Space syntax as a theory

as well as a method

Bill Hillier
Space Syntax Laboratory
University College London

 Space syntax is commonly thought of as a set of techniques for analysing architectural and urban space and
foreseeing functional outcomes.
It is of course both, but it aspires to be more than this: a theoretical model of human space: how it is structured,
how it works, how it is understood, and how it is part of the thing we call society.

The underlying belief is that you cant have the first without the second: foreseeing functional outcomes at the
design stage depends on having a theory that connects the two: a structure-function theory, or form-function theory
if you prefer.

 Part of the theoretical model, as it has so far developed, is a theory of the city as a spatial system.
In this paper, I will explain the space syntax theory of the city, and how it reflects aspects of societies spatially
But first I will look at the theoretical foundations of space syntax, as these will play an important role in the theory of
the city.
So forgive me if talk about architecture for a while because in the last analysis, space syntax is an architectural
theory of the city. I hope you will see what I mean by this as the argument progresses.

 Theoretically, space syntax is rooted in a problem that seems a first sight to forbid a structure-function theory for
either buildings or cities.
It is the simple fact that most of the basic elements that make up buildings and cities rooms in buildings and
streets in cities are, give or take size and shape - more or less, the same kind of thing.

This suggests that the relation between space and human activity must be pretty well indeterminate: you can do
most things in most spaces.

10
1

6
9

10

9
6

10

The answer that space syntax proposes to this question is the foundation of space syntax: that spatially speaking
buildings and cities are configurations.
What is missing in our descriptions of spaces in terms of shape and size is a configurational description of the
position of each space in relation to all the others.

Configuration is defined not simply as connections, but as relations that take into account other relations
For example, we can graph what the simple layout above looks like as a pattern from the point of view of two spaces
We see that in one case it is relatively shallow, or integrated, in the other, deep, or segregated

salle
commune

salle
commune

salle
commune
2

10
1

6
9

salle
commune

9
6

salle
commune

salle
commune

10

By summing the depth of each space from all the others, we can mathematically place every space on a continuum
from integration to segregation. We can then make these values visible by colouring from red for integrated through
to blue for segregated. So we see a pattern.
We can then investigate, say, a regional housing culture, by comparing the functions of spaces with their integration
values. To the extent that different functions are statistically associated with different values, we can say that a
spatial meaning has been given to the idea of function. A form-function relation exists because a particular function,
or range of functions, has been realised spatially through the positioning of the function in the layout as a whole. So
the spatial configuration of the layout can be said to have, within that culture, a social meaning.

people
spaces

people move in lines

interact in convex spaces

see changing visual fields as they

move around built environments

But once we have the concept of configuration, we encounter another fundamental problem in
developing a structure-function theory: people use space in different ways for different kind of activity.

For example, we move linearly, interact in convex space, and experience space as isovists with a convex
core and more linear spikes.
However, each of these can form the basis of configurational analysis.

 For example, on the right above is a visual integration analysis meaning using visual isovists as the basis for
integration calculation of the Tate Britain Gallery. On the left are traces of 100 visitors entering the gallering
and moving for ten minutes. Although each individual takes a different route, the two patterns are remarkably
similar, and in fact statistical analysis shows a 70% correlation between the two patterns. The spatial structure
of the building is shaping, not individual routes, but emergent patterns of movement densities and copresence. It is mixing people up and bringing them together again in terms of the spatial structure of the
building. We have found another structure-function relation.

 But the two structure-function relations we have found are different in

important respects, and in fact illustrate two fundmentally different ways in
which space can relate to society, which I call, on the one hand, the conservative
or reproductine, and on the other the generative or morphogenetic.
A spatial layout can reflect and embody a social pattern, as in the case of the
house, where space was laid out and categorised to give reality to a culturally
given pattern of activity, and so reinforce and reproduce it. We call this the
conservative use of space, since space is being used to reflect and so reproduce
a given social pattern by the control of co-presence.
But space can also shape a social pattern, as in the case of gallery, since by
shaping movement, space also creates a pattern of natural co-presence in space,
and new possibilities of social interaction. We can call this the generative use of
space, since we are using space to create new patterns of co-presence and
potentially new elements in social patterns.
Both of these will turn out be vital in understanding the processes through
which collections of buildings become living cities.
And this in turn reflects distinctive properties of cities as spatialized societies.

So what about cities ? Cities are large collections of buildings held together by a network of space: the street
network. The network is the largest thing in the city. It is what holds it all together. It has an architecture, that is a
certain geometry and a certain topology and a certain scaling. But does it matter?

part of TOKYO

part of LONDON

Lets look more closely. Looking at least line maps syntactic representation of street network as the
fewest lines that cover their structure - of organic grids like London and Tokyo, we begin to find
some interesting geometry. First, the eye intuitively picks out line continuities, that is lines joined by
nearly straight connections. If we move along one of these we are very likely to find another at the
end of the line, and then another. This happens at all scales, but at each scale the lines are locally
longer than lines which lack this kind of angular connection. Probabilistically, we can say the longer
the line, the more likely it is to end in a nearly straight connection to another line.

part of TOKYO

part of LONDON

Lets look more closely. Looking at least line maps syntactic representation of street network as the
fewest lines that cover their structure - of organic grids like London and Tokyo, we begin to find
some interesting geometry. First, the eye intuitively picks out line continuities, that is lines joined by
nearly straight connections. If we move along one of these we are very likely to find another at the
end of the line, and then another. This happens at all scales, but at each scale the lines are locally
longer than lines which lack this kind of angular connection. Probabilistically, we can say the longer
the line, the more likely it is to end in a nearly straight connection to another line.

part of TOKYO

part of LONDON

We also see a large number of shorter lines with near right angle connections, forming
more local grid like patterns. Again if you find one then there are likely to be several
others in the immediate neighbourhood. We can also say the shorter the line, the more
likely it is to end in a right angle or near right angle. These are the opposite properties to
those we find in highly formal cities, like Brasilia or pre-Columbian Teotihuacan, where the
longest lines end at right angles on the most important buildings. Organic grids have the
inverse properties.

part of TOKYO

part of LONDON

It goes further. If we make least line maps for a number of real cities we find other
consistencies. At all scales, from the local area to the whole city, we find cities are made
up of a very small number of long lines and a very large number of short lines, so much
so that in terms of the line length distributions in their least line maps cities have been
said to have scale-free properties (Hillier 2002, Carvalho & Penn 2004). This means that
wherever we are, we are not far from a line much longer than the one we are on.

part of TOKYO

part of LONDON

So geometrically, cities seem to acquire a dual structure, in the form of a global foreground
network and a local background network, with the former nested in the latter, and each
with its own metric, geometric and topological properties. This poses a puzzle. How and
why do such network consistencies emerge from decades or centuries of activity by
innumerable uncoordinated agents acting in very different social, economic and cultural
situations and working with very different, and highly variable, overall geometries ?

We can take steps towards understanding this by showing how space syntax techniques identify configurational
structure in urban street networks through the UCL DepthMap software (which is freely available).
The fundamental element is the street segment between junctions. This is the graph element, and we are interested in
its relations to all other segments in the system.

DepthMap allows 3 definitions of the distance between each segment and each of its neighbours: metric, that is the
distance in metres between the centre of a segment and the centre of a neighbouring segment; topological,
assigning a value of 1 if there is a change of direction between a segment and a neighbouring segment, and 0 if not;
and geometric - assigning the degree of the angular change of direction between a segment and a neighbour, so
straight connected are 0-valued and a line is a sequence of 0-valued connections, so that the linear structure of cities
is captured

So using the etric definition of distance we find the system of shortest path maps for integration and choice, with the topological definition we find the system of fewest turns maps, and with the geometrical definition we find the system o

It then uses these 3 concepts of distance to calculate two measures: syntactic integration, or mathematical
closeness, which measures how close each segment is to all others under each definition of distance; and syntactic
choice or mathematical betweenness, which calculates how many distance-minimising paths between every pair of
segments each segment lies on under different definitions of distance. So using the metric definition of distance, we
find the system of shortest paths for integration and choice, with the topological definition we find the system of
fewest turns maps, and with the geometrical definition we find the system of least angle change maps.

Each of these measures can then be calculated at any radius from each segment, using any of the three definition of
distance to define the radius. So a typical measure would be least angle choice at a radius of 2 kilometres.

Each measure will identify structure in the network which can be made intuitively clear by using colours to represent
mathematical values, as usual from red for high through to blue for low.
The most powerful measures are, as we will see, those based on least angle distance, metric radius and a
combination of integration and choice we call normalised choice. We can think of this measure as indexing the
movement potential of each space, reflecting both through- and to-movement. We can use it, for example - least
angle normalised choice at radius n, meaning no radius restriction - to compare the global structures of cities.

LONDON and its region within the M25, with its strong centre and strong
radials, but weak lateral connections between the radials

BEIJING with its relative weak centre and weak radials, but strong
lateral structure between radials

TOKYO with its fairly strong centre, strong radials and strong laterals,
generating the strong sub-city structure characteristic of Tokyo

So the analysis can identify different kinds of structures, global and local, in the network, and makes
them visible by colouring segments red for high movement potentials through to blue for low.
On the left we see the movement potentials of each of the 285000 segments of London within the
M25 for large scale movement. It predicts the real main movement arteries. On the right we see
much finer scale structure for local movement potentials up to 750 metres. The red pattern you see is
essentially Londons urban villages and the links between them.

The configurational approach has

led to a key discovery about cities:
that in an of itself the architecture
of the street network shapes
movement flows.
Research shows that between 50%
and 80% of the movement flows
on streets are due to the structure
of the network itself, that is to the
potential flows identified
mathematically.
This does not mean that space
determines individual movement.
It means that if people go under
their own volition from
everywhere to everywhere else,
some spaces get more used than
others. The relation between grid
structure and movement is an
emergent effect, as it was in the
Tate. It happens whether we
intend it or not.

But by examining real movement patterns in cities, we have shown that people move by reading the angular
geometry of the network, not actual metric distances. So by analysing the network in terms of its least angle
change paths from all street segment to all others, we can approximate movement potentials from the
architecture of the network, and of course for new designs inserted into the network. But more importantly,
once the influence of the grid on movement is understood it opens then way for a new theoretical
understanding of the city as a self-organising system through what we call the city-creating process. On the
right are the 168 largest local centres in London. There are 10 times as many smaller ones. They reflect the
structure of the network. How does it happen?

It works like this. Because the network structure shapes flows, it also shapes land use patterns, in that
movement-seeking land uses seek locations that the grid has already made movement-rich, while others,
often including residence, migrate to less-movement rich parts of the network. Economic values follow this
process. With feedback and multiplier effects once one shop appears, others follow -this is the
fundamental city creating process by which cities evolves from collections of buildings to living cities,
with busy and quiet zones, often in close juxtaposition, and with differentiation of areas according to the
detail of how they are embedded in the larger scale grid. So the pattern of centres (right above) is shaped
by the structure of the urban grid (left above).

This leads us to a definition of the spatial form of cities. Cities in general not just organic ones - selfevolve into a foreground network of linked centres at all scales, from a few shops and a caf through
to whole sub-cities, set into a background network of largely residential space. The two networks have
different geometric and metric properties. The foreground network has longer lines, nearly straight
connections and route continuity, the background network shorter lines, right angle connections and
more local grid-like structures.
Good cities, we suggest, have pervasive centrality in that centrality functions diffuse throughout the
network. The pattern is far more complex than envisaged in theories of polycentrality. Pervasive
centrality is spatially sustainable because it means that wherever you are you are close to a small
centre and not far from a much larger one.

This process of self organising into the dual network form with pervasive centrality is not
just found in organic cities. Suzhou, for example, is planned on a grid, but we can
detect the same process.

At radius-n the whole Suzhou region the

North- south alignment is secondary and
fades as it goes north

At radius 5000 metres, however, the whole

alignment becomes on of the strongest in
Suzhou. However, the northern end, where
main cluster is found, is weaker

But at radius 3000, the main alignment

fades, and the northern end of the line, where
the centre is, becomes one of the strongest
lines in the whole of Suzhou. It is the combination of local and global strength that creased
the centre in that location.

What about the wonderful shopping

streets in the south east of the old city.
At radius-n these do not appear at all.

At radius 5000 the local group of street begin

to appear.

But at 2000 the analysis seems to identify a

But at 2000 the analysis seems to identify a
central seed for all three areas.
central seed for all three areas.

BRASILIA choice radius 5000 metres

BRASILIA choice radius 1600 metres

All of these patterns are bought to light by least angle analysis of the segment network. Does this mean that metric
analysis has no role ? On the contrary, metric analysis brings to light not linear structures, but area patchworks,
reflecting the functional pattern of centres, with their smaller scale grids, and non-centres with larger scale grids. On the
left we see, for Istanbul, the patchwork produced by metric segment analysis metric depth analysis at a radius of 500
metres. On the right we represent this as a scattergram, with metric depth at radius n, that is, for the city sa a whole, on
the horizontal axis, and metric depth at radius 500 metres on the vertical. Each of the mountains represents a local
area with the peak defininig the centre of the area, and so potentially the most intensive activity. This periodic structure
is found in most cities, showing once again the generation of spatial structure by functional factors.

In general we can say that the foreground network is generated by micro-economic activity, and because this tends
to follow invariant principles (you want to maximise the number of people), foreground grids tend to take a generic
form, for example the deformed wheel you see in Nicosia above. But the background network is shaped by sociocultural factors and these will be different and so differentiate the form of the network.

So in the case above, we see two very different residential networks, the Turkish in the north-east and the Greek in
the south west, but held together by the foreground network.

 Certain general principles about space in cities follow from this theory of city formation..
The first is that the grid-movement relation we call it the law of natural movement - is fundamental to
cities. Most relations between the form of the city and the way it functions pass through this relation in
some way.
Most significantly of all, it is the grid-movement relation which generates the life of the city. For much of the
twentieth century we thought of place as one thing and movement between places as another. Now we see
that in the ways cities self-organize to create the endless diversity of places which is their most amazing
characteristic, movement is the heart of place.

 The second is that the fact that syntax models have the ability to bring to light structures
at many scales, from the most local to the most global, is vital, because research has
amply shown that most urban phenomena are multi-scale. For example, the degree and
nature of the movement passing along a street will be shaped by how the street is
embedded in both the global as well as the local network.
The same applies to local centres. These will occur and grow to the degree that that are
embedded in a local metric system (brought about by small blocks) and a more global
least angle system - in other words, with how they are linked to both components of the
dual grid.
Similarly, public squares are affected in their functioning not only by how the space is
defined by the surrounding buildings, but also by how it is spatially embedded in the
larger scale system of space.

 The third is that the relations between form and function in cities are generic, not specific.
Cities appear to us as patterns of activity related to patterns of space. This is how the task of design is
presented to the designer: how a specific pattern of activity is to be related to a specific pattern of space.
But theoretically it is not like that, and this is not how cities become as they are. Space is created not directly
by the inter-related demands of specific activity patterns, but indirectly by the different demands that kinds
of activity place on the movement and co-presence that is created by space. This is why good city form can
adapt easily to new patterns of use.

 Finally, space in cities works in more that one way. The foreground network is structured
to maximise movement, and it is so because it is driven by micro-economic factors which
benefit from high levels of movement, while the background network restricts and
structures movement, and does so because it is driven by social and cultural factors
which find expression in the way residential space is structured. So the dual network in
cities reflects functional as well as spatial processes
This is an instance of the more general potential of space to operate in two ways. Space
can be use generatively to create new patterns of movement and so co-presence and
potential relations in the social system, or it can be used conservatively to express and
so reproduce existing social patterns and structures. The former is associated with
spatial integration, the latter with spatial segregation.
The difference between the foreground and background networks is the difference
between more morphogenetic and more conservative space. The former focuses
movement to create development and change, the latter diffuses it to keep things as
they are.

 This generic structure seems to underlie all cities in some sense, and we must conclude
that beneath the individuality and cultural typing of cities, there is a universal generic
city which makes the city what it essentially is.
All societies must in some sense be morphogenetic in order to cope with changing
technological and social circumstances, and all societies must also act in ways that
reproduce their structures hence the dual use of space
Cities are spatially massive morphogenetic machines that produce change, set into a
conservative background which stabilises their structure. This is the generic city. It was I
believe the discovery of the generic city that first made urban societies possible.

 So we have a testable theory of how urban space is organised and how it works. It brings
to light structures of various kinds in the urban grid, and these structures are powerfully
related to functional patterns.
What can we do with it ? Well first we can use it as a frame for all kinds of urban data by
simply adding data to the model segment by segment: movement flows, land uses,
densities, demographic information, land and rental values and so on.
We then have a tool for asking spatial questions of the city, of the form: is there a spatial
dimension to this or that urban problem to social malaise, to migration patterns, crime
distributions, to the emergence of centres, to the success or failure of areas all these
are areas we have investigated using space syntax.

But more importantly, perhaps,it can be used in design

Essentially, the procedure is this:

- we build a model of the site and its context, usually the whole city these days
- we test the model against existing movement flows and land use pattern
- we can then use the verified model to test out designs by inserted them into the model
and re-running the analysis
- and suggest new design ideas from the analysis.
Let us look at real examples of where space syntax is being used, or has been used:

- to support the development of whole cities

- to design urban areas
- to redesign public squares
- and to create new urban links

Effects of growth on the centre show former plans are unsustainable

Existing

Historic
Core

Former
plan

Historic
Core

We can show the structural

isolation of the historic centre in
two more ways First we analyse
the whole structure of Jeddah as it
has evolved. We see the
integration core of Jeddah has
moved to the north and east of the
old centre, which is now a green
patch rather than a focus of red
and orange lines.

New
plan to
We then expand
the analysis
include all current plans for
extending the city. And re-run the
analysis. We see that the structure
of the city has moved even more
decisively north and east and the
old centre has become blue as
well as green. In other word,
current plans will exacerbate the
structural problems of the city

We can show this dramatically by

colouring up all the street
segments in Jeddah according to
how much urban integration
each gains by the planned new
developments.

Historic
Core

Red means high. We see that all

the gains are away from the old
centre, and none are in the centre
itself. The plans would make
the structural problems of the
city worse rather than better.

The whole
structure of the
city will be
drawn back from
the east and
north towards
to historic
centre.
At the same
time, the
patches of
unplanned
settlement will
become more
integrated into
the urban fabric.

Existing

Old local plans

New plan

Spatial accessibility Existing City-wide

Deira

Dubai International
Airport

Dubai Creek
Downtown Dubai

Business Bay

Spatial accessibility
high

NAChn

low
00

22

4km
4km

Site location

Spatial accessibility Proposed masterplan City-wide

Deira

Dubai International
Airport

Dubai Creek
Downtown Dubai

Business Bay

Spatial accessibility
high

NAChn

low
00

22

4km
4km

Site location

Option testing Experiment 1 Spatial accessibility City-wide

Deira

Dubai International
Airport

Dubai Creek
Downtown Dubai

Business Bay

Spatial accessibility

NAChn

high
00

22

4km
4km

low

Option testing Experiment 2 Spatial accessibility City-wide

Deira

Dubai International
Airport

Dubai Creek
Downtown Dubai

Business Bay

Spatial accessibility

NAChn

high
00

22

4km
4km

low

Option testing Experiment 3 Spatial accessibility City-wide

Deira

Dubai International
Airport

Dubai Creek
Downtown Dubai

NAChn

Business Bay

00

22

4km
4km

Option testing Experiment 4 Spatial accessibility City-wide

Deira

Dubai International
Airport

Dubai Creek
Downtown Dubai

NAChn

Business Bay

00

22

4km
4km

Option testing Experiment 4b Spatial accessibility City-wide

Deira

Dubai International
Airport

Dubai Creek
Downtown Dubai

NAChn

Business Bay

00

22

4km
4km

Option testing Experiment 5 Spatial accessibility City-wide

Deira

Dubai International
Airport

Dubai Creek
Downtown Dubai

NAChn

Business Bay

00

22

4km
4km

Option testing Comparison

Mean
Maximum
City-wide accessibility City-wide accessibility
(NACh_n)
(NACh_n)

Proposed

0.89

1.29

Option 1

1.10

1.38

Option 2

1.09

1.38

Option 3

1.09

1.38

Option 4

1.14

1.40

Option 4b

1.14

1.40

Option 5

1.14

1.39

Option 6

1.15

1.41

MultiScale core
%

0.18%

Option 6b

1.42

4.80%

Option 6c

1.42

5.63%

This is lunchtime

non_tourists_12to2wd = -1241.28 + 486.386 * sp. integration; R^2 = .89

p = 0.0047
1800

non_tourists_12to2wd

1600
1400

Blackfriars Bridge

1200

Westminster Bridge
Waterloo Bridge

1000

London Bridge

800

Southwark Bridge
Hungerford Bridge

600
400
200
3

3.5

4.5
5
5.5
spatial integration (local)

6.5

Let me end with an advertisement. Space syntax is now being applied to the
complete street and road network of whole countries. The first research on this will
be presented on Friday afternoon by Miguel Serra.

So if you would like to know if least angle analysis can predict movement for
thousands of locations at the level of a whole country, dont miss it !
Thank you for your attention.