a single edge whose removal C o------0 d
a/ ~ a
d o------0 e
disconnects the graph- it represents a single partition wall serving to divide
the polyomino into two separate parts: level 3: embeddings
b<?• b7d
~c~
C
o---o---o--0
level 4: shape
~
Matela and O'Hare present statistics for the total number of bridges in
adjacency graphs for polyominoes, up to n = 9. Notice that the graph b d e
C
of the 'square' polyomino for n = 4 contains neither a cut point nor a
I~
rn GhhJ
bridge: level 5: dimensions- -
ITT
EB □
Going back to some of the general questions about the classification of Figure 8.11. Five-level hierarchy for the classification of plans. At the highest level
plans are considered as unordered sets of rooms. At the second level , relations of
dissections which were raised at the beginning of this chapter: once a
adjacency between rooms are introduced. At the third level the resulting adjacency
catalogue, say of dissections, such as Bloch's, or an equivalent catalogue of graphs are embedded. At the fourth level shape is introduced, in the form of some
polyominoes, is completed, to what kinds of use can it be put? overall geometrical discipline , that is, the plan is organised on a rectangular grid, a
It is possible to imagine that a designer working on a particular planning triangular grid, it is composed of rectangular rooms, etc. At the fifth level, dimensions
problem might consult the catalogue to find an arrangement or some few are considered.
arrangements corresponding to his requirements. The better the catalogue 16
~ ) G~erra (1977) has hinted at some kind of parallel between such a hierarchy of
is organised, in terms of a hierarchy of progressively more precise mcreasmgly abstract representations for architectural plans, and the famous classification
specifications, the easier it will be for him to find the configurations of geometries by Felix Klein in his 'Erlanger Programme' of 1872 (see Klein, 1939).
he is seeking. The exact organisation of the classificatory hierarchy Compare also March and Steadman ( 1971, pages 19-2 S), and March and Steadman
would depend, as mentioned, on the particular interests of the users. (1979).
134 Chapter 8 Properties and classification of rectangular arrangements 135
dimensions would be considered. As we shall see in the next chapter, it is which his search may take. Simon (1975) has even suggested that style in
quite possible that a given geometrical arrangement, which meets specified design may be connected with the sequence in which constraints are
adjacency requirements, may not be capable of being dimensioned to the applied, and so to the nature of the route which the search process takes
sizes which are perhaps required for each room. through the space of possibilities considered.
Bloch (1979a) himself has described (although not implemented) a Some of these questions will be taken up again in the next chapter. It is
computer system which would allow the operator to make a search for not in such direct applications in design methods, ne_vertheless, that I myself
dissections in essentially this manner. The computer would display on a believe the greatest usefulness of catalogues, and an enumeration approach
screen some representation of the 'solution space' for dissections of some to plan arrangements in general, lies. Within the limitations imposed by
given order or orders, such as for example the Combes diagram. The user the definition of some class of designs- dissections, polyominoes, or
would then point to and so select, with a light pen or cursor, subsets of whatever- all permutations of arrangement are exhaustively covered.
arrangements, perhaps specifying further properties, until some single Hence general statements may be made about the range and behaviour of
dissection was reached, or some smaU set defined, which could then be all possible designs, within the given definition.
enumerated. (In this case Bloch imagines the specified dissections being We have been looking- in March, Matela, and O'Hare's work on
generated 'to order', and not being retrieved from some stored catalogue. polyominoes, and Bloch's on dissections- at some theoretical generalisations
The effect is the same, although the computing implications, for storage of just such a statistical kind. These studies are only a beginning, as their
space required , and the speed of response of the system, are very different.) authors would admit, and they are by no means complete; but they do
In the chapter which follows we will look at an actual existing computer begin to suggest how a comprehensive view might be built up of the total
program which does something similar. The system in question is perfectly extent of plan forms of certain types, of how the behaviour of plans varies
automatic. It is not interactive, nor does it display dissections or groups systematically with changes in properties of different kinds, and of how rare
of dissections graphically in the course of operation. Also it works with a or frequent plans with given characteristics are in relation to the whole
stored and unordered list of all dissections up to n = 8, and for this 'universe' of possibilities. (We have noticed, for example, how very scarce are
reason is slower than it might be in running. However, it does move in arrangements with any symmetry among all dissections for larger n, however
effect through the same five-level classificatory hierarchy as just described, common bilateral symmetry may be among actual large building plans.)
producing at the end all plans, of rectangular dissection form, which What is more, a statistical treatment offers the prospect of extending
conform to specified requirements of adjacency and dimension. the scope of this kind of treatment beyond the practical limits imposed on
It might with justice be argued, on the other hand, that this suggested exhaustive enumeration by the accelerating growth in combinatorial
form of classification of plans has a distinctly 'functionalist' flavour. It possibilities. It may not be feasible to catalogue all dissections, say, for
echoes the typical modern movement or 'design methods' approach, much larger values of n than 10 (although it might be possible to count
starting from a 'programme of requirements' stated as a list of rooms, their total number). But it is quite conceivable that general findings on
which are then related together by means of a 'functional diagram' or dissections up to this order could be extrapolated to larger values of n, or
'bubble diagram' (the graph), and finally given precise shapes and sizes. that a sampling approach might be used at higher levels.
In other approaches to design, or at other periods, the architect might It is worth pointing out one typical feature of dissections in connection
well have started with geometry, with some specific form , perhaps a with this problem of how to treat larger configurations. This is that certain
historically evolved 'type-form', and would have tried to fit the necessary large dissections are capable of being separated into two or more smaller
accommodation as best he might into this chosen shape. In the Beaux dissections, either at cut rectangles, or where internal walls run across the
Arts system of composition, the starting point was always a skeleton of whole width of arrangements (they can be cut as though by guillotine):
what were simultaneously circulation routes and axes of symmetry, around
which the rooms were then arranged to form the geometrical parti. Even
in a functionalist approach, it may be that certain dimensions and shapes
are necessarily fixed right from the beginning, as when a building is to be
fitted exactly onto a given site. In such a case the shape and size constraints
may well have their effects back on possible patterns of adjacency.
All this raises deep and difficult issues about how computer systems
intended to aid the designer in producing plans might be organised For larger values of n a great proportion of all dissections are of this
hierarchically, if they are not to be too constraining on the directions character.
136 Chapter 8 Properties and classification of rectangular arrangements 137
Thus we can identify a much smaller set of 'irreducible' or 'uncuttable' The rooms in a plan might not be simple rectangles, but could take L, U,
dissections in this sense, from which others can be made up by a process T, +, or more complicated shapes. Beyond this, the spaces in a plan
of aggregation (see Bloch, 1979a, chapter 7 ; Earl, 1980). For example, in might not even be completely enclosed, but only partially delimited with
the following figure all such dissections are given for n = 6, of which freestanding screen walls- as, for example, in plans characteristic of Mies
there are only four: van der Rohe's early work, or the architectural projects of the de Stijl
group (figure 8.12). •
There are no reasons in principle why such classes of plan might not be
enumerated by techniques similar to those described for generating
dissections-although the numbers of possibilities for given n ( or, say, for
given numbers of walls) would naturally, for the more general classes of
arrangement, be very much larger. Such enumerations remain to be made.
Earl ( 1980) though has proposed a classification which embraces all
What matters then is the pattern of adjacencies which occur across the these and other types of rectangular configuration, and which is based in
boundary where the dissections abut. Configurations with very large n the first place on the nature of the endpoints of walls in an arrangement.
might be described by decomposing them by the converse process into a Such an endpoint may be one of three types:
series of smaller dissections. (Every polyomino may be cut into smaller (a) not coincident with any other wall in the shape,
polyominoes; nor need the cuts be 'guillotine' cuts.) Another possibility (b) coincident with another wall at its endpoint (that is, at a 'two-way
is the hierarchical nesting of dissections, where a component rectangle of junction'),
a dissection at one scale, is itself made up from a complete rectangular (c) coincident with another wall not at its endpoint (that is, at a three-way
dissection at a smaller scale. junction).
In chapters 10 and 11 l will propose that the theoretical work described (Notice that at a four-way junction, neither of the walls involved can have
here on plan morphology has many applications in building science and in its endpoint).
architectural history. It may be worth noting now some seemingly Six types of shape can be defined in these terms, depending on whether
promising areas for further studies in 'pure morphology'. For example walls in those shapes have endpoints of type (a) only; of type (b) only
some of the analyses descri bed in this chapter could themselves be extended [it is impossible to have shapes with only endpoints of type (c)]; of types
or completed. In particular, the relationship of rectangular dissections to (a) and (b), (a) and (c), or (b) and (c); or of all three types (a), (b), and
their adjacency graphs remains to be explored, in the way in which Matela (c) together.
and O'Hare have done for polyominoes. Shapes with wall endpoints only of type (a) are close to the Mies van
An obvious and powerful objection to all the work presented so far, is der Rohe and de Stijl plans [although those illustrated in figure 8 .12 also
that even within the general limits of rectangular geometry, dissections and show some wall endpoints of types (b) and (c)] . A rectangular dissection
polyominoes represent only particular and restricted classes of designs; has wall endpoints all of type (c), with the exception of the four walls
and there are many real architectu,al plans which, though all their walls which form the outside perimeter and meet in points of type (b). Shapes
are straight and lie in one or other of two perpendicular directions, do with type (b) wall endpoints only, are like the outside boundaries of
not correspond to either of these types. There are plans, for example, polyominoes. A further subclassification is possible in terms of the types
corresponding to packings of rectangles with outside perimeters which of intersections of walls.
are not themselves simple rectangles (that is, they are not rectangular In the same paper Earl goes on to discuss the relations between some of
dissections), but which have internal three-way junctions (that is, they are the different graph-theoretic representations of rectangular arrangements,
not polyominoes). Some examples are which I have presented here as a rather heterogeneous range of possible
approaches, but which Earl is able to show can be connected together by
considering their underlying structures at a more general theoretical level.
So Earl relates together the forms of representation used variously by
Grason (1970c), Mitchell et al (1976), Flemming (1977), the 'electrical
network' representation of Brooks et al ( 1940) and as generalised to three
dimensions by Teague (1970), and his own work with March (March and
Earl, 1977).
138 Chapter 8 Properties and classification of rectangular arrangements 139
Exercises
8.1 Prove that Combes's formulae relating to numbers of external and
(a) internal wall segments in an arrangement, apply to polyominoes (see
page 127). (You can follow the general lines of the demonstration for
dissections on pages 114-115, but you will need to take account of the
numbers and types of junctions which can occur in the perimeter of a
polyomino.)
8.2 For the polyominoes illustrated, measure the value of March and
Matela's 'shape index', the value of their 'perimeter index', and the
distances between the pairs of cell centres indicated, as straight-line, taxi-
cab and graph distances in each case.
(b) • • • •
•
• • •
8. 3 Figure 8.10 shows the relationship of polyominoes for n = 4 to their
weak dual adjacency graphs. Explore the relation of rectangular dissections
to their weak dual adjacency graphs in a similar way, up to n = 6, using
the catalogue given in the appendix. You should be able to compile a
table showing the number of polyominoes and number of weak dual
adjacency graphs for each order, equivalent to that for polyominoes given
on page 131.
8.4 Using the catalogue of dissections in the appendix, find all 'irreducible'
or 'uncuttable' dissections (see page 135) for n = 7.
(c) (d)
Figure 8.12. Plans in which spaces are only partially delimited, by means of free-
standing screen walls: (a) Mies van der Rohe, brick villa project 1923; (b) Mies van
der Rohe, German Pavilion, International Exhibition, Barcelona 1929; (c) Theo van
Doesburg and C van Eesteren, private house project 1923, first floor plan; (d) Theo
van Doesburg, 'Counter-construction'-axonometric of private house project, Paris 1924.
Floor plan morphology in design 141
9
Floor plan morphology in design definition, and can then apply further criteria of his own for selection
within this range. Or else he can, as a result of seeing the possibilities, go
back and change the initial constraints so as to generate some different set.
"This work is strictly non-architectural, ... it has nothing to do with architecture." The earliest 'design methods' for plan layout were all of the first,
The Editor The Architects' Journal heuristic type. Many of these were concerned with producing plans in
which the amount of pedestrian traffic generated w,ould be a minimum.
N~merous efforts have been made over the last fifteen years to devise Thus for every pair of rooms or 'zones' in the building a figure would be
mathematical or computerised methods for generating architectural plans worked out- perhaps derived from surveys- for the typical frequency of
automatically. Some of this work has involved the kinds of graph-theoretical journeys made between those rooms per day or per week. In any actual
techniques and geometrical ideas which have been described in previous layout, this figure could be multiplied by the distance separating the rooms
chapters. Despite the strong criticisms which I expressed in chapter 1 of in question. Then the total of all such products (trip frequency times
the theoretical basis of much of this work, and in spite of reservations distance) could be summed for all pairs of rooms. The design methods
indeed as to its practical usefulness, I nevertheless feel that it will be were intended to find arrangements in which this sum was minimised.
helpful to describe some of it, if briefly, here. The reasons are that The typical form of representation used for plans, as mentioned in
certain of the methods introduce new geometrical techniques for descrioing chapter 2, was a regular square grid of fixed grid dimension. Thus
plans; or else they throw light on how various morphological issues of a dimension was not separated from shape description, and all rooms had to
general kind might be investigated. be approximated both in shape and in area as a number of unit grid cells.
There is a key distinction to be made right at the start. A method Difficulties arose in the measurement of distances between cells or rooms,
might be intended to generate just one, or a few plans, in which certain either as Euclidean, taxicab, or possibly graph distances, as described in
stated requirements of adjacency between rooms, and perhaps also the last chapter. The question of fixing values for the frequency of trips
constraints on the dimensions and shapes of rooms, are satisfied. Such between rooms was also problematic, since it is quite plausible to suppose
methods are typically intended to mimic the human designer in the that this frequency might itself vary to some extent with the distance
processes which he might be assumed to go through in producing a layout. separating the rooms. The algorithms for generating layouts consisted in
In computing language they are heuristic methods, in which strategies are effect of techniques for packing together polyominoes (the rooms), or of
employed which are believed to lead towards some desired result, but are rearranging those packings in such a way as to reduce the value of the
by no means guaranteed to do so. circulation measure.
On the other hand methods might be devised which would produce all Such methods have been discussed and criticised very widely elsewhere,
possible plans (in the sense which will have become clear from previous and so I do not propose to embark on any more detailed account.
chapters) conforming to the given requirements. These can be termed Comprehensive reviews of the subject have been made by Tabor ( 1970),
exhaustive methods. by Eastman (1975), and by Mitchell (1977, chapter 13). The largest
The distinction is crucial, since in the first case the method or the objection to these methods centres on the fact that they attempted to
computer program is itself effectively exercising choice, beyond the constrain the process of 'composition' of a plan by means of one single
explicit specifications laid down by the user. The criteria for this choice evaluative criterion of functional performance- that of circulation- and to
are built into the way the method is structured, either deliberately or else optimise the arrangement (in a mathematical sense) on that basis alone.
more or less fortuitously, as, for example, when the first layout which is Meanwhile, other workers had come to realise that at a smaller scale the
found, conforming to the given specifications, is presented as 'the solution'. absolute distances between rooms are not of such very great functional
These criteria may be hidden from the architect who is supposed to make significance compared with the patterns of adjacency between rooms, and
use of the technique; and in any case will very likely not correspond to the that these adjacencies could be represented by graphs. The topology of the
criteria which he himself would apply, which would come out of a much plan might be manipulated by means of graphs, more or Jess independently
broader set of considerations- to do with aesthetics, structure, heating, of its shape and dimensions.
lighting, or whatever-quite outside the scope of the specific method. The first published application of graph theory to small architectural
In an exhaustive approach the situation is very different, since there is no plans was made by Levin (1964). Levin confines his attention to the
choice exercised by the computer program or the technique itself, beyond access graphs of plans, where the vertices represent rooms and the edges
the constraints which are fixed initially. The architect (or theoretician) is the existence of doors between those rooms. He discusses the possibility
presented with the entire set of feasible alternatives under the specified of enumerating all those planar connected graphs which could represent
142 Chapter 9 Floor plan morphology in design 143
the access graphs of plans. In the first place he considers only outerplanar This method of Levin's, since it aims to optimise the plan on the basis
graphs. An outerplanar graph (see chapter 7, pages 90 and 91) is a graph of a measure of pedestrian movement, is open to all the criticisms made of
which possesses at least one and possibly several embeddings such that similar circulation-minimising techniques. Nevertheless it points the way
every vertex lies on the exterior face. Such embeddings therefore represent to some exhaustive enumeration methods which do not have the same
plans with no interior rooms. There is a finite number of ways, as Levin limitations. And in general a great deal of subsequent research in
goes on to show, in which n labels, signifying different functions for the applications of graph theory can be seen as extensioll$ to or mechanisations
rooms, might be assigned to then vertices of each outerplanar graph. He of this original work of Levin's.
enumerates all such graphs and their possible labellings up to n = 4. [His Two authors following Levin's lead were Cousin, and Friedman. Cousin
enumeration thus corresponds to that of Korf ( 1977), and of March and (1970) describes some of the general ideas of the graph-theoretic treatment
Earl (1977) (compare figure 7.1), but for outerplanar graphs only. Also of plans, but he does not suggest the idea of an enumeration of possible
the interpretation of the graphs is different- in Levin's case as access graphs or plans, nor does he propose any explicit design method.
graphs, in the other authors' case as adjacency graphs. I Friedman (197 5) introduces a somewhat eccentric and erratic series of
Levin mentions the problem that planar graphs in general may have graph-theoretic ideas about the planning of flats into a much larger work
more than one embedding. He also introduces the subject of nonplanarity, on 'scientific' architecture and city planning. His is an exhaustive
and shows that certain access graphs (specifically K 5 ) are not capable of approach: exhaustive with a vengeance. Like Levin he considers the access
being realised in plans, and that in such cases one or more of the access graphs of plans, with the vertices labelled to signify the functions of rooms.
requirements must be dropped. In addition Friedman allows the possibility that each room might
Levin in this early paper thus adumbrates an exhaustive approach to the (independently of all others) take a series of geometric shapes, might be
generation of small plans, or at least their graphs; although it should be · oriented to face different directions, and might contain different types of
said that he himself expresses doubt as to whether an enumeration of equipment such as plumbing fixtures or furniture. He envisages a computer
access graphs beyond n = 5 would serve any practical purpose. Also he system-the 'flatwriter'- which would allow the user- not an architect,
deals only with graphs, and does not make any systematic approach to the but the prospective inhabitant of the flat- to permute together a set of
geometry of plans. such options in all combinations, to give what Friedman enthusiastically
In the second half of his paper, however, Levin goes on to describe a recognises would be a truly astronomic number of alternative 'designs'
design method based on circulation criteria which is much closer to the (figure 9.1).
heuristic techniques mentioned earlier. The method consists in building Of course, the procedure completely ignores the fact that many,
up an access graph edge by edge. Those edges are placed first which probably the great majority, of these combinations would be incompatible,
correspond to pairs of rooms between which the frequency of trips is because the arbitrarily chosen shapes and sizes of the rooms would not
greatest. Edges are added in descending order of their trip-frequency pack together geometrically, the required orientations would be impossible
values, and planarity is preserved at the later stages by judiciously omitting to achieve together with the adjacency requirements, and so on. The only
edges with low values. constraint which Friedman allows is that embodied by the access graph,
The rationale for this procedure is, clearly, that rooms between which and the recognition that it might be nonplanar.
trips are frequent, should be made directly accessible one from another, The user of the flatwriter would in theory be helped in his choice from
but that this is less important where the traffic is infrequent. The this vast range, by being provided with a figure for each plan, for what is
assumption has difficulties, ~ince if a corridor or other circulation space essentially its circulation 'cost' (Friedman calls it an 'effort value').
were introduced rooms could still perhaps be made very close though not A set of trip frequencies is specified, not just between adjacent rooms,
adjacent; and many of the values for trip frequencies between pairs or but between all pairs of rooms. For a given access graph the shortest path
rooms would alter if the new corridor were included as a 'room'. between each pair of rooms is calculated as a number of edges, and this
It is possible that the final resulting access graph might have several number is multiplied by the corresponding trip frequency. All such
planar embeddings. Levin argues that the designer can choose between products are then summed to give the total 'cost' of the plan.
these on the basis of other considerations. For example, he may want to Quite apart from any doubts as to whether this particular measure
put rooms which require natural lighting on the perimeter of the plan. provides any realistically useful basis for evaluating plans, the main
Again, the relative sizes of rooms may mean that certain embeddings of criticism to be made of Friedman's suggested approach concerns the way
the graph are impossible, such as when a cycle of vertices representing in which huge numbers of combinations of plan elements are generated
small rooms encloses a vertex or vertices representing very large rooms. blindly with very little account taken, beyond the access graph, of the
144 Chapter 9 Floor plan morphology in design 145
internal relations between those elements< 11>. The flatwriter is a kind of particular context. They consider adjacency graphs and their dual
architectural counterpart to the proverbial monkeys typing Shakespeare, or relationship with plan graphs. Their main attention is devoted, however,
the Lullian combinatorial machines for literary composition which Gulliver to the possibility that a requirement graph might not be planar, that is,
found at the Academy of Lagado. Other workers, as we shall see, have given that the problem is overconstrained and adjacency graphs are to be sought
more consideration to the structure of the geometrical problem, and in as subgraphs of the requirement graph.
this way have placed severe restrictions on the generation of combinations, They propose firs t making a test for planarity< 18>. • If the requirement
so as to bring an exhaustive enumeration of plans within manageable bounds. graph fails this test, they then go through a process of identifying some
Several authors have come to the study of spatial layout problems out minimum ' resolving' set of edges, whose omission will make the graph
of a management science or operations research, rather than a properly planar. The method which they use here applies only to graphs containing
architectural interest. They have been concerned with the arrangement of one or more Hamiltonian cycles. (To make a formal test for the presence
machines, equipment, or industrial activities in factories. Seppanen and of Hamiltonian cycles as such- which this method does not attempt-
Moore ( 1970) were among the first to propose the use of graphs in this would be difficult.)
Remember that a cycle is a continuous sequence of vertices and edges
l'I TTIN4 which does not double back on or intersect itself, and which returns to
(a)
-~
<:>) .._
.. ..,_,1, ..
-·Jl&ll. Ml l'. *TI. 'YiL ca,.100.
f§J§[Q]~[[]EJ BB~[] □
I I I • I I
IB ➔ N
(b) (c)
LOIIDOII ,
MUSKED UO SOLD ll'KOlESAlt II JOH" IAQU[S AND SON, IOI KATTOII GAIDEN;
Figure 9 .1. The 'flatwriter' computer system for generating floor plans, proposed by AJID fO II WAD U M0IT or,., u ..O 111c; , ., .CT IUOIIT0llU
Friedman ( 1975). It is imagined that the user would select out of a variety of options TKlOICNOWT Tift ClflOl>ONt
(a) for room arrangement, room shape, positions of equipment , orientation, etc some (a) (b)
set of particular specifications (b) for his desired layout. The program would then
generate a plan (c) or plans, by permuting these elemen ts together. Figure 9.2. The 'icosian game' devised by the Irish mathematician Sir William Hamilton.
The problem is to find a sequence of edges in the graph illustrated, which passes
0 7) Readers should also be warned that Friedman's arithmetic is unreliable. His through every vertex once, and ret urns to the original vertex (a Hamiltonian cycle).
calculation of the number of possible combinations of a hypothetical set of options The game was made in two versions: in the form (a) of a wooden dodecahedron, and
offered by the flatwriter (Friedman, 1975, page 36) is wrong by a factor of six. And ( b) a projection of the same solid onto the plane.
his example of the frequency with which 'Mr X' moves between the rooms of his
house (page 4 I) implies that the average number of times the man enters a room per (IS) Actually this is strictly unnecessary to their purposes, since the procedure which
day is not equal to the number of times he leaves that same room. follows, itself acts effectively as a planarity test.
146 Chapter 9 Floor plan morphology in design 147
(Unlike some of the rather disappointing graph theory puzzles mentioned A second pair of authors, Lepine and Nakajima ( 1973), took over this
in chapter 6, this one is actually soluble.) work of Seppanen and Moore, to apply it to architectural design, and
It is possible for a graph both to be nonplanar, and not to contain a suggested some elaboration of the edge-elimination algorithm, by giving
Hamiltonian cycle, in which case the method described by Seppanen and weights to all the edges in the requirement graph to express their relative
Moore cannot be used. If the assumption is made, however, that a functional importance. They posed the problem then of finding that
Hamiltonian cycle does exist, then it can be embedded so as to divide the 'resolving' set of edges whose omission would make the graph planar, and
plane into an interior and an exterior region. The remaining edges of the meanwhile whose total value was a minimum.
graph can then be reintroduced, and embedded either inside or outside the Another worker who used similar techniques to attack problems of
Hamiltonian cycle. Seppanen and Moore describe an algorithm for industrial plant layout was Krejcirik ( 1969a; 1969b). His was a heuristic
identifying incompatibilities between these embeddings of edges, and, method designed to produce one optimal layout satisfying both adjacency
further, for identifying sets of edges whose removal will make the graph and dimensional criteria.
planar. In Krejcink's method the edges of the requirement graph are again
Figure 9.3 illustrates the basic idea with a simple example of a graph on weighted to represent their relative importance. It is assumed that the
six vertices, which possesses a Hamiltonian cycle drawn here in regular final plan will consist of rectangular spaces packed into a rectangular shape
hexagonal form. The interior of this hexagon can be 'triangulated' with overall. So four exterior regions are defined on the 'north' 'east' 'south'
additional edges, and the exterior face can also be 'triangulated', up to the and 'west' sides, and requirements for adjacencies of spaces' to th~se '
point where the graph becomes maximal planar. Any edges which still regions can be introduced.
remain to be placed must be incompatible with those already introduced. Krejcirik's next step is to look for Hamiltonian cycles in the requirement
(In this case there remain only three positions in which edges might be graph. (If no such cycle exists the method is presumably inapplicable.)
added. An edge in one of these positions gives K 5 , and an edge in either That cycle is chosen, for which the total weight of the edges which it
of the other two positions gives K 3 3 .) contains, is greater than in any other Hamiltonian cycle. Meanwhile, the
In a graph with more than one Hamiltonian cycle, Seppanen and order in which the rooms occur around the cycle must be compatible with
Moore's method may find different sets of incompatible edges for different the requirements for their adjacency to the exterior regions.
cycles. So all cases would have to be investigated to find the overall The idea here, speaking rather loosely, is to find a way in which all the
minimum 'resolving' set. rooms can be placed in a 'ring' (the Hamiltonian cycle), which can then be
Seppanen and Moore recognise that there are further steps, from a set around the perimeter of the plan so as to satisfy the requirements for
planar adjacency graph to a finished plan layout, but make no proposals adjacencies of rooms to the exterior. Meanwhile, the order of the rooms
as to how these might be carried out. on the 'ring' is such that adjacency requirements of great functional
importance between them are satisfied. Where rooms are not required to
be on the perimeter, the 'ring' can be folded back on itself, so that the
rooms in question are placed in the interior of the plan. Figure 9.4 gives
a notional illustration.
Now remaining edges representing further requirements of adjacency are
added back in where possible, in descending order of value. (The most
important requirements are satisfied first.) These edges can be added
back, up to the point, in principle, where the region inside the embedded
Hamiltonian cycle is completely triangulated; and to the point where the
region outside the cycle is also as completely triangulated as is compatible
with the adjacencies to the north, east, south, and west points. Any edges
remaining at this stage must be omitted from the graph.
(a) (b) A test is applied to the resulting embedding of the requirement graph,
Figure 9.3. (a) A graph on six vertices containing a hexagonal Hamiltonian cycle to detect the presence of subgraphs which cannot be realised in plans
(marked in heavier line). The interior and the ex terior of the hexagon can be triangulated made up of rectangular rooms ( compare the figure on page 100). Where
with additional edges, up to the point where the graph becomes maximal planar. Only such subgraphs occur, appropriate adjustments are made, presumably by a
three positions then remain (b) for the addition of edges (dotted lines), yielding in one suitable rearrangement of edges.
case Ks (left) and in the other two cases K 3, 3 .
148 Chapter 9 Floor plan morphology in design 149
Krejcirik's method then proceeds to take account of dimensional adjacency requirements for small house plans which turn out to comprise
requirements, and finally produces a complete dimensioned layout. We will K 5 or K3, 3 (as in the example of figure 6.6). Much more often the process
come back to this part of his work shortly. Meanwhile it should be said of moving from requirements to an adjacency graph of a feasible plan will
that Krejcirfk's approach can once again be criticised on the same grounds involve the addition of edges rather than their subtraction; although
as all such 'optimising' methods. (Nor does his program necessarily produce obviously these new edges must be added in ways such that planarity is
the optimum arrangement, even in its own terms, although it is claimed to always preserved.
come near this.) What is more, it is not directly applicable to certain This point was well recognised by Grason (1968 ; i 970a; 1970b; 1970c),
kinds of adjacency requirement graphs (those without Hamiltonian cycles). whose computer method for producing small plans conforming to adjacency
A more serious and general criticism of many of these authors- Levin, and dimensional requirements is among the most sophisticated of those
Cousin, Seppanen and Moore, KrejciHk- is that they really fail to make devised and implemented to date. Grason confined his attention to
the distinction, which I emphasised in chapter 6, between an adjacency rectangular plans made up of rectangular rooms, that is, rectangular
requirement graph, and the final adjacency graph of some plan in which dissections. His was the first method to enumerate exhaustively all
those requirements are met. Or at least, they imply that typically the solutions to such problems.
designer will specify a large number of adjacency requirements, sufficient The representation which he uses throughout to manipulate arrangements,
to make up a triangulated or maximal planar graph; and that the plan is an augmented dual adjacency graph in which the edges are coloured and
graph can then be derived directly from an embedding of this adjacency directed to represent the orientation of the corresponding walls in the plan,
requirement graph, through the dual relationship. either 'north-south' or 'east-west', as described in chapter 7 (compare
Thus they overlook the strong possibility that the requirement graph figure 7 .8). Weights are then placed on the edges to represent the lengths of
may only be a spanning subgraph of the final adjacency graph, and that the wall segments (compare figure 7.9). A complete dimensioned plan
additional edges might be introduced into the embedded requirement would thus be represented in Grason's system by an adjacency graph in
graph, simply to represent adjacencies arising through close-packing. Their which the edges are coloured, directed, and weighted as in figure 9 .5.
concern is on the contrary devoted to the possibility that the requirement
graph will contain too many edges, that is, it is nonplanar and the problem
is overconstrained, so that certain edges must be eliminated.
One might question whether the subject of the planarity of requirement /
/
graphs really deserves all this attention, however. If one considers typical, /
/
/
realistic sets of adjacency requirements for architectural plans, it is
arguably rather rare that they form graphs which are nonplanar. It takes //,,, I \ ' ,
n n
w
;..~.............-:y··
o,....
' .... "••·
'
3
····)
t /t'--
- -41-
\
I
2')
1-:. 1
.~.1 7>•~....... ..
1
~,...r,._ ...__-----,'
13
}'>, .. ···•...~.:-::--
7 ........... ,oe
3~······
1
.... , .. •:,
,
',,
,, ···············\///
•~. . ·. s> ......o'············1,,, //
w e ' - _ __,..,_~__...,,_r_ - -
I /
~
w
' \ I /
',, ~ ~ ,,✓
'' \ I /,,,
' \ I ,,,
'- \ I /
(a) (b) (c) '
'o, \ I ,,,
Figure 9.4. Notional illustration of the operation of a computer program for industrial
plant layout due to Krejcirik (1969a ; 1969b). A Hamiltonian cycle is found (a) in the Figure 9.5. Representation of a plan of rectangular dissection form in the computer
adjacency requirement graph. Further required adjacencies are added in (b), in system for plan generation due to Grason (1970c). The edges of the augmented dual
descending order of importance, up to the point where the interior of the cycle is adjacency graph are coloured ( dotted or broken lines) according as to whether they
completely triangulated and the exterior of the cycle is also triangulated so far as is represent adjacencies across north-south or east-west walls, respectively. Edges are
compatible with required adjacencies to exterior regions. A plan is derived (c) as the directed as they cross corresponding wall segments from south to north or from west
dual of this adjacency graph. to east, and weighted to represent the lengths of these wall segments.
Chapter 9 Floor plan morphology in design 151
150
Notice that the external walls of the plan running east-west are extended Each of these embeddings is tested for what Grason has called its 'well-
by convention beyond the four corners, and the adjacencies of the four formedness'. If an embedding is 'well-formed', this means it is a subgraph
exterior regions at north, south, east, and west are shown across these of an adjacency graph or graphs which in their turn are capable of having
extended walls: colours and directions assigned to their edges in such a way as to represent
a rectangular packing of rectangular rooms. The tests for well-formedness
0
'' apply both to the vertices and to the faces of the embedded graph. A
, /
,'T
, I
, ',,
/
(at least four) edges incident with that vertex, which correspond to the
segments of wall making up the four sides of that room. We saw in
o, ,o
' /
chapter 7 (compare the third figure on page I 03) how this implies that
' /
the edges in question must be capable of being grouped into four groups,
'' /
/
'' /
/ such that in each group all edges are coloured the same, and adjacent
'0 /
groups are coloured differently. Suppose that, as before, the two 'colours'
Grason has a special 'graph grammar' which is designed for carrying out are indicated by dotted and broken lines. Dotted edges in the adjacency
operations on graphs. The significance of this grammar is that it represents graph correspond to walls running north-south in the plan, and broken
the graph in an hierarchical manner, corresponding to the connectedness edges to walls running west-east. The edges are directed, as they cross
of its components. So the representation distinguishes in effect the the walls from west to east, and from south to north. It follows that the
!-connected, 2-connected, and 3-connected components of the graph, and edges incident with any room vertex must be arranged in their four groups
identifies the cut vertices or pairs of 'hinge' vertices (Grason calls them in order, starting from the west side, and reading clockwise: dotted,
'pivot pairs') by which these components are connected together (co~pare inwards; broken, outwards; dotted, outwards; broken, inwards:
chapter 7, pages 86-87). (Grason allows the possibility that a requrrement \ I
graph as a whole might not actually be connected.) ·•...11\. 7'
The first application of the grammar is in simply putting together the ~-., \ I
·o-....), ....
set of edges corresponding to the required adjacencies between rooms, to ...::;··~
form the adjacency requirement graph itself. This is done edge by edge; I
I
and because of the hierarchical nature of the representation in terms of
connectedness, it is possible to test without undue complication, at the Going back now to the embedded requirement graph: any 'well-formed'
insertion of each successive edge, whether or not planarity is preserved. room vertex must either fulfill the above conditions, or else it must be
With the assumption that, with all edges inserted, the resulting such that the conditions can still be satisfied when further edges are
requirement graph is planar, Grason's next step is to produce, again with added, in completing the adjacency graph.
the aid of the graph grammar, all possible embeddings of this graph. If the The conditions for well-formedness of faces are of an equivalent kind.
graph as a whole is disconnected, or if it is only I-connected, it is first In the completed adjacency graph the faces can only be of two kinds:
modified, by 'anchoring' and 'bracing' the disconnected or !-connected triangles (corresponding to three-way junctions in the plan) or quadrangles
components, in such a way that the graph is made 2-connected. The g'.aph (corresponding to four-way junctions). Again there are conditions governing
grammar then acts to produce systematically all embeddings, by choosing permissible colourings and directions for the edges of these faces, as
different faces for the exterior face , and by folding and twisting the indicated in chapter 7 (first and second figures on page l 03). Any face in
2-connected components about their 'hinge' vertices into different faces, in an embedded requirement graph must either be of one of these two types,
all permutations, as described here in chapter 7 (pages 89-91 ). or else it must be capable of being ' filled' with such faces.
Notice that the 'frame' consisting of the adjacencies between the four Grason shows how these conditions can be satisfied by setting conditions
exterior regions in the figure shown above is fixed for all embeddings, and on permissible sequences of edges in the boundary of any face. If one
that the remainder of the graph must obviously be embedded inside this imagines oneself taking a walk around the cycle of edges around a face,
frame. If then there are adjacencies specified between rooms and the then each vertex will potentially represent a 'turn' in that walk, depending
exterior regions, the possibilities for embedding are correspondingly on how the edges are coloured and directed. The only circumstance in
restricted by the fact that the remainder of the graph is 'tied' by these which there is no 'turn' is when two successive edges share the same colour
edges to the frame.
Floor plan morphology in design 153
I"-
"'
-
.,_
"'
and the same direction:
m
" A
···7····0•••;)>-•••·
N
-) - o- ) - A
I
.,_ I"-
"'
"' But two successive edges of the same colour and differently directed, must
correspond to a 180° turn in the plan:
M
" R
··-->-··-0···<··· I \
M
"'
.,_
I"-
-
"'
- )-o - (-
*'
And two successive edges of different colours must represent a 'turn'
through 90° , which depends on the direction and colours of the edges:
A A
M
"
•··}•··O - 7 -
···<··-0 -)-
· ·>EI G·>·
M ···)···-0 -(- · ·->EJ @·>·
"'
I"- ···<···O -(- A
I
A
I
.,_
- "'
Grason defines certain sequences of 'turns' which are inadmissible in the
cycle of edges around a face, if it is to be capable of being realised in a
rectangular plan.
" These conditions for well-formedness are applied as the embedded
requirement graph is built up; and all embeddings which meet the tests
are retained for the next part of the process, which completes possible
adjacency graphs, and simultaneously assigns dimensions. Figures 9.6
and 9.7 illustrate a worked example given by Grason, showing a set of
adjacency requirements between five rooms (table 9.1 ), and an embedding
of the resulting requirement graph [figure 9.6(a)] . See how the colours
" (dotted or broken) and directions of edges representing adjacencies to the
M
- N
Table 9.1. Adjacency and dimensional requirements for Grason's worked example.
M I"-
.,_ "'
Room Required adjacencies Allowed sizes
C - "'
a tow, s, E 6 X 3, 4 X 3
(3 to n, 6, e 4 X 3, 3 X 3
" to e, s, 6, E 3 X 3, 3 X 2
'Y
6 to (3, 'Y 3 X 2, 3 X 1
€ to a, (3, "f, s 6 x l,3xl
Chapter 9 Floor plan morpholog y in design 155
154
illustrate d in figure 9.6(d). There are only four other possible solutions to
exterior regions are necessarily fixed at this stage, and that this in turn
this particular problem, and they are illustrated in figure 9.6(e).
implies the colour and direction of other 'internal' adjacencies; but that
certain edges (solid lines) still remain uncolour ed and undirected. Grason's program deserves recognition as the first successful method for
generating this type of plan automati cally and exhaustively. Without
Dimensional constrain ts in Grason's system are specified in terms of a
denigrating that achievement, there is a serious criticism to be made
number of precise alternative sizes which each room may take. Thus in
the worked example illustrated, room a can take either the size 6 x 3 however; indeed it is a drawback which Grason h,imself was quick t~
or the size 4 x 3 (in metres). Possible sizes for other rooms in the example point out. This is that by computin g standards , the program is very slow.
are listed in table 9. 1. It is as though the rooms were jigsaw pieces of For instance the worked example shown here required 23 minutes of
fixed shape and area. (The orientatio n of nonsquar e rooms is not computin g time on an IBM 360/ 67, to produce the five possible plans
specified, and they may be placed with the Jong dimension either north- with five rooms as illustrated. In other tests, the program took still larger
south or east-wes t.) In other design methods, as we shall see, dimensional amounts of time- I 00 minutes for a six room plan, 210 minutes for eight
constrain ts are expressed rather as maximum and minimum values for rooms- without in either of these cases completing an exhaustive search.
Grason suggested that the program might be rewritten to be more efficient.
lengths, widths, or areas of rooms.
There is now a further series of tests for well-formedness applied to However, it seems that much of the reason for its slowness is intrinsic, to
the embedde d requirem ent graph, relating to the consistency of the do with the complex alternating sequences of constructive steps, and tests,
which are needed to exhaust all possibilities.
dimensioning process, in which weights are attached to existing edges and
new weighted edges are introduced to complete the adjacency graph. A Another minor shortcom ing is the fact that the method accepts
first type of test applies to the edges incident with each vertex, and dimensional constrain ts, as we have seen, only as fixed sizes on rooms,
where more realistically it might be imagined that architect s would be
requires that the 'Kirchho ff law' is obeyed for the sums of weights on
prepared to allow any dimension between certain limits. For example in
those edges, as explain~d in chapter 7 (compare figure 7 .9). Thus the
total value of all inward-directed 'dotted' edges at a vertex must equal Grason's illustration, room a must be either 6 x 3 or 4 x 3; where perhaps
the total value of all outward-directed 'dotted' edges; and similarly for feasible plans exist, besides those produced , in which this room takes size
'broken' edges. A second type of test relates to the plan dimensions
5 x 3 or even 4 x 4 which lies between the same effective area limits.
overall , and checks that, for example, the sum of all areas assigned to Grason's own reaction to the slowness of his program was an inclination
rooms is compatible with the area of the plan as a whole, and that the to turn from an exhaustive approach , which he imagined had only a
sum of dimensions assigned to rooms across the plan in one direction is research and not any practical interest, to heuristic techniques. However,
compatible with the total width of the plan in that direction. methods have since been devised which are still exhaustive and are capable
Edges are added one at a time to the embedde d requirem ent graph, so of producing similar plans with at least seven or eight rooms, in radically
as to fill its faces up to the point where all faces are simple triangles or reduced amounts of computin g time. Before describing these systems, I
quadrang les; and tests for well-formedness as described are carried out at should mention several other authors whose work is comparable with
Grason's in various respects.
each step. This is done in all possible ways, and the corresponding
dimensio ned plans are then derived from these complete d adjacency graphs. The approach taken by Krej cirik (1969a; I 969b) to the dimensioning of
Figure 9.6(b) gives a 'snapshot ' view of an intermed iate stage in the rectangular plans was to use, like Grason, a represent ation consisting of an
completi on of one adjacency graph for Grason 's worked example, and adjacency graph with coloured , directed, and weighted edges. Dimensional
figure 9.6(c) gives the complete d graph. The corresponding plan is constrain ts in Krejcirik's method are stated as maxima and minima on the
lengths and widths of rooms, and dimensions are allowed to take any
modular value between these limits. Krejcirik's method involves testing
9 3
the adjacency graph for 'well-formedness' in terms of the colouring and
direction of the edges incident with each vertex.
Figure 9. 7. An example of a dimensioned plan produced However, there are no tests equivalent to Grason's for 'well-formedness'
by the computer program of KrejciHk (1969a; 1969b), in terms of the weights on edges (that is, the Kirchhof f law). Instead the
in which adjacency requiremen ts are satisfied, and program merely assigns weights to edges in such a way as to represent
room and plan dimensions lie within specified maxima rooms which are adjacent in the required relationships, and have acceptable
and minima, but the rooms are not close-packed and
sizes, but which are not necessarily close-packed (figure 9. 7). Within the
the plan contains 'holes' (shaded). It must be adjusted
by hand to give the final arrangeme nt. given dimensional constrain ts on rooms the program derives a plan whose
Chapter 9 Floor plan morphology in design 157
156
overall east-wes t dimension does not exceed some stated maximum , and cycle repeatedl y through possible combinat ions of these values, according
whose north-so uth dimension is minimised ; but this may be a plan which to essentially the same criteria for dimensional 'well-formedness' as
contains several 'holes'. The idea is that this arrangement can then be Grason's (the Kirchhof f law), until all dimensionally feasible solutions are
adjusted by hand to give the final layout. exhaustively generated. As with Pereira's method, 'dimensionless' plans of
A group of workers led by Pereira (LNEC, 1972; Pereira, 1974) dissection form are derived to meet the adjacency requirem ents and are
then dimensioned. '
developed a program which is in certain respects very close to Grason's. is that adjacency
The represent ation used is the same type of coloured , directed, weighted Tl~e distinctive feature of Grason's approach by c~ntrast
requITem ents are put together to specify only a partially complete d
adjacency graph. A requirem ent graph is tested for planarity , its different s of some edges left
adjacency graph, with the colours and direction
embeddings are produced , and each of these is coloured according to
criteria similar to those expressed in Grason 's tests for 'well-formedness', undecided; and then additiona l edges are inserted to complete the graph,
to yield all possible corresponding plans of rectangular dissection type. simultaneously with the attempt to satisfy the dimensional constrain ts.
However, it is not clear from the published accounts exactly how
Exhaustive enumerat ion of dimensioned plans
embeddings are generated for 2-connec ted requirem ent graphs (this was
one of the principal functions of Grason's 'graph grammar'); and details So far the design methods described have all been of what might be called
are described only of the embeddin g of 3-connected graphs, by the choice a 'construc tive' character. Graphs, or plans, are built up piece by piece so
of different faces for the exterior face. as to satisfy the given requirements of adjacency and dimensio n. A process
As for the dimensioning part of the system of Pereira and his colleagues, which is in some ways the opposite of this- a 'selective' approach -was
the dimensional constraints are expressed as maxima and minima on the suggested by myself (Steadman , 1973), and implemen ted as a compute r
lengths and widths of rooms; and they discuss the general theory of system with Mitchell and Liggett (M itchell et al, 1976).
assigning colours and weights to the adjacency graph of a rectangular The essential idea behind our method is that if it is decided that a plan
dissection , under conditions for ' well-formedness'. However , it seems that be of rectangular dissection form, and is to contain up to say seven or
their program as actually implemen ted was again not designed to produce eight rooms, then a technique for enumerat ing dissections exhaustively (in
close-pac ked plans; and that the operator of the program was to be shown our case the technique described in chap ter 4) will in effect provide all
partial solutions - with overlapping rooms, and holes-and allowed to potential 'dimensio nless' solutions in advance. The dissection s may simply
interact with the program to modify the dimensio nal constraints, so as to be stored in a computer file, and a set of adjacency and dimensio nal
produce finished plans. Thus the system cannot be claimed to provide requirements may then be tested against all of these possibilities in turn.
exhaustive solutions under stated sets of adjacency and dimensional It is a 'brute force' method, and admittedly lacks a certain intellectual
constraints, in the same way as Grason's. elegance on this count, but it has the virtue of being highly effective.
Another author who developed a compute r program for generating small The problem of satisfying a set of adjacency requirements now becomes
plans of rectangul ar dissection type was Gilleard ( 1978; 1980). Gilleard's a matter of taking the adjacency graph of each dissection in turn, and
method for completing a set of adjacency requirem ents so as to give a testing to find all the possible ways (if any) in which the requirement
'well-form ed' embedded adjacency graph differs somewha t from Grason's. graph can form a labelled spanning subgraph of the adjacency graph.
He considers in the first place all ways in which additiona l edges can be Suppose the rectangles in a specific dissection (the rooms) are labelled
added in, up to the point at which the graph becomes a triangulation of a, b, c, ... and the vertices of the requirem ent graph (the functions or
the quadrilateral represented by the adjacencies of the four exterior regions. room names) are labelled a , {3, 'Y, .... We want to know in what permutat ions
Those triangulations which cannot correspon d to packings of rectangles, a, /3, 'Y ... can be matched with, or assigned to a, b , c, ... . In the absence
because of subgraphs which are not 'well-form ed', are then rejected; and of any adjacency requirem ents (that is, if the 'requirem ent graph' consisted
quadrilat eral faces- corresponding to four-way junctions in the pla~- are j ust of n isolated vertices) we would need to consider all permutat ions of
produced by the omission of suitable edges from the triangulations, again assignme nt of a, {3, 'Y ... to a, b, c, ... . The number of these pennutat ions
in all permissib le 'well-formed' ways. The resulting embedde d adjacency would be factorial n, conventio nally symbolised as n ! , a figure which grows
graphs are converted to all correspon ding rectangular plans by deriving their grows rapidly with increasing n <19 >.
duals (the plan graphs) and colouring these plan graphs directly. (The
process is perfectly equivalen t to Grason's colouring of the adjacency graph.)
19 I
< > Factorial n is given by II x (n - J) x ... x 2 x I. Vaueso f n.' for n = 1, ... , 8 are:
Gilleard's approach to dimensioning is to set a range of integer values n= l 2 3 4 5 6 7 8
for the lengths and widths of each room, and then to have the program n! = I 2 6 24 120 720 5040 40 320
158
Chapter 9 T Floor plan morphology in design 159
However, this theoretical maximum number of assignments becomes The computing time here is a function both of the number of rooms n
progressively reduced, as adjacency requirements are introduced, and and of the strictness of the adjacency requirements. (With fewer '
the requirement graph acquires structure. For example, a vertex in the requirements set, the process takes longer.) With typical and realistic sets
requirement graph can only be assigned to a vertex in the adjacency graph of requirements, assignments to six rooms can be made in a few seconds
of a particular dissection which is of equal or higher valency. This condition with an IBM 360/91. For seven rooms, less than a minute is generally
forms the basis for a test which is applied in our assignment procedure. required , whereas the eight-room case often needs several minutes
The valencies of vertices in the adjacency graph of the dissection are listed in Figure 9.8 illustrates a worked example, which u~es the system ;o design
descending order of magnitude, and similarly for those in the requirement a small apartment of seven rooms. Figure 9.8(a) shows the graph of
graph. The two lists are then compared, and if any value in the second required adjacencies (edges drawn as solid lines) including adjacencies to
list exceeds the corresponding value in the first list, this is sufficient to the exterior regions; and a second graph of prohibited adjacencies (edges
establish that no assignment can exist. An example is drawn as broken lines). Successful assignments of room names to rooms
in dissections in this case resulted in 504 different 'dimensionless' plans,
dissection b f d g a C h
/3 of which a small sample is illustrated in figure 9.8(b). Notice that it is
ml :5t::
valencies ® 4 3 3 2 2 2
assumed that the apartment is entered from the north side and is flanked
C requirements 6 Cl •t /3 'Y T)
i (3) 2 2 2 1 1 1 by neighbouring apartments on east and west. '
valencies
g It is interesting to compare this number of 504 with the maximum
h 0 indicates values being compared number of potential assignments in the absence of any adjacency constraints.
t
If this test is passed, the program proceeds to make assignments of the To get that number we must take 7! (= 5040) assignments for each
requirement graph into the dissection. No vertex in the requirement graph isomorph of each dissection; multiply by eight for the number of isomorphs
is assigned to a vertex in the adjacency graph of the dissection with lower (this is an overestimate, since some dissections will have symmetries; but
valency than itself. Within this limitation, the requirement vertices ex, /3, 'Y, ... as indicated in figure 8.7, the percentage of dissections with any symmetries
are assigned to the room vertices a, b, c, ... in all permutations; and is only 12 · 5 for n = 7); and then multiply again by the number of
assignments in which edges of the requirement graph match those of the dissections with seven rectangles, which is 735. This gives a grand total of
adjacency graph, are retained. The following example illustrates a 29 635 200. Thus successful assignments in this (fairly typical) illustration
successful assignment for the requirement graph given above: amount to only some 0 · 00 I 7% of the theoretical maximum.
In the second part of our system, dimensional constraints are satisfied.
These are stated as maximum and minimum values on the dimensions and
~ - 'Y. areas of rooms. Also constraints can be placed on the proportions which
rooms may take. The process involves finding sets of values for the
~
Cl 6 •
dimensioning vectors in the x and y (east-west and north-south) directions,
~T) t
so as to give actual sizes to the intervals of the grating on which the
We will see shortly how, with realistic sets of adjacency requirements, this dissection is represented, as described in chapter 2.
process cuts down radically the numbers of alternatives under consideration. It is very possible that certain sets of dimensional requirements may not
It is possible to set requirements that rooms not be adjacent. Assignments be capable of being met at all in a given 'dimensionless' dissection, despite
in which these prohibited adjacencies occur are all rejected. Since the fact that the adjacency requirements are satisfied. If they can be
adjacencies may be stipulated to the exterior regions on the north, east, satisfied, however, and if dimensions are allowed to vary continuously
south, and west, it is necessary to rotate and reflect every dissection, and between certain fixed limits, then in principle there will be infinitely many
to make assignments to all isomorphs in each case. The process, applied to 'possible solutions', varying only by minute increments of size. The notion
all dissections with n rectangles, gives all plans in 'undimensioned' form of exhausting all possibilities must clearly be interpreted rather differently
which meet the adjacency (and 'nonadjacency') requirements <20). Should in this context.
it happen that the requirement graph is not planar, naturally no successful
assignments will result. <20) (continued)
nonaligned dissections, only the equivalent aligned versions. Thus certain possibilities
<20) One particular failing in the method should be mentioned, which means that it is for the dimensioning of these dissections were potentially excluded, as explained in
not strictly exhaustive- although it is nearly so. The list of dissections used in the chapter 3 (compare the first figure on page 29). Clearly this is a shortcoming which could
first stage was not complete (compare page 39), and in particular did not include all easily be remedied, by making suitable amendments to the file of dissections.
Chapter 9 Floor plan morphology in design 161
160
Grason's approach and Gilleard's too, in effect, was to consider only a in any detail. A basic introduction to the most general and widely used
small number of fixed modular sizes for each room, and to exhaust all technique, that of linear programming, is given in Grawoig (1967). Some
possible arrangements of those specific sizes. This might be reasonable in applications to architectural problems are described in Mitchell ( I 977).
the design of buildings using a modular system of prefabrication say, with In all cases some objective capable of being expressed as a simple
a rather large unit module. In general, however, it seems more plausible function of a set of variable quantities is either minimised or maximised.
that architects would want to allow any variations of size or proportion Thus, for example, a monetary cost might be minimised, or a profit
within acceptable upper and lower bounds. Our own approach has bee.n maximised, subject to a series of constraints which place limits on how much
to devise a method whereby certain dimensional properties of the plan of the objective can be achieved. The constraints are expressed as a series of
overall can be 'optimised' in a mathematica! sense; meanwhile dimensional inequalities (or perhaps also equalities), for example, 'x must be greater
limits set on individual rooms are all met. than or equal to some value', 'y must be less than or equal to some value'.
To do this we make use of various techniques of mathematical
programming, which originated in the subject of operations research (see area: 764 ft'
area: 654 ·4 ft'
Churchman et al, 1957). I do not propose to explain these methods Ot -y •
8 ·5
n s ·5
Ot -y
•
0
I 3. 0 0 6
I ,., 5. 6
I 0-y 3. 0 {3
e e 0-- - -
--- ::cf,,,,,--....-.. .oer__ - - -;;;."""
'x,... ---~ow 10 ·O
t
10 -0
{3 Tl t
w
{JO~-- -:_o_-:-:.--
o- 6·0 10-0 5.5 9-6 60 75 5530 90
Tl
3 ·O ' 3 -0 t
{3 ,., /J ,.,
10 .o I 10 ·O
6·0
Tl
6-0
Tl
1·0
•
5.5 3 0 8 7
(b) (c)
figure 9.8. Worked example of the operation of a computer system for plan generation Figure 9.8 (continued)
due to Mitchell et al (1976). Plans of rectangular dissection type are produced for an (c) Results of applying the dimensional constraints on rooms set out in table 9.2, by
apartment with seven rooms, Cl'. to 71. (a) The graph of required adjacencies (solid lines) means of a nonlinear programming technique in which the total areas of the plans are
and the graph of prohibited adjacencies (broken lines). (b) A sample of the 504 minimised. Only these six arrangements, out of the 504 'dimensionless' plans, are
I
successful assignments of room names to rectangles in dissections for n = 7. capable of satisfying the specified constraints.
162 Chapter 9 Floor plan morphology in design 163
Linear programming may only be used where the constraints and the It transpires that out of 504 'dimensionless' plans, only six are capable
objective can both be expressed in linear form, and is inapplicable to of being dimensioned as specified: all six are shown in figure 9.8(c). The
equations and inequalities involving powers, for example, x 2 or y 3 • fact is that the patterns of 'overlap' or alignment in a dimensionless
Objectives which fulfil this condition in the case of a (rectangular) plan dissection, without fixing the absolute sizes of rooms, have strong
would be, for example, the total length of the perimeter of the plan, its implications for their relative dimensions. They imply that certain rooms
proportion, or its overall dimension of width or depth. In each case the must have equal lengths or widths, that some must .be smaller or larger
objective might be either minimised or maximi~ed. The constraints could than others, and that the sums of widths of certain groups of rooms must
be expressed as limits on the widths, lengths, or proportions of individual equal the sums of widths of other groups. (These are the kinds of
rooms- for example, that the length of room r should not be greater than relations expressed in effect by weights attached to the .edges of a
l feet, its width not less than m feet, or its shape not more elongated than coloured adjacency graph.) It is for these reasons that typically, as in the
m: l. All these constraints and objectives involve only numerical ratios or worked example, permissible topologies are severely reduced in number by
linear dimensions. realistic dimensional requirements.
In Mitchell et al ( 1976) we demonstrated linear programming applied to It is interesting to see how some of the plans with smaller total areas in
the dimensioning of a floor plan, by taking the example of a mobile the example, are ones where the hall space is considerably larger than in
trailer home whose overall width was assumed to be fixed at the plans of greater area. (This hall appears in the centre of the plan in each
maximum dimension allowable on the roads. It was further assumed that case, since it is- perhaps unrealistically- not constrained to be adjacent to
the dimensions of width of the rooms were all fixed. The total length of the an exterior region. Access to the apartment from the outside at the north
trailer was then minimised, for a given plan arrangement of rectangular is assumed to be through the living room.) The total computing time
dissection form , and subject to constraints on the lengths of separate rooms. taken for both stages of this seven-room worked example, the assignment
This is obviously a slightly artificial example, though, and in general of the requirement graph and the dimensioning process, was some three
one wou ld want to place limits on the areas of rooms, and perhaps to minutes on an IBM 360/91 <22>.
minimise or maximise the area of the plan overall. However, this means The dimensioning procedures used in Mitchell et al (1976) have been
that constraints and objective cannot then be expressed linearly (areas further developed by Gero (1977), who used for illustration the same
being measured in ft2 or m 2 ), and so it becomes necessary to use nonlinear worked examples of trailer home and apartment plan. Gero, however,
programming. (The same would apply in any attempt to optimise the used dynamic programming (see Bellman, 1957), a technique which in this
three-dimensional form of a building, where measurements of volume, ft 3 context has several advantages.
or m 3 , enter into objective or constraints.) Inside the optimal solution to a dimensioning problem of this kind
Table 9.2 shows the dimensioning constraints on the various rooms, set there is liable to exist a certain flexibility, in that say the dimension of
for our worked example of the apartment plan as illustrated earlier. In Table 9.2. Dimensional requirements for Mitchell et al's worked example.
this case we used a nonlinear programming algorithm due to Clasen et al
(I 974) and implemented for the present problem by McGovern (1976). Room Length (ft) Width (ft) Area (ft 2 ) Maximum
proportion ratio
Besides the constraints listed, it was stipulated that where two rooms were min max min max min max
required to be adjacent, their overlap should not be less than three feet-
assuming the adjacency being to allow for a door. In the worked example a living room 8·0 20 ·0 8·0 20·0 150 ·0 300·0 l ·5: I
the total areas of plans were minimised under the given constraints<21 >. {3 kitchen 6·0 18·0 6·0 18 ·0 50·0 120·0
-y bathroom 5.5 5.5 8-5 8-5
Notice nevertheless that this does not imply some value judgement to ~ hall 0 15-0 3.5 6·0 0 72·0
the effect that a plan of minimum area is necessarily the 'best' in a E bedroom I 9-0 20-0 9-0 20·0 100·0 180·0 I -5: I
functional or economic sense; since it would be possible to derive both t bedroom 2 8·0 18·0 8·0 18 -0 100 ·0 180·0 I -5: I
the minimum area, and the maximum area for the plan, under the given 11 bedroom 3 10·0 17-0 10.0 17•0 100 ·0 180-0 I ,5: I
constraints. Then it is known that all possible dimensional solutions- be
they in theory infinitely numerous- must lie between these extremes. 22
< > Any comparison, however, with the times quoted above (page I 5 5) for the
<21 >In Mitchell et al ( 1976) it is incorrectly stated that the objective optimised in this solution of comparable problems via Grason's method should take account of the fact
same worked example is a notional cost of construction , measured as a function of that the machine used by Grason was slower, perhaps as much as one hundred times
floor area, in which the cost per square foot for kitchen and bathroom differs from ~lower, than the IBM 360/91. And Grason's program was executed interpretively, not
that for other rooms. In fact the objective is simple total floor area, as here. m a compiled language.
Chapter 9 Floor plan morphology in design 165
164
one room might be increased and the dimension of an adjacent room A further possibility with dynamic programming is that whole classes of
correspondingly reduced, up to certain limits, without the overall length, dimensioning problems may be solved simultaneously through a process
area, or whatever other feature of the overall plan is optimised, being known as 'invariant embedding'. Figure 9.9(c) illustrates a dimensionless
affected. It is obviously of interest to the designer or building scientist to plan containing L-shaped rooms, and alternative positions for two bedrooms,
know precisely the extent of this flexibility ; and this the use of dynamic which corresponds in its adjacency relations to all six plans shown in figure
programming allows. Figure 9.9(a) shows for example one of the 9.8(c). Those plans may each of them be derived from the arrangement of
dimensioned solutions to the apartment problem from the top right figure 9.9(c) by appropriate assignments of dimensions, such that either one
diagram of figure 9.8(c), with exactly the same total plan area, but with a of the x or one of the y dimensions of the grating is set to zero, certain
number of changes made to the dimensions of individual rooms. alignments of walls are offset, and the rooms all become simple rectangles.
Gero also shows how in this same case, if the limits on the dimensions Finally, in this account of design methods for rectangular plans subject
of the bathroom 'Y are relaxed, and its area is allowed to increase above the to adjacency and dimensional constraints, there is the method devised by
effective maximum specified in table 9 .2, then a dimensioned solution of Flemming (1977 ; 1978) and based on his 'wall representations' as
lower total plan area (627 ft2 ) is possible. In this arrangement the long described here in chapter 4. Flemming has developed a computer system,
dimension of the bathroom is oriented north-south instead of east-west the DIS program, with which it is possible to enumerate exhaustively
[ figure 9. 9( b) ] . solutions to problems of the same kind addressed by Grason and by
Mitchell, Steadman, and Liggett.
area: 654-4 ft' area· 627 ft2 Flemming shows how trivalent rectangular dissection-type plans
I I
- --r f CT-plans') may be derived, by means of the DIS program, through successive
"' 1' • 8- 5
"'
5 .5 operations of the types depicted in figure 4.1 5. [Notice that at this stage
6
6
the plans do not have dimensions, nor is the pattern of overlap of rooms
5 -6 3. 5
across walls represented-compare figure 4. l 3(a) and the first figure on
/J TJ t /J TJ t
page 45.J These geometric operations on the plan are carried out in the
10 -0 10· 0
program through equivalent logical operations on the wall representations
. themselves. Figure 9 .10 illustrates an example for four rooms in which
I I 60 I
75 / I \ 90
I
70 65 4540 90
(b) 3-0 3-0 five required relations of adjacency are specified. The program adds rooms
(a)
one at a time, such that at every step two general conditions are met. The
first is that no required adjacency which has already been satisfied at an
Y, "' I,- 1' •
or t i earlier stage is disrupted by the introduction of the new room. The second
- - 6 is that all adjacencies required between the new room and those already
Y,
placed, or with the exterior regions at n, e, s, and w, are satisfied .
Y, /J TJ •or t Flemming ( 1977) is able to prove that there exists a unique sequence of
Y,
operations by which each 'topologically feasible solution' may be reached.
x, I x,
Figure 9.10 shows only part of the whole tree of alternative sequences by
X x, x,
I
which all such solutions meeting the stated adjacency constraints may be
(c)
produced. (Other solutions would be possible via the unfinished branches
Figure 9.9. Flexibility of dimensioning of individual rooms within an overall set of on the right of the figure.)
dimensional constraints. (a) One of the dimensioned plans from figure 9.8(c) (top For the dimensioning of these topological solutions Flemming makes
right), with the same total area (654-4 ft 2 ), but with changes made to some of the
room dimensions within the given constraints. (b) Demonstration by Gero (1977), use of linear programming. He argues that although it will often be useful
using dynamic programming, that if the constraints set on the dimensions of the to specify constraints involving areas, these can nevertheless be approximated
bathroom -y in the same example (see table 9.2) are relaxed, a solution of lower total linearly. It will be clear that constraints on the lengths and widths of
area (627 ft1) is possible. (c) Dimensionless plan corresponding in its adjacency rooms or on dimensions of the plan as a whole, can be expressed directly
relations to all six plans shown in figure 9.8(c) (from Gero, 1977). Each of the six can in relation to wall representations, since each of these lists rooms in order
be produced from this arrangement through appropriate assignments of dimensions, along either side of every wall. Constraints on the 'overlapping' adjacency
such that one of the x-dimensions or one of they-dimensions of the grating is set to
zero, some wall alignments are offset, and the L-shaped rooms become rectangles.
of two rooms across a wall can also be expressed, such that, for example,
a sufficient dimension is allowed for a door.
Chapter 9 19-47 19-47 19-47 19-88 19 -88
166
Figures 9.11 and 9.12 illustrate a worked example which uses the DIS
program, in which an eight-room apartment is planned within an L-shaped
19-88 19 -88 19·88 I 9-88 19 -88
area. A ninth dummy space is used to fill out the corner of the L.
Figure 9. 11 shows the adjacency requirement graph, and table 9.3 Jays out
/1
Figure 9.10. Generation of trivalent rectangular dissection-type plans with the DIS
program of Flemming ( I 977). Five adjacencies are required between four rooms a , {3,
'Y, 6 and the exterior regions w and e, as shown in the graph. The program adds
20-65 20- 73 21 ·08 21-25
rooms one at a time through operations on wall representations (see chapter 4). No
required adjacency, already satisfied at an earlier stage, is disrupted by the introduction
of the new room. Meanwhile the new room is positioned such that its required
adjacencies to those already placed, and to the exterior regions, are satisfied. The figure
shows only part of the tree of sequences of operations leading towards feasible solutions.
n /1
facade
apartment
3
~ -
- · 00
_3 3
facade 0
g
21 -25 21 ·28 22-45
8
3" ~
w e
Q
12 m /1
maximum
I
Figure 9.11. A worked example of the use of Flemming's (1977) DIS program, for the Figure 9.12. The twenty-seven feasible plans which meet both adjacency requirements
planning of an eight-room apartment whose overall perimeter is L-shaped. The general (figure 9.11) and the dimensional constraints set out in table 9.3. The sum of the overall
arrangement is shown of the apartment in relation to the access corridor and building dimensions of the plan (in effect, half the perimeter length) is minimised in each case
facades, with upper limits given on overall dimensions. The L-shape of the apartment by means of linear programming. The figure over each plan gives this value. Notice
is filled out to form a rectangle by the addition of a ninth dummy exterior space a that two adjacent halls t and ri are specified, which together form a variety of shapes
(shaded). Required adjacencies are shown in the graph. of circulation area.
168 Chapter 9 Floor plan morphology in design 169
the dimensional constraints. These correspond to current standards for A second type of objective function used by F lemming for dimensioning
public housing schemes in the Federal Republic of Germany. Notice that has the effect of keeping certain rooms (for example, circulation) to the
two rectangular halls are specified, and are required to be adjacent, so as minimum areas allowable within a configuration, with any available 'slack'
to allow for the possibility of either a T-shaped or an L-shaped circulation area concentrated into other rooms (for example, living rooms).
area. The adjacency requirements of figure 9.11 represent a selection F lemming's system is capable of exhausting all solutions where constraints
from a number of alternative possibilities (for adjacencies between rooms, as here are reasonably tight, and the number of spaces not too large. But
and to the exterior regions) which would be equally acceptable within it can still be useful as a design tool, or for exploratory exercises, where
similar standards. the numbers of potential solutions are so great as to make enumeration
In this case the objective which is minimised by the linear program is impractical. The DIS program has the virtue of displaying to the operator
the sum of the north - south and east-west overall dimensions of the plan. the products of each geometric operation, as in the successive levels of
(In effect the perimeter length is minimised.) Figure 9.12 shows all twenty- figure 9. I 0. Thus he can decide not to explore all branches of the solution
seven plans capable of meeting adjacency and dimensional constraints. tree which would be generated in theory by his initial set of constraints.
Flemming calls these 'principal options'. The figure over each plan gives Instead he can pick what seem to be promising partial solutions, on selected
the sum of its overall dimensions in metres (equals half the perimeter length). branches, and develop only those. Such an approach is only possible with
Of course, as already discussed, there would be in principle infinitely many procedures, such as Flemming's, which are of a 'constructive' character.
solutions differing only slightly in their dimensions, between the extremes In general all these methods- Grason's, that of Mitchell, Steadman, and
determined by say minimising and maximising perimeter length; as well as, Liggett, Flemming's- come up against limits on the number of rooms
most probably, a certain flexibility in room dimensioning even at the which may be treated in the exhaustive enumeration of plans, because of
extremes. Flemming speaks of the particular dimensioning solutions the intrinsically explosive nature of the combinatorial problem. Any
produced by linear programming as 'representatives' of these potential technique along these lines which attempted to deal with many more
ranges of dimensional possibility. See how several plans in figure 9.12, rooms, would be obliged to subdivide the problem hierarchically in some
with the same perimeter length, differ widely in their internal disposition way, and deal only with a subset of rooms at one time. On a single floor
and room dimensions. level, large plans might be subdivided into discrete zones or bays. Perhaps
special conditions could be imposed on the adjacencies of rooms across
Table 9.3. Dimensional requirements for Flemming's worked example. the boundaries of adjacent zones.
Room Dimension (m) Area (m2 ) Multistorey plans with large numbers of rooms might be treated, one
floor at a time, if the number of rooms on each floor was limited. Special
min max min max
precautions would have to be taken to ensure that vertically interpenetrating
a (dummy) 3-60 elements such as stairs, lifts, or double-height spaces were properly aligned,
(3 living room 3·60 22-00 and to ensure that each floor as a whole had the correct dimensions and
r master bedroom 3-30 5-40 14-00 was oriented correctly over the lower floors. These possibilities are
& bedroom 2-40 4-20 7-20 9.99
discussed in Mitchell et al (1976).
€ bedroom 2-40 4· 20 7 -20 9.99
A special difficulty posed by all these techniques concerns circulation
1 hall 1 ·20 6-00
Tl hall 1-20 6-00 spaces. When an architect makes a plan in the normal way, he often
8 kitchen 2· 10 5-40 7-20 introduces circulation areas as he proceeds, because, as he thinks, the
l bathroom l ·80 4-20 4-20 required relations of adjacency and access cannot be met without them.
Total area has a maximum dimension east-west of 12 ·00 m , and a maximum dimension The present methods, by contrast, require that all spaces, including
north-south of 18 ·00 m. circulation spaces (whether 'necessary' or not) be specified, together with
Minimum lengths of common wall to be shared by rooms or spaces which are required their adjacencies and dimensions, at the outset. Solutions which might
to be adjacent , are: exist with fewer circulation spaces, or with more, are not found. It is
a tow 3-60 m a to e 0 -90m & to 11 0 -90 m 1 to 0 0·90m difficult to see how this objection can be overcome within the terms of
a to s 3 ·60 m (3 to 1 0·90 m € to Tl 0·90 m 11 t o t 0·90m the methods, other than by exhausting plan solutions with successive
a to (3 2 ·40 m r to 11 0-90m 1 toe l ·20 m 8 ton 0-90m numbers of circulation spaces.
a to & 0·90 m r to n 2-40 m 1 to 11 1 -20 m
Baybars and Eastman ( 1980) have outlined a graph-theoretic design
method, which consists essentially of a technique for generating maximal
170 Chapter 9 10
planar graphs exhaustively, and for picking out those which possess Plan morphology and building science
spanning subgraphs corresponding to a specified set of adjacency
requirements. The selected maximal planar adjacency graphs are then
embedded, and their duals taken to produce different plans. Baybars and " ... our own study of organic form, which we call by Goethe's name of Morphology ,
Eastman argue that by subsequently deleting edges corresponding to is but a portion of that wider Science of Form which deals with the forms assumed
by matter under all aspects and conditions,· and, in a still wider sense, with forms
'optional' rather than to required adjacencies from the maximal planar
which are theoretically imaginable." •
graph, it is possible to enumerate systematically all possible positions for
D'Arcy W Thompson (1961)
extra internal spaces in the plan (in addition to those represented by the
vertices of the adjacency graph). The extra spaces might be interpreted
either as 'courtyards' or as circulation spaces. However, at the time of This chapter presents some applications of the ideas about floor plan
writing Baybars and Eastman have not extended this method, as they morphology which have been explored so far in the book, to certain
intend to, to incorporate a dimensioning stage. topics which are traditionally thought of as falling within the scope of
No comparable methods exist for geometries other than rectangular, 'building science' - questions of natural lighting, circulation, the adaptability
although there are no reasons in principle why they should not be possible. and flexibility of plans. There is more of promise to report here, than
Their applications would be rather limited. Korf ( 1977) has criticised the solid progress. Nevertheless this promise is very great, much greater
restriction of existing methods to rectangular arrangements, and points out indeed, as I have previously argued, than the kinds of 'design method'
(as we have seen in chapter 7) that certain combinations of adjacency which formed much of the subject of chapter 9. The equivalence in
requirement cannot be met in plans whc•se rooms are all simple rectangles. length of that chapter and this one should be taken therefore as a
His proposal involves drawing the duals of embedded adjacency graphs as reflection of how work has been divided so far, rather than as a token of
'bubble diagrams' which he says are 'entirely independent of shape'. the importance which 1, at least, would assign to the two directions of
However, any actual drawing of the dual must assign shapes and sizes to research.
the ' bubbles', and any actual architectural plan would obviously have to I say questions which are traditionally thought of as those of building
be realised with definite room shapes and dimensions; so that this science, since part of the philosophy of this morphological or configurational
suggestion of Korfs really just ducks many of the issues which we have work is that the scope of a 'science of building' should encompass not just
been examining in this chapter. the study of building materials, structures, and buildings as environmental
enclosures, but that it should include, and indeed be founded on, a study
of the forms and arrangements of buildings- their geometry and their
topology. This is surely Rittel's conception when he speaks of one of the
first chapters of a 'theory of architecture' dealing with 'the theory of cell
configurations'. As Hawkes (1980, page 14) has expressed it:
" ... the role of scientific research in architecture, as opposed to building
science, is primarily to work upon what have been called the 'animals of
architecture' (March and Matela, 1974): first to establish a morphology
of architectural form and then to use the tools of building science and other
methods of analysis to make explanatory statements about the relationship
between form and performance. As a practical programme, this has the
attraction of offering a more structured basis for the application of
traditional analytical research."
At the beginning of chapter 8 I raised some issues posed by the
'combinatorial explosion' in numbers of possibilities for rectangular
arrangements with increasing numbers of rectangles n. It appears that for
values of n much greater than 10, the extent of combinatorial variety- in
rectangular dissections, in polyominoes- becomes so great that a complete
enumeration is of little practical purpose; and that indeed for values of n not
much larger than this, enumeration itself becomes a practical impossibility.
Chapter 10 Plan morphology and building science 173
172
Meanwhile there are many real buildings, obviously, which contain more It is difficult to find evidence, for example, about even such a simple
than ten or a dozen rooms on a single floor. matter as what proportion of existing buildings possess a rectangular
I suggested in chapter 8 that it might be feasible to treat this very great geometry. The only studies known to me are one made by Bemis ( 1936)
theoretical range of possibility, at these higher levels, by statistical methods. of American houses, and one by Kruger ( 1977; 1979a) of buildings in the
city of Reading, Berkshire.
This problem of combinatorial variety can, however, be viewed in quite a
Bemis was an early and influential promoter of industrialis ation and
different way. This involves looking rather more closely at what we mean
by 'possible' and 'possibility' in this building science context . prefabrica tion in the building industry, and he was ~oncerned to know the
The reason for examining the properties of say rectangular dissections as extent to which systems of prefabricat ion with a rectangular geometry
a class of geometrical 'objects' in the first place, was that arguably they would restrict variety of design in houses, or alternatively what proportion
serve as fair geometric models, at a certain level of abstraction , for the of 'special' nonrectang ular componen ts would be needed to supplemen t
kinds of wall and room arrangeme nt which are typically to be found in such systems. Thus he did not categorise houses as a whole as to whether
the plans of actual buildings. Rectangular dissections are chosen for their geometry was rectangular or nonrectangular. Instead he took one
study, that is, because they are realistically 'plan-like'. Equally polyomino es house at a time and measured the 'proportio n of rectangula rity' of its form.
He calculated two figures, one for the rectangulari ty of the whole house
might be chosen-al though I have suggested (page 16) one respect in
which polyomino es are not like the plans of many buildings. by volume (thus the volumes of pitched roofs, curved bay windows, etc
In the end then these are empirical matters. The practical relevance of were deducted from the total volume of otherwise rectangular houses),
examining the properties of rectangular dissections turns on how many and the second, since he was interested in constructio n, for that proportion
buildings presently existing, or which have existed, in fact do possess a of the physical fabric of the house which was rectangular (again by volume).
geometry of rectangular dissection form. This perhaps would give some His basic survey was made in Boston, but the results were extrapolat ed to
guide to the number of such buildings likely to be designed and constructe d the entire United States. He found that 88 · 5% of all American housing
was rectangular by total volume, and 98 ·3% rectangular in its constructio nal
in the future. lf many buildings were made up of hexagonal rooms, then componen ts.
that would provide practical justificatio n for studying the geometry of the
Kruger worked at a much less detailed level, analysing from large-scale
packing of hexagons. Should an architect take it into his head to design a
maps the perimeter outlines of the plans of all buildings in the entire city
building whose rooms were all irregular pentagons, then of course it would
of Reading (see figure 11.14). He found that 98% of those plan shapes
not be included in these enumeratio ns. (Although all subdivisions of the
were of rectangular geometry.
plane into regions, that is, all plans treated at the topological level, are
Equivalent surveys of room shapes and plan shapes, and extended to
exhaustively covered by the enumerati on of plane maps.) Such a design
ot~er types of building; would show the extent of application , for building
would not be 'impossibl e' in any final or absolute way. lt would just be
science, of rectangular dissections , polyomino es, or other classes of
rather improbable.
arrangements as geometric models for plans.
When we speak of 'possible designs', or 'possible plans', it is thus the
To come back to the problem of the very large numbers of these kinds
actual world of real buildings, with all the practical and functional and
of geometric arrangeme nt for large n : I suggest that, despite the merits of
technologi cal constraints applying to their forms, which serves to define
the (theoretica l) realms in which these 'possibilities' lie. What we are rectangular dissections as models of smaller plans, there is an increasing
proportion of 'theoretica l possibilities' for larger dissections which
asking, when we talk of 'possibility ' in this context, is " How many o ther
theoretical designs are there , similar in some sense to those which have nevertheless become rather unlike the plans of buildings, and hence begin
to Jose their practical interest .
been actually made?" The problem is that systematic analyses of the
geometric forms and arrangements of the mass of actual buildings are Such dissections consist, certainly, of rectangular componen ts
rather rare. There does not exist, as there does for natural species, a corresponding to rooms, packed toge ther in different configura tions. But
descriptive anatomy or 'natural history' of architecture< >.
23 these configurat ions are not at all probable architectur ally, in ways which
are hard to pinpoint precisely, but are no less real fo r that. It is something
<23> Perhaps the closest approximati ons in a single volume to a ' natural history' of to do with such facts as that real buildings tend to have a limited depth,
architecture , are provided by Durand's R ecueil et Paralle/e des Edifices (1 801), which because of the needs of daylighting and natural ventilation, so that when
as I have suggested elsewhere (Steadman, 1979) may well have had a conscious model large they become organised into regular patterns of wings and courts.
in biological classification ; and Banister Fletcher's History of Architectur e (I 896) on Or that rooms are set along relatively simple and coherent circulation
'the comparative method'-th is a 'comparative anatomy' perhaps? See also my
systems consisting of a few branching corridors which extend along the
comments on Pevsner ( 1976) in chapter 11.
174 Chapter 10
1I Plan morphology and building science 175
building's whole length. There are many large dissections which are made dissections, is purely configurational, and quite independent of the specific
up, by contrast, of a deep maze-like agglomeration of overlapping rectangles, dimensions which the structural members or their relative spacing might take.
many of them completely internal, and through which any linking pattern of Consider, however, the more elementary and conventional way in which
circulation routes would be circuitous and confusing. An example is the properties of individual beams are studied , by determining the
maximum loads above which those beams will deflect beyond acceptable
limits, and ultimately fail. This in effect defines for given materials cross-
1---
sectional shapes of beams, etc, the limits of the 'u~iverse of possibl;
I-- - beams', in dimensional terms.
We can make the same sort of distinction for beams, between their
L--L-
configuration and their dimensions, as was introduced here in chapter 2
for room arrangements (compare Spillers, 1974). Indeed this distinction is
manifested in the standard format of handbooks of structural steel tables.
Each possible cross-sectional shape of steel beam- or at least those which
are manufactured- 'T', 'L', 'H', square-section, circular section, etc is listed
If we could capture properties like these in explicit geometrical measures, separately.
then we might be able to limit the study of dissections, for example, to a Then for each shape the structural properties are tabulated for different
much reduced class of arrangements which would all be 'building-like' in actual sizes of flange, web, etc. The engineer can only choose his designs
some such well-defined sense. The identification of these properties can first of all from among the given configurations. And he is limited a
only come from a programme of empirical work, which takes the plans of second time, within these shapes, to certain maxima or minima of size-
existing and historical buildings and devises measures whereby these that is, to maximum spans say, or minimum structural depths of beam.
aspects of their characteristic geometry and topology can be described. These limits, and similar ones applying to columns, will in turn have
What follows is largely speculation then: but one would expect practical their implications for the possible dimensioning of different framework
limitations on 'possibility' at this larger architectural scale to be of two configurations as a whole.
general kinds, the one configurational or topological, the other dimensional There are obviously many comparable ways in which function ,
(although dimensional and shape or adjacency constraints are always technological means (that is, available tools, machines, or manufacturing
closely interrelated, as we have seen in the previous chapter). techniques), and the properties of materials place ultimate dimensional
bounds in this sense on the realm of 'possible buildings'. The strengths of
Dimensional limitations on building fonn materials limit the maximum heights of buildings. The size of the human
Let us look first at some dimensional limitations which are likely to body limits the minimum dimensions for a door, or the maximum height
· apply equally to almost all kinds of building. So far in the book the for a step in a staircase. The necessity or desirability of admitting natural
emphasis on possibility in design has been placed very largely on possibility light to buildings limits their maximum depth ; or if artificial lighting and
of arrangement in 'undimensioned' terms, with the understanding that the ventilation are introduced, then another limit of depth is set perhaps by
variety of possibilities for assigning dimensions to these arrangements is the need for windows solely to give a view of the outside world. Such
effectively infinite. However, in real architectural or engineering terms, limits cannot admittedly be defined with absolute precision. The lines at
these dimensional possibilities, although they may be infinitely numerous, which they are drawn in practice tend to represent some compromise
are certainly not unbounded. between considerations of comfort and convenience on the one hand and
To take an example from the science of engineering structure: it would limitations of materials, technique, and cost on the other. '
be possible in principle to tabulate the numbers of distinct ways in which Thus it would very likely be possible to construct buildings taller than
columns and beams may be put together into structural frameworks- the highest presently existing, with the expenditure of more money, or
frames with different numbers of those elements, or frames with different with the development of new materials or techniques. Buildings might be
numbers of bays and floors- very much as we have done for plan made indefinitely deep, if the occupants were prepared to sacrifice the
arrangements. Within this, certain subclasses of framework with given convenience of daylighting, or the pleasure of a view. Limits of these
structural characteristics might be identified, as, for example, in the work kinds on 'possible buildings' are therefore established largely by judgements
of Bolker and Crapo ( 1977) on all those structures which are completely of value. In this sense they are not like the absolutely necessary and
braced against lateral loading. This enumeration, like that of rectangular unavoidable constraints of topology and geometry. This does not, however,
176 Chapter 10 Plan morphology and building science 177
alter their force, or the fact that they serve to explain many of the extreme limit is represented by bamboo. Equivalent considerations of
observable restrictions on the variety of building forms which are actually strength in relation to weight limit the maximum height of trees to
seen in practice. approximately 300 feet. Similar reasons applying to bones mean that it is
The kinds of dimensioning technique described in the last chapter might impossible for terrestrial animals to be much larger than the mammoth or
be used for investigating some of these questions experimentally. In the the prehistoric dinosaurs.
method devised by Mitchell, Liggett, and myself and applied by way of Other features of the design of organisms besides their 'structural
illustration to the example of a small flat, there was a whole series of engineering' have analogies with the design of buildings. The ratio of
limits set- in a rather arbitrary way- on the minimum and maximum sizes surface area to volume is important in both cases, since it affects the rate
and proportions for rooms. A nonlinear programming technique was then of heat loss or gain, to or from the surroundings. In buildings this same
used to produce plans with overall minimum area, within these constraints ratio of surface to volume is significant, of course, for natural lighting and
on room shape and dimension. It would equally have been possible to ventilation. Animals and buildings are both served by different kinds of
minimise or maximise other properties of the plans, such ai the lengths of circulation systems-networks of tubes or passageways, respectively, which
their external perimeter walls, or their linear dimensions of width or depth. must penetrate at a more or less uniform density to every part.
[Indeed· some studies along precisely these lines are reported in Mitchell If organisms of differing absolute size were to possess exactly the same
(1975a) l. shape (simply scaled up in direct proportion), then they would have very
The implication was that the constraints of room shape and size were different properties in these respects. This follows from the way in which
such as might typically be set by a designer wishing to produce plans on areas vary with the square of linear dimensions, and volumes vary with the
some particular occasion. And that by means of the programming method cube. An animal of double the linear dimensions, for example, would
he would be able to explore the limits of overall dimensional variation (in have only four times the cross-sectional area of legs with which to support
total area, length of perimeter, overall length or width) allowable for those eight times the body weight (assuming weight to be a simple function of
specific plans. volume).
Imagine, however, a similar method being used by a building scientist. What happens in fact is that these effects are compensated for by
He by contrast would be interested in the absolute extremes of size and changes in shape. Such changes may be observed in comparisons of the
shape allowable for each type of room - insofar as those are determinable. forms of adults of different but related species which vary in their
One might imagine, for example, that lower limits of area or dimension absolute size. They also occur in the development of individuals of some
could be set on certain room types such as bathrooms, kitchens, or given species from embryonic to adult form - in which case they are
corridors, on the basis of anthropometric standards and the requirement to brought about by appropriate adjustments in the differential rates of
accommodate certain furniture or fittings. The building scientist would growth of the parts. Thus the relative shape and proportions of head,
then want to know the complete extent of the range of plan types resulting, trunk, and limbs in human babies are quite different from those in the
with in each case their maximum or minimum overall areas, depths etc. grown man or woman.
Conversely it might be that limits on these overall dimensions of depth The study of relative growth rates and their effect on shape is termed
themselves could be established say in terms of maximum allowable a/lometry. After D'Arcy Thompson's pioneering work the subject was
structural spans, or some maximum acceptable depth for daylighting-and greatly developed, by Huxley (1932) and others. There has recently been
that these fixed global constraints would in turn have their effect back on something of a revival of interest in these questions of size and shape in
individual room shapes and sizes. the investigation of biological form (for example, see Alexander, 197 l ).
The designer looks for a particular plan, or some few plans, with desired One study of building forms using allometric techniques borrowed from
properties. The building scientist seeks to map out the whole field within biology has been made by Bon (1971 ; 1972a; 1972b; 1973).
which the designer's search is conducted. Bon took a sample of forty residential buildings-houses, apartments,
hotels, including even a houseboat and a railway sleeping car built for a
Allometric studies of building form nineteenth-century financier- of various dates and from different parts of
There is a tradition of research in biology, starting from D'Arey Thompson's the world. Bon's notion here was that, of all functional types, it is
On Growth and Form, which has attempted to define limits on organic dwellings which are the most dependent on natural lighting. Also he was
structure and the 'design' of animals and plants, in a very similar way. able to find buildings of this general class, of very widely varying sizes.
Thompson shows for instance how the strengths of woody fibres put a Bon measured for each building its volume V (in cubic feet), and its total
minimum ratio of slenderness on the stems of plants, of which about the surface area of external wall S (in square feet). These values are shown
178 Chapter 10 Plan morphology and building science 179
plotted on a graph in figure 10.1. The ratio of surface area to volume The faster observed rate of growth in wall area, relative to growth in
decreases generally with increasing absolute size of building, as would be volume, constitutes a case of positive allometry.
expected. What is interesting here is the comparative rate of growth in The consequences can be observed in practice in the way in which the
each of the variables. forms of large daylit buildings are flattened in the vertical plane into slabs,
Imagine a building of simple cube shape, with length of side d. The and in the horizontal plane become spread out into winged, branching, or
volume V of the building is d 3 , and its total surface area of walls S courtyard configurations. The actual buildings in Bon's study would not
( discounting roof and floor) is 4d 2 • The continuous curve in figure 10.1 themselves necessarily lie on or near the 'extremes of possibility' which
represents values of volume and wall surface area of such a cube-shaped the constraints of daylighting allow. No doubt the standards of lighting
building, for increasing d. Notice that the points representing the larger in them vary considerably, a supposition which is given weight by the
actual buildings in Bon's sample all lie above this curve. The implication, departure of the observed values of S and V somewhat from a smooth
clearly, is that in the actual buildings the ratio of surface area to volume curve, and even more so by the fact that the absolute ratio of S to V does
has been increased over that of the cube, by some elongation of the shape, decrease quite markedly with increasing building size, despite the positive
or perhaps by some corrugation of the wall surface. This effect is to be allometric effect. However, the very fact of this positive allometry points
explained by the way in which residential buildings-of any size-are so to the existence of some kind of dimensional limit on acceptable depths
designed as to give a large proportion of the rooms an adequate area of for residential buildings, and an attempt to escape, in the shapes of larger
window for the purposes of daylighting, views, and natural ventilation. buildings, from the effects of a decreasing ratio of surface area to volume<24>_
It would be expected that if the rates of growth in V and Sin the Dimensional constraints of this kind have their effect back on the
sample of actual buildings were constantly related, then this relation properties of the plan and adjacency graphs of arrangements. The
would be of the general allometric form S = b va, where a and b are significance of the depth limit is that in all probability most of the rooms in
constants. With our cube-shaped buildings, for example, V = d 3 , S = 4d 2 , these buildings are on the perimeter, and only circulation spaces and perhaps
which gives S = 4 V'1,, that is, S = 4 v 0 · 66 . The general equation can be bathrooms, storerooms, etc are windowless and in the centre of the plan.
restated by means of logarithms, as
106
logS = 1ogb+a logV.
0
Thus if the logarithms of the two variables are plotted, and the values fall
10'
on or near a straight line, this shows that the variables are related
allometrically. This is indeed the case here, as figure I 0.2 demonstrates.
The slope of the line is given by a = 0 · 77. Had the buildings all been of
a similar shape (as, for example, cubes or any other fixed form) the slope
of the line would have been 0 · 66. (That line is called the isometry line.)
~ "1
10'
0
0
0 0
.:::
~
._ 7 - 10• 10' 10• 10'
0 logV (log ft')
!I
'g 4 Figure 10.2. The data of figure I 0.1 plotted on logarithmic scales, to show the
:§ allometric relationship of the two variables S and V (from Bon, 1973). The isometry
line is marked. The line fitted to the observations has a steeper slope: the faster rate
"' of growth in wall area relative to growth in volume is a case of positive al/ometry.
4 7 10 13 16 19 22 25
V (in units of 10' ft') <24> Bon suggests that similar allometric techniques might be used to investigate
structural questions such as the relation between the weight of buildings (as a function
Figure 10.1. Graph of total surface area of external wall S (ft2 ) against volume V (ft3 ) of the volume of their materials of construction) and the total cross-sectional area of
for a sample of forty residential buildings [frnm data given in Bon ( I 972b)]. The their structural elements. This could provide a means for determining the height limits
continuous curve represents (theoretical) buildings of simple cubic form. imposed by certain types of building material.
180 Chapter IO I Plan morphology and building science 181
This naturally tends to result in the tree-like circulation systems typical of Weeks (see Llewelyn-Davies et al, 1975), and others; in university buildings
apartment blocks or hotels, with rooms branching off central corridors, or (UCERG, 1968; Bullock et al, 1970); and in schools (see Fawcett, 1976b).
clustered around regularly spaced staircase wells. There is some tendency for larger buildings to contain larger rooms,
It was noted in chapter 7 that there exists a class of graphs which are especially those which consist of only a few principal spaces, or just a
capable of being embedded in the plane such that all vertices lie on the single space, such as auditoria, churches, or markets. One would not,
exterior face. These are outerplanar graphs. Considered as adjacency however, expect the range of room sizes in any one.residential building to
graphs they correspond, in their outerplanar embeddings, to those plans in be very great, nor would one expect larger residential buildings to contain
which all rooms lie on the perimeter. All trees, and all graphs with only larger rooms, only a greater number of them. The same would be true of
one cycle, are outerplanar. office buildings (made up of individual private offices, that is). In both
Lynes (1977) has commented on the significance of outerplanarity in cases the rooms tend to take their sizes, roughly speaking, from those of
adjacency graphs for the planning of naturally lit buildings. Harary ( 1969) the pieces of furniture which they accommodate. In this there is a loose
has stated the criterion for outerplanarity of a graph: that it must not analogy perhaps with the way in which animal bodies of all sizes are made
contain as a subgraph any graph homeomorphic to K 4 or K 2 , 3 . These up from cells which do not themselves vary greatly in size, only again in
subgraphs are: their numbers<25>.
Bon (1972a) made a second allometric study of a smaller sample, of
twenty buildings, again all houses or dwellings. In this case he measured
the total floor area F (in square feet) and what he called the total
'communication network length' L (in feet). This latter he defined as
"the minimal orthogonal network connecting all the room centroids in a
building". By 'room centroid' he means a point at the notional centre of
See how in all embeddings of these graphs, one vertex remains in the gravity of the room's shape in plan. Bon forms a circulation network of
interior, and not on the exterior face. The addition of one more vertex tree form joining all these points, in which the routes run in one or other
connected to all other vertices in each case will produce K 5 and a graph of two perpendicular directions, parallel with the walls of the building-
• containing K3 , 3 , respectively (you might like to check this). that is, the distances along these routes are taxicab distances-and such that
Lynes makes the point that the presence of K4 or K 2, 3 in an adjacency the total length of all routes is minimised. The following figure gives an
graph does not mean that some particular room has to be internal. There example:
will be several embeddings, in which in each case a different room will fall
in the interior. Lynes also remarks on the way in which the placing of
windows in a building restricts the ways in which it may be extended in
the future. He suggests that this problem too could be investigated by a
graph-theoretic approach.
In chapter 8 I listed a whole series of the properties of adjacency graphs
which might have practical architectural significance. One of these was It is thus not a representation of the actual circulation system, but some
the distribution of the valencies of vertices- the number of edges incident measure of the theoretical minimum length of circulation needed.
with each vertex in the graph. There is plainly some relation, if not a If log F is plotted against log L for Bon's observations, a straight line of
necessary or simple one, between this distribution of valencies and the slope 1 · 07 can be satisfactorily fitted (figure 10.3). Had similar shapes of
distribution of room sizes in a building- since rooms which are relatively building plan and communication network been preserved- which would
large will tend to be adjacent to more other rooms. The two measures imply that the number of rooms remained the same with increasing
together might go some way, together with the structural properties of the building size- the isometry line in this case would have taken a slope of
access graph, towards the classification of functional building types on a 0 · 5. This is therefore a very marked case of positive allometry, arising
out of the differentiation of the internal plans of the buildings into ever
geometric and topological basis.
Some empirical studies exist of the distribution of room sizes in selected <25 > It was proposed, as early as 1893, by Ryder, that it is the ratio of surface to
types of building, made mostly out of an interest in the potential flexibility volume which puts a maximum size limit on the individual organic cell itself. As
and adaptability of plans, or the timetabling of their use. Thus there are D'Arcy Thompson (1961) says, "Nature has her materials ofpredeterminate dimensions,
surveys of room sizes in hospitals made by Cowan and Nicholson (1965), and keeps to the same bricks whether she build a great house or a small".
Chapter 10 Plan morphology and building science 183
182
greater numbers of rooms, all of which must be reached by the 'minimal And in residential buildings, as we have seen, the majority of the rooms
network' of circulation. are on the perimeter anyway.
Such a general result is only to be expected. What is perhaps more In chapter 7 we saw that in large maximal planar graphs the average
particularly interesting is that the ratio FIL for Bon's sample varies only valency of vertices approaches the value 6. It would be interesting to
slightly, and lies for the great majority of cases between 8 and 12. In a know how closely the adjacency graphs of actual buildings approach to
minimal circulation system with the form of a tree, the number of edges e this theoretical limit. Two empirical studies have qeen made whose
must be one less than the number of vertices v, that is, e = v- I. This is authors had rather different purposes in mind, but which all the same
a general property of trees. It follows that the number of centroid to throw light on this question.
centroid distances measured by Bon for each building must be very close Both studies were of houses. The first was made by myself (Steadman,
to the number of rooms. This suggests that the mean room dimension for 1976): my interest was in finding where the plans of typical small modem
these residential buildings is of the order of ten feet, which is again what houses would lie in the whole space of theoretical possibility represented
might be expected. by Combes's diagram (compare the figures at the beginning of chapter 8)
for dissections. I took for my sample house plans illustrated in the English
0 National Building Agency's publication Generic Plans (NBA, 1965). This
13 j \\"'"
13 -i
7
'l,"1 <>
;!; 2
02
~ 10 ' 0 O 10 -,
...~ ~ 3 1 02
0 ,
~ 1- 1-
.D
E
::,
z
4 ~ 4 "
T --r-r-r-r r-r-1 l I I -r -,-- r- ,--,--,.-r r r I I I I "l ' r l •-----y- r-T
1 4 7 10 13 16 4 1 10 13 16
Number of partitions, p Number of partitions, p
-, I
{b)
10' 10' '10' (a)
logF (log ft 2 ) \\"'" 2 -
13 J ,....i<> 13 -
Figure 10 .3. Graph of total 'communication network length' L (ft) against total floor ;!; 1 l olo l o I
area F (ft 2 ) for a sample of twenty residential buildings, plotted on logarithmic scales
to show the allometric relationship of the two variables (from Bon, 1973). The
=; 10 ~
JO 3
Io I
0 02 10 -
isometry line is marked. The line fitted to the observations has a much steeper slope: 'o ~
the faster rate of growth of the communication network relative to growth in floor .,
~
1-
.D
area is a strong case of positive allometry. E
::,
z
4 -
The adjacency and acces.5 graphs of residential buildings IT T I I Tl T I I I r'!'"" +---,--,-,- l • l .--, _
10 13 16
As for the adjacency graphs of residential buildings, one would not expect 7 10 13 16 1 4 1
Number of partitions, p Number of partitions, p
large variations in their vertex valencies precisely because of this fact that {d)
(c)
the order of room size is the same throughout- assuming the rooms to be
of roughly comparable shape, convex for the most part and not excessively Figure 10.4. Plans of small modern houses of rectangular dissection type , plotted on
elongated. (In a building such as a large hotel, with rooms planned along Combes's graph of numbers of internal and external wall segments (compare figure 8.1)
a series of corridors, these conditions would not apply.) There would (from Steadman, I976). The plans are taken from a catalogue, Generic Plans, published
by the English National Building Agency (NBA, 1965). They are classified as four-
be a systematic difference certainly between the valencies in the weak person, five-person, and six-person types: these are shown plotted separately in (a), (b),
dual graph, of vertices representing , respectively, rooms in the interior and (c), respectively. The number of plans falling on each point is indicated. Both
and rooms on the perimeter. But this 'boundary effect' is to some ground-floor and first-floor plans are included. {d) shows all 164 plans combined in a
extent compensated for by including adjacencies to the exterior region. single diagram. See how of these the great majority (117) fall on the line 2w = 11 +p.
184 Chapter 10 Plan morphology and building science 185
is a catalogue of model plans for the use of architects and house builders. Of course, in the small adjacency graphs of the dissections the effect of
However, they are for the most part derived from a survey of actual plans the lower vertex valencies around the perimeter means that the average
in use in local authority and privately developed housing schemes. figure for these is much less than 6.
I took only those in which the shape of the plan overall was rectangular, Bon investigated the same phenomenon in a sample of seventy house
and approximated the layout in each case as a rectangular dissection. I plans drawn from the Central Mortgage and Housing Corporation catalogue
then plotted the numbers of instances of each dissection on the relevant Small House Designs (CMHC, 1958). He measurep vertex valencies in the
points in Combes's graph, according to their numbers of internal and adjacency graphs of these plans, including this time adjacencies to the
external wall segments p and w [figure I 0.4(a)-(c)]. Generic Plans classifies exterior region, and obtained a mean value of 4 · 24. He also measured
houses as four-person, five-person, and six-person types. These were plotted vertex valencies in the plan graphs and obtained a mean value of 2 · 65.
separately, and the ground-floor and first-floor plans were both included. The mean number of rooms in a plan in the sample was 9. Bon
The results for all plans together were further summarised in a single compared his survey figures with the same values calculated for a regular
diagram [figure I0.4(d)]. plan graph consisting of nine hexagons packed as in the left-hand part of
There is a distinctly marked clustering, as will be observed, along a figure 10.5. His purpose here was to allow for equivalent boundary effects
straight line corresponding to the equation 2w = 11 + p. Out of a total of in the actual plans and this ideal configuration. The corresponding
164 plans, 11 7 fall precisely on this line. If the weak dual adjacency adjacency graph has a mean vertex valency of 4-45 , and the plan graph a
graphs of dissections on this line are examined, they are found to be of mean vertex valency of 2-53. (Notice that vertices of valency 2 are
the following types: counted at the angles between the external wall segments.) Bon took these
results to show a close approximation, in these respects, of his sample of
plans to a theoretical 'hexagonal' pattern c26>.
The adjacency between two rooms provides the opportunity for access
between them, but this opportunity is not always exploited. That is to
say, the access graph is generally a spanning subgraph of the adjacency
(Recall that the number of edges in such a graph equals the number of graph. Bon asks the interesting question, what proportion of edges in the
internal wall segments p in the plan.) These are in fact triangulations of adjacency graphs of plans are simultaneously edges in the access graphs?
polygons- triangle, quadrilateral, pentagon, hexagon, with the interior in In the sample just mentioned the mean vertex valency of the access graphs
each case subdivided into triangles in all distinct possible ways. They are of the house plans was 1 · 92, by comparison with 4 · 24 for the adjacency
all of them outerplanar graphs (as might be expected for small houses). They
graphs. This means that for only 45% of adjacencies between rooms, or
can be conceived of as fragments of a larger regular triangulated graph which between rooms and the exterior, were there actually doors connecting the
is maximal planar and in which the mean vertex valency tends towards 6: rooms or regions in question.
Bon showed that this percentage, calculated for small house plans, does
not seem to vary greatly when calculated for residential buildings of widely
differing sizes. He took his sample of forty buildings, already referred to
in connection with the allometric studies, and counted in each case the
numbers of edges e and vertices v in the access graph. The results are plotted
C26> In point of fact, most of the actual plans would have been rectangular, and it
The dual of this graph, that is, the corresponding 'plan', is a regular packing might have been more reasonable for Bon to take for comparison a packing of nine
of six-sided faces, which may be drawn in rectangular form as a brickwork- rectangles in 'brickwork' arrangement. This would be no different in its adjacency
type pattern: graph from the hexagon packing. However, there is a slight difference in the mean
valency of the vertices of the plan graph due to the counting of vertices with valency 2
on the perimeter. The resulting revised figure for the mean vertex valency of the plan
graph is 2 · 57. Notice also that Bon's method of calculating the mean vertex valency
in adjacency graphs is slightly odd, in that for rooms on the perimeter he includes
adjacencies to the exterior region, but he excludes the vertex representing the exterior
region itself. Thus for the adjacency graph of the hexagonal plan of figure 10.5, the
exterior region, whose vertex has valency 8, is not counted in taking the mean. Where
all edges and all vertices are counted, the mean vertex valency is given by 2e/v.
186 Chapter 10 Plan morphology and building science 187
in figure 10.6. See how the ratio of e to v remains remarkably constant shown in figure 10.6. Subsequently Bon (1972b) extended this same
with the increase in size of the graph. A line which represents the best fit analysis to a much larger sample, of 500 buildings, and obtained a
st atistically to the observed values has the equation e = 1 · 30 v - 3 · 04, as similarly consistent ratio e/v although the slope of the allometry line was
somewhat shallower (the equation was e = I · 13v - 1 · 34 ).
Rather than compare these values of e with the corresponding values for
the adjacency graphs of the actual plans, this time, Bon chose rather to
make comparisons with maximal planar graphs. The real adjacency graphs
would not necessarily be maximal planar, of course, although they would
probably approach close to that point, as we have just seen.
For maximal planar graphs e = 3v - 6. This follows from Euler's
formula v- e + f = 2, applied to plane triangulations. If every face is a
t riangle, and since every edge must belong to just two faces, then f = je .
Substituting in Euler's formula gives 3v - 3e + 2e = 6 , that is, e = 3v - 6.
See how in the adjacency graph of the hexagonal arrangement of figure 10.5,
for example, which is maximal planar, e = 24, v = I 0 , and 3v - 6 = 24.
(a) (b) For each plan, Bon compared the actual number of edges in the access
graph with this theoretical maximum for planar graphs on the same
Figure 10.S. (a) Theoretical nine-room plan of hexagonal geometry , f~r comparison in
terms of mean vertex valencies of adjacency graphs and plan graphs, with a sample of
number of vertices. This ratio, e : (3v - 6), expressed as a percentage, has
seventy actual house plans whose mean number of rooms is nine (from Bon, 1971 ) , been called the 'gamma index' (Garrison and Marble, 1961). It measures
and (b) the same plan drawn with a rectangular geometry (see footnote 26). t he degree of connectedness of a planar graph. The mean value of the
gamma index for Bon's sample of forty buildings was 43%. For the
0 seventy small house plans t he mean value was 42%.
Bon concludes from this that the 'hexagonality' of the plans, as
100
evidenced by the properties of their adjacency graphs, is more a simple
consequence of the close-packing of rooms, rather than a means of
maximising access between them. This seems to follow since less than
80
.._...,,
!,
/" half ( 42 or 43%) of the theoretical opportunities for providing access are
~
'7 actually made use of. This is a complex matter, however, since t here is
t
60 liable to be a marked variation in vertex valency in the access graphs-
" which the mean figure conceals- from private rooms such as bedrooms or
ii bathrooms with a valency of 1, to circulation spaces such as halls or
~ 0
40 landings, with much higher valencies, which give access to several rooms.
It is these latter types of space which are placed, of course, in the centre
of plans, and the former around the perimeter.
20 l One does not expect there to be doors in very many of the exterior
walls of the plan (especially not on floors above the ground!) and it is
0
i ➔
therefore these walls which would account for at least some of the 'missed
opportunities' for access. More would be revealed by counting separately
0 20 40 60 80
Vertices, v
and comparing the percentages of external and internal wall segments which
are or are not penetrated by means of access. In the ideal arrangement of
Figure t 0.6. Graph of number of edges e agains! num_be_r of vertices v in the access nine hexagons in figure 10 .5, for example, there are eigh t adjacencies of
graphs of a sample of the plans of forty residential buildings (from ~on, 1971 ) . (Two rooms to t he exterior region, compared wit h sixteen interior adjacencies
observations are omitted here.) A straight line fitted to the observations has the
equation e = 1 . 30v - 3 •04. Also shown are straight lines corresponding to trees,
between rooms. Thus one-third of the theoretical opportunities for access
e = v - l and maximal planar graphs, e = 3v- 6 . The access graphs of plans, since in t his case are in the perimeter. It may be that the 'hexagonality' o f
they must be connected, and planar, must lie between these limits.
188 Chapter 10 Plan morphology and building science 189
plans reflects rather an effort to achieve high levels of interior access, only two staircases, connected at each level horizontally, in the central
between rooms. A measure of general interest for plans would be one corridor or 'balcony access' types of arrangement. High-rise office
showing how adjacencies between rooms and with exterior regions are buildings may be served by a single circulation 'core' from which all rooms
divided between those made use of for access and those made use of to on each floor are reached. They may like blocks of flats have a central
provide windows. corridor, in which case the access graph is tree-like; two parallel corridors
It is worth making the point that the number of edges in an access with a central row of artificially lit service and storage rooms; a continuous
graph must have a lower bound, if that graph is to be connected, as in cycle of corridor in the 'racetrack' type of plan, and so on (figure I 0. 7).
practice it would tend to be. (Even the access graphs of pairs of semi- Some workers who have applied graph theory to describing networks in
detached or rows of terraced houses are connected, at least at ground subjects other than architecture- in biology (Shimbel, 1953), in transport
level, if access to a common exterior region is included.) The graph which studies (Garrison and Marble, 1961 ; Kansky, 1963), in geography (Haggett
connects v vertices with a minimal number e of edges is a tree, in which and Chorley, 1969)- have devised a variety of quantitative measures of
case e = v - 1 as we saw earlier. The line representing this equation has graph properties which may be relevant here (compare Tabor, 1976). We
been plotted on Bon's diagram in figure 10.6. The gamma index for trees have already looked at the gamma index for connectedness. Another
must always equal (v - 1) : (3v - 6) which for large graphs tends to the property is the cyclomatic number, which for a connected graph is given
value 33%. This then is the lower bound on the possible value of the by e - u +I. For trees, where e = v - I , it follows that the cyclomatic
index. The distance which the graphs in Bon's sample lie above the line number is always zero. The addition of further edges to a tree while
e = v - I must represent the extent to which they contain cycles, and so keeping the same number of vertices, so that it becomes a graph with
are to a degree 'redundant' in providing means of access between rooms. cycles, increases the cyclomatic number by I for each edge.
Figure I 0.8 illustrates the process for a tree on twelve vertices. Notice,
Gross measures of the properties of adjacency and access graphs however, that the cyclomatic number does not count the number of cycles
With the exception of certain work on access graphs by Hillier and in a graph. In a connected plane graph, the cyclomatic number is equal to
colleagues, to be described in the next chapter, there are no comparable the number of faces (excluding the exterior face). This follows from
empirical studies known to me of the properties of the graphs of building Euler's formula. A cycle may pass round any number of faces sharing
types other than residential. However, one can imagine such studies edges in common. Where a plane graph contains adjacent faces the number
providing an important part of the formal classification of building types, of distinct cycles may thus be much greater than the number of faces.
as I suggested earlier. In the access graph of an architectural plan, the presence of cycles means
The adjacency graphs of residential buildings as we see are characterised that there exist two or more different routes between certain pairs or groups
by no great variations in the valencies of vertices (with the exception of of rooms. The legal requirement in large modern buildings to provide
communal access corridors in flats or hotels), and with a high proportion alternative means of exit in the case of fire means that their access graphs
of rooms having adjacency to the exterior. The graphs of other building must at least comprise cycles to this extent. The same would apply in any
types might display large variations in vertex valency (the graph partition grand house with 'back stairs' or servants' stairs. Any plan with a racetrack
or distribution of edges between vertices would be much more uneven)- arrangement or with continuous circulation around courtyards would equally
this perhaps reflecting large differences in room sizes, as in, for example, have an access graph with cycles on each floor considered separately.
big auditoria with their surrounding lobbies, bars, backstage facilities, and The diameter of a graph is found by taking the shortest paths between
other service spaces. Again the distribution of adjacencies to the exterior all pairs of vertices (measured as a number of edges in each case) and
region might typify such buildings as recording studios or certain parts of taking the longest of these- it is the 'longest shortest path'. This is some
hospitals, in which rooms are deliberately isolated from the outside very rough measure, for a graph with a given number of vertices, of how
environment. spreading or how compact it is.
As for access graphs, it is certainly possible to imagine larger buildings Applications of these and similar measures to real plans are discussed
being classified by the general network properties of their circulation further in chapter 11. They give gross indications of the overall structural
systems. To give some specific examples: the access graphs of blocks of character of adjacency and access graphs. At a more detailed level it
flats consist generally of repeated identical 'knots' of private circulation would be possible to describe plan types in terms of classes of specific
within the flats , all connected to some larger linear public circulation labelled graphs (the labels signifying room functions). Thus one can
routes. This public circulation could consist of a series of separate imagine small house plans being categorised by distinct patterns of access
staircases along a block . Or it might be that a block was served by say between living room, dining room, kitchen, bedrooms, etc.
Plan morphology and building science 191
Some statistical work along exactly these lines was carried out by Bon
(1972b) on his sample of seventy small houses. He categorised their
access graphs by numbers of vertices, the disi.ribution of valencies of those
vertices, and the numbers of cycles in the graphs (for 27% of the plans the
(a) number was zero- the graphs were trees- and in a further 67% there was
one cycle only). He measured for the access graphs the values of some of
the structural indices described above. He identified characteristic recurrent
r-,
= = configurations of labelling of the graphs, that is, typical patterns of access
□~~ I
0 l_ll,
0 .
t=l
between rooms of different functions. And he measured the mean
accessibility of rooms with different functions, that is, the average distance
(as a number of edges) from one room to all other rooms- with high values
t=TTT resulting, as would be expected, for halls and corridors.
Besides the functional significance of adjacency between rooms or
(b) regions in terms of access or lighting, there are the further factors of
privacy and acoustic separation, which could serve to account for the fact
\l of rooms being made deliberately not adjacent in certain plan types.
So far in the book we have only considered the functional reasons for
• • • • the direct binary relation of adjacency or nonadjacency between pairs of
l~I rooms or regions. However, there are certain instances, particularly with
m
access relationships, where three or more rooms are made adjacent and
• • accessible one to another in some specific pattern or sequence. This
occurs, for instance, where a private office of a manager or a doctor is
• • reached from a public corridor only through a waiting room, or through
the intermediate office of a secretary or receptionist. In hospitals, in
[], [J,[J,[J,[J,[), ~
public baths, or in factories, where people or goods are involved in definite
0, n, [}[}Crfl' sequences of activities or processes, then the circulation system joining
rooms or spaces will naturally follow these sequences. The layouts of
p, museums or exhibitions provide further examples. These topics are also
taken up again in the next chapter.
(c)
---- ·-· r~ I
I
[. ··· ·~ ·- .. .._0~~ -, --, --r7·
, :~.-~ -
H-H R-H FfH Em
e = 11
cyclomat ic
e = 12
cyclomatic
e = 13
cyclomatic
e = 17
cyclomatic
Figure 10.8. The cyclomatic numbers of sample graphs on twelve vertices (after
(d)
Tabor, 1976). For a connected graph the cyclomatic number is given by e - v+ 1:
Fi~re 10.7. Typical examples of plans of office buildings (from J?~dicke, 1962,. thus for a tree (first graph), where e = v - I, the cyclomatic number is zero. The
figures 48, 55, 58, and 62): (a) double-zone layout with central utility core-Equ~table addition of each new edge increases the cyclomatic number by one: in connected
Savings and Loan Association building, Portland, Oregon (Pietro Belluschi); (b) tnple- plane graphs, as here, it counts the number of interior faces.
zone layout with parallel corridors- competition design for Phonix-Rheinrohr AG
building, Dusseldorf (Paul Schaeffer-Heyrothsberge); (c) open layout with internal
utility core-Mile High Center, Denver (IM Pei); (d) single-storey courtyard layout-
Schlumberger building, Ridgefield, Connecticut (Philip Johnson).
192 Chapter 10 Plan morphology and building science 193
Pedestrian circulation
In chapter 9 we examined a number of automatic methods for producing
plans supposed to be 'optimal' in terms of their circulation patterns, and I
offered several criticisms of the shortcomin gs of these methods. In each
case some regular frequency of daily or weekly pedestrian journeys was
assumed between all pairs of 'activities' in an organisation ; and the effort
was then made to find an arrangeme nt of rooms or positions for those
activities such that the total distance travelled in the given time, or else
some notional total travel cost, was minimised.
The methods were open to three objections in particular. First, the
1:::::::::::s::::11:::::~
(a)
a
minimisation of circulation was not in general balanced against other
functional (let alone aesthetic) criteria in design such as those of natural III IIl IIl I 111111111 Il 11 I 11111 l II IIII111111111 IIl I
lighting, structure, etc. Second, it was difficult to make realistic
measureme nts of distances between pairs of rooms without specifying in
advance some general pattern of circulation routes, this implying in tum
an approxima te overall form for the building. Thus the problem had a
certain chicken-and-egg character, since the whole purpose of the automatic
layout method was precisely to generate this building form and circulation
system in the first place. Third, such methods were implicitly designed to
fit the plan of a building as tightly (hence as 'efficiently ') as possible to some
fixed pattern of activities taking place in and between rooms. Whereas with
the design of many building types a certain openness to change, a degree
of adaptabilit y and flexibility in the plan, would be more desirable.
In response to these criticisms some authors working on the subject of
circulation in buildings, notably Tabor ( 1976) and Willoughby (1975a;
1975b) have taken a wholly opposite view of the question, and turned
from a 'design methods' to a more properly 'building science' approach.
They have started instead with specific building plans, with their layout of
rooms and circulation routes completely given, and then have made
quantitativ e compariso ns between these forms on the basis of their relative
'circulation performan ce'. The main focus of their work has been on office
buildings, hence their chosen forms are several of them similar to those
identified above as representat ive of office types (compare figure I 0.7).
Tabor compares slab, cross, and court forms [ figure I 0.9(a)]. Both the
slab and the court may have rooms single-banked or double-ban ked along
a corridor. Willoughby in addition examines a 'fishbone' or branching,
tree-like plan type [figure 10.9(b)].
It is clear that these type-plans have been chosen to be representative in
a diagrammatic way of actual designs of office buildings as found in {b)
practice. Room sizes are not widely variable- indeed Tabor and Willoughby
assume for the sake of simplicity uniform room areas throughou t. And Figure 10.9. Theoretical types of plan for office buildings, each containing the same
each layout as a whole acknowledges implicitly the constraints of natural number of similar-shaped rooms, for comparison in terms of circulation performance.
(a) Slab, cross, and court plans (from Tabor, 1976). (b) Slab, court, cross, 'fishbone',
lighting to the majority of these rooms, the desirability of a simple and and open plan types (from Willoughby, 1975a).
coherent circulation system, and perhaps also the constructio nal and
structural virtues of a regular, modular, rectangula r plan.
194 Chapter 10 Plan morphology and building science 195
Willoughby comments for instance on the relation of his 'fishbone' type At the simplest level it would be possible to measure the shortest
to the arrangement of linked Nissen huts put up by the armed forces in distances between all pairs of rooms and plot the distribution of these
many parts of the United Kingdom during the Second World War, and values. This, or the mean of all these distances, would give some indication
subsequently converted for use as local government offices, university of the concentrated or dispersed nature of the plan as a whole. The
laboratories, and even hospital buildings. For further comparison measure of the diameter of a graph can be extended to networks, giving
Willoughby introduces a continuous deep office plan which would need the 'longest shortest distance' as a length in feet o,r metres.
permanent artificial lighting and ventilation, and would correspond to the In chapter 8 we saw such calculations made by March and Matela for
familiar Burolandschaft type of the I 960s. polyominoes in their undimensioned form (the distances there were
Thus the forms have been chosen to be 'building-like' in the senses expressed as numbers of edges). It would be possible to make similar
vaguely adumbrated earlier<21>. Beyond that the choice is perhaps somewhat analyses of the plans of real buildings; or as Tabor has done, of hypothetical
arbitrary. Any attempt to take such exercises further would do well in 'built forms' with specified dimensions. In March and Matela's work three
my view to go back to an empirical survey of actual office buildings (or different kinds of distance measure- straight-line, 'taxicab' , and graph
other functional types), to determine from them a realistic range of distances-were employed. With actual or schematic architectural plans it
'possible' forms and dimensions, rather than to go on ad hoc comparing obviously makes most sense to measure distances as travelled along the
more hypothetical forms. access network itself. These will usually be greater than the corresponding
It is not my intention to report the results of these circulation straight-line distances. The difference between the two can be expressed
'experiments' in detail, if for no other reason than that they have been as a diversion factor, or detour index (see Tabor, 1976).
fully published elsewhere (refer to Tabor, I 976 ; Willoughby, I 975a ; If the access network contains cycles, and hence alternative routes are
197 Sb). It is, however, well worth looking at their general methodology, possible, it seems reasonable to make the behavioural assumption that
to see how th.is relates to some of the other morphological issues which people will tend to follow the shortest of these routes (although for
we have already touched on. visitors unfamiliar with a building's layout this would by no means always
I have suggested in previous chapters that the exact pattern of relations be true). Several algorithms exist for determining shortest paths in a
of adjacency and of access within some group of rooms only assumes network [Dreyfus (1969) gives a survey). A question arises with multi-
functional significance at a small scale, say that of ten or twelve rooms at storey buildings, as to how to compare distances travelled vertically by
the most. In larger plans the properties of access graphs joining hundreds stairs or lifts, with horizontal distances along corridors. The problem can
of rooms, one would expect to be hierarchical. The smaller functional be circumvented by the measurement of average travel times instead of
groupings of rooms- be they departments or separate tenancies in office distances, the assumption of certain walking and climbing speeds, etc,
buildings, separate flats in a block of flats, or whatever- would be linked although this calculation too has its difficulties, especially in relation to lifts.
to each other and say to the main entrance and any other shared facilities Tabor is able to rank the various type-forms which he considers, by
by a system of specialised circulation spaces- corridors, staircases, lifts. taking the mean of the distances between all pairs of rooms in each case.
The structure of the access graph at this level of the hierarchy would The forms are constructed so as to be directly comparable, in that they
therefore need to be interpreted both in terms of the overall form of the each contain the same number of rooms of the same size throughout.
building, and in terms of the resulting (dimensioned) distances separating The implication is that those forms in which the mean distances are lower
pairs or groups of rooms along this system. That is to say at this larger have an inbuilt circulation advantage. It would be possible to arrange
scale, the relation of proximity between rooms comes to have functional accommodation for an organisation with a given pattern of trip frequencies
relevance. So the access graph needs to be considered as a network in in these plans such that the total distance travelled was less than in other
which weights are attached to the edges to represent dimensions of length plans, simply by virtue of their shapes. With multistorey buildings the
along corridors etc. (At the scale say of a single house, these distances results can be appreciably affected, for the same form, by the numbers
are effectively negligible since all rooms are 'close' to all others.) and positions of lifts.
As Tabor allows, however, these demonstrations assume that all distances
<27 > The contrast of inward-looking courtyard buildings with outward-looking slab or between pairs of rooms are equally significant functionally. In a well-
'pavilion' forms is also in part related to a theoretical argument about the general way planned building, as opposed to one in which functions were simply
in which buildings of these generic forms make use of land, put forward by Martin assigned to rooms at random, those activities for which proximity was
and March (1966; 1972). That argument in its turn was based nevertheless on some desirable would naturally be placed closer together. Tabor approaches this
considerations of the spacing apart of buildings, having to do at least by implication
with lighting, or ventilation, or views.
problem theoretically by positing a measure of the 'propensity' of the
Chapter 10 Plan morphology and building science 197
196
occupants of a building to make shorter journeys. As this value increases 1972, chapter 2 ; Mitchell, I 975b), but other factors such as orientation,
towards shorter and shorter trips (the building is better planned), so the area of glazing and hence solar gain, ventilation , and the temporal patterns
ranking of the building forms by mean journey times alters. At the of occupancy and control of the building's mechanical systems in relation
extreme, where all journeys are very short, the effect of the overall building to daily and seasonal climatic fluctuations. It is dynamic simulation
shape is negligible and the forms rank equally. Conversely, it is where models which are the 'tools of building science' needed in this case.
journeys are longest that the building form has the greatest relative effect A designer could make use of the knowledge gained from these
on the resulting mean travel times. experiments, when choosing a form to suit some actual project. Those
Willoughby tackles the same problem from a slightly different direction. whose interest is rather in describing and explaining the forms and plans
He supposes that a large office building may be divided up into varying of existing buildings could use similar measures or models to predict, or
numbers of areas or zones, each of which is occupied by a separate rather 'retrodict', their past and present performance. To what extent
department or firm or tenant. He assumes contact between the occupants buildings are actually planned to be optimal or close to optimal in say
of all rooms within each of these zones, and no contact at all between circulation or energy conservation terms, or to what extent these are
zones. He therefore calculates the total of journey times between all pairs traded off against other considerations, are things which remain to be seen.
of rooms within the same zone, and sums these values for all zones. For all To go back to circulation networks, the real functional significance of
his experiments, the total number of rooms is again fixed throughout. He distances along a route system would clearly be highly dependent on the
has two variables then: the range of building forms on the one hand, and the use of a building and the pattern of movement which it accommodated.
grouping of rooms into a greater or smaller number of zones on the other. In a block of flats, for instance, the distances between separate flats might
The results are dependent to quite a large extent on the accidents of not be so significant as the distance of each flat from the main entrance.
the geometric fit of room groupings with building forms. For instance, a In public buildings accommodating large numbers of people for short
division of the rooms into three equal-sized groups does not fit neatly into periods such as auditoria or the concourses of stations or airports, the
a four-storey plan, nor into a single-storey building in the shape of a four- distances travelled in these mass movements would naturally be of more
armed cross. At least one department or tenant must have accommodation importance than the distances separating staff offices or service rooms in
split between wings or floors, and this dislocation increases the distances those same buildings. A very specific constraint on maximum travel
travelled. Willoughby uses a modified form of one of the automatic distances in many public buildings, a real limit on 'possible' dimensions
layout planning techniques to allocate the room groupings to the building and one indeed which is widely embodied in legislation, is that imposed
forms in an optimal way, such that the total of travel times is always by the requirements for means of escape in the case of fire.
minimised, so as to make the comparisons on a fair basis. So we could imagine the access networks of large buildings being
It would be misleading to list the results of these experiments in detail subjected to several levels of analysis: a detailed consideration of the
without giving a more precise account of the working assumptions. patterns of purely topological access relations within small and relatively
However, in general summary, over all groupings of the rooms into different isolated groups of rooms within the plan ; consideration of the overall
numbers and sizes of zone, it is the single-storey open plan which because graph structure- whether tree-like, containing cycles, etc-of the
of its compactness gives the lowest total of travel times, followed by the components of the network corresponding to principal circulation routes;
one-storey and two-storey slab and cross forms. Meanwhile, a decrease in and measurement of distances or travel times along this network, depending
the size and corresponding increase in the number of room groupings on the characteristic movement patterns of the occupants or users of
results in smaller absolute differences in travel time totals between forms, buildings of that functional type.
much as in Tabor's demonstration - although this decline is not a completely Before leaving the subject of networks, it is worth mentioning the
regular one. possibility of using the type of 'electrical network' described in chapter 7,
The purpose of the experiments both by Tabor and by Willoughby, in or the extension of the principle by Teague ( 1970) to three dimensions,
effect, is to determine how functional characteristics, in this case circulation for the purposes of certain analyses of building performance. Since these
properties, vary across the range of morphological possibility- to analyse networks represent the systematic relations occurring through the plan
'the relationship between form and performance' in Hawkes's words. between room dimensions, wall and floor areas, and volumes of rooms
Hawkes himself (1980) has recently proposed, in general terms, an they might be used in connection with finite-element methods of structural
equivalent kind of approach to the question of the relationship of building analysis, as Teague ( 1970) suggests, or in modelling the flow of heat
shape and energy use. This work would take account not simply of the through the internal structural subdivisions of a building as well as through
insulation properties of the building envelope (compare Martin and March, its external surfaces.
198 Chapter 10 Plan morphology and building science 199
Adaptability and flexibility Suppose, however, that certain activities in an organisation were required
I mentioned, but have not yet enlarged on, one criticism of the automatic to be adjacent- that an adjacency requirement graph could be specified.
layout planning methods: their failure to allow for flexibility and Then, clearly the organisation could be accommodated only in those plans
adaptability in plans. There are many features of buildings affecting the arrangement of which meets the relevant adjacency requirements. We
adaptability, and indeed several senses in which a plan may be said to be can illustrate the situation with a simplified example which uses rectangular
adaptable. At a small scale at least, however, the work on plan enumeration dissections. Suppose that we ignore all considerations of room size, shape,
already described provides the means for a precise measurement of certain servicing, etc and concentrate solely on the questio'n of 'fit' in relation to
geometrical or topological features of adaptability. room adjacency. Figure 10.10 shows all dissections for n = 4, together
Both terms, 'adaptability' and 'flexibility', are often used loosely. It is with their weak dual adjacency graphs.
best to define several distinct measurable properties of plans vis-a-vis the There are only four different graphs as shown: call them (a), (b), (c),
organisations which they house, which affect the capacity for future and (d). Notice that (a) contains three edges, (b) and (c) each contain
changes in use or in form: four edges, and (d) contains five edges.
( 1) There is the capacity of the same fixed plan to accommodate the same Take one of these graphs (c). Label the vertices ex, (3, -y, o as shown.
organisation, but in different arrangements. Now let us imagine that the graph of adjacency requirements for the
(2) There is the capacity of the same fixed plan to accommodate different organisation which is to be housed is as in figure I 0.11 (a), with vertices
organisations. labelled a, b, c, and d. It is in fact the same graph as (c). However, since
(3) There is the ease or otherwise with which the interior configuration of the vertices are labelled, they may be mapped onto the vertices of (c) in
some given plan may be physically altered- by removing, moving, or two distinct ways, as figure I 0.11 (a) shows. The adaptability of the plan
introducing interior walls or wall segments, etc. vis-a-vis the organisation is two. Take a second adjacency requirement
(4) There is the ease with which a building may be extended- its capacity graph, for a different organisation, as in figure I 0.11 (b). It may be
for growth. mapped into the vertices of (c) in six ways. A third requirement graph as
Suppose that an organisation can be separated into distinct 'activities' or in figure 10.l l(c) may be mapped into (c) in ten ways. In each case it is
'functions' each suitable for accommodation in a single room. (There are a matter of counting permutations of assignment of a, b, c, d to ex, (3, -y, li
admittedly dirnculties in this assumption, in many cases.) Assume further such that the required adjacencies are satisfied.
that a plan consists of a series of separate rooms with different attributes. We can move th.is study to a more general level, and consider
Then it can be asked of each activity, will it 'fit' into each of the rooms systematically all possible adjacency requirement graphs for organisations
in turn? This might be a matter of simple room size- a class of twenty which might occur. If we continue to take it that there is one vertex or
pupils needs a classroom with space for at least twenty desks. It might be activity to a room, then we need to consider all graphs on four vertices.
a matter of the fixtures and fittings provided in a room, the services
6 6
1' /J
6
,,
6
,,
The first of our four kinds of adaptability measures then the total
number of distinct ways in which some given list or schedule of activities
may be assigned to rooms "in a given plan, such that each activity has a room
in which it fits. If there are no constraints on the adjacency or proximity
of activities, then th.is question may be studied independently of the
"'D/J
6 1'
geometrical organisation of the plan. Such an approach has been taken by
Fawcett (1976a, 1978), who counts simply the permutations of arrangement 6
However, there is no point including graphs with more than five edges, in This leaves the graphs numbered I to 10 depicted in figure 10.12. Half of
the present context, since as figure 10.10 shows there is no rectangular these are disconnected graphs.
dissection in which so many requirements could simultaneously be satisfied. Suppose that the vertices of all these graphs are labelled, and that they
are mapped in all allowable ways into the graphs (a), (b), (c), (d) as before.
~"':
Table 10.1 gives the results. Notice that graphs 1 to IO are listed in order
of increasing numbers of edges, as are graphs (a) t.9 (d). In general, two
[~"':
~d
C
'Y
C
6
d
'Y
d
6
C
trends emerge-though they are not completely smooth ones. Looking
down the table, we see that the number of fits decreases for a given plan
(a) as the number of edges in the requirement graph increases. Obviously
there comes a point where the number of adjacency requirements exceeds
the actual number of adjacencies in the plan, and so where no fit is possible
A
c d
at all (bottom left-hand corner of the table). Meanwhile, looking across
the table we find that for a given organisation and set of requirements the
number of fits increases as the number of edges in the adjacency graph of
the plan increases.
Each entry in the table then represents (in this adjacency respect)
adaptability of the first kind, of a particular plan vis-a-vis a particular
organisation. Adaptability of the second kind is measured by counting
the number of nonzero entries in each column, since this corresponds to the
total number of distinct (unlabelled) requirement graphs, that is, the
(b) number of possible different organisations, which the plan in question can
Q accommodate. Notice how this value increases in moving across the table
oa Q oa "' oa "' oa "'oa oa
from (a) to (d). These are in fact the numbers of (unlabelled) spanning
subgraphs of the adjacency graph.
I\
c d
Table 10.1. Number of allowable ways graphs I to 10 of figure 10.12 can be mapped
into plans given by graphs (a) to (d) of figure 10.10. In the lower left-hand corner of
the table. the number of adjacency requirements exceeds the number of adjacencies
in the plan.
graph number graph (a) graph (b) graph (c) graph (d)
number of edges 3 edges 4 edges 4 edges 5 edges
(c)
1 0 24 24 24 24
Figure 10.11. Mappings of adjacency requirement graphs into rectangular dissection
2 I 12 16 16 20
type plans. (a) A set of required adjacencies between four room functions a, b, c, d,
3 2 4 8 10 16
as shown in the graph, may be mapped into the four rooms of the given plan a, 13, -y, l!
4 2 8 16 8 16
in two possible ways: the adaptability of the plan vis-fl-vis the organisation is two.
5 3 2 8 4 12
(b) A second adjacency requirement graph may be mapped into the plan in six ways.
6 3 0 0 6 12
(c) A third graph may be mapped into the plan in ten ways.
7 3 0 0 6 12
8 4 01 0 2 8
9 4 o la o 8
10 5 0 0 0 i 4
2 3 4 5 6 7 8 9 10 column total 50 80 76 132
nonzero entries
Figure l 0.12. All possible adjacency requirement graphs for rectangular dissection in column 5 6 8 10
type plans, on four vertices. They are numbered 1 to 10 for reference in table 10.1.
202 Chapter 10 Plan morphology and building science 203
If we count the column totals, these give the numbers of labelled spanning It is these types of graph, or their presence as subgraphs themselves in
subgraphs for each adjacency graph. These figures express some combination adjacency graphs for larger n, which determine the maximum possible
of the first two types of adaptability-the number of different organisations number of labelled spanning trees. Each of the perfectly cyclic graphs
which can be housed multiplied by all the ways in which those organisations corresponds to a unique polyomino as shown in the figure. Within the
can be arranged in the plan. In this diagrammatic example it is adjacency limited terms of the exercise, then, it is possible to say that there are
graph (d), the graph with the greatest number of edges, which scores the only a very few of these highly compact forms in which maximum
highest on all these adaptability measures. adaptability of the first two types can be achieved. (On the other hand a
Matela and O'Hare (1976a) discuss this question of the subgraphs of an compact form has few external walls, and there are thus fewer positions
adjacency graph, in relation to polyominoes, and remark too on the on the boundary at which additional rooms might be attached. So its
significance for adaptability. They were not able to count all spanning 'external adaptability', its potential for growth, is, by contrast less than
subgraphs of larger polyominoes by computer. However, as a step towards that of more elongated, branching forms.)
this, they did count numbers of labelled spanning trees. For a given number Another author who has studied the subgraphs of adjacency graphs in
of edges, it is in general those graphs with many cycles which possess the this connection is Marsh (1976). Because, as we have seen, it is plans
greatest numbers of labelled spanning trees as subgraphs. Figure 10.13 whose adjacency graphs are most highly connected which offer the greatest
shows all those 'perfectly cyclic' graphs which may be the adjacency (internal) adaptability , Marsh confined his attention to counting the
graphs of polyominoes up to n = 9. In each case all edges lie on cycles. numbers of subgraphs of maximal planar graphs. And because he made
this enumeration by hand, he considered only such graphs on five and six
g
n=4 n=6 n=7 n=S n=9 vertices:
tml
cyclomatic
number
C - 1
I ~
Marsh's results confirm the general trends suggested by our exercise with
graphs on four vertices. Marsh also notes the importance for the fit of
requirement graph with adjacency graph, of the number of cycles in both,
C - 2 and the lengths of those cycles.
Notice that part of the system of Mitchell, Steadman, and Liggett
rt _t i-;> carries out automatically this exact same process of mapping labelled
•
•>-1--- ) ).
- spanning subgraphs into graphs, and could in principle be used for
•r-1- >-t-<) measuring adaptability.
All this, of course, takes no account of the other factors governing
>-I-) ;-1-)
adaptability such as room size, servicing, etc which would need to be
C =3 introduced into a more realistic exercise (refer to Fawcett, l 976a; 1976b).
llJ
The enumeration of rectangular dissections suggests ways for measuring
the ease with which plans of this form might be changed internally, or
extended on the perimeter. Indeed certain of the original generating
algorithms proceed by exactly these operations: · an interior room is
c=4 divided into two parts, or a new room is added along an outside edge of
Figure l 0.13. All 'perfectly cyclic' graphs which can be the adjacency graphs of the plan. For every plan with n rooms there is a certain number of its
polyominoes, up to n = 9 (from Matela and O' Hare, 1976a). All edges in these 'descendants' with n + I rooms produced through each of these operations:
graphs lie on cycles. Such graphs have the maximum number of labelled spanning These numbers could be interpreted as measures of possibilities for physical
trees for given n: the highly compact forms of the corresponding polyominoes, alteration of. that given plan by the subdivision or addition of rooms.
considered as plans, offer maximum internal adaptability.
The possibilities for removal of internal wall segments, thus reducing the
204 Chapter 10 Plan morphology and building science 205
number of rooms, are given conversely by the number of 'parent' dissections dining room, kitchen, living room; hall, we, corridor, and staircase. He
with n - I rooms from which the dissection in question can be independently assumed that the first three of these could be approximated as two
generated. rectangles in plan, either living room plus combined kitchen/dining room,
No general investigations along these lines have so far been undertaken. or dining room plus kitchen/living room. (The possibilities that all three
In practice the possibilities for subdivision or extension of building plans rooms might be separate, or all three functions combined in a single room,
are crucially dependent on dimensional as well as on adjacency factors. were also left open.)
One specific study has been made, however, of the adaptability of Bailey then enumerated all distinct ways of placing these two rectangles
house plans, which takes account of realistic room size and adjacency in a rectangular shell under certain adjacency constraints, such that a third
constraints together. This is the work of Bailey ( 1977), who examined connected polyomino-shaped area remained for the planning of all the
rectangular plans with room sizes based on the recommendations of the other spaces. These schematic plan outlines were then combined with a
Department of Environment handbook Space in the Home (DoE, 1968) systematic permutation of positions and orientations for staircases- either
with plan areas and overall plan dimensions corresponding to the 'metric dog-leg or straight flight types. Figure I 0.14 shows some examples.
house shells' proposed by the National Building Agency (NBA, 1969). Certain of these plan diagrams were selected for dimensioning within the
Again because he was working by hand and without the benefit of specified standards, and were planned in detail on both floors. In each
computer methods, Bailey was obliged to simplify the problem. He case Bailey determined whether within the dimensional constraints it was
considered two-storey, nominally 'five-person' houses, each containing a possible to offer certain very specific options for adaptation, such as the
minimum of three bedrooms. He assumed the ground floor to comprise a ability to offer an extra bedroom , a study, or utility room, or various
combinations of dining, kitchen, and living areas. Figure I 0. 15 gives
2 3 4 s 6 7 8 etc examples of some of these options for a given 'shell' size, and which
correspond to some of the different schematic arrangements of the ground
A floor which were illustrated in figure I 0.14.
Such an exercise could be completed and carried further by means of a
B gmill ~
m ~~~
~ m ~
computer method such as that of Mitchell, Steadman, and Liggett. Room
arrangements on both floors (or indeed on more than two floors) could be
exhaustively enumerated under adjacency constraints, and under dimensional
C I Md constraints both on the rooms and on the plan overall. The resulting variety
of plans could be interpreted in two ways.
It could be studied as Bailey does in the context of specific forms of
D Ell adaptability- how many ways in which functions could be assigned to
rooms in some given plan, or how many useful ways in which that plan
could be physically altered by moving say one, two, or more internal wall
E trl segments, or altering certain services, doors, or windows.
Alternatively the range of such plans could be viewed much more
generally as plotting the total extent of possibilities in design within some
~ set dimensional 'discipline' - such as those imposed by the recommendations
of Space in the Home and the NBA's 'metric shells' . It is clear that such
etc disciplines must limit plan variety-and in practice these limits become
Figure 10.14. Permutations of general arrangement for the ground floor plans of small partly known with time, in an empirical and anecdotal sort of way, to
houses, from a study of adaptability by Bailey (1977). The houses are two-storey, architects who have to work within them. A systematic approach, however,
nominally 'five person ' with at least three bedrooms. The ground floor comprises in would measure these effects precisely. Such exercises as the NBA's
each case a dining room , kitchen , living room, hall , we , corridor , and staircase. The tabulation of Generic Plans (NBA, 1965) could be automated. It should
first three of these are assumed to be planned within a single rectangle or two
be possible to identify the effect, on increasing or reducing the number of
adjacent rectangles. The figure shows some examples of possible placings A, B, C, ... of
these rectangles (unshaded) within the rectangular house 'shell', leaving a circulation resulting plans, of changes in specific dimensional or adjacency constraints.
and service area (shaded); combined with possible placings l, 2, 3, ... of staircases- in Some dimensional disciplines in housing design, notably that put forward
these instances a dog-leg type. by Habraken ( 1961) in his 'supports' concept for separating the structural
206 Chapter 10 Plan morphology and building science 207
El6 E7 frame and services in a block of flats from the detailed layout of internal
partitions, have been intended to increase the initial choice of plans
available to the tenants, and to increase their options for making subsequent
alterations to those plans. The Greater London Council's PSSHAK system
(Hamdi and Wilkinson, 1971) is another case in point. Again an exhaustive
enumeration of those possibilities would give an exact measure of the
success of the underlying system of dimensional constraints in increasing
the range of plan options.
w~dbb u)J ones which may move as technique develops or new materials are
discovered- something which events in the history of architecture have
C sb
ba
[ ]ttrJ repeatedly shown.
This view of building science leads us to see the study of architectural
history then, insofar as it has to do with building forms, materials, and
: ...
.:•
, or ------~
' ' constructional methods, as a description and analysis of where the actual
lies, or has Iain, within the possible. At the same time, it is only through
:Bl·
EiJEfJ
empirical study of actual buildings and their properties, as I have argued,
\ dk w that these theoretical realms of 'possibility', at any period in history, can
: C in the last analysis be defined.
'
.. :'
'------ ..I
..
Key:
I
k
d
dk
living room
kHchcn
dining
dining/kitchen
sb single bedroom
db double bedroom
ba bath
n.
EiC
Exercises
Rather than specify formal exercises relating to this chapter (and the next),
I suggest that you might like to use the material presented here for the
basis of your own research projects.
For instance you could take samples of plans, for example, those of
small houses or apartments, and make analyses following the lines of my
Figure 10.1 5. Some of the basic dimensionless plan arrangements of figure 10.14, own work (pages 183 -184 ), plotting the occurrence of dissection type
dimensioned according to British governmental recommendations (DoE, 1968; NBA,
1969), from Bailey's ( 1977) study of adaptability in house plans. Plans on both floors
arrangements on Combes's diagram; or following the lines of Bon's work
are illustrated, together with various options for alternative room uses or additional (pages 185-187), drawing and classifying their labelled access graphs,
rooms. Plan dimensions overall are S · 7 x 8 · 1 metres.
208 Chapter 10
11
measuring the valencies of vertices, and measuring the numbers of faces or Plan morphology and architectural history
cycles in the graphs.
You could study the relationship for actual plans of access graphs to
adjacency graphs (see pages 187-188). What proportion of adjacencies "Archaeological studies and the history of science are concerned with things only as
between rooms or with exterior regions are used for access? What technical products, while art history has been reduced to a discussion of the meanings
proportion of adjacencies with the exterior are used to provide natural of things without much attention to their technical and formal organization. The
task of the present generation is to construct a history of things that will do justice
light and ventilation? both to meaning and being, both to the plan and to the fullness of existence, both
You might take the access graphs of larger buildings of some given to the scheme and to the thing."
functional type, for example, offices or blocks of flats, and try to classify George Kubler ( 1962)
them, using some of the gross measures of graph structure described here,
and making also some measures of distances along the circulation network, With a few outstanding exceptions, applications to architectural history of
as suggested on page 189. techniques for morphological analysis are, so far, fewer even than those to
You might explore the questions of adaptability of rectangular the traditional subjects of building science. Therefore this final chapter
dissection type plans by internal rearrangement or by extension on the will be even fuller of speculative suggestions for directions in research,
perimeter (independent of any consideration of dimensions), as suggested than the last.
on pages 203-204, and making use of the catalogue of dissections in the Many of the descriptive methods and analytical tools listed in the last
appendix. chapter may be applied as well to old buildings, of course, as they may
to new- although a certain caution needs to be taken, that modern
functionalist preoccupations, tacitly embodied in some of these measures,
are not carried over to historical and social conditions where they do not
apply. For example, the idea of different rooms in a house being allocated
to distinct 'activities' was less developed in say the seventeenth century
(compare Barley, 1963) than it is today. (Even in modern housing design ,
functional room uses are often much less tidy in practice than the architect
originally supposes.) Again, the typical modern hierarchical classification
of roads related to their capacities for vehicular traffic will hardly be
applicable to the streets and alleys of the mediaeval city.
I have already made some suggestions for ways of classifying building
types on a morphological basis. The kind of multilevel hierarchic taxonomy
described in chapter 8, involving a conceptual separation of dimensional,
shape, and topological properties, could be applied in architectural history.
Indeed something very much along these lines has been proposed by
Guerra (I 977) in a paper on the rehabilitation of old buildings. Guerra
suggests that the access graphs of older public buildings such as churches,
convents, or theatres might be compared with the organisational or access
requirement graphs of modern cultural or educational institutions, to
measure their suitability for accommodating these new uses.
Guerra further imagines a kind of historical/geographical classification of
vernacular architectures in which the occurrence of buildings with similar
plan layout and construction might be plotted on maps. Contours drawn
on these maps could mark the boundaries between distinct local types, and a
series of maps made for different time periods could in principle show the
movement of these boundaries and types. This is a very grandiose project of
course, and one for which the basis in empirical survey work, in the case
of many building types and geographical regions, simply does not exist.
210 Chapter 11 Plan morphology and architectural history 211
Nevertheless there have been collected together a few archives of plans, 'probable', and might be expected to occur with similar frequency. For
as, for example, the specialised collections of British vernacular house example, since there are five polyominoes with four cells, so the probability
plans to be found in what is by now an extensive literature, which overlap of occurrence of each would be 1/5 or 0·2. If differently handed
the fields of architectural history, historical geography, and archaeology isomorphs were regarded as distinct, then there would be seven shapes,
[see Brunskill (I 971) for a bibliography), as well as in the inventories of and the probability of each would become 1/7 or 0· 1429. Other ways of
the Royal Commission on Historical Monuments for England and the measuring the geometrical probability of the occurrence of the same
Royal Commission on Ancient Monuments for Wales. It is perhaps shapes, which give slightly different results, are also possible.
unfortunate that the very extensive records of British houses brought Of course, the actual planning of the houses was by no means random,
together under the original initiative of Cordingley at the University of and the observed frequencies of occurrence of plan shapes in Dickens's
Manchester, do not always include full details of plan forms. Brunskill sample are very different from these notional probabilities- indeed certain
and others working from the Manchester materials have prepared maps of 'theoretically possible' shapes as we have seen do not appear at all.
exactly the kind envisaged by Guerra, although these largely relate to the Dickens proposes that these differences might be interpreted as evidence
materials of construction used in domestic buildings, rather than to their of two factors at work. The actual plans are chosen as being more
plan types.
For other building types, there are few if any comparable systematic
collections of plans. Most of what does exist-for 'polite', Western
architecture- has been listed by Pevsner in the encyclopaedic bibliography
and notes to his History of Building Types ( 1976); although that book in
itself is largely confined to names, dates, and stylistic comments, and
strangely, although it includes many illustrations of plans, hardly addresses
the question of plan form at an analytic level at all.
It is fair to say that in all this work any classification of plans which is
attempted in geometrical terms, tends to be on the basis of rather loose
and informally defined 'family resemblances' of shape and layout. Thus
Brunskill ( 1971) distinguishes such types of vernacular house plan as the
'hall', the 'two-unit', the 'inside cross passage', the 'central fireplace', and
the 'double-pile' families (figure I 1.1) (compare also Faulkner, 1958; (a)
Pan tin , I 962-1963).
One author who has attempted a greater precision in the geometrical
definition of types is Dickens ( 1977), who examined a sample of seventy-
four small Cambridgeshire house plans from the inventory of the Royal
Commission (1968). He approximated all of these plans as polyominoes
(taking the principal rooms as the cells of the polyomino, and ignoring
circulation spaces); and he drew the access graph in each case, labelled by
room functions. The houses are all two-storey and the graph describes
access on both levels, with the staircase shown as a zig-zag line. The
resulting types are illustrated in figure 11.2. They correspond to six
different polyominoes with one, two, three, and four cells, and the total
number of different labelled graphs is 21. The column of figures
accompanying each graph shows the number of occurrences of that type
found in each of six successive fifty-year periods, from 1550 to 1850.
Notice that all polyominoes are represented up to three cells, but that
only two appear out of the five possible polyominoes with four cells. Figure 1I.I. Basic types of English vernacular house plan (from Brunskill, 1971 ):
If the plan shapes had been arrived at by some random process, then (a) the hall type; (b) the two-unit type; (c) the inside cross-passage type; (d) the
for a given number of cells n each polyomino could be said to be equally central fireplaces type ; (e) double-pile plans.
Chapter 11 Plan morphology and architectural history 213
212
compact, this resulting in a reduction of the ratio of wall surface to plan frequencies of occurrence as plans- although it should be said that the
area (or to volume), and a minimisation of distances within the plan. Or sample is too small for statistically meaningful correlations to be made in
the chosen forms might be those which are easier to construct- specifically, any case.
Dickens argues, those in which the plan perimeter has fewer re-entrants, and For each overall polyomino shape there are several alternative positions
where as a consequence the roof construction in particular would be simpler. for the entrance to the house and several positions for the staircase. These
He u_ses geometric measures of shape designed to capture these properties, give rise to a number of theoretically possible combjnations, which again
very similar to those used by March and Matela (1974) (see page 127), may be compared with those actually occurring.
and ranks the polyominoes on those bases. The square polyomino with . ~ickens u~es standard statistical tests to determine whether there is any
four cells scores highly on both counts as would be expected, and it is s1gmficance m the relative frequencies, in each time period, of the
indeed the most frequently observed four-room plan shape. The ranking occurrence of plan shapes, plan sizes (numbers of rooms), and types of
of the other four-cell shapes does not, however, correspond with their access graph. These tests reveal certain significant associations of types
□ □□ □□□ OD □□ □□□
□ □□ □
b b b O b b b
~I
b 0
~·! ~ ·
0
b b 0
0
0
s 4
b b b b
¥-1
0
~·!
I k
I
0 b 0
5 0
I
b b 0
b b b 0
I 0
0 1
~·
0
2 0 s 0
0 0
0 1 k
0 0
b b C b b b
0
~: ·f-°· !
0 number of plans
3 occurring in period:
0 living room 0 1551-1600
0
0 s service room 0 1601 - 1650
0 k kitchen 0 1651-1700
b b c b I
0 b bedroom 0 1701-1750
·=¥=·!
0 c circulation 0 1751-1800
0
0 o outside 0 1801-1850
b b
0
9
O
f staircase
·F·i
0
Figure 11.2. Classification of a sample of seventy-four small Cambridgeshire house Figure 11.2 (continued)
plans (from Dickens, 1977). The plans are approximated as polyominoes (with the ~hown, and with the staircase indicated by a zig-zag line. Against each graph is
principal rooms taken as the cells of the polyomino and with circulation spaces md1cated the number of occurrences of that type in each of six fifty-year periods
ignored). The access graphs of all plans are illustrated, with access on both storeys between 1550 and 1850.
214 Chapter 11 Plan morphology and architectural history 215
with periods, as, for example, a large number of instances of the straight of one or more structural bays, and the plan is extended linearly by the
three-cell plan in the period 1650-1700, and high frequencies of occurrence addition of bays at each end.
from 1650 to 1700 for the first two access graphs shown below and from The later houses move across the space of Combes's diagram towards
17 50 to 1800 for the third one. 2w = 11 + P where modern plans are found. This could be interpreted in
terms of several factors. The masonry techniques serve to free the plan
from the constramt of a narrow linear shape which, was imposed by cruck
construction. The increasing differentiation of fun ctions within the house
~esuIt.s m
.
a greater complexity of plan form, more rooms, and hence more
mternal wall segments. And the growing demands for privacy require that
roo~s should be accessible from common circulation spaces without one
A second rather comparable piece of work on vernacular house plans havmg to pass through one room to reach others. In turn this requires
was carried out by Arbon (reported in Steadman, 1976), and follows plan forms with 'overlapping' adjacencies between rooms, of the kind
exactly my own exercise in locating dissection type plans on Combes's occurring with the greatest freq uency towards the lower boundary of the
diagram of internal and external wall segments as described in the last diagram.
chapter. Arbon's sample of thirty-eight plans dating from 1450 to 1690 These two studies, Arbon's and Dickens's, would have to be taken much
was taken from Fox and Raglan's (1951-1954) work on the houses of further and applied to larger samples, to produce historically conclusive
Monmouthshire in Wales. The results are illustrated in figure 11.3 where ~esults. _However, they do indicate a general approach , whereby some
the houses are classed into three groups: cruck houses built between 1450 population' of plans of real buildings is compared with the whole range of
and 1520, pre-Renaissance masonry houses of 1540 to 1580, and plans which are 'theoretically possible' on some appropriate geometric
Renaissance masonry houses of 1600 to 1690. definition. In this way those features of their spatial arrangement which
Recall that modern house plans from the NBA's Generic Plans were are contingent on the necessities of the geometrical discipline are identified
heavily clustered along the line 2w = 11 + p, marked again in figure 11.3. and separated out i_n the analysis. Any further concentration or clustering
It will be seen that the earliest, cruck houses lie by contrast predominantly of the sample w1thm the space of geo metrical possibilities, can be attributed
along the upper limit of the diagram at w = 2p + 4. The dissections on to the operation of additional constraints- to the fact of architects or
this line are those consisting of straight rows of rectangles, corresponding builde~s limiting their choice (whether consciously or unconsciously),
to the characteristic planning of cruck houses in which each room consists accordmg to more restricted criteria. In the work just described these
l
were !nferred to be technical and functional criteria, to do with structure
• and circulation.
13
14
0 It is in th~s ki~d of process, of setting the actual against the possible, as
12 o• I have descnbed 1t, that I believe much of the promise of a morphological
11 I •• analysis of historical buildings lies. The mere act of transforming the
~ -1~ ~ ~ean,s _by wl~ch a ~Ian is rep_resented, from a traditional scale drawing,
mto d1mens~onless form or mto the form of an adjacency graph say, does
~ 8 not necessanly of itself throw any light on historical issues. In many cases
7
6
as much can be 'read', and as easily , from a simple inspection of the
5 conventionally represented plan. Nevertheless there are some buildings
where a transformation of their plan representation does have the effect of
I I I I l l I -r-, 7 l I - , -, illuminating properties which are by no means otherwise apparent.
1 2 3 4 5 6 7 8 9 10 11 121314151617 An outstanding example of this is provided by the work of Hillier and
Partitions, p
colleagues (Hillier et al, 1978a; 1978b; Hanson and Hillier, I 979; Hillier
Figure 11.3. Classification by Arbon of thirty-eight Monmouthshire house plans and Hanson, forthcoming) who drew the 'permeability' structures- in our
(from Steadman, 1976). The plans are plotted on Combes's graph of numbers of
terms the access graphs- of a variety of buildings ancient and modern.
internal and external wall segments (see figures 8.1 and 10.4). The sample is classified
into three groups: cruck houses, 1450-1520 (light circles); pre-Renaissance masonry !~e method is most revealing with large and complex buildings. However,
houses, 1540-1580 (heavy circles); Renaissance masonry houses, 1600-1690 (solid ~t 1s perhaps best explained first in relation to small plans in order to
circles). illustrate the concepts involved.
216 Chapter 11 Plan morphology and architectural history 217
The figure below on the left shows a rectangular dissection type plan of rooms are interconnected by relationships of access, by doors or openings.
three rooms labelled a, b, and c as shown: In the examples of figure 11.4 it is the same dissection which provides the
basis of the plan throughout, whereas the depth of the plan in this measure
a C of Hillier's varies between I and 3.
2···y""'"'
I ............ . A n:ean depth figure d0 ; may be derived for the plan as a whole, by
0 ..... ~ b..... .
summmg the depths of all spaces from the exterior, and dividing by their
number. These values are given for the plans illustrated in figure 11.4.
One possible access graph for this plan is illustrated, with a vertex to They may be compared directly since the number of rooms is the same
represent the exterior region o. The figure on the right shows the same throughout. In general though, values of mean depth are dependent
access graph drawn out again but in a specially standardised (Hillier calls it naturally on the size (number of vertices) as well as the structure of an
a 'justified') format. The vertex for the exterior o is placed on the lowest access graph. However, it is possible to obtain a relative or normalised
level in the diagram, level 0<28>. Those spaces which are directly accessible measure of mean depth by means of the expression
from the exterior- in this case only room b-are placed on the first level 2(d°' - I)
up, level I. All spaces to which the shortest route from the exterior v-2
contains two edges- the remaining rooms a and c- are placed on level 2.
For graphs with more edges the process can be continued up to any The maximum depth for graphs on v vertices occurs with graphs which
required number of levels. consist of a simple linear tree, in which the exterior region o is represented
The next figure shows all possible access graphs for the same arrangement by one of the end vertices (as in graphs I to 6 of figure 11.4 ). In such
of rooms, the assumption being such graphs are connected (and, of course, cases the maximum depth must be v -1 , and the mean depth ½v. The
planar). minimum depth, and minimum mean depth, which graphs of any size can
take is I. These correspond to plans where all rooms are directly accessible
from the exterior_(as in graphs 26-28 and 35-38 of figure 11.4).
Substituting for d 0 ; in the above expression gives
3
,,~~~~'1J,~~
·····ic·············.. ··ii>"················i,,················1a·················ii,·················ia·········
~ : : : ~:: : :·: : : : : ~::::::::::::::::: ~:: : : : : : : : ~:: : ::::::::::: ~:: : :::::::::::· ~::::::::::::::fJ.~:::::::::::il:.~::::::::::lfi:..~::::::::::rii.~:
2 . .... ·················· .•...•........ .... ········•········· ........•..•.....• ··········•···•· · · ......•.•......... ... • .. . .............. . . . ......... ·· ····•·•··•··•·• . .....•..
b C a a b C C b C a
O O O O O O O O O 0
diameter: 3 3 3 3 3 3 3 3 3 3
cyclomatic 0 0 0 0 0 0 0 0 0
number: 0
mean depth: 2 2 2 2 2 2 1·33 1· 33 1.33 1. 33
II 12 13 14 15 16 17 18 19 20
,~,~~~,~[J~
2 b ··~·················~·················1·
I •. •..
a b · c ············1··
a ············y__··
c a b·································%···········~···········~············~·
b b ca c ab a
.. .... ... . . •.• •....•........ . .•..•..•......•.. ...•.....•..•. . ..• •.•.•. . ..•. . ... ••• •. ..•••.... ••.......•.• .. .... .••...•..•. ••••.. . ..............
······
..
o .~... o··~··········~·· ;;.:.............. : ................. : ................ : ............~ :........... : ................. : ................. : .............~... ,;··~·
diameter: 3 3 2 2 2 2 2 2 2 2
cyclomatic 0 0 0 0 0 0
number:
1 ·66 1-66 l ·66 l ·66 1 ·66 1 •66 l ·33
mean depth: 1·33 1 · 33
Figure 11 .4. All thirty-eight possible access graphs for the three.room plan illustrated. The graphs are set out in Hillier's 'justified' format below the plan
in each case: the exterior region o is set on level zero, any space directly accessible from the exterior on level 1, any space accessible from the exterior
21 22 23 24 25 26 27 28 29 30
0 .•~••.• ~........•..•~ .••.•: . .. ••. . .•.•a••• .••~ ••••..••••• ~. . .~.•••••••• ••••~. .~ ••••••••••• ~.. . ... ~ .... ~ .•••••.•.~ .••.•~ ..•.....•.. ~•••• ~•.•
1 •• · ···················· ········ ···· ····· ····· ·· .. . ......•. .•• ...... ... . ... ... .......• ... ..... ... ... ... . ..
O O O O O O O O O 0
diameter: 2 2 2 2 2 2 2 2 2 2
cyclomatic
number: 1
mean depth: 1 ·33 l ·33 1-33 1-33 1. 33 l ·33 1 · 33
31 32 33 34 35 36 37 38
number, e - v + 1, is equal to the number of faces (not counting the Plane graphs on a given number of vertices v may contain any number
exterior face). Cyclomatic numbers are given for all the graphs illustrated of interior faces from zero to a maximum given by 2v - 5. This follows
in figure 11.4. It does not help especially in counting the number of from the fact that for maximal planar graphs e = 3v - 6. Euler's formula
faces in a graph to draw it in the justified format. Consider, however, graphs gives f = e - v + I if the exterior face is excluded. So for maximal planar
with no interior faces, that is, trees, where the cyclomatic number is zero. graphs f = 3v - 6 - v + I = 2v - 5. A relative measure comparing the
The implication in access terms is that there is a unique route from the number of faces in a plane graph with the maximum possible, is therefore
exterior to any room, and that shallower rooms lie on this route and given by the ratio f /(2v - 5), which can take values between O and I.
hence control the access to deeper rooms. The justified version of the Values for this ratio are given for all the access graphs in figure 11.4,
graph shows this very clearly. which as will be seen, vary between O for the first graphs (the trees) and
I. . ¥. . ..
For example, in the access graph on the left below, room b controls for graph number 38 which is maximal planar and in which the number of
access to room c, and room a controls access to room b. interior faces is given by 2v - 5 = 3. In this way , graphs with differing
numbers of vertices can be compared for their degree of connectivity, by
3 .......
2 .......
b
,............
.. .. .. 2 ...
a...~..
y.. l'. J
2 ... 2 ... ·i;'""
a measure which is not unlike Garrison and Marble's gamma index <29>.
Individual vertices in the access graph, that is, rooms, may be characterised
by the number of faces on which they lie. At a maximum this number
,
I ..b.... 1 ..b.... / .. . I ... ...... .. . can equal the valency of the vertex (the figure on the left below gives an
a a b
example):
0 ............ . 0 .............. 0 .............. 0 ............. .
···1
. . . . ..
0 0 0 0
.... ..........
The second graph shows a case in which a controls access simultaneously
... i ...
both to b and to c. Hillier would describe the vertices b and c in this
second case as being symmetric with respect to a: neither of the spaces b ....... ....... .
or c controls access to the other from a. By contrast in the first graph, b i
and c are asymmetric with respect to a: b controls access to c from a. . .. . ... . . . . . .. . . .... . .. ..................... .
0 0
Wherever faces and hence cycles are introduced into the access graph, it
follows that there must be alternative routes from the exterior to certain The maximum value which any vertex valency in a graph can take is
rooms (and alternative routes between pairs of rooms). The third graph clearly v - I , where the vertex in question is joined directly to all other
gives an example. In many cases control of deeper spaces by shallower vertices (see the figure on the right above).
ones is lost. In the last graph, c is accessible from the outside either To turn now to the actual uses made by Hillier and his coworkers of
through a or b. these methods and measures: in one study Hanson and Hillier (I 979)
Hillier has a special term to distinguish those graphs which possess have examined the plans of small houses, specifically a sample of
cycles passing through the exterior region. He calls these distributed seventeenth-century houses drawn from the work of Wood-Jones ( 1963).
graphs. The last graph above is a case in point. Graphs without such Figure l l .5(a) shows the ground-floor plans of twenty-one houses in the
cycles (that is, trees, or with cycles which do not pass through the area of Banbury, Oxfordshire, dating from 1575 to 1700, and described by
exterior, as in the first three graphs above) are referred to as nondistributed. Wood-Jones as being variations of a 'through-passage' type. The through-
It is clearly possible for distributed graphs to contain subgraphs, each passage, as its name implies, runs from front to back of the house and
connected at a single vertex , which are in themselves nondistributed. The connects a front entrance from the road directly to a back entrance from
.....
J.
following graph is distributed in these terms, but the subgraph bcde is a a yard or garden. Rooms in the house are set either side of the passage,
nondistributed subgraph: typically with a kitchen and service rooms on one side, and a hall and
parlour on the other. [The through-passage is thus a classic example of a
J ... 'cut' room, in Bloch's terminology (compare figure 8.9 on page 127).]
2 ... ~.~... c...~..
29
I ... . .... .. < > The gamma index by contrast relates the number of edges in a graph to the number
b of edges in a maximal planar graph on the same number of vertices. Hillier refers to
o.................. the degree of connectivity of a graph measured as described, as its 'relative ringiness'.
0
n
~;'--·~
k
Ground-floo r plans are sho~ with lower ::ds
•ght The road frontage ts shown ~Y
:;t;nded broke n line. Basement locations are
hatched.
~~..
Plan morphology and architectural history 223
.i. ] ~
c 1603 Warmington Manor
T:]El
1654 Poplars Farm,
Chacombe
~u -
c 1640 Bloxham
p . k
ington Manor
p °'{···ph /k
0 ~ Poplars farm, Chacombe
P~ _.- ,., k
y Bloxham
·./;k sy p
.' 'Jo//
1607 Hornton Manor I 65 5 Deddington c 1640 Bennetts, h 0 ~ Hornton Manor : %".. Dcddington · ~§ Bennetts, Bloxham
~~ p\J.
I
----~ ~ -
1699 Home Farm,
-y
h .:
'k
h •...
P~
h . :
c1615 Cromwell Cottage, 1685 Manor Farm, 0 Cromwell Cottage, o Manor Farm, 0 • :omc Farm, Williamscot
Great Bourton
ib:J
Hornton Hornton Great Bourton
~~ ~ __ _
W.~ l:l _ [JG__ s .
k
-
P~
originally : h .· k
p ~r t or: h ~ p
c !61 3 Sunnyside Cottage, c 1685 Little Bourton c !680 Mansion House Farm,
0
Swerford Mollington
Sunnyside Cottage, Swerford Little Bourton Mansion House Farm,
p
stairs (? )
h r:71
i.J.
p
.
Pvk
Mollington
s s~ ... \
\ok
!g
/
Sycamore Terrace,
··..
Y
p s
k
Springfield House,
. k:
P\0 h
:
O
s/ k
The Mount, Hornton
Bloxham Middleton Cheney Homton Bloxham Middleton Cheney
---E-1.r:r ff:i.:il
c l 640 McGreals Farm, 1685 Friar's Cottage, :c 1695 Milcombe
.(p h~<\_ ..
- Friar's Cottage,
P k s
. .. ~
s
2 _3 unit plans, 1603 - 1640 2-u nit pla:s, 1654 - 1685 3-4 unit plans, 1640-1699
stairs k kitchen
• circulation, storeroom p parlour
h hall s service
(b)
Hanson and Hillier drew the access graphs of these plans according to
the conventions described earlier (see figure 11.S(b ). Vertices representing
principal rooms are shown as hollow circles and vertices representing
circulation spaces and storerooms as solid circles. The through-passage
appears immediately , as a pair of edges joining the exterior space o at
level O to the vertex representing the passage itself at level I . A cursory
look serves to show, as Hanson and Hillier point out, that roughly one- .,,
third of Wood-Jones's plans in figure I 1.5 do not in fact contain a
..... . .¥. . . .
through-passage. Those which do may be characterised as belonging to
one of the two following basic arrangements, or minor variants on these:
~p
.......................
k h p
k
h
..................... ........
-=9°0
...
0.
~
The structure of the access graph in itself is the same in both these cases,
.,,~.-=
•• 0
0
~
with long and short arms leading off the through-passage vertex. The 0.
0.
u
.,,>
::,
of the early seventeenth century: some examples are shown on the left of c
8 ..c
figure 11 .6. Notice that the parlour here is generally a deep, hence
0.
-5
~
relatively private, space, and often with only one door (that is, vertex
C
valency I). Overall, these plans are deep by comparison with later types,
"
..
~
C
... 0.
0
~
service rooms are often directly accessible from the exterior, whereas the
parlour remains a deep nondistribut ed space with single access and not on
the main distributed cycle of rooms.
Hanson and Hillier discuss the possible relations of these evolutionary
changes in plan form to economic and social changes, and in particular to
226 Chapter 11 Plan morphology and architectural history 227
changes in family structure, village life , and the relations between the
sexes. In general. and not just in this piece of work on domestic space
1638 organisation , the primary interests of Hillier and his colleagues are
Thenford, Northants
sociological and anthropological, rather than technological. It is their
late 16th and 17th century
Abernodwth Llangadfan, belief that social relations are not (or not just) expressed or communicated ,
Montgomeryshire but that they are actua lly embodied in or constiru1ecl by the spatial
C1650
Down End, Hook Norton, organisation of buildings and settlement patterns. Thus they are concerned
Oxon to find formal properties of these access graphs which might coincide with
l social distinctions, or which might affect the ways in which contact
between individuals or social groups is allowed or prevented.
L....______
er:"'•
I
1658
Charlton, Northants
C1610
North Leverton, Notis There is not the space here, nor would this be the place, for me to try to
do justice to the range and depth of Hillier's sociological thesis. I will try
simply to indicate some of the broad lines along which the argumen t is made.
~
For many if not the majority of building types it is possible to make a
C1625 fundamental distinction between ' inha bitants' and 'strangers'. The
t~ C1680
Hook Norton, Oxon
Bayton, Notis ' inhabitants' are either the permanent occupants of spaces, such as
residents in houses or the occupants of private offices; or else they may
be those whose social status or occupation is identified with or 'mapped
into' a building or room, such as the priest in a church, the players in an
1636 1673 auditorium, or the shopkeeper in a shop. 'Strangers', on the other hand,
Dial House, Sulgrave, Burringham, Lines
-·qrm
--------· Northants are obviously the members of the public, who might attend a church
service, go shopping, or listen to a concert.
.. H : P : • h k k
For certain buildings there would be a subset of 'strangers' - they may
be called 'visitors' -who are still in a kind of subordinate position vis-a-vis
t~
1658 the 'inhabitants', but nevertheless come to occupy the building for
w..,.,.., ·•~.
Northants h p extended periods. Besides visitors to private houses or guests in hotels,
IP ' I J-~fu•::,m,
- Northants
o...m...,,
C 1636
The House of Dennis Jackson,
Yeoman, Wigston Magna,
Leics
this category would include pupils in a school, convicts in a prison , or
patients in a hospital. The last case, that of a hospital, is especially
complicated since there exists a whole gradation of 'strangers' from casual
visitors, through out-patients, to long-stay in-patients or permanent inmates.
To return to the distinction between the distributed and nondistributcd
parts of the access graph: this distinction has the effect of partitioning the
vertices of the graph into two sets, as illustrated in figure 11.8, those
which lie on cycles passing through the exterior, and those which do not.
C 1640 It is Hillier's proposal that the distributed vertices will in many cases
Fir Cottage, Church lcomb, correspond to those parts of a building to which strangers may gain access.
Glos
C 1660 Meanwhile, the undistributed vertices, which lie on what amounts to a set
Laurel Farm, of trees or tree-like subgraphs attached at their 'roots' to the distributed
Shotteswell, Warks
h hall system , will tend to be the preserve of the inhabitants. Where this
k kitchen correspondence does obtain , then social relations between inhabitants will
10 0 10 20 30 feet p parlour
be mapped into the nondistributed parts of the graph, whose structure
differentiates or separates them from strangers, and ensures their privacy.
This is the case with blocks of flats or office buildings (other than
Figure 11.7. Seventeenth-century house plans from various counties of central open-plan offices), as some of the plans illustrated in the last chapter
England, of the 'lowland' type (from Wood-Jones, 1963). suggested (compare figure 10.7), and it is also true for many larger
228 Chapter 11 Plan morphology and architectural history 229
mean
sd depth sd coals
....a-.~ ..
: : Coach houst : :
Coufl Wood
carrier
ho usekeeper's room
10 ?O 30 'l ~o leel
drawing
c- , , s a 10 12 u u mette room
I
(a)
(b) 7 6 4 3 2 0
Figure 11.9 . (a) Ground-floor plan of Buchanan House, Stirlingshire 1851 -1853
(William Burn) (from Girouard, 1971, figure 2, page 23). (b) The corresponding justified Figure 11.9 (continued)
access graph (after Hillier et al, 1978b, page 25). At the bottom of the graph are
the private quarters of the family ; at the top the servants' quarters and service rooms. number of spaces 78 cyclomatic number 14
It is set on the page such that depth increases from right to left. The shallow, distributed depth of deepest space 7 measure of relative connectedness 0 ·093
vertices with high valencies correspond for the most part to circulation spaces. On the mean depth of all spaces 3·5 depth of deepest undistributed 4
family side these link the rooms in which visitors would be entertained--drawing room, relative mean depth 0-066 space
morning room , dining room. On the servants' side they are where goods are delivered to maximum vertex valency 12 mean depth of distributed spaces 2-4
the house , and through which the servants penetrate to the family rooms. Deeper in the mean vertex valency 2-2 mean depth of undistributed spaces I ·5
plan are more tree-like, undistributed parts of the graph, corresponding on the family side
to the most private rooms: the Duke's private sitting room , the boudoirs, and bedrooms.
Access to these is strictly controlled by shallower spaces. The deepest space on the
servants' side is the plate safe. The solid vertical line marks the mean depth of the plan,
and the broken lines, the standard deviation of the depth of all rooms from that mean.
232 Chapter 11 l Plan morphology and architectural history 233
mean depth
~ prisoners
i
middle
yard ~ ::9
8
untried
prisoners
~
.."
-0
E
E
~
I "8-5..
-8=
> chapel
.;i ya1d
13 12 11 10 9 8 7 6 5 4 3 2 0
(b)
Figure 11.10 (continued)
prisoners are the deepest spaces of all. The shallower, distributed parts are the
preserve of the inhabitants-the warders. Besides the prisoners who here constitute the
visitors in Hillier's terms, there are visitors in the more usual sense, who are allowed in
under strict supervision, and who meet the prisoners in interface spaces- the visiting
rooms, situated at about half the depth of the plan.
number of spaces 124 cyclomatic number 12
Figure 11.10. (a) Newgate Gaol , Lon don. prior to alterations . circa 1836. (b) The depth of deepest space 13 measure of relative connectedness 0-049
corresponding justified access graph (after Hillier et al , 1978b. page 72). Here, mean depth of all spaces 6-4 depth of deepest undistributed 9
obviously, the control of deep spaces by shallower ones constitutes the whole purpose relative mean depth 0·088 space
of the building. The graph as a whole is extremely deep. and highly undistributed. maximum vertex valency 8 mean depth of distributed spaces 2·7
The cells all lie in the tree-like depths o f the graph ; and those for condemned mean vertex valency 1-8 mean depth of undistributed spaces S·O
t0
w
t
.i,.
~ I
~
~ ; {- "11\;,r
~ , po ...
iJ'dtUaGf"
_,
: ~t\~
r~ z~~ r ~~
I ~ =·. '
=
~ 1. /!Jf,j !O,.
c•ou• t
...,..,.:, co= < 11
~-+ II ·. _ L4
I--t-. ~]!?.-~I[~-
M CON•
1
o~,.,,.. ?1
·
h
1,1 ""· .. 11 I
t -
~~ ~
I
vi ,[.)ii, t
~
~
l l T( >l(N
·~. .-rr,, ._
~ .:.
ETIQ ~ •
;- ~~ ~
r::; >M< ~
11
~!:' "
j;,Eroro;;----
(a)
Figure 11.11. (a) Ground-floor, fust-floor, second-floor, and third-floor plans of Compton Primary School, Clerkenwell, London, as
remodelled 1970 (from Architects' Journal, 1971 , pages 708-712). (b) The corresponding justified access graph for all four floors (after
Hillier et al, 1978b, page 68). The graph here is much more distributed generally than that for Newgate- there are many more cycles.
The visitors- in this case the schoolchildren- are much less rigorously supervised than are the prisoners. But it is still true that the
inhabitants- here the teachers- occupy some of the shallower spaces, whereas the classrooms are deeper in the plan and lie in the
undistributed parts of the graph. (Since this particular graph is not planar, 'the measures of cyclomatic number and relative connectedness
are inapplicable.)
number of spaces 91 relative mean depth 0 -092 depth of deepest undistributed space 3 .,9
depth ,of deepest space 9 maXllllum vertex valency JO mean depth of distributed spaces 3 ·6 "S.
mean depth of all spaces 5· l mean vertex valency 2-6 mean depth of undistributed spaces 1 •S ..,
(1)
::;:i
..,, "'::,
3
0
00 ..a
g-
~ g
"'::,0-
°' ~
(")
[
"' I oo / _ .. o 0 - ~
o--o o c)L ~ -- :=..:o/ ::1,
1 3 - -- - - -t- "' "~
8. s-
Q. 3
(1)
(')
[ ~
~
e.
::,-
;;;·
0
w
r, - - - - c. '<
~
N
8
.l
:,
s.,
8
-----------4-,/...a.-------------------------'.\
(b)
t0
..,,
w
236 Chapter 11 Plan morphology and architectural history 237
At the very deepest levels in access graphs it seems that the spaces the mean valency over all vertices,
associated with the most elevated social status are to be found. In several the cyclomatic number (that is, the number of interior faces f),
examples of villages and palaces which Hillier and colleagues have taken the measure of relative connectedness f/(2u - 5),
from anthropological and archaeological sources, it is the deepest spaces the deepest undistributed space, where depth is measured from the nearest
which are found to be occupied by headmen, chiefs, or the most essential distributed space,
functions of government. The access to these elevated levels is controlled the mean depth of distributed spaces, and ,
by whole sequences of antechambers, spaces occupied by subsidiary the mean depth of undistributed spaces, where depth is again measured
officials, guards, etc. from the nearest distributed space.
Spaces containing the most sacred objects seem to have high values for Hillier and his colleagues have as their long-term aim to work towards a
the relative mean depth of all rooms measured from that room-they are theory of building types based on their spatial organisation, whose
remote, that is to say, detached from the remainder of the complex. classifications will follow the lines of these analyses. They suggest that a
Among modern buildings, one notable example of a deep (perhaps even primary dimension of classification will be according to variations in the
sacred?) space of a kind which is at least comparable in its structural relationships between inhabitants, strangers, and the interfaces at which
position in the access graph, is the operating theatre in a hospital. the two meet. Extensions of these same methods to larger samples of the
I referred to the notion of patients, pupils, or prisoners forming a class same functional building type may make it possible to identify repeated
of 'visitors' in institutional buildings. Here, as Hillier points out, the structural patterns in access graphs, with characteristic values for at least
spatial relation of visitors in this sense to inhabitants is inverted. It is now some of the various measures described. It may be that structural
the inhabitants- the teachers, hospital staff, warders-who occupy and similarities will be found not always at the level of whole buildings, but
supervise the distributed spaces, so that they can control entry to and sometimes for 'subcomplexes' within buildings. It may transpire that the
from the nondistributed spaces- the classrooms, the wards, the cells-by access graphs of actual buildings are constituted out of unique combinations
the 'visitors'. of relatively standardised or regular subcomplexes.
Figures 11.9, 11. l 0, and 11.11 illustrate some examples of plans, together This static analysis of buildings at fixed dates could be extended to the
with their justified access graphs, taken from a study of twenty large study of historical processes of change. Two types of process can be
buildings made by Hillier et al (1978a). The buildings analysed are of very distinguished here. There is the alteration of a single building by
varied kinds, containing anything between 30 and 300 spaces in each. conversions or additions, the sorts of changes which were discussed under
Half of these are palaces, temples, and tribal settlements from the the heading of 'adaptability' in the last chapter. Hillier and colleagues
anthropological and archaeological literature. The remainder are modern address this question in a preliminary way by drawing, in the case of two
. (nineteenth-century and twentieth-century) buildings including a school, a of their buildings, Newgate Gaol and Compton Primary School, a series of
prison, a hospital, and a range of other types. This is not in any sense graphs corresponding to the original plans and to succeeding stages in their
therefore a statistically random sample, but rather an attempt to explore later remodelling.
the dimensions of variation in their permeability or access structure, of The second and generally much more interesting process of change is
individual buildings of widely different character, age, and function. the evolution of a design as it is materialised in successive individual
Figure 11.9 shows the Buchanan House built in Stirlingshire, Scotland, buildings of the same functional or morphological type. Hillier uses a pair
in the early 1850s. Figure 11.10 shows Newgate Gaol in London as of biological terms here, to distinguish the genotype from the phenotype.
originally built in 1836. And figure 11.11 shows Compton Primary School The genotype in biology is what is transmitted in heredity by the set of
in Clerkenwell, London ( 1971 ). The figures at the bottom of the access instructions embodied in the genetic code. It is the type or design of the
graph in each case mark the depths dot of spaces, as before. The vertical class of organisms- so by analogy the type-design of a class of buildings.
solid line marks the mean depth dot, and the two broken lines the standard The phenotype refers to the actual individual, the organism, or in our case
deviation from this mean. Also tabulated for each graph are the values of the building, in which that type-design is realised. Phenotypes, actual
the measures explained earlier, specifically: buildings, may undergo processes of growth or adaptation during their
the number of vertices u, that is, rooms or spaces, lifetimes. It is only genotypes or designs which can properly be said to
the depth of the deepest space dgr", 'evolve'.
the mean depth dot, Empirical studies of the first kind of process, growth and change of
the relative mean depth 2(dot - I )/(u - 2), individual buildings, remain very few in number. Most of these, since they
· the maximum vertex valency, have been made out of an interest in adaptability, have tended to
238 Chapter 11 Plan morphology and architectural history 239
concentrate on changes in room use or room size without reference to the therefore quite close (Steadman, I 979). In the modern 'self-conscious'
spatial organisation of the plan as a whole (see Cowan, l 963; Cowan and process of design carried on by professional architects, the role of direct
Nicholson, l 965; Llewellyn-Davies et al, l 973). One exception is a study copying is obviously much reduced. Nevertheless all architects still work
by Bon ( l 973) who was able to find data on the stages of development on within a tradition, and still rely heavily on precedent and on recourse to
one building, a palace in Ugarit (Ras Shamra). He counted the numbers tried solutions. This can result in a continuity at the level of spatial
of edges and vertices in the plan graph of this building for eleven stages of organisation and underlying geometric form, while sµrface features and
growth, and in the access graph for nine stages of growth. He found that stylistic treatment are more rapidly transformed.
the ratio e/v remained remarkably constant throughout, that is, a precise For example, Hawkes (1980), in a historical review of the designs of
allometric relationship applied (figure 11.12). auditoria, points to the persistence of a very few basic auditorium forms,
As for evolutionary studies of types of building characterised in detailed from antiquity right up to this century. The Greek and Roman type, as
topological and geometric terms, I know of none other than the work on described by Vitruvius, is carried through to the Renaissance. From the
houses already cited in this chapter. Hawkes (1976b), however, has touched late seventeenth to the nineteenth century, a horseshoe-type plan deriving
in a more discursive way on some of the issues which might arise in this principally from Fontana's Teatro Tor di Nona in Rome (1671) dominates
kind of 'geometrical history' of building types. He emphasises the role European practice, and remains largely unchanged behind Baroque, Rococo,
of what he calls 'stereotypes' in architectural design. A stereotype is and Classical revival treatments. During the nineteenth century a new type
"a generally held notion about the nature of a good solution to any of rectangular 'shoebox' plan, with one or two balconies, emerges to satisfy
recurrent building design problem" (page 465). There is clearly a certain the demand for a type of hall specially suited to musical performances.
affinity between this idea of a stereotype, and Hillier's genotype- although As Hawkes ( l 980, page 6) says of the horseshoe form, "There were,
Hawkes's interests are more those of a building scientist than of a social clearly, deviations in detail from the precedent in response to local
scientist. (Also he sees the stereotype as something of which designers are conditions, particularly of size and social organisation, and also to
consciously, if only perhaps vaguely, aware, which is not necessarily true development in the style and scale of productions, particularly to
of the genotype in Hillier's conception.) accommodate the needs of opera ...". This is precisely the distinction of
ln primitive and vernacular architectures it is arguable that designs for phenotypes (the 'deviations in detail') from the genotype (the 'precedent').
buildings are transmitted and reproduced through repeated copying of the Similar themes, of the influence and persistence of stereotypes, are
existing type, or through repetition of a fixed set of inherited constructional pursued by Hawkes (1976b) in his 'evolutionary tale' of the history of
procedures, with only small changes to the design being introduced in each central city office building forms from the beginning of this century to the
'generation'. The parallel with biological evolution in this respect is present.
120 Hawkes attributes the longevity of some of the earlier forms in both
350 -
building types to the fact that they were known to be functionally
300 · 100 satisfactory, but the basis for this success was not understood at a
250 - 80
theoretical level. The horseshoe auditorium form, for example, had been
., ., arrived at empirically by a rather slow process of trial and error. Later
.,; 200 -
&l, ~ 60 0 architects tended to adhere to the stereotype since they had little scientific
~
iZ 150 - l:.J basis, specifically little acoustical theory, with which to predict the results
40 of radical departures from that form. The same is true, according to
100 "
20
Hawkes, for the stereotypical, naturally lit, and naturally ventilated, urban
50 - office building of the beginning of this century.
0 ~ -~~ 0 Departures from the stereotype may be occasioned by changes in social
0 50 100 150 200 250 0 20 40 60 80 100 structures and functional demands, as, for example, the way in which the
Vertices, v Vertices, v
(a) (b) classical theatre was 'brought indoors' in the Renaissance, and how a social
segregation of the theatre audience was subsequently effected by the
Figure 11.12. Numbers of edges e and vertices v in the plan graph (a) and the access separation of boxes from stalls and pit. Later technical innovations, for
graph (b) of a single building-a palace in Ugarit (Ras Shamra)-at successive stages in
its growth (from Bon, 1973). Plan graphs are shown for eleven stages of development,
example, those in artificial lighting and structural materials, provided the
and access graphs for nine stages. In both cases the ratio e: v remains remarkably opportunity for morphological change, by allowing wider and deeper
constant. balconies and freeing the stage from the constraints of natural lighting
240 Chapter 11 Plan morphology and architectural history 241
from top or side. The realms of 'technical possibility' were correspondingly composition and style in many areas of design and the plastic arts<30>.
extended. Stiny and Mitchell in the project just mentioned, for example, have
In the case of modern office buildings, Hawkes points to the comparable developed a grammar which constructs and enumerates not just the plans
effects of the constructional innovation of lightweight curtain walling, and of Palladio's (1570) actual projects as illustrated in the Quattro Libri, but
as a not unrelated development, the introduction of full air-conditioning. all those other plans, conforming to the same compositional rules, which
A new situation arises in the twentieth century with the emergence of Palladio never designed (see figure 11.13). •
building science itself as an academic discipline. The findings of building The subjects of settlement or urban morphology (see Conzen, 1960;
science become part of the mental equipment of architects, and a resulting Clarke, 1977; Hillier and Hanson, forthcoming; Hillier et al, 1976) are
emphasis is placed in design on those aspects of performance which can be strictly beyond the scope of this book. It is worth making the point,
predicted and controlled- first historically, in the case of office buildings, all the same, that no very distinct line of demarcation can be drawn
on daylighting and sunlighting, and later on the complete artificial control between the architectural and the settlement scales: indeed the size of many
of temperature and ventilation. As Hawkes (1980, page 471) puts it: a single building can approach that of a village, with an organisation along
"Along with the parallel and frequently related developments in the covered 'streets' or around internal courts. The real distinction is between
technologies of construction and equipment of buildings, this growth in enclosed space within buildings, and open spaces between buildings.
our skills of specification and prediction has been a fundamental force in
the determination of the nature of the built form".
It ought to go without saying that all these proposals for a morphological □□□
history of buildings and building types are made in the frank recognition □□□
that such a history would be a partial one, focussing on geometrical, □□□
4 5
material, and technological constraints, on functional performance, and as
~ ~
in the case of Hillier's work, on the relation of spatial to social organisation.
None of this denies the creative capacities of individual architects, the
pervasive influence which can be exerted by the example of great buildings,
or the status of architecture as an art. The hope is rather that such a
lt:jj
6
D7 8 9 10
programme may to some extent counterbalance or complement the
exclusive concentration by some architectural historians and critics on
personalities, styles, 'influences' in the narrowest sense, and especially
today on questions of semiotics and iconology.
This is not to suggest by any means that matters of architectural style, 11 12 13 14 15
and specifically as expressed in floor plans, are beyond the reach of formal
analysis, as the work of Stiny and Mitchell (1978a; I 978b) on the plans of
Palladian villas amply demonstrates. This work makes use of the logical
apparatus of the shape grammar, which has been developed by Stiny (1975)
16 17 18 19 20
and Gips (1975). A shape grammar can be imagined as analogous to a
grammar for an ordinary language. But where the latter specifies the Figure ll.13. Part of a catalogue of possible room layouts (in dimensionless form) for
permissible combinations in which words may be assembled in that Palladian villas (from Stiny and Mitchell, 1978b). The figure shows all schemes on
language into the one-dimensional sequences constituted by grammatical 3 x 3 grids- including that of Palladio's actual plan for the Villa Angarano (number 4).
Stiny and Mitchell also enumerate all schemes on 3 x 5 grids (not shown here).
sentences, a shape grammar specifies rather the combinations in which
two-dimensional or three-dimensional shapes may be assembled, to form
complex compositions in the plane or in space.
It is not my intention to go into the subject of shape grammars here,
( 30)As a matter of fact shape grammars have already made a covert appearance in
for the simple reason that they are the subject of the books by Stiny and
earlier chapters, since the method for generating rectangular dissections devised by Earl
Gips already cited, as well as being discussed in Stiny and Gips (1978), to (chapter 4) was formulated by him precisely as a shape grammar; and Krishnamurti's
which readers are strongly recommended. There is no doubt that the method of colouring on gratings ( chapter 5) also effectively simulates the operation of
shape grammar is a powerful tool for the investigation of questions of a shape grammar, with colouring rules taking the place of 'shape rules'.
242 Chapter 11 Plan morphology and architectural history 243
Figure l l .14(a) shows this type of graph for a suburban area of Reading
in which detached, semidetached, and terraced houses are all to be found.
The second type of graph represents relations of adjacency of buildings
to surrounding open spaces. It is assumed for the sake of simplicity,
working at this scale, that minor indentations of the perimeter of a
building can be ignored (according to definite rules which Kruger specifies).
For instance, a pair of semidetached houses with extensions at the back as
shown above is approximated by two simple rectangles. In the graph, the
vertices represent buildings again, and also exterior regions; and each edge
represents the adjacency of a building to an exterior space across a wall in
the building perimeter. Figure l l. l 4(b) shows graphs of types 1 and 2
combined (in which vertices representing exterior regions have been omitted
for clarity).
Two more types of graph in Kruger's analysis relate to the road system,
and to the means of access to buildings. Here the vertices correspond to
buildings once more, and to junctions between roads, driveways, paths, etc
(where these give access to buildings); and the edges represent the roads
and paths themselves. Figure 1 l.14(c) illustrates, for the same area of
Reading, a combination of these access graphs with the first graph (of
adjacency between buildings). The edges shown as broken lines connect
each building to its access point on the road network.
244 Chapter 11 Plan morphology and architectural history 245
The road network divides the whole urban area into a series of bounded building (a 'perimeter measure'), the average number of external walls per
regions or city blocks (they are the faces of the embedded road access array (a 'compactness measure'), a ratio of the number of external walls to
graph). Kruger describes a fifth and final type of graph in which the the number of party-walls in an array (a 'shape measure'), and the number
vertices correspond to these blocks, and the edges represent the adjacencies of party-walls per array (a 'connectivity' measure).
between blocks. Obviously it is only possible for buildings which are With these sorts of measures Kruger examines relationships, for example,
within the same block to be adjacent. Figure 11.15 shows the definitions between the connectivities of the building arrays and.the connectivity of
of all five of Kruger's graphs in diagrammatic form. the road system (these relations are shown to be weak); or the effect on the
Kruger's purpose in all this is to relate the building morphology to the connectivity of the building arrays, of an increase in the number of buildings
structural characteristics of the road network, and beyond these to other per unit of land area, that is, an increase in density of ground coverage.
properties of the urban system more traditionally dealt with by urban In this case, as would be intuitively expected, the connectivity of buildings
geographers and planners, such as floor space and population densities, increases, and the graph of type I comes to contain more cycles.
land use, and patterns of travel behaviour. To do this he makes use of Figure 11.16 illustrates such a graph for an area in the centre of Reading,
measures on all these graphs, relating their numbers of vertices, edges, where the housing is mostly terraced, and commercial and industrial
cycles, and (for disconnected graphs) components, of the kinds mentioned buildings are packed closely together. [Over the greater part of the city,
here in previous chapters. For example, he computes their cyclomatic by contrast, type 1 graphs are highly disconnected, and without cycles, as
numbers, and the values for various connectivity measures such as the in figure l l. l 4(a).] Kruger points out how in general, in Western industrial
gamma index of Garrison and Marble referred to in chapter 10. He also cultures at least, the graphs of adjacency of arrays of buildings, up to the
calculates for the buildings themselves the values for some measures which point where they are maximally connected, must still preserve outer-
are rather similar to those defined for polyominoes by March and Matela planarity- since it is usually required that every separate building have
(1974) and described here in chapter 8. independent access from the road network (graph type 3), and not access
A set of mutually adjacent buildings, that is, for which Kruger's graph of solely through another building.
type I is connected, forms what Kruger calls an 'array', in which the party Kruger's empirical work in determining values for and relations between
walls between buildings are analogous to the partitions between rooms in a certain of these measures, enables him successfully to simulate, by means
single building plan. Kruger takes the average number of external walls per of a probabilistic model, the distribution of numbers and types of building
arrays (types, that is, distinguished by their connectivities), throughout a
series of zones or cells representing the map of the whole city of Reading.
type 4 type S
Figure 11.16. KrOger's type I graph for a central area of Reading. The density is
Figure 11.1S. Diagrammatic representation of all five of KrOger's (I 977) graph types higher and hence there are more adjacencies between buildings, and more cycles in the
for urban areas. In the type 5 graph a vertex represents a city block, and an edge the graph, than say in a suburban area [compare figure l 1.14(a)] .
adjacency of two blocks across a street.
246 Chapter 11 12
Although Krilger's analysis is made of the contemporary city, and·applies Afterword: prospects for an architectural morphology
to one point in time, it is obviously highly suggestive for similar approaches
to the morphology of towns and settlements at earlier dates in history; and
indeed Krilger himself mentions possible extensions of his methods to treat In the introduction I mentioned the relation of the present book to The
the processes of morphological change in cities taking place over time. Geometry of Environment (March and Steadman, 1971 ), and I talked
about the purposes which the two books have in CO!JlmOn, including that
of introducing the architectural designer to some mathematical ways of
Exercises conceiving and manipulating the geometric forms of buildings.
As in the previous chapter, I suggest that you might like to use the work At that time many workers who were enthusiastic about the new
described here as the starting point for your own investigations of historical application of computers in architecture saw the promise of the machine
plans: analyses of house plans following the approaches taken by Dickens as a means for automating the process of designing (the computerised
or Arbon; analysis and classification of buildings of different functional layout methods described in the first part of chapter 9 were a product of
types by drawing their access graphs in Hillier's 'justified' format, and by that philosophy). This was not our view. Rather we believed that the
measuring the properties of those graphs; or analyses of the building most immediately promising use for the computer was as a tool for the
morphology and road and land-use patterns of villages or parts of towns representation of designs. The computer would, in this view, replace not
by means of the approach of Kruger. the designer, but his pencil and his drawing board. The computer was a
powerful tool for building models of buildings. If this idea was correct,
then mathematical tools would be needed for describing and operating on
the two-dimensional or three-dimensional geometry of building form, of a
kind which could be expressed in one or another computer language.
Developments in computer systems for building description since then
have done much to vindicate this belief (for example, see Eastman et al,
1975; Mitchell, 1977.)
Such computer models can serve two kinds of purpose, however. They
may be employed directly by the designer, not to generate his design for
him automatically, but to represent what he designs- to model it and
draw it- and to evaluate it, as, for instance, by calculating its cost, its
thermal behaviour, or some other aspect of performance.
Alternatively, the same kinds of model may be used by the building
scientist to make some more general experiments on whole ranges of
hypothetical building designs. The product of this experimental work is a
body of knowledge- a building science-which is useful to the designer at
a more strategic level. It comes to form part of his education, part of his
background understanding of the nature of buildings and their performance,
out of which his design ideas may be formed and against which they can
be criticised and assessed.
It is fair to say, however, that much of building science up to ten years
ago, by concentrating attention on the properties of materials and the
structural and physical behaviour of the building fabric, seriously neglected
the study of building form. (This was perhaps partly a consequence,
ironically enough, of the functionalist attitude-form would come out of
function, and so it was function which ought to be studied-and partly a
result of the fact that many building scientists were physicists or chemists
by training, and not very interested in questions of shape or spatial
arrangement at an architectural scale.)
248 Chapter 12 Afterword: prospects for an architectural morphology
The situation had been reached where it was possible to model say the aid of mathematical models- most notably the 'method of coordinates' -
flow of heat, the transmission of structural forces, or the penetration of which Goethe had excluded from his botany (although Thompson too had
natural light in some detail; but to do this for only the most elementary his blind spots, especially when it came to biochemistry , genetics, and the
shapes of room or building [see the discussion by Hawkes (1976a)]. theory of evolution).
Alternatively, the results gained from experimental work with such models The foregoing chapters record the tentative beginnings, then, on the
referred only to a small number of very particular building designs. So it part of a number of workers, to take up, however belatedly, Goethe's
was difficult to generalise any findings to other buildings with different challenge for architecture. There seem to be two large and immediate
geometries, and the work remained at an anecdotal level. research tasks. The first is an empirical enterprise: to begin to compile
This book has reported a research effort which has been going on over that body of systematic records of the plans of existing and historical
the last ten years to try to lay the foundations of a unified geometrical buildings, classified somehow both geometrically and by functional type,
theory of architectural plan form- a 'theory of cell configurations'. The which would correspond to a 'natural history' of architecture. The second
work has made use of mathematical tools of the kind discussed in The is to integrate the models of the geometric and topological properties of
Geometry of Environment , much elaborated, as well as other approaches. plans, treated in the earlier part of this book , together with representations
The computer methods which have been built on the basis of those of physical characteristics such as those relating to structural strengths,
descriptive techniques are needed, as we have seen , because of the large flows of heat, light, and sound , or to the movement of people and goods
combinatorial difficulties which arise when all permutations of arrangement through plans, which are more familiar from traditional building science.
or pattern are enumerated. (They have nothing to do directly with Hawkes (1980, page 14) speaks of the priorities being first "to establish
'designing by computer'.) a morphology of architectural form" and then "to use the tools of .
There are ways in which the geometrical findings have already found building science ... to make explanatory statements about the relationship
immediate applications to certain long-standing problems in building between form and performance". But perhaps, as I have argued here in
science- for instance the measurement of flexibility and adaptability of chapter I0 , there cannot be quite such a clear distinction between
plans- as we saw in chapter 10. In the meantime there has been the 'possible geometries' and functional or structural factors- since these latter
growth of interest- which we have followed in the last chapter- amongst considerations themselves put dimensional and shape limitations on those
some architectural historians, and those whose concerns are more forms which are 'technologically and functionally possible'.
sociological and anthropological, in the spatial analysis of building types Once such a morphology can explore, for buildings, that variety of
and settlement patterns. Connections are emerging, as we have seen, forms which is 'theoretically imaginable', then the history of actual
between the two strands of work, geometrical and historical. buildings may be studied as a kind of evolutionary process (in the case of
'Morphology' is the word which Goethe coined to signify a universal primitive or vernacular architectures at least) through these spaces of
science of form and spatial structure. Goethe's method in botany, where hypothetical possibility. (At the same time all this analogy from biology
his first morphological interests Jay, was intended not just to provide must not be allowed to encourage any of the confusions- the 'biological
abstract representations, and a classification, of the variety of existing fallacies' equating organic evolution with cultural evolution- which have
plants, but to extrapolate beyond these and to show how recombinations bedevilled the architectural theory of the modern movement.)
of the basic elements of plant form could create theoretical species The study of form in biology itself at the scale of organs or of the whole
unknown to nature. Goethe's morphology remained nevertheless at a body , and of the relation of form to organic function - an 'engineering' of
metaphysical, transcendental level, and he did not refer organic forms to animal or plant design- although it flourished in the nineteenth century, has
the mechanics of their anatomical structure, to functional analyses of their languished somewhat in this century, with the rise in prestige of molecular
physiological working, or to the growth processes by which they were biology and genetics. Is it too bold to suggest that an architectural
produced. morphology might in time have something in turn to offer to biology, in
In the biology of this century it was D'Arey Thompson who accepted geometrical models of cellular assemblies, under constraints of function and
most wholeheartedly Goethe's challenge to develop an organic morphology engineering structure, and undergoing long-term processes of evolutionary
as part of "that wider Science of Form which deals with the forms change?
assumed by matter under all aspects and conditions, and, in a still wider
sense, with forms which are theoretically imaginable" (D'Arcy Thompson,
1961, page 269). D' Arey Thompson moved from a purely descriptive to an
explanatory morphology, seeking causes in physical forces and invoking the
Appendix Appendix 251
~ ~
Enumeration and Classification, and it includes dissections with alignments
and dissections with four-way junctions, but omits dissections for which
Im= n. 12300
12300
D,.. iden
Order 5, Grating (2.4)
~~
Order 3, Grating (2.2)
3 fronts 5 fronts
03 1310
~mm~
D, 02400
20400
Ea x.
~
Order 4, Grating (2.3)
1130
4 fronts 03210
121
D, .• 11310
iden
§
~~~
0221
022
Order 6, Grating {3.3)
5 fronts 02220
01410
~~~~mm
iden
0401 5 fronts
~~Eill§J~
04101
x. 0401 iden
63 C2
Ed 02301
1211 D, iden
D,
[E@][B
§
252 Appendix 253
Appendix
irnJ~~rfil~rnJ i~ffiJ~~~§J@l~~
S fronts
10320 ·ct ~~~~~~ .
03030
~tgj~
[m ~ till till
i t§Hill §J IBH§ rgJ ~ @HB 04101
8] rm
rnJ ~ Lil @d @J ~ 12201
CT]
ffu ~ B:IHfil ~ rnHm fil] tQJ LIB
6 fronts
04200
c, i [§ Qj ~ §J] rnHDHnl rnHrn
~
13110 ~~□□
D1,v Order 7, Grating (2.4) Order 7, Grating ( 3.3J
~
6 fronts
;;
7 fronts
5~~~~
034000
03210
~~~
7 fronts
042100
m~
042100
iden
11310
iden ~
~§
12120
iden
033010
~8]
042001
~
K•
§]
254 Appendix Appendix 255
~
K•
§
133000
133000
I
034000 ~;
~~
214000
D1 , v
~
;;
115000
123 100
123 100
~~~
i~ 204100
D,.•
i''~~~~rnmm~~~rn~ ~
113200
~~~
00
~~mm
!f~ rn ~ ~ rn
~ ~
033010
rawrn
256 Appendix Appendix 257
114010 022300
iden iden
~ ~ HU ~ ~ ml ml ffiJ ~H~B
012400
iden
§~~tm
013210
iden
@d ~ lid
iden
~~
014200
006001
~BE~~~~tIE
iden
~@jtE§~§jlffi@J[mE§tm I 7 fronts
042100
~~[lli@j~~ iden
~~
123100
iden
~
258 Appendix Appendix 259
024100 031210
~fH~~HIE ~ ~ ~
032200
~aHm ~HBHID ~
042001
Ol ~ ~ tiiHI!HaH~ ~ ~ ~ H]ffij~Q]
tili~~~~~ffij~~tm · ~· t6d ~ ff] ~ filJ §'.E rm @Hm
fu]
040300 ~
IBJ~
022300
Order 7, Grating (3.5)
4 fronts
002320 .
~~~
033010
~ooOOfIDOO~oo
5 fronts
101 500
~ ~ QJHi~H~~HilHilH!±HilHill ~rnJi
~~~~~□
003310
041110
iden
~ rnHm rnHID ~Hill t§] ~ [ill
~~
1221 IO
D, .•
tm rnHm rnHm rm
~
0231 IO
~~[1]~~0:!H~~
260 Appendix Appendix 261
011410
iden
~~~
202300
K.
~
103300
iden
~~@]§]
111400
iden
~@][§~
020500
C2 iden
~[§~OO~~~rnJ[§lliJ
104110
iden
~ ~ rnHm
262 Appendix Appendix 263
7 fronts
052000 022300
c, iden
~ I ~H~H~~H~~ ~ ~ ~ ~Hill ~
~~[§~~@J~~
034000
iden
~~~ 114010
iden
~H~
141100
D,.,
~
123100
122110
iden
iden
~§~ rnnrn
023110
iden
~~~~~~~~@J~
024100
C2 iden
~~~~~~
131200
iden
§]ruJ§5~~~gj~WHHJ
~
032200
iden
QH~~ ~2H~HDH~ ~ g
031210
~ ~ ~H~H~~~
iden
113200
iden
~Hlli ffIHI~ @Hra ~Hill tfil ~
§§~~§]@]~
121300
. ~ ~ @HOHfil @Hlli ~
iden
~
264 Appendix Appendix 265
042001 003310
D,,h C, iden D1 idcn
ffiH[]HrJHDHrn ~~~~~~~~
123001
D,,h
IT§j~L§~illfjITQd·~Qill
~ ~~L§ITBJ[@ ~~~
.
024001
~ITB~~~~~~
idcn
Bl~[Tirra[l]~ □□~~
~~
QE~[lli~~~[§ffiJ~tij 011410
iden
~~~Q]d~Qfil[§~
~□
Order 7, Grating (4.4) ~LEQbEj~~~~~
4 fronts
001510 6 fronts
11 1400
iden
~~
020500
iden
5 fronts
102310
~~[ill~~~
023110
D, iden
D1 iden
I t@j ~ ~
266 Appendix Appendix 267
006001
iden
~~lQj~~~tQ]~
7 fronts •
042100
D, iden
~~
032200
idcn
~~~~~~
121300
D, iden
~~
022300
iden
~~@][rn@~g]j~
t[Q~~~~
132010
D,
~
033010
D, iden
~~~~~
114010
, iden
~~
041110
iden
~ ~ §Jj
Appendix References
268
tEm Almgren F J Jr, Taylor J E, 1976 "The geometry of soap films and soap bubbles"
Scientific American 235 82-93
Bailey N I, 1977 Small House Plans: Classification and Adaptability DipArch thesis,
School of Architecture, University of Cambridge, Cambridge, England
Barley M W, 1963 "Glossary of names for rooms in houses of the sixteenth and
seventeenth centuries" in Culture and Environment Eds I LI. Foster, L Alcock
(Routledge and Kegan Paul, Henley-on-Thames, Oxon) pp 4 79- SOI
Baybars I, Eastman C M, 1980 "Enumerating architectural arrangements by generating
their underlying graphs" Environment and Planning B 7 289-31 0
Bellman R E, 1957 Dynamic Programming (Princeton University Press, Princeton, NJ)
Bemis AF, 1936 The Evolving House (Technology Press, MIT. Cambridge, MA)
Bernal J D, 1937 "Art and the scientist" in Circle Eds J L Martin, B Nicholson,
N Gabo (Faber and Faber, London) pp 119-129
Biggs N L, 1969 "Rectangulations" Proceedings of the Cambridge Philosophical Society
65 399-408
Biggs N L, Lloyd EK, Wilson R J, 1976 Graph 171eory 1736-1936 (Clarendon Press,
Oxford)
Blair D J, Bliss T H, 1967 " The measurement of shape in geography" Quantitative
031210 Bulletin 11, Department of Geography, University of Nottingham , Nottingham,
England
Bloch C J , I 976 "On the set and number of minimal gratings for rectangular
dissections" Environment and Planning B 3 7 1- 74
Bloch C J , 1978 "An algorithm for the exhaustive enumeration of rectangular
dissections" Transactions of the Martin Centre for Architectural and Urban Studies
3 5-34
Bloch C J, 1979a A Formal Catalogue of Small Rectangular Plans: Generation,
Enumeration and Classification PhD thesis. School of Architecture, University of
Cambridge, Cambridge, England
Bloch CJ, 1979b "Catalogue of small rectangular plans" Environment and Planning B
6 155-190
024001 Bloch C J, Krishnamurti R, 1978 "The counting of rectangular dissections" Environment
and Planning B 5 207 - 214
Bolker E D, Crapo H, 1977 " How to brace a one-story building" Environment and
Planning B 4 125-1 52
Bon R, 1971 "Presentation to the Morphology Group" mimeograph. Department of
City and Regional Planning, Harvard University, Cambridge, MA
Bon R, 1972a "Allometry in micro-environmental morphology" Harvard Papers in
Theoretical Geography Paper E, Special Papers series. Laboratory for Computer
Graphics and Spatial Analysis, Department of City and Regional Planning, Harvard
University, Cambridge, MA
Bon R, 1972b An Introduction to Morphometric Analysis of Spatial Phenomena on a
Micro-environmental Scale master's thesis, Department of City and Regional Planning,
Harvard University, Cambridge, MA
Bon R, 1973 "Allometry in the topologic st ructure of architectural spatial systems"
Ekistics 215 October issue, 270-276
Bonta J P, 1979 Architecture and its Interpretation: A Study of Expressive Systems in
Architecture (Lund Humphries, London)
Broadbent G, Bunt R, Jencks C (Eds), 1980 Signs, Symbols and Architecture (John
Wiley, Chichester, Sussex)
270 References References 271
Brooks R L, Smith CA B, Stone AH, Tutte WT, 1940 "The dissection of rectangles Eastman CM, Lividini J, Stoker D, 1975 "A database for designing large physical
into squares" Duke Mathematical Journal 7 312 - 340 systems" Proceedings of the 1975 National Computer Conference (AFIPS Press,
Brunskill R W, 1971 Illustrated Handbook of Vernacular Architecture (Faber and Montvale, NJ) pp 603-61 I
Faber, London) Eberhard V, 1891 Zur Morphologie der Polyeder (Teubner, Leipzig)
Bullock N O A, Dickens PG, Steadman JP, 1970 A Theoretical Basis for University Euler L, 1736 "Solutio problematis ad geometriam situs pertinentis" Commentarii
Planning Land Use and Built Form Studies Report I , Department of Architecture, Academiae Scientiarum Imperialis Petropolitanae 8 128-140
University of Cambridge, Cambridge, England Faulkner PA, 1958 "Domestic planning from the twelfth tb the fourteenth centuries"
Bunge W, J966 T11eoretical Geography (C W K Gleerup, Stockholm) Archaeological Journal 11 S 150- 183
Carter H, J975 "The town in its setting: the geographical approach" in 171e Plans and Fawcett W H, 1976a "Measuring adaptability" Transactions of the Martin Centre for
Topography of Medieval Towns in England and Wales Ed. MW Barley, RR-14, Architectural and Urban Studies 1 5 -40
Council for British Archaeology, London, pp 7- I 9 Fawcett W H, 1976b "School management and school design" Journal of Architectural
Casalaina V, Rittel H, 1967 "Morphologies of floor plans" paper presented at the Research 5(3) 10- 21
Conference on Computer-Aided Building Design, University of California, Berkeley, Fawcett W H, 1978 A Mathematical Approach to Adaptability in Buildings PhD thesis,
CA (copy available from J P Steadman) School of Architecture, University of Cambridge, Cambridge, England
CMHC, J 958 Small House Designs (Central Mortgage and Housing Corporation, Ottawa, Flemming U, 1977 Automatisierter Grundrissentwurf Darstellung, Erzeugung und
Ontario) Dimensionierung von dicht gepackten, rechtwinkligen Fliichena11ord11ungen doctoral
Churchman C W, Ackoff R L, Arnoff L, 1957 Introduction to Operations Research thesis , Department of Building Planning and Construction, Technical University of
(John Wiley, New York) Berlin, Berlin
Clarke D L (Ed.) , 1977 Spatial Archaeology (Academic Press, Lond~n) . Flemming U, 1978 "Wall representations of rectangular dissections and their use in
Clasen R J, Graves G W, Lu J Y, 1974 "Sortie allocation by a non-hnear programm~ng automated space allocation" Environment and Planning B S 215 - 232
model for determining a munitions mix" Report R-1411 -DDPAE, Rand Corporation, Fletcher B, Fletcher B F, 1896 A History of Architecture for the Student, Craftsman
Santa Monica , CA and Amateur: Being a Comparative View of the Historical Styles from the Earliest
Combes L, 1976 "Packing rectangles into rectangular arrangements" Environment and Period ( Batsford, London)
Planning B 3 3 - 32 . . Fox Sir C, Raglan Lord, FRS, I 951 - 1954 Monmouthshire Houses; A Study of
Conzen M R G, 1960 Alnwick, Northumberland: A Study m Town-Plan Analysis Building Techniques in the Fifteenth to Seventeenth Centuries 3 volumes (National
(Institute of British Geographers, London) ,, . . Museum of Wales, Cardiff. Wales)
Cousin J. 1970 "Topological organization of architectural space Architectural Design Frew R S, 1973 Towards a Theory of Systems Architecture PhD thesis, Department of
40 491-493 Systems Design, University of Waterloo, Ontario, Canada
Cowan P, 1963 "Studies in the growth, change and ageing of buildings" Transactions Frew R S, Ragade R K, Roe P H, 1972 "The animals of architecture" in Proceedings
of the Bartlett Society I 55-84 of the EDRA 3/AR 8 Conference Ed. W J Mitchell, School of Architecture and
Cowan P, Nicholson J, 1965 "Growth and change in hospitals" Transactions of the Urban Planning, University of California, Los Angeles, CA, pp 23-2- 1- 23-2-7
Bartlett Society 3 63 - 88 Friedman Y, 1975 Towards a Scie11tific Architecture translated by C Lang (M IT Press,
DoE, 1968 Space in the Home Department of the Environment (HMSO, London) Cambridge, MA)
Dickens P, J977 "An analysis of historical house-plans: a study at the structural level Garrison W L, Marble PF, 1961 The Structure of Transportation Networks Report to
(micro)" in Spatial Archaeology Ed. D L Clarke (Academic Press. London) pp 33-45 the US Army Transportation Research Command, by the Transportation Department,
Dreyfus S E, 1969 "An appraisal of some shortest-path algorithms" Journal of Northwestern University, Evanston, IL
Operations Research 17 (3) 395-412 Gero J S, 1977 "Note on 'Synthesis and optimization of small rectangular floor plans'
Durand J N L, 1801 Recueil et Parallele des Edifices (Paris) of Mitchell, Steadman and Liggett" Environment and Planning B 4 81-88
Diirer A, I 528 Hierinn sind begriffen vier Bucher von menschlicher Proportio11 Gilleard J D, 1978 "LAYOUT- a hierarchical computer model for the production of
(Nuremberg) architectural floor plans" Environment and Planning BS 223-241
Earl CF, 1977 "A note on the generation of rectangular dissections" Environment and Gilleard J D, 1980 A Hierarchical Computer Model for the Production of Architectural
Planning B 4 24 1- 246 Floor Plans doctoral thesis, Department of Civil Engineering, University of Salford,
Earl C F, 1978 "Joints in two- and three-dimensional rectangular dissections" Salford, England
Environment and Planning B 5 179- 187 Gips J , I 975 Shape Grammars and their Uses (Birkhauser, Basel)
Earl C F, J980 "Rectangular shapes" Environme11t and Planning B 7 311 -342 Girouard M, 197 1 The Victorian Cou11try House (Clarendon Press, Oxford)
Earl C F, March L J , I 979 "Architectural applications of graph theory " in Applications Golomb S W, 1966 Polyominoes (George Allen and Unwin , London)
of Graph Theory Eds R J Wilson, L W Beineke (Academic Press, London) pp 327-355 Goodman P, Goodman P, 1947 Communitas (University of Chicago Press, Chicago, IL)
Eastman C M, 1970 "Representations for space planning" Communications of the ACM Grason J, 1968 "A dual linear graph representation for space-filling location problems
13 242 -250 of the floor plan type" mimeograph, Carnegie-Mello n University, Pittsburgh, PA
Eastman C M, 1975 Spatial Synthesis in Computer-aided Building Design (John Wiley, Grason J, 1970a "Fundamental description ofa floor plan design program" in EDRA 1,
New York) Proceedings of the First Annual Environmental Design Research Association
Conference Eds H Sanoff, S Cohn, School of Design, North Carolina State University,
Raleif,h, NC, pp 175-182
References References 273
272
March L J (Ed.), 1976 The Architecture of Form (Cambridge University Press, Palladio A, 1570 I Quattro Libri dell'Architettura (Venice)
Cambridge) Pantin W A, I 962-1963 "Medieval English town-house plans" Medieval Archaeology
March L J, Earl C F , 1977 "On counting architectural plans" Environment and 6-7 202-239
Planning B 4 57-80 Pereira L M, 1974 Layout Schemes from Adjacency Graphs: A Case Study in Problem
March L J Matela R 1974 "The animals of architecture: some census results on Solving by Theory Building PhD thesis, Department of Cybernetics, Brunel
n-omin~ populati~ns for n = 6, 7, 8" Environment and Planning B 1 193-216 University, Uxbridge, Middx
March L J, Steadman J P, I 971 The Geometry of Environment (RIBA Publications, Pevsner Sir N, 1976 A History of Building Types (Thames and Hudson, London)
London) Preparata F P, Yeh RT, 1973 Introduction to Discrete Structures for Computer Science
March L J, Steadman J P, 1979 "From descriptive geometry to configurational and Engineering (Addison Wesley , Reading, MA)
engineering" Proceedings: International Conference on Descriptive Geometry Rittel H, 1970 "Theories of cell configurations" in Emerging Methods in Environmental
Engineering Graphics Division, American Society for Engineering Education, Design and Planning Ed. GT Moore (MIT Press, Cambridge, MA) pp 179-181
Vancouver, 14-18 June Rosen J, 1975 Symmetry Discovered: Concepts and Applications in Nature and Science
Marsh J , 1976 Towards the Adaptability Constraints on Design DipArch thesis, School (Cambridge University Press, Cambridge)
of Architecture, University of Cambridge, Cambridge, England Royal Commission on Historical Monuments, England, 1968 An Inventory of Historical
Martin J L, March L J, 1966 "Land use and built forms" Cambridge Research April Monuments in the County of Cambridge (Historical Monuments Commission, Lo ndon)
issue , 8-14 Ryder J A, 1893 "The correlations of the volume and surfaces in organisms"
Martin J L, March L J (Eds), 1972 Urban Space and Structures (Cambridge University Contributions from the Biological Laboratory of the University of Pennsylvania l 3-16
Press, Cam bridge) Sauda E J , 1975 Computer Program for the Generation of Dwelling Unit Floor Plans
Matela R J, 1974 An Analysis of the Animals of Architecture: A Complete Enumeration MArch thesis, School of Architecture and Urban Planning, University of California,
of Polyominoes by some of their Architectural Properties MEd thesis, Department Los Angeles, CA
of Architecture, Yale University, New Haven, CT Seppanen J, Moore J M, 1970 "Facilities planning with graph theory" Management
Matela R, O'Hare E, 1976a "Graph-theoretic aspects of polyominoes and related spatial Science 17 ( 4) B-242-253
structures" Environment and Planning B 3 79-110 Shimbel A , I 953 "Structural parameters of communication networks" Bulletin of
Matela R, O'Hare E, 1976b "Distance measures over polyomino populations" Mathematical Biophysics l S SO 1-507
Environment and Planning B 3 111 -131 Shirey R W, 1969 Implementation and Analysis of Efficient Graph Planarity Testing
McGovern I, 1976, untitled master's thesis, Graduate School of Management, University Algorithms PhD thesis, Department of Computer Sciences, University of Wisconsin ,
of California, Los Angeles , CA Madison , WI
Ministry of Housing and Local Government, 1961 Homes for Today and Tomo"ow: Shubnikov AV, Koptsik VA, 1974 Symmetry in Science and Art (Plenum , New York)
Report of a Sub-committee of the Central Housing Advisory Committee [ Parker Simon H A, I 969 The Sciences of the Artificial Karl Taylor Compton Lectures, 1968
Morris Report/ (HMSO , London) (MIT Press, Cambridge, MA)
Mitchell W J, 197 Sa "An approach to automated generation of minimum cost dwelling Simon HA, 1975 "Style in design" in Spatial Synthesis in Computer-aided Building
unit plans" International Technical Cooperation Center Review (Tel-Aviv, Israel) 4 (3) Design Ed.CM Eastman (Applied Science, London) pp 287-309
I 16-139 Spillers W R, 1974 "Some problems of structural design" in Basic Questions of Design
Mitchell W J, 1975b "Automated generation of minimum energy cost building designs" in Theory Ed. W R Spillers (North-Holland, Amsterdam) pp 103 -117
Responding to Social Change Ed. B Honikman (Halstead Press, New Yo rk) pp 117_- 132 Steadman J P, 1973 "Graph-theoretic representation of architectural arrangement"
Mitchell W J , 1977 Computer-aided Architectural Design (Petrocelli/Charter, New York) Architectural Research and Teaching 2 I 61-172; reprinted 1976 in The Architecture
Mitchell W J, Dillon R L, 1972 "A polyomino assembly procedure for architectural floor of Form Ed . L J March (Cambridge University Press, Cambridge) pp 94-11 S
planning" in Proceedings of the EDRA 3/AR 8 Conference Ed. W J Mitchell, School Steadman JP, I 976 "A note on Combes's classification for rectangular dissections"
of Architecture and Urban Planning, University of California, Los Angeles, CA, Environment and Planning B 3 33-36
pp 23-5-1-23-5-12 Steadman JP, 1979 The Evolution of Designs: Biological Analogy in Architecture and
Mitchell W J , Steadman JP, Liggett RS, 1976 "Synthesis and optimisation of small the Applied Arts (Cambridge University Press, Cambridge)
rectangular floor plans" Environment and Planning B 3 37-70 Stiny G, 197 5 Pictorial and Formal Aspects of Shape and Shape Grammars (Birkhauser,
Mondrian P, 1937 "Plastic art and pure plastic art (Figurative art and non-figurative art)" Basel)
in Circle Eds J L Martin , B Nicholson, N Gabo (Faber and Faber, London) pp 41-56 Stiny G, Gips J, I 978 Algorithmic Aesthetics: Computer Models for Design and Criticism
Moore G T (Ed.), 1970 Emerging Methods in Environmental Design and Planning (MIT in the Arts (University of California Press, Berkeley, CA)
Press, Cambridge, MA) Stiny G, Mitchell W J, 1978a "The Palladian grammar" Environment and Planning BS
NBA, 1965 Generic Plans: Two and Three Storey Houses (National Building Agency, 5-18
London) Stiny G, Mitchell W J, 1978b "Counting Palladian plans" Environment and Planning B
NBA, 1969 Metric House Shells (National Building Agency, London) S 189-198
Newman M H A , I 964 Elements of the Topology of Plane Sets of Points (Cambridge Stravinsky I, 1970 Poetics of Music Charles Eliot Norton Lectures, 1939-1940 (Harvard
University Press, Cambridge) University Press, Cambridge, MA)
Norberg-Schulz C, I 974 Meaning in Western Architecture (Praeger, New York) Summerson J, 1949 Heavenly Mansions, and other Essays on Architecture (Cresset Press,
Ore 0, 1963 Graphs and their Uses (Random House, New York) London)
276 References Index
Architectural Morpholngy
A n introduction to the geometry of building pla ns
J P Steadman
Centre for Configurationa l Studies, The Open University
ISBN O 85086
Just sixty years ago the architect and writer W R Lethaby called for a
programme of theoretical work on the geomet ry of architectural plans,
which wou ld as he said " ... cover the field by a systematic research into
possibilities". "The possibilities of walls and vaults, and of the relations
between the wa lls and the cel l, and between one cell and another, want
investigating, as Lord Kelvin investigated the geometry of crystal line
structures and t he 'packing of ceils' ."
This book is the first introduction to an area of research whi ch has
grown up over the last ten years, and which has begun to answ<·r Lethaby's
call. The author shows how, given suitable geometrical definiti,rns of
certain classes of plans, systematic methods can be devised for enumerating
all possible plans of each type. Particular attention is devoted t o plans
consisting of rectangular rooms, set within rectangular boundaries- so-calied
rectangular dissections- since the plans of many actual small buildings,
especial ly houses, approximate to this kind of geometrical arranqement.
Computer methods for generating rectangular dissections are dl"s:::ribed
in some detai l and classes of plans with other geometries are also discussed.
Mathematical techniques are introduced for t he representation of plans
and their properties, and, in particular, topological properties of the
adjacencies between rooms are represented by using the theory of graphs.
The author goes on to show how these plan-generating methods, and
the catalogues of plans which they can produce, may be applied in three
areas: in design, in building science, and in the study of architectural
history. Design methods are described by which it is possible to enumerate
exhaustively all plans for small houses or apartments, conforming to given
adjacency and dimensional requirernent s. In bui lding science, applications
of this morphological work are suggested to t i1c study of ci rculation and
environmental performance, to the subject of adaptability and flexibility
in plans, and in the formal classification of building types. In architectural
history, co nnections are discussed between this approach to the
representation and classification of historical plans, and their interpretation
in terms of construction, social function, and artistic style.
P. pion publication