Graphs and Networks

Graphs and Networks

Multilevel Modeling

Second Edition

Edited by
Philippe Mathis
First edition published 2007 by ISTE Ltd
Second edition published 2010 in Great Britain and the United States by ISTE Ltd and John Wiley &
Sons, Inc.

Apart from any fair dealing for the purposes of research or private study, or criticism or review, as
permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced,
stored or transmitted, in any form or by any means, with the prior permission in writing of the publishers,
or in the case of reprographic reproduction in accordance with the terms and licenses issued by the CLA.
Enquiries concerning reproduction outside these terms should be sent to the publishers at the
undermentioned address:

ISTE Ltd John Wiley & Sons, Inc.

27-37 St George’s Road 111 River Street
London SW19 4EU Hoboken, NJ 07030
UK USA

www.iste.co.uk www.wiley.com

© ISTE Ltd 2007, 2010

The rights of Philippe Mathis to be identified as the author of this work have been asserted by him in
accordance with the Copyright, Designs and Patents Act 1988.

Library of Congress Cataloging-in-Publication Data

Graphs and networks : multilevel modeling / edited by Philippe Mathis. -- 2nd ed.
p. cm.
Includes bibliographical references and index.
ISBN 978-1-84821-083-7
1. Cartography--Methodology. 2. Graph theory. 3. Transport theory. I. Mathis, Philippe.
GA102.3.G6713 2010
388.01'1--dc22
2010002226

British Library Cataloguing-in-Publication Data

A CIP record for this book is available from the British Library
ISBN 978-1-84821-083-7

Printed and bound in Great Britain by CPI Antony Rowe, Chippenham and Eastbourne.
 Table of Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv

PART 1. GRAPH THEORY AND NETWORK MODELING . . . . . . . . . . . . . . . 1

Chapter 1. The Space-time Variability of Road Base Accessibility:

Application to London . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Manuel APPERT and Laurent CHAPELON
1.1. Bases and principles of modeling . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1. Modeling of the regional road network . . . . . . . . . . . . . . . . . 3
1.1.2. Congestion or suboptimal accessibility . . . . . . . . . . . . . . . . . 6
1.2. Integration of road congestion into accessibility calculations . . . . . . 10
1.2.1. Time slots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.2. Evaluation of demand by occupancy rate . . . . . . . . . . . . . . . . 11
1.2.3. Evaluation of demand by flows. . . . . . . . . . . . . . . . . . . . . . 12
1.2.4. Calculation of driving times. . . . . . . . . . . . . . . . . . . . . . . . 15
1.3. Accessibility in the Thames estuary . . . . . . . . . . . . . . . . . . . . . 19
1.3.1. Overall accessibility during the evening rush hour (5-6 pm) . . . . 21
1.3.2. Performance of the road network between 1 and 2 pm and
5 and 6 pm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.3. Network performance between 1 and 2 pm. . . . . . . . . . . . . . . 23
1.3.4. Network performance between 5 and 6 pm. . . . . . . . . . . . . . . 25
1.3.5. Evolution of network performances related to the Lower Thames
Crossing (LTC) project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
vi Graphs and Networks

Chapter 2. Journey Simulation of a Movement on a Double Scale . . . . . . 31

Fabrice DECOUPIGNY
2.1. Visitors and natural environments: multiscale movement . . . . . . . . 32
2.1.1. Leisure and consumption of natural environments . . . . . . . . . . 32
2.1.2. Double movement on two distinct scales . . . . . . . . . . . . . . . . 33
2.1.3. Movement by car . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.1.4. Pedestrian movement . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.2. The FRED model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.1. Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.2. Structure of the FRED model . . . . . . . . . . . . . . . . . . . . . . . 36
2.3. Part played by the network structure . . . . . . . . . . . . . . . . . . . . . 37
2.4. Effects of the network on pedestrian diffusion . . . . . . . . . . . . . . . 39
2.4.1. Determination of the potential path graph: a model of
cellular automata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.4.2. Two constraints of diffusion . . . . . . . . . . . . . . . . . . . . . . . 40
2.4.3. Verification of the model in a theoretical area . . . . . . . . . . . . . 42
2.5. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

Chapter 3. Determination of Optimal Paths in a Time-delay Graph . . . . 47

Hervé BAPTISTE
3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2. Floyd’s algorithm for arcs with permanent functionality . . . . . . . . . 49
3.3. Floyd’s algorithm for arcs with permanent and
temporary functionality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3.1. Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3.2. Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.4. Conclusion: other developments of Floyd’s timetable algorithm . . . . 60
3.4.1. Determination of the complete movement chain . . . . . . . . . . . 60
3.4.2. Overview of all the means of mass transport . . . . . . . . . . . . . . 62
3.4.3. Combination of means with permanent and
temporary functionality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.4.4. The evaluation of a timetable offer under the constraint of
departure or arrival times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.4.5. Application of Floyd’s algorithm to graph properties. . . . . . . . . 65
3.5. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

Chapter 4. Modeling the Evolution of a Transport System and

its Impacts on a French Urban System . . . . . . . . . . . . . . . . . . . . . . . 67
Hervé BAPTISTE
4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2. Methodology: RES and RES-DYNAM models . . . . . . . . . . . . . . 68
4.2.1. Modeling of the interactions: procedure and hypotheses . . . . . . . 68
 Table of Contents vii

4.2.2. The area of reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

4.2.3. Initial parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.3. Analysis and interpretation of the results . . . . . . . . . . . . . . . . . . 79
4.3.1. Demographic impacts . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.3.2. Alternating migrations revealing demographic trends . . . . . . . . 82
4.3.3. Evolution of the transport network configuration . . . . . . . . . . . 84
4.4. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.5. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

PART 2. GRAPH THEORY AND NETWORK REPRESENTATION . . . . . . . . . . 91

Chapter 5. Dynamic Simulation of Urban Reorganization of

the City of Tours . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Philippe MATHIS
5.1. Simulations data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.2. The model and its adaptations . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.2.1. D.LOCA.T model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.2.2. Opening of the model and its modifications . . . . . . . . . . . . . . 101
5.2.3. Extension of the theoretical base of the model . . . . . . . . . . . . . 102
5.3. Application to Tours . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.3.1. Specific difficulties during simulations . . . . . . . . . . . . . . . . . 103
5.3.2. First results of simulation . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.4. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.5. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

Chapter 6. From Social Networks to the Sociograph for the Analysis

of the Actors’ Games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Sébastien LARRIBE
6.1. The legacy of graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.2. Analysis of social networks . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.3. The sociograph and sociographies . . . . . . . . . . . . . . . . . . . . . . 119
6.4. System of information representation . . . . . . . . . . . . . . . . . . . . 127
6.5. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

Chapter 7. RESCOM: Towards Multiagent Modeling of Urban

Communication Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Ossama KHADDOUR
7.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
7.2. Quantity of information contained in phatic spaces . . . . . . . . . . . . 132
7.3. Prospective modeling in RESCOM . . . . . . . . . . . . . . . . . . . . . . 136
7.3.1. Phatic attraction surfaces . . . . . . . . . . . . . . . . . . . . . . . . . 136
7.3.2. Game of choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
viii Graphs and Networks

7.4. Huff’s approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

7.5. Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
7.6. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
7.8. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

Chapter 8. Traffic Lanes and Emissions of Pollutants . . . . . . . . . . . . . 147

Christophe DECOUPIGNY
8.1. Graphs and pollutants emission by trucks . . . . . . . . . . . . . . . . . . 147
8.1.1. Calculation of emissions . . . . . . . . . . . . . . . . . . . . . . . . . . 150
8.1.2. Calculation of the minimum paths . . . . . . . . . . . . . . . . . . . . 152
8.1.3. Analysis of subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
8.2. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
8.2.1. Section of the A28 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
8.2.2. French graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
8.2.3. Subset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
8.3. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

PART 3. TOWARDS MULTILEVEL GRAPH THEORY. . . . . . . . . . . . . . . . . 175

Chapter 9. Graph Theory and Representation of Distances:

Chronomaps and Other Representations . . . . . . . . . . . . . . . . . . . . . 177
Alain L’HOSTIS
9.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.2. A distance on the graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
9.3. A distance on the map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
9.4. Spring maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
9.5. Chronomaps: space-time relief maps . . . . . . . . . . . . . . . . . . . . . 186
9.6. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
9.7. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191

Chapter 10. Evaluation of Covisibility of Planning and

Housing Projects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
Kamal SERRHINI
10.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
10.2. The representation of space and of the network:
multiresolution topography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
10.2.1. The VLP system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
10.2.2. Acquiring geographical data: DMG and DMS . . . . . . . . . . . . 197
10.2.3. The Conceptual Data Model (CDM) starting point of a graph . . . 197
10.2.4. Principle of multiresolution topography
(relations 1 and 2 of the VLP) . . . . . . . . . . . . . . . . . . . . . . . . . . 198
 Table of Contents ix

10.2.5. Need for overlapping of several spatial resolutions

(relation 2 of the VLP). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
10.2.6. Why a square grid? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
10.2.7. Regular and irregular hierarchical tessellation: fractalization . . . 202
10.3. Evaluation of the visual impact of an installation: covisibility . . . . . 202
10.3.1. Definitions, properties, vocabulary and some results . . . . . . . . 202
10.3.2. Operating principles of the covisibility algorithm
(relations 3 and 4 of the VLP) . . . . . . . . . . . . . . . . . . . . . . . . . . 205
10.3.3. Why a covisibility algorithm of the centroid-centroid type? . . . . 212
10.3.4. Comparisons between the method of covisibility and
recent publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
10.4. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
10.5. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220

Chapter 11. Dynamics of Von Thünen’s Model: Duality and

Multiple Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
Philippe MATHIS
11.1. Hypotheses and ambitions at the origin of this dynamic
von Thünen model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
11.2. The current state of research . . . . . . . . . . . . . . . . . . . . . . . . . 227
11.3. The structure of the program . . . . . . . . . . . . . . . . . . . . . . . . . 227
11.4. Simulations carried out . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
11.4.1. The first simulation: a strong instability in the isolated state
with only one market town . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
11.4.2. The second simulation: reducing instability . . . . . . . . . . . . . 235
11.4.3. The third simulation: the competition of two towns . . . . . . . . . 237
11.4.4. The fourth simulation: the competition between five
towns of different sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
11.5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
11.6. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244

Chapter 12. The Representation of Graphs: A Specific Domain of

Graph Theory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
Philippe MATHIS
12.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
12.1.1. The freedom of drawing a graph or the absence
of representation rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
12.2. Graphs and fractals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
12.2.1. Mandelbrot’s graphs and fractals . . . . . . . . . . . . . . . . . . . . 248
12.2.2. Graph and a tree-structured fractal: Mandelbrot’s H-fractal . . . . 251
12.2.3. The Pythagoras tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
12.2.4. An example of multiplane plotting . . . . . . . . . . . . . . . . . . . 256
x Graphs and Networks

12.2.5. The example of the Sierpinski carpet and its use in

Christaller’s theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
12.2.6. Development of networks and fractals in extension . . . . . . . . . 258
12.2.7. Grid of networks: borderline case between extension
and reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
12.2.8. Application examples of fractals to transport networks . . . . . . . 260
12.3. Nodal graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
12.3.1. Planarity and duality . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
12.4. The cellular graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
12.5. The faces of the graph: from network to space . . . . . . . . . . . . . . 296
12.6. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299

Chapter 13. Practical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 301

Philippe MATHIS
13.1. Premises of multiscale analysis . . . . . . . . . . . . . . . . . . . . . . . 301
13.1.1. Cellular percolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
13.1.2. Diffusion of agents reacting to the environment . . . . . . . . . . . 303
13.1.3. Taking relief into account in the difficulty of the trip . . . . . . . . 304
13.2. Practical application of the cellular graph: fine modeling of
urban transport and spatial spread of pollutant emissions . . . . . . . . . . . 305
13.2.1. The algorithmic transformation of a graph into a cellular graph
at the level of arcs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
13.2.2. The algorithmic transformation of a graph into a cellular graph
at the level of the nodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
13.3. Behavior rules of the agents circulating in the network . . . . . . . . . 309
13.3.1. Strict rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
13.3.2. Elementary rules. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
13.3.3. Behavioral rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
13.4. Contributions of an MAS and cellular simulation on the basis of
a graph representing the circulation network . . . . . . . . . . . . . . . . . . . 311
13.4.1. Expected simulation results . . . . . . . . . . . . . . . . . . . . . . . 311
13.4.2. Limits of application of laws considered as general . . . . . . . . . 312
13.5. Effectiveness of cellular graphs for a truly door-to-door modeling . . 314
13.6. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
13.7. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315

PART 4. GRAPH THEORY AND MAS . . . . . . . . . . . . . . . . . . . . . . . . . . 317

Chapter 14. Cellular Graphs, MAS and Congestion Modeling . . . . . . . . 319

Jean-Baptiste BUGUELLOU and Philippe MATHIS
14.1. Daily movement modeling: the agent-network relation . . . . . . . . . 320
14.1.1. The modeled space: Indre-et-Loire department . . . . . . . . . . . 320
 Table of Contents xi

14.1.2. Diagram of activities: a step toward the development

of a schedule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
14.1.3. Typology of possible agent activities . . . . . . . . . . . . . . . . . 322
14.1.4. Individual behavior mechanism: the daily scale . . . . . . . . . . . 323
14.2. Satisfaction and learning . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
14.2.1. The choice of an acceptable solution. . . . . . . . . . . . . . . . . . 324
14.2.2. Collective learning and convergence of the model toward
a balanced solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
14.2.3. Examination of the transport network . . . . . . . . . . . . . . . . . 327
14.3. Local congestion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
14.3.1. The peaks represent different types of intersections . . . . . . . . . 329
14.3.2. The emergence of congestion fronts on edges . . . . . . . . . . . . 330
14.3.3. Intersection modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
14.3.4. Limited peak capacity: crossings and traffic circles . . . . . . . . . 336
14.3.5. In conclusion on crossings . . . . . . . . . . . . . . . . . . . . . . . . 351
14.4. From microscopic actions to macroscopic variables a global
validation test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
14.4.1. The appropriateness of the model with traditional throughput-
speed, density-speed and throughput-density curves . . . . . . . . . . . . . 352
14.4.2. The distribution of traffic density over time . . . . . . . . . . . . . 356
14.4.3. The measure of lost transport time by agents because of
congestion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
14.4.4. Spatial validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
14.5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
14.6. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360

Chapter 15. Disruptions in Public Transport and Role of Information . . . 363

Julien COQUIO and Philippe MATHIS
15.1. The model and its objectives . . . . . . . . . . . . . . . . . . . . . . . . . 364
15.1.1. Public transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
15.1.2. Hypotheses to verify . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
15.2. The PERTURB model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
15.2.1. Theoretical fields mobilized . . . . . . . . . . . . . . . . . . . . . . . 367
15.2.2. Working hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
15.2.3. Functionalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
15.3. The simulation platform. . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
15.4. Simulations in real space: Île-de-France . . . . . . . . . . . . . . . . . . 373
15.4.1. Disruptions simulated in the Île-de-France public transport . . . . 374
15.4.2. Node-node calculations: measure of the deterioration of
relational potentials between two network vertices . . . . . . . . . . . . . . 375
15.4.3. Unipolar calculations: measures of the deterioration of
traveling opportunities from a network vertice . . . . . . . . . . . . . . . . 381
xii Graphs and Networks

15.4.4. Multipolar calculations: global measures of structural

impacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
15.5. Simulations in theoretical transport systems . . . . . . . . . . . . . . . . 388
15.5.1. The initial network and line creation . . . . . . . . . . . . . . . . . . 388
15.5.2. Studied disruption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
15.5.3. Multipolar calculations . . . . . . . . . . . . . . . . . . . . . . . . . . 391
15.5.4. Simulations integrating capacity constraints . . . . . . . . . . . . . 396
15.6. Discussion on hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
15.6.1. Field of structural vulnerability . . . . . . . . . . . . . . . . . . . . . 401
15.6.2. Field of functional vulnerability . . . . . . . . . . . . . . . . . . . . 402
15.7. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
15.8. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405

Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407

List of Authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
 Preface

This work is focused on the use of graphs for the simulation and representation
of networks, mainly of transport networks.

The viewpoint is intentionally more operational than descriptive: the effects of

transport characteristics on space are just as important to the planner as the transport
itself.

The present work is based on the research conducted at Tours since the 1990s by
various PhD students who have become researchers, lecturer researchers or
professionals.

The book is structured in four parts following an introductory chapter which

contains a reminder of the necessary definitions from graph theory and of the
representation problems.

Part 1 presents the traditional applications of graph theory in network modeling

and the improvements required for their use as a planning tool.

Part 2 tackles the problem of the representation of graphs and exposes a certain
number of innovations as well as deficiencies.

Part 3 considers the prior achievements and proposes to develop their theoretical
justifications and fill in some gaps.

Part 4 shows how we can use micro-simulations with MAS models with the help
of cellular graphs reversing the original top down viewpoint for multi-scale spatial
and temporal bottom up models, partially integrating information and learning.

Philippe MATHIS
 Introduction

Strengths and Deficiencies of Graphs for

Network Description and Modeling

The focus of this book is on networks in spatial analysis and in urban

development and planning, and their simulation using graph theory, which is a tool
used specifically to represent them and to solve a certain number of traditional
problems, such as the shortest path between one or more origins and destinations,
network capacity, etc. However, although transportation systems in the physical
sense of the term are the main concern and will therefore form the bulk of the
examples cited, other applications, such as player, communication and other
networks, will, nevertheless, be taken into account and the reader is welcome to
transfer the presented results to other domains.

All of the examples presented below essentially correspond to a decade of

research and some ten PhDs. of the Modeling Group of the Graduate Urban
Development Studies Center of Tours. Obviously, these will be supplemented by
other contributions.

In network modeling, which is a field stemming from operational research, a

certain form of empiricism tends to dominate, in particular in the intermediate
disciplines between social sciences and hard sciences, such as urban development,
which, essentially, borrow their tools. However, their specific needs are barely taken
into account by fundamental disciplines, such as mathematics, or more applied ones,
such as algorithmics, undoubtedly simply because the dynamics of research are very
different. We will try to contribute to the mitigation of this difficulty.

Introduction written by Philippe MATHIS.

xvi Graphs and Networks

The modeling and description of networks using graphs: the paradox

The aim of this work is, among other things, to highlight a paradox and to try to
rectify it. This paradox, once identified, is relatively simple. Since Euler’s time
[EUL 1736, EUL 1758] it has been known how to efficiently model a transport
network by using graphs, as he demonstrated with the famous example of the
Königsberg bridges and, following the rise of Operations Research in the 1950s and
1960s, a number of optimization problems have been successfully resolved with
efficiency and elegance.

According to Beauquier, Berstel and Chrétienne: “graphs constitute the most

widely used theoretical tool for the modeling and research of the properties of
structured sets. They are employed each time we want to represent and study a set of
connections (whether directed or not) between the elements of a finite set of objects”
[BEA 92]. For Xuong [XUO 92]: “graphs constitute a remarkable modeling tool for
concrete situations” and we could cite numerous further testimonies.

The power of the method increased considerably with the fulgurating

development of computers and microcomputers1. However, although graphs are a
powerful tool for the modeling and resolution of certain problems, they otherwise
appear unable to represent and describe precisely and without implicit assumptions a
network of paths on the basis of elements which are needed for the calculation such
as, for example, minimal path or maximum flow, etc. Since, on the basis of a matrix
definition2 of the graph, all the plots (i.e. representations) are equal and equivalent in
graph theory.

We thus have a method that is simultaneously very simple and has great
algorithmic efficiency, but is otherwise deficient, unless it were only to model a
network represented on a roadmap, on which basis it delivers knowledgeable and
powerful calculations. It does not satisfy the two essential criteria of all scientific
work: reproducibility and comparability, particularly with respect to network
modeling and the production of charts and/or synthesized images. It also does not
allow for the ongoing movement between graph and cellular in an algorithmic
fashion, or the use of multi-agent systems. Finally the theory of traditional graphs
makes a congestion approach, still limited to network edges, difficult, since the
peaks are neutral by definition.

1 Has the generation of 50 year-olds not also been called the Hewlett-Packard generation? Its
ranks remember calculations with a ruler, with logarithmic tables or with the
electromechanical four operations machine, etc.
2 See below for the definition of the adjacency matrix, often referred to as associated matrix
in the works from the 1960s, and of the incidence matrix.
 Introduction xvii

At first we propose to show the effectiveness of graph theory in the field of

calculation, which we could quickly call of optimization. Then, we propose to
demonstrate that the practice of modelers anticipated the theorization with
pragmatism and efficiency, and, finally, to suggest some solutions and research
paths to establish and generalize what has been conjectured by usage. In the last
section, we will discuss in detail the Bottom Up approaches with multi agent
systems which can learn and partially use information, moving in cellular graph. We
will show that the capacity of peaks is clearly more limited than that of edges and
consequently its non application in the urban transport systems of the Ford-
Fulkerson theorem and the importance of learning to avoid the biggest congestions.
Similarly, we will show with the help of the Ile de France transit system example
that the problem can be handled from two different points of view which are both
legitimate and inseparable and that the information and capacity constitute criteria of
differentiation points of view.

Strength of graph theory

Simplicity of the graph

A graph can be defined as a finite set of points called vertices (i.e. nodes)and a
set of relations between these points called edges (i.e. arcs).

Graph theory relates primarily to the existence of relationships between vertices

or nodes and, in the figure that represents the graph, the localization of nodes is
unimportant unless otherwise specified, and only the existence of a relationship
between two nodes counts.

Formally, the graph G = (V, E) is a pair consisting of:

– a set V = {1,2,…, N} of vertices;
– a set E of edges;
– a function f of E in {{u, v}⏐u, v∈V, u ≠ v}.

An element (u, v) of VxV may appear several times: the arcs e1 and e2, if they
exist, are called multiple arcs if f(e1) = f(e2). The graph will then be a multigraph or
p-graph, where the value of p is that of the greatest number of appearances of the
same relation (u, v), i.e. the number of arcs between u and v.
xviii Graphs and Networks

If the arcs are directed, we will then talk of a directed graph or digraph. If the
arcs are undirected, we are dealing with a simple graph that can be a multigraph3.

The graph G is similarly characterized by the number of vertices, the cardinal of

the set X, which is called order of the graph.

The total number of arcs between two nodes has a precise significance with
regard to the definition of the graph only if: p ≠ 1.

When p > 1, the number of relations between two nodes i and j may be between
0 and p. The graph is then called p-graph and multigraph when the arcs are
undirected.

In order to know the number of pairs of connected nodes it is therefore necessary

to have the precise definition of the relations, i.e. an integral description of E which
is generally expressed in the shape of a file or a table4.

If the graph admits loops, i.e. arcs, whose starting points and finishing points are
at the same node, and it admits multiple arcs, we call it a pseudo-graph, which is the
most general case.

Graph theory only takes into account the number of nodes and the relationships
between them but does not deal with the vertices themselves. The only exception to
this rule is the characteristic of source or (and) wells which is recognized at nodes in
certain cases, such as during the calculation of the maximum flow for Ford-
Fulkerson [FOR 68], etc.

However, merely taking into account the existence of nodes, their number and
the relationships between them in graph theory is insufficient for network modeling.
A better individual description of network vertices is an important problem that
graph theory must also tackle to enable certain microsimulations, such as the study
of flows and their directions within the network crossroads, or the capacity of the
said crossroads, etc.

Thus, graph theory only deals with relationships between explicitly defined
elements which are limited in number. Indeed, in order to determine certain
traditional properties of graphs, such as the shortest paths, the Hamiltonian cycle,
etc., the number of nodes must necessarily be finite.

The graphic representation of G is extremely simple: “it is only necessary to

know how the nodes are connected” [BER 70]. The localization of the nodes in the

3 See below the definition of the simple undirected graph and the multigraph.
4 See section 11.1.2.
 Introduction xix

figure, i.e. implicitly on the plane, the representation or drawing of the graph do not
count, nor does the fact that the latter has two, three or n dimensions.

This offers great freedom in representing a graph. On the other hand, for the
reproduction of a transport network, for example, and if we wish the result to
resemble the observation, in short, if we want to approximate a map, this
representation will have to be specified. This is done by associating to it the
necessary properties or additional constraints, so that the development process of the
representation can be repetitive and the result reproducible (for example, definition
of the coordinate type attributes for the nodes), which is what Waldo Tobler requires
for maps.

Simplicity of the methods of definition and representation of graphs

Let us consider the associated matrix or adjacency matrix A of graph G. It is the
Boolean matrix n × n with 1 as the (i, j)-ith element when u and v are adjacent, i.e.
joined together by a edge or a directed or undirected arc and 0 when they are not
[COR 94].

Other authors [ROS 98] generalize this notation by accepting the loop (by noting
it 1 at the (i, i)-ith position) and multiple arcs, thus considering that the adjacency
matrix is then not a zero-one or Boolean matrix because the (j, i)-ith element of this
matrix is equal to the number of arcs associated to {ui, vi}. In this case, all the
undirected graphs, including multigraphs and pseudo-graphs, have symmetrical
adjacency matrices.

The problem of the latter notation is that it can be difficult to distinguish, unless
we define beforehand a valuated adjacency matrix when the valuations are expressed
as integers and small numbers.

The list of adjacency

The use of the adjacency matrix is very simple. However, it may be
cumbersome, in particular in the case of a large graph whose nodes are only
connected by several arcs which is, for example, the case of a road network or a
lattice on a plane. In this case, the matrix proves very hollow and the majority of the
boxes are filled with zeros. To optimize the calculation procedures we then use
methods which make it possible to remove these zero values and to only retain the
existing arcs.

One of the simplest ways of describing a graph, in particular by using a machine,

is to enumerate all its arcs when there are no multiple ones or to enumerate them by
identifying [MIN 86] those whose origin and destination are identical when we are
xx Graphs and Networks

dealing multigraphs or directed p-graphs, which constitutes an arcs file5. The writing
can be simplified by using an adjacency list.

This adjacency list specifies the nodes which are adjacent to each node of the
graph G. We can even consider for a Boolean adjacency list of a p-graph or of a
multigraph that the number of times where the final node is repeated indicates the
number of arcs resulting from the origin node and leading to the destination node,
half a bipolar degree. If the description of the graph is not only Boolean, it might
then be necessary to identify each arc between the same two nodes, in particular, by
their possible valuation, weighting or another characteristic, such as a simple
number.

The incidence matrix

For a graph without loop, the values of the incidence matrix “vertices-edges”
Δ(G) are defined [BEA 92] by:
– 1 if x is the origin of the arc;
– -1 if x is the end of the arc, 0 otherwise.

In order to avoid confusion let us recall that it is completely different from the
“node-node” adjacency matrix whose valuation is equal to 1 when the two nodes
considered are connected by an arc. It is this latter matrix, which in certain works is
referred to as the associated matrix.

The algorithmic ease has already been underlined and the methods of description
of graphs listed above, which are naturally usable by a machine, do nothing but
amplify it.

The adjacency matrix enables a simple usage of numerous algorithms, as well as

numerous indices, as we will be able to see. It also makes it possible to use sub-
tables, etc. However, the description by using an adjacency list enables a greater
processing speed due to the absence of zero values tests and the possibility of using
pointers6.

Hereafter we will establish that with some supplements this description of graphs
enables us to describe representations and reproducible plots, and that it is
sufficiently flexible to extend the formalism of graphs to other fields.

5 See in Chapter 12 an example of time-lag graphs.

6 It can be defined as the address of an element.
 Introduction xxi

Glossary of graph theory for the description of networks

The definitions of graph theory are commonly allowed and scarcely leave ground
for ambiguity. However, certain terms have evolved through time, just as it happens
in any active field. We propose to develop the representations of graphs by
considering them as strictly belonging to the theory and to express other
representations in the form of graphs. Therefore, we must now specify the
definitions of the most used terms.

Indeed, since the fundamental work of Berge [BER 70] was published in France
30 years ago a certain number of definitions have evolved through use (see below).

Directed graph
A directed graph (V,E) consists of a set of vertices V and a set of edges E, which
are pairs of the elements of V [ROS 98].
“Pseudographs form the most general type of undirected graphs, since they can
contain multiple loops and arcs. Multigraphs are undirected graphs that may
contain multiple arcs but not loops. Finally, simple graphs are undirected graphs
with neither multiple arcs, nor loops” [ROS 98].

Arc and edge

An arc is a directed relation between two nodes (U, v) of the set of nodes of G.
An edge is always an undirected arc between two nodes (U, v) of G.

Adjacency
Adjacency defines the contiguity of two elements. Two arcs are known as
adjacent if they have at least one common end. Two nodes are adjacent if they
are joined together by an arc of which they are the ends. The nodes u and v are
the final points of the arc {u, v}.

Incidence
Incidence defines the number of arcs, whose considered node is the origin
(incidence towards the exterior: out-degree) or the destination (incidence
towards the interior: in-degree). Since the degree of a node is equal to the
number of arcs of which it is the origin and/or destination, each loop is counted
twice.

Regular graph
When all the nodes have the same degree, the graph is known as regular.

Degree of a node
The degree of a node in an undirected graph is the number of arcs incidental to
this node, except for a loop that contributes twice to the degree of this node. The
degree of this node is noted by deg(v).
xxii Graphs and Networks

Symmetric graph Complete graph

A graph is known as symmetric, if A complete graph is a graph where each
each node is the origin and node is connected to all the other nodes by
destination of the same number of exactly one arc. A complete graph with n
arcs. nodes is noted by Kn. A complete directed
graph is a digraph where each node is
The adjacency matrix of a connected to all the others by two arcs of
symmetrical graph is symmetrical. opposite directions.

Subgraph
A subgraph is defined by a subset A⏐A⊂ V of nodes of G and by the set of arcs
with ends in A⏐UA⊂ U, GA = (A,UA). For example, the graph of the Central
region is a subgraph of France. It is fully defined by an adjacency submatrix.

Partial graph
A partial graph is defined by a subset of arcs H⊂E/GS = (V,E). A partial graph
may be a monomodal graph of a multimodal graph as well as a graph of trunk
roads within the graph of all the roads in France. The adjacency matrix of a
partial graph has the same size as the adjacency matrix of the complete graph.
For example, if the partial graph is a modal graph (i.e. defined by a specific
means of transport), the adjacency matrix of the complete graph (i.e. of the
transportation system) is the sum of all the adjacency matrices of the partial
graphs (various means of transport).

Partial subgraph Chain

A partial subgraph combines the A chain is a sequence of arcs, such that
two characteristics mentioned each arc has a common end with the
above: it is formed by a subset of preceding arc and the other end is in
nodes and a subset of arcs GSA = common with the following one. The
(A,V) such as, for example, the cardinal of the considered set of arcs
partial subgraph of TGV cities. defines the length of the chain. In a
transport network where the arcs are, by
definition, directed, the chain only makes
sense only if the arcs are symmetrical, i.e.
directed both ways.

Path Circuit
A path is a chain where all the arcs A circuit is a path whose origin coincides
are directed in the same way, i.e. with the terminal end.
the end of an arc coincides with the
origin of the following one.
 Introduction xxiii

Cycle Eulerian cycle

A chain is called a cycle if it starts An Eulerian cycle in a graph G is a simple
and finishes with the same node. cycle containing all the arcs of G. An
Eulerian chain in a graph G is a simple
chain that contains all the arcs of G.
A chain is known as Hamiltonian, if it
contains each node of the graph only once.

Connected graph Strongly connected graph

An undirected graph is connected if A graph is described as strongly
there is a chain between any pair of connected if, for any pair of nodes,
nodes. there exists a path from the origin
A directed graph is known as connected node to the destination node.
if there is there a path between any pair In other words, in a strongly
of nodes. connected graph it is possible to go
from any point to any other point and
to return from it, which is one of the
essential properties of a transport
network.

Quasi-strongly connected graph Bi-partite graph

A graph is known as quasi-strongly A graph G is bi-partite if the set V of
its nodes can be partitioned into two
connected if for any pair of nodes u, v,
non-empty and disjoined sets V1 and
there is a node t, from which a path
V2 in such a manner that each arc of
going to u and a path going to v start
simultaneously. the graph connects a node of V1 to a
A strongly connected graph is thus node of V2 (so that there is no arc of
quasi-strongly connected. G connecting either two nodes of V1
or two nodes of V2)7.

A joint or pivot
A node is a joint if upon its suppression the resulting subgraphs are not
connected.

Isthmus
An isthmus is an edge or an arc whose suppression renders the resulting partial
subgraphs unconnected.

Articulation set
By extension, a set UA ⊂ U is an articulation set if its withdrawal involves the
loss of the connectivity of the resulting subgraphs G.

List 1. Essential definitions

7 See below the K3,3 graph.

xxiv Graphs and Networks

Description, representation and drawing of graphs

For the majority of authors the term representation indicates the description of
the graph by the adjacency matrix and the adjacency list or the incidence matrix and
the incidence list, as well as that the graphic representation of the considered graph
in the form of a diagram, whose absence of rules we have seen8.

For representations in the form of a list or a matrix table we will use the term
description, possibly by specifying computational description and by mentioning the
possible attributes of the nodes, such as localization, form, modal nature9,
valuations10 of the arcs, etc.

For graphic, diagrammatic representation we will use the term (graphic)

representation or drawing of the graph.

This notation appears more coherent to us since, in the first case, we describe the
graph by listing all of the nodes and arcs, possibly with the attributes of the nodes
and the characteristics of the arcs: modal nature, valuation, capacity, etc., which are
necessary for computational calculation. For the computer the representation of arcs
has neither sense nor utility.

On the other hand, in the second case, we carry out an anthropic representation
of the graph, possibly among a large number of available representations according
to constraints that we set ourselves, such as planarity, special frame of reference,
isomorphism with a particular graph, or geometrical properties that we impose on a
particular plot, such as linearity of arc, etc.

Isomorphic graphs

The simple graphs G1 = (V1,E1) and G2 = (V2,E2) are isomorphic if there is a

bijective function f of U1 in U2 with the following properties: u and v are adjacent in
G1 if and only if f(u) and f(v) are adjacent in G2 for all the values of u and v in E1.
Such a function f is an isomorphism.

8 See section 1.1.1.1.

9 Here the term indicates the means of transport which is possibly assigned to the arc:
terrestrial, such as a car, a truck, a train, or maritime, by river or air.
10 Indicates a qualifying value allotted to an arc or an edge, such as distance, duration, cost,
possibly modal capacity, etc. Two arcs stemming from the same node and having the same
node as destination can have different valuations, for example, distance by road and rail
between two cities.
 Introduction xxv

V2

U1 U2 V1 V2
U2 V3
V1
U3
U1

U5 U4 V5 V4 U3 U4 V3 V4

Figure 1. Example of isomorphic graphs

Plane graph

A plane graph is a graph whose nodes and arcs belong to a plane, i.e. whose plot
is plane. By extension, we may also speak of a plot on a sphere, or even on a torus.

Two topological graphs that can be led to coincide by elastic strain of the plane
are not considered distinct.

All the graph drawing are not necessarily plane; they can be three-dimensional
like the solids of Plato, or like a four-dimensional hypercube traced in a three-
dimensional space and projected onto a plane as the famous representation of The
Christ on the Cross of Salvador Dali.

Planar graph

It is said that a graph G is planar if it is possible to represent it on a plane, so that

the nodes are distinct points, the arcs are simple curves and two arcs only cross at
their ends, i.e. at a node of the graph.

The planar representation of G on a plane is called a topological planar graph

and it is also indicated by G.

Any planar graph can be represented by a plane graph, but the reciprocal is not
necessarily true.

Saturated planar graph

A planar graph is described as saturated when no arc can be added without it

losing its planarity. In a saturated planar graph the areas delimited by arcs are
triangular.
xxvi Graphs and Networks

Christaller’s transport network (Figure 3) [CHR 33] is a plane graph based on

triangular grids it is neither planar nor saturated because some arcs do not only cut
across each other at the nodes and some areas are quadrangular.

Figure 2. European11 quadrimodal graph

11 Graph plotted by CESA Geographical position working group 1.1 Study Program on
European Spatial Planning, December 1999. An extended version integrates the ferry boat
into this graph which represents four modes of transport.
 Introduction

Figure 3. Walter Christaller’s network

xxvii
xxviii Graphs and Networks

Road network
Highways and
Ch. DECOUPIGNY, K. SERRHINI, CESA April 2000 express routes
Route 70
Route 60 and 50

Figure 4. Multimodal graph of France12

12 Planar multimodal saturated graph of France. Plotted by CESA, this graph is truly
multimodal because, due to the zoom method of Laurent Chapelon (see below), in its
extended version it can integrate regional, departmental, agglomeration and even intra-urban
graphs with their specific modes (1999).
 Introduction xxix

The search for the planarity of graphs led to famous publications and numerous
algorithms. Among the best known results let us quote two traditional properties:
– Euler’s formula:
- let G be a connected planar simple graph with e edges and v vertices,
- let r be the number of regions (or areas) in a planar representation of G.

Then r = e – v + 2

– Kuratowski’s theorem: a graph is not planar if and only if it contains a

homeomorphic subgraph with K3,3 or K5 (see Figure 5).

V2

V3
V1
U1 U2 U3

U6 U5 U4 V5 V4

Figure 5. K3,3 graph and K5 graph [ROS 98, p. 479 and 419]

We add two definitions to these traditional ones in order to specify graph

plottings with more than two dimensions.

Graph with geographical reference

A graph with geographical reference “GGR” is either a plane graph or a graph

plotted on a simple surface, for which it is possible to perform a bijection between
the nodes of the plotting and those of its projection onto a plane. In other words, no
node or face of a GGR must have a double point as a projection.

The typical example will be a digital terrain model like that in Figure 6 [BRU
87]: this graph, whose areas are quadrangular and thus not necessarily plane,
xxx Graphs and Networks

belongs to a surface defined in a three-dimensional space which is qualified by the

use of the 2.5 dimension in landscape analysis13.

Graph with spatial reference

We will also call a graph with space reference, or GSR, a graph plotted on a
convex surface or in a three-dimensional space or more, such as, for example, the
solids of Plato but also the GSR like those used by Kamal Serrhini to define co-
visibilities [SER 00]14. In this case, when the graph is projected onto a plane there
does not have to be a bijection between the drawing of the graph in the space and its
plane projection because certain points, nodes or areas can be doubled15. Let us note
that the use sometimes qualifies the graphs drew on a “planar” sphere by extension,
since in this case the plane is considered as a sphere with infinite radius.

Figure 6. Digital terrain model of the Mount Blanc [BRU 87]

13 This qualification, i.e. dimension 2.5, is usual and established by the use in co-visibility. It
should be specified that it is used to differentiate this type of GGR grid from the GSR defined
hereafter, which, in turn, is not limited by the constraint of bijection. This qualification of
dimension 2.5 has nothing to do with a non-integer dimension of the fractal type.
14 See below Part 2, Chapter 10.
15 For example, several points of a façade or a work are projected orthogonally onto the same
point in the ground.
 Introduction xxxi

Dual graph

There exist many definitions of duality16. We will retain the one used by the
majority of authors. The duality of a graph consists of associating each area of a
graph called primal to a node of the dual graph. Berge provides the following
definition for it [BER 87]: “let us consider a planar graph G, which is connected and
without isolated nodes. We make it correspond to a planar graph G* in the following
manner: inside every face s of G we place a node x * of G *; we make every edge e
of G we correspond to an edge e* of G*, which connects the z nodes x* and y*
corresponding to the faces s and t that are on both sides of edge e. The graph G*
thus defined is planar connected and does not have an isolated node: it is called the
dual graph of G” (see Figure 7).

Figure 7. Primal graph and its dual [BER 70]

This definition is insufficient and, for example, Aldous and Wilson specify
[ALD 00]: “let G be a connected graph. Then a dual is constructed from a plane
drawing (italics added) of G, as follows…”. The definition is more exact, as they
demonstrate, by defining the dual of a convex three-dimensional polyhedron, as we
examine in Part 3.

Similarly, in certain, even very simple, cases the dual is not of the same type as
the primal in the sense that a primal 1-graph can have a p-graph as a dual, as we will
see. This may present a problem in terms of graph description, since the adjacency
matrix of the dual then has to be an extremely hollow p-dimensional matrix…

16 For example, the definition of the dual graph of a grid (adjacency matrix) provided by
Pumain will not be retained here; [PUM 97], page 31.
xxxii Graphs and Networks

The concept of a dual graph is very rich, but insufficiently used. Indeed, it makes
it possible to make areas correspond to networks and vice versa, which is one of the
fundamental problems of synthesized images and of specialized network
representations among others. Thanks to the duality and to what stems from it, it is
possible to bypass some of the limitations facing modeling.

A particular type of graphs: the tree and tree structure

A tree is a connected undirected graph without a simple circuit. An undirected

graph is a tree, if and only if each pair of nodes is connected by a simple and single
circuit.

Tree and tree structure: we can call a particular node of the tree the root and then
attribute a direction to each node in such a way that there is a only path from the root
to each node. We then obtain a directed graph which is referred to as a tree structure
(see Figure 8).

f g

d
b

e
a

T
a c

b a
c
d e
b
f g e d

f g

With root a With root c

Figure 8. Examples: a tree and two tree structures [ROS 98, p. 506]
 Introduction xxxiii

If v is a node other that the root, the father of v is the single node u, such that
there is a directed arc from u to v. If u is the father of v, v is called the child of u.
Nodes with the same father are called siblings. The ancestors of a node other than
the root are nodes of the path leading from the root to this node. The descendants of
a node v are the nodes that have v as ancestor. The node of a tree that does not have
offspring is called a leaf. Nodes that have offspring are called internal nodes. A
subtree is the subgraph comprised of a node of the tree, its descendants and all the
arcs leading to its descendants.

A tree structure is described as m-ary if each internal node does not have more
than m offspring. The tree is called a complete m-ary tree if each internal node has
exactly m descendants. An m-ary tree with m = 2 is called a binary tree.

A tree structure will be quasi-strongly connected if there is a node u, from which

we can reach all the others via a path. The node u is known as the root of the tree: it
is the common ancestor of all the nodes.

An example of a tree structure in networks is a monopolar access map, such as

those plotted by Laurent Chapelon17, or Legrand’s star map presented below (see
Figure 9)

Some additional theorems deserve being mentioned:

– a tree with n nodes contains n-1 arcs;
– a complete m-ary tree with i internal nodes contains n = mi + 1 nodes;
– a complete m-ary tree with:
- n nodes contains i = (n-1)/m internal nodes and l = [(m-1) n + 1]/m leaves,
- i internal nodes contains l = (m-1) i + 1 leaves,
- l leaves contains n = (ml-1)/(m-1) nodes and i = (l-1)/(m-1) internal nodes.

17 Chapelon Laurent; see Chapter 1.

xxxiv Graphs and Networks

Figure 9. Merchandise traffic by rail in France. Annual throughput by line section

(in effective thousands of tons) in 1854. Extract from Renouard, Les transports
des merchandises depuis 1850, Armand Colin

The height or depth of a tree structure is the maximum level of the nodes or the
length of the longest path between the root and any node. An m-ary tree structure of
height h is known as balanced if all the leaves are at the levels h or h-1. There are at
most mh leaves in an m-ary tree of height h.
 Introduction xxxv

Spanning tree

Let G be a simple graph. A covering tree of G is a subgraph of G, which is a tree

containing each node of G. Finally, a simple graph is connected if and only if it has
a spanning tree.

The great algorithmic simplicity of graph theory is the heritage of operational

research. Numerous problems that arise in the network simulation have been solved
by operational research [FIN 08, KAU 64], which developed many algorithms first
for a manual and then for a computerized resolution.

Among many publications that have made it possible to obtain effective

algorithms it is necessary to cite Ford and Fulkerson for the flow algorithm in
graphs, Dantzig and then Ford for the first shortest paths algorithms, etc. These
works paved the way for many later achievements that have been encouraged and
reinforced by developments in data processing, which provided the necessary means
of computation.

Graph description methods for computer by list or by adjacency or incidence

matrix, and all the methods derived from those easily lend themselves to the
computerized processing and thus to the resolution of problems involving large
graphs.

The use of these methods of computerized graph description has been made even
simpler since the 1990s in the domain of transport networks by the development of
data capture methods and the development of adjacency list-type files through the
digitalization of maps or direct on-screen capture and modification (Chapelon, Les
logiciels MAP et NOD [MAP and NOD Software]; see Chapelon, L’Hostis). The
rapid development and generalization of GIS (geographical information systems) or
of SRIS (spatial reference information systems) reinforce and accelerate this
evolution.

Once an algorithm has been developed, the size of the graph or manageable
network depends only on the data, the characteristics of the machine and the
available computing time. However, it has to be said that the computing time of
certain algorithms grows very quickly with the size of the graph. This is a problem
of algorithm efficiency, of calculation complexity: “let us consider a given class of
problems, whose size is given by the integer n. We say that a solution algorithm A
has a complexity of the order f(n) – noted by Of(n) – if the asymptotic growth of the
computing time TA(n) according to n, which is the size of the problem, is of the
order f(n) at the most (where f is an increasing positive function of n, which is
generally a polynomial function). An algorithm A is known as polynomial and of
complexity O(nk) if, in the worst case scenario, the resolution time grows as the kth
xxxvi Graphs and Networks

power of n, which is the size of the problem, when n→∞…” [MIN 86]. For
example, the time needed for the research of the minimal paths between any pair of
nodes in a graph by Floyd’s algorithm [COR 94, p. 550] grows according to the
cube of the number of nodes. In this algorithm that has an exceptionally beautiful
symmetry of expression, there are three overlapping loops and when the number of
nodes doubles, the computing time increases eightfold. It is an algorithm of
complexity O(n3), that is, a polynomial complexity [ROS 98, p. 97 et seq.].

On the other hand, although computers have progressed considerably in their

capacities to rapidly produce high definition images, this problem has practically not
been tackled by graph theory whose development logic remains largely marked by
its history and dependant on the research logic of mathematicians and operational
researchers who are practically unaware of this aspect.

However, the fastest method to apprehend networks and results expressed as

networks is the synthesized image, i.e. an effective and precise graphic
representation, which is reproducible and verifiable to the same extent as the
optimization calculations of operational and therefore algorithmic research. In
spatial analysis and urban development planning the synthesized images must
become a measurement and diagnostic instrument, as in many other disciplines:
medicine, physics, biology, etc.

The representation of networks by graphs

Network modeling by graphs constitutes a relatively recent application for an old

requirement. Although the network in its current meaning is a relatively modern
concept, the need to represent a set of roads and terrestrial or maritime routes, on the
other hand, is very old and has become accentuated with time. Indeed, this need
develops along with the progressive interpenetration of societies witnessing the
development of their exchanges. Little by little, tradesmen, sailors, soldiers will feel
a greater need for not being dependant on local guides, for no longer being limited
by local knowledge and for being able to consider all of the characteristics of
various displacements: difficulty, length, duration, risk, locally available resources,
etc. That explains the development of cartographic representation methods rather
than networks and spaces.

From functional representation to resemblance

One of the most ancient functional representations of a road network is the

“Peutinger Table” (Figure 10 of the central book) from the 2nd or 4th century, (that
 Introduction xxxvii

reached us thanks to a medieval copyist) which represents a map of ancient routes

from city to city with the indication of distances.

Figure 11. Layout of the network of the Peutinger Table

against the background of a modern map

This table is very interesting because it does not constitute a map but a functional
representation corresponding to the drawing of a graph. It makes it possible to
identify stages (poles or nodes of the graph), to calculate distances (valuation of the
arcs) and, thus, the duration of the journey.
xxxviii Graphs and Networks

Indeed, we note that it strictly corresponds to the definition of a graph plotting

provided by Berge: “it is only necessary to know how the nodes are connected”. The
Peutinger Table only slightly resembles a map and its representation against the
background of a map18, whose form corresponds to that of the territory as above,
makes it possible to see the differences (see Figure 11). These two representations
are isomorphic but neither one nor the other has the properties of reproducibility and
comparability. We could think that the second is very close to having these
properties, but it is simply the force of habit that makes us forget all the implicit
notions that it uses, such as, for example, the localization of nodes identical to the
localization of cities, which results from these “stages mentioned in the table…”.
The proof of this assertion is very simple and to observe it would suffice to ask a
student, who is unaware of all the graphic semiology, to draw this map.

A millennium later, starting from the 13th and 14th centuries, the first portulans19
established by the Genoese and Venetian and then by Arabic and Portuguese
navigators, illustrated the famous pilot’s logbooks thus showing that the description
of the possible route must be accompanied by a representation, a graphic drawing
and that the text is not enough20.

This need for cartographic representations supplementing the “nautical

instructions” kept developing with the great discoveries and would constitute an
extraordinarily important tool at sea and then on solid ground. Was it not Napoleon
who considered the map as a weapon?

Increasingly scientific mapping

A considerable progress in precision, in semiological and mathematical rigor in

the increasingly objectified maps has been carried out, among others, under the
influence of the geometricians, of whom the Cassinis21 were undoubtedly the most
famous. The triangulation of a space thus corresponds to defining a saturated planar
graph in this space.

18 This map has been plotted in 2001 by Olivier Marlet, a student of “Urban Sciences” DEA
archeology option at Tours.
19 From the Italian portulano: pilot.
20 Indeed, if machine descriptions of graphs as adjacency matrix, for example, enable many
calculations including those of certain “morphological” indicators, they do not make it
possible for the human mind to visualize the specific network, which is described without
additional data…
21 From 1696, which is the date of the first planisphere plotted by Gian Domenico Cassini, to
the large map known as Cassini’s with a scale of 1/86,400, which was introduced in 1789 to
the French National Assembly by his great-grandson Jean Dominique Cassini. This is one of
the first topographic maps based simultaneously on a complete triangulation of France and
geodetic readings taken on the ground.
 Introduction xxxix

However, between the sea or terrestrial maps and the Peutinger Table there is a
fundamental difference: maps have the aim of representing the territory (i.e.,
amongst other things, a surface) in the most precise possible way: the relief of the
depths or altitudes, its form (plateaus, mountains, peaks, reefs, cliffs, crossing
points, major and minor rivers, currents, etc.), whereas the Peutinger Table is only a
functional representation and does not resemble road networks.

Even if the map resulting from the Cassinis features the road communication
networks, its aim is different: one of its constraints is that of “resemblance” to the
territory because it must describe it and not be limited to networks, to “paths and
circuits” in the sense of graphs, but also make it possible for someone to orientate
himself in this terrestrial or maritime space. It is a tool to define the position in space
in the most precise possible way.

This constraint of resemblance between the terrestrial surface and a plane

representation was the object of many mathematical works and the development of
the most traditional projection methods as those of Mercator22 and Lambert.

This objective of resemblance that the maps are given with the aim of
representing the space or the territory implies the existence of homeomorphism
between the space represented and the representation of space23. This property of
homeomorphism doubles on maps due to a resemblance resulting from graphic
semiology codes: the forests are in green, the rivers, the lakes and the seas in blue,
etc. Even on a small scale Michelin map, the winding roads are represented with
many turns indicating their sinuosity, etc. We are relatively far from Peutinger’s
functional representation, even though it features some zigzags, stylized mountains
and urban monuments.

Already overloaded with stylized decorations, fantastic characters and animals to

fill in the blanks and to compensate for the unknowns, the map is overloaded with
descriptive details of the territory by means of graphic semiology: the map has to
account for the territory and it must make it possible to apprehend it as a whole, its
evolution and its significant details. We refer to this as the constraints of
resemblance.

22 Mercator Gerhard 1521-1594 Tabulae geographicae ad mentem Cl. Ptolemaei (1578) -

Mercator readopts the projection invented in the 1st century by Martin de Tyr. This projection
on a cylinder tangent to the equator is improved by Lambert who uses a cone tangent to a
central meridian line.
23 This increasingly strong homeomorphism risks creating confusion between the map and
the territory, which Korzyski denounces in: Science and Sanity, an Introduction to Non-
Aristotelician Systems and General Semantics (1933): “the map is not the territory”.
xl Graphs and Networks

Networks, roadmaps and graphs: the constraints of resemblance

The maps representing networks must give the same impression of immediate
resemblance, thus enabling an instantaneous comprehension. Undoubtedly, it is for
this reason that the layout of a road on the Michelin map evokes the real layout of
the road: straight where it is actually straight and curving when the layout is
winding. Having said that, there is no strict relation between the two layouts because
the width of the route is not connected with that of the representation but only bears
a similarity: bolder lines for highways, thinner for minor roads, etc. Also, the
smaller the scale of the map, the larger the disproportion becomes, which necessarily
requires a simplified layout (Figure 12).

Figure 12. Evolution of a coastal layout through the reduction of the scale of representation

Networks, tree structures, flow charts: the constraints of hierarchy

The described networks can be of particular types, such as flow charts retracing a
hierarchical system in a structure. The usage dictated by habit encourages a
representation of this type of dissymmetrical network in a pyramidal form, with the
most important element in the hierarchy sitting atop and the subordinate levels
following each other in order of decreasing importance.

These tree structures use a technique which is found in scientific applications, in

physics and data processing: diagrams blocks ([CHO 64], p. 91 et seq.), which will
 Introduction xli

hereafter be referred to as data flow charts [FAU 75], which are graphic translations
of a program or of a part of a program.

Many domains are still insufficiently explored

Graph theory primarily developed in two research directions: mathematical

research and operational research, according to the specific dynamics of these two
fields. The concerns and the questions posed by mathematicians are very specific,
but often have very little to do with those of the developers of transport networks,
space analysts and urban planners.

The concerns of the operational researchers are much closer to them. Some are
obviously common, but urban planners, as those for whom their work is meant, i.e.
contractors and the public, have a visualization requirement that cannot be
circumvented and a need for illustration that, except in rare cases, is not present for
operational researchers.

Urban planners need representations that are repetitive and verifiable, as well as
comprehensible for all the public, in particular, within the framework of public
interest investigations. Moreover, we have seen that graph theory is absolutely not
preoccupied with the representation in the sense of graphic plotting of a graph. In a
certain manner Berge eliminates the problem by writing: “it is only necessary to
know how the nodes are connected. The localization of the nodes in the figure, the
representation or plotting of the graph do not count” [BER 70, BER 87].

It is neither considered nor even mentioned that the representation can be

“plotted” on a plane or in a three-dimensional space or be the projection of one onto
the other [CAU 76, CAU 86], except in the case of a planar graph and that of the
construction of a dual graph, which can be plotted on a plane, a sphere or a torus
[HEA 76, HEA 86, LHU 76, LHU 86, POI 76, POI 86]. The concerns of the
founding fathers of graph theory have been simultaneously developed and reduced.

The drawing of graphs and their constraints, network image or images

This absence of rules of graph drawing is a considerable difficulty for the
representation of modeled networks, whereas for network cartography there is a
graphic semiology.

We pose as an axiom that the rules of the effective, material layout and of a
representation form an integral part of graph theory and that the necessary and
sufficient conditions for this realization to be reproducible and verifiable must be
established.
xlii Graphs and Networks

We propose as a working hypothesis that this definition will enable us to

integrate various geometrical results into graph theory, in particular, certain basic
elements of fractals that will make it possible to provide the rules for the extension
or reduction of graphs, to ground in theory the already operational concept of zoom
by using the property of self similarity or strict or probabilistic internal homothety
and to provide global morphological indicators of networks, such as the fractal
dimension [GEN 00].

The problem of graph extension and reduction and the scales of representation
under constraints are very important problems in network modeling. We have
mentioned the complexity of the algorithms for the resolution of traditional
problems, such as that of the search for the shortest paths in a graph, and also
mentioned that it was of the O(n3) level. It is then easily conceived that two
problems present themselves acutely: that of the reduction of large graphs and that
of the rule to be used for the process not to be left completely to the free
interpretation of the modeler, but to be, as we have already stated, reproducible and
verifiable; basically, to be more objective, more scientific.

This problem of reduction of graphs and of the size of graphs to be processed,

even though it evolves rapidly with the performance of machines, ipso facto poses
the problem of the levels of calculation or the scales in spatial analysis. The problem
of the existence of various network apprehension and modeling scales immediately
leads to the question of the existence of internal homothety between the various
levels considered. Can they be treated in the same manner and up to which point,
which limit?

We pose the working hypothesis that at a certain level the logic and the
hypotheses change. For example, the reflection on the saturation of a transport
system can be solved by calculating the capacity of one or several cross-sections of
the network by using the Ford-Fulkerson theorem. However, this is done at the cost
of very strong assumptions and, in particular, those according to which the node is
strictly neutral, which is the general assumption in graph theory that only deals with
the relationships between nodes and not with them as such. Consequently, the
saturation of an arc is related to its capacity. This is obviously at the very least an
abusive assumption and completely opposed to the reality of the flux of flows. Any
car driver knows from experience that the saturation of an axis almost always results
from the saturation of a crossroads, i.e. of a node, except for cases that constitute an
accident, like the reduction of the number of lanes, and even in this last case,
normally, the capacity of an arc should equal that of its narrowest portion.

Similarly, the total disinterest of the theory in the vertex or node, structure means
that the description of certain network elements is practically impossible. For
 Introduction xliii

example, the highway half-interchange24 obviously exists and yet it is impossible to

represent it by a node under the current state of the theory.

Therefore, it is necessary to improve network modeling, to call into question this

assumption of the neutrality of nodes and to consider their structures, the possibility
of describing them while respecting the axioms and fundamental properties of graph
theory.

We put forward another working hypothesis: the development of drawing rules,

of representation as well as a reflection on the duality must make it possible to pass
from the traditional graph to percolation and the use of MSA (multi-agent systems)
in a cellular network, with all the modifications of hypotheses with respect to
application that this enables and implies. Indeed, to describe and to simulate, even to
optimize, as a whole,, a transport system is a type of problem that implies certain
assumptions that we will clarify; on the other hand, describing the individual
behavior of an agent in a network is another problem and does not necessarily
require the same assumptions, since sometimes it renders them particularly illusory.
Similarly, even thought graph theory indeed makes it possible to represent a network
with facility and flexibility and with a certain degree of generality in order to solve
some of the problems mentioned above, it enables very little with respect to the
graph area which is enclosed between its various arcs.

That is obvious because graph theory is only preoccupied with the relations
between nodes. However, the property of duality makes it possible to take this space
into account. Nonetheless, this property is closely related to the planar
representation of the graph and therefore requires a certain number of additional
hypotheses and the contribution of additional precise details in order to answer the
questions posed above. This problem will be tackled later in the book.

Lastly, one of the problems that dynamic network modeling comes across is the
development of networks, the extension of the graph and of its grid, either
simultaneously or subsequently.

We will see that the possibility of describing simple fractals as graphs makes it
possible to find a solution for these problems.

If we wish, as is the case, to be able to achieve comparable and repetitive

drawing and, thus, the layout of maps in Waldo Tobler’s sense, we will note that
additional assumptions are necessary to specify the type of possible layout.

24 See Part 3: a half-interchange does not enable the exit and entry to the highway in both
directions.
xliv Graphs and Networks

Similarly, we observe that with some additional assumptions and constraints we are
able to drow graphs that have all the properties of fractals.

An explicit assumption must be stressed due to its consequences on later

applications: in graph theory we adopt the hypothesis of perfect rationality and total
knowledge of the network. Indeed, let there be a graph corresponding to the
definition above, where all its elements are known to us via a complete and finite list
of nodes and that of the relations between nodes, which is described by the list or the
adjacency matrix. Such a definition is absolutely necessary to be able to deal with
the basic problems of graph theory and, in particular, the problems of optimization
of paths and maximum flow. The demonstration is obvious: if a single arc was
missing, it would be impossible to obtain the matrix of the minimal path, since at
least one of them might not exist if an arc was added. A fortiori if a node is missing.

This significant hypothesis is no longer necessary for the simulation of agents in

a “cellular graph”: in order to find a path, the local knowledge such as, for example,
the number of arcs starting at the node or the next node and their orientations with
respect to a reference direction may suffice. The set of necessary assumptions will
thus have to be specified for each application and each representation of the graphs.

The plan
The plan of the book follows the development of the questions that we have
touched upon above and comprises three parts: the first is entitled Graph theory and
network modeling; the second, Graph theory and network representation and the
third, Towards a multilevel graph theory.

Bibliography

[ALD 00] ALDOUS J.M., WILSON R.J., Graphs and Applications, an Introductory
Approach, Springer-Verlag, London, Berlin, Heidelberg, 2000.
[BEA 92] BEAUQUIER D., BERSTEL J., CHRETIENNE P., Eléments d’algorithmique,
Masson, 1992.
[BER 70] BERGE C., Graphes, Dunod, Paris, 1970.
[BER 87] BERGE C., Hyper graphes, Bordas, Paris, 1987.
[BRU 87] BRUNET R., La carte mode d’emploi, Fayard/Reclus, Paris, 1987.
[CAU 86] CAUCHY A.L., “Recherches sur les polyèdres”, Premier mémoire J. Ecole
Polytechnique 9 (Cah.16) (1813), p. 68-86., in N.L. Biggs, E.K. Lloyd and R.J. Wilson,
Graph theory 1736-1936, Oxford University Press, 1976, 1986.
[CHA 97] CHAPELON L., L’HOSTIS A., MATHIS P., Les logiciels MAP et NOD, 1997.
 Introduction xlv

[CHO 64] CHOW Y., CASSIGNOL E., Théorie et application des graphes de transfert,
Dunod, 1964.
[CHR 33] CHRISTALLER W., Die zentralen Orte in Sûddeutschland, G. Fischer, Paris,
1933.
[COR 94] CORMEN T., LEISERSON C., RIVEST R., Introduction à l’algorithmique,
Dunod, Paris, 1994.
[EUL 1736] EULER L., “Solutio problemats ad geometriam situs pertitnentis”, Commemtarii
Academia Scientiarum Imperialis Petropolitanae, 8, p 128-140, Opera Omnia, vol. 7,
p. 1-10, 1736.
[EUL 1758a] EULER L., “Demonstreatio nonnullarum insignium proprietatum quibus solidra
hedris planis inclusa sunt praedita”, Novi Comm. Acad. Sci. Imp. Petropl. 4 (1752-1753,
published in 1758), 140-160, Opera Omnia (1), vol. 26, 94-108, 1758.
[EUL 1758b] EULER L., Elementa doctrina solidorum. Novi Comm Acad. SCI. Imp.
Petropol. 4 (1752-1753, published in 1758), 109-140, Opera Omnia (1) vol. 26, 72-93,
1758.
[FAU 75] FAURE R., Eléments de recherche opérationnelle, Collection “Programmation”,
Gauthier-Villars, Paris, 1975.
[FIN 08] FINKE G., Operations Research and Networks, ISTE, London, John Wiley &Sons,
New York, 2008.
[FOR 68] FORD L.R.Jr., FULKERSON D.R., Les flots dans les graphes trad, J.C. Arinal
(dir.) Gauthier-Villars, Paris, 1968, Flows in networks, RAND Corporation, 1962,
Princeton University Press.
[GEN 00] GENRE GRANDPIERRE C., Forme et fonctionnement des réseaux de transport :
approche fractale et réflexions sur l’aménagement des villes, PhD Thesis, Besançon,
2000.
[HEA 86] HEADWOOD P.J., “Map Colour Theorem”, Quaterly Journal of Pure and Applied
Mathematics 24 (1890), 332-338, Graph theory 1736-1936 Biggs N.L., Lloyd E.K.,
Wilson R.J., Oxford University Press, 1976, 1986.
[KAU 64] KAUFMANN A., Méthode et modèles de la recherche opérationnelle, Vol. 1 and
2, Dunod, Paris, 1964.
[LHU 86] LHUILLIER S.6A.6J., “Mémoire sur la Polyédrométrie”, Annales de
Mathématiques 3 (1812-3), p. 169-189, in Biggs N.L., Lloyd E.K., Wilson R.J., Graph
theory 1736-1936 Oxford University Press, 1976, 1986.
[MIN 86] MINOUX M., BARTNIK G., Graphes, Algorithmes, Logiciels, Dunod, Paris, 1986.
[POI 86] POINSOT L., “Sur les polygones et les polyèdres”, J. Ecole Polytechnique 4
(Cah.10) (1810), p. 16-48., in Biggs N.L., Lloyd E.K., Wilson R.J., Graph theory 1736-
1936, Oxford University Press, 1976, 1986.
xlvi Graphs and Networks

[PUM 97] PUMAIN D., SAINT JULIEN T., Analyse spatiale 1 Localisations dans l’espace,
Armand Colin, 1997.
[ROS 91] ROSEN K.H., Mathématiques discrètes, Chénelière/MCGraw-Hill, 1991.
[SER 00] SERRHINI K., Evaluation spatiale de la covisibilité d’un aménagement, sémiologie
graphique expérimentale et modélisation quantitative, PhD Thesis, Tours, 2000.
[XUO 92] XUONG N.H., Mathématiques discrètes et informatique, Masson, Paris, 1992.
 Figure 10. Peutinger table

Created by M. Appert, L. Chapelon, 2001, UMR 6012-ESPACE, Montpellier Created by M. Appert, L. Chapelon, 2001, UMR 6012-ESPACE, Montpellier
NOD/MAP software: L. Chapelon, L’Hostis, Ph. Mathis, CESA, 1993/2002 NOD/MAP software: L. Chapelon, L’Hostis, Ph. Mathis, CESA, 1993/2002

Figure 1.3. Graph of the road network in the Figure 1.4. Sum of minimal driving times
East Thames corridor between 5 and 6 pm

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
Created by M. Appert, L. Chapelon, 2001, UMR 6012-ESPACE, Montpellier Created by M. Appert, L. Chapelon, 2001, UMR 6012-ESPACE, Montpellier
NOD/MAP software: L. Chapelon, A. L’Hostis, Ph. Mathis, CESA, 1993/2002 NOD/MAP software: L. Chapelon, A. L’Hostis, Ph. Mathis, CESA, 1993/2002

Figure 1.5. Performance of the road network Figure 1.6. Performance of the road network
between 1 and 2 pm (off-peak) between 5 and 6 pm (rush hour)

Created by M. Appert, L. Chapelon, 2001, UMR 6012-ESPACE, Montpellier

NOD/MAP software: L. Chapelon, L’Hostis, Ph. Mathis, CESA, 1993/2002

Figure 1.7. Simulation of the performance results obtained

with the construction of the Lower Thames Crossing in 2015
 Accessible space of the DTM Example of return paths
of the natural park of Tanet et obtained between the car parks
Gazon du Faing and the natural attractions on
the DTM graph of Tanet

Figure 2.2. Realization of the graph of the path from DTM

Figure 2.4. Geomorphology of the theoretical DTM graph

Impacts for a 60-minute single movement Impacts for a 60-minute return movement

Figure 2.5. Simulation of the movements on Figure 2.6. Impacts of the movements on the
the Tanet and theoretical space theoretical graph summing up the
specificities of the Tanet space
Figure 6.4. Representations of social Figure 6.6. Set of links between individuals
relationship with Krackpot software

Figure 6.7. Social relationships Figure 6.8. Social relationships

after a first “purification” after reorganizing the nodes manually

Figure 6.10. Conical tree

On the right: map of the quantity of
information contained in the phatic
spaces mentioned by the inhabitants of
the town center of Latakia. On the
right: various views in perspective
projection with a rendering showing
the town center and the quantity of
information in the phatic spaces in
Latakia. The values are represented by
cylinders. Everything is represented
against the cartographic background
of topographic services.

Design and plotting: O. Khaddour, Ph.

Mathis. CESA, 2003.

Figure 7.1. Various representations of the attractiveness of phatic

spaces according to Reilly’s model
Figure 8.9. Accessibility by road from the Ruhr to the large basins of activities
Figure 9.7. Stages of the construction of a map in space-time relief
Figure 9.8. Space-time relief of the road and motorway
 Figure 10.3. Method of modeling a multiresolution topography

Ph. Mathis, K. Serrhini in collaboration with M. Mayaud (cartographer), CESA, January 2001

Figure 10.5. Covisibility from the green mesh of the A28 section
or from the space before and after the construction of the bridge
(section of the A28 at Chanceaux-sur-Choisille, Indre-et-Loire, 37)
 Ph. Mathis, K. Serrhini in collaboration with M. Mayaud (cartographer), CESA, April 2000

Figure 10.6. Space seen along the A28 section. Generalized visibility from the A28
section reported (in %) on the smoothed square (32 x 32) meshes of space
(section of the A28 at Chanceaux-sur-Choisille, Indre-et-Loire, 37)

Figure 10.7. Determination of the Figure 10.8. Anthropic generalized visibility.

topographic transversal (view: Plan (XY)) Percentage or number of people in the space
having a view on each mesh of the A28 section
 Figure 10.9. Potential generalized visibility, from the space, of an A28 section balanced by
the distance from the installation (A28 section at Chanceaux-sur-Choisille, Indre-et-Loire, 37)

Figure 10.10. Directed, dynamic and static Figure 10.11. Primal graph, dual graph and
generalized visibility along the A28 section covisibility. Covisibility starting from the
deferred (in % or in time) on the square green mesh of the A28 section according to
space meshes the number of tops per square space mesh
 Figure 11.9. Culture areas in a Christaller space with
two cities of equal population (potential market)

Figures 11.11 and 11.12. For each year: on the left, production surfaces for each product
(in red for product 1, in green for product 2 and in blue for product 3); on the right,
market place where production is transported (all the domains that sell
any type of products in the same town are of the same color)
Figure 12.21. Example of a network with internal homothety [CHA 97]
CESA, Transports Group. Created by L. Chapelon, Ph. Mathis, in collaboration with H. Baptiste, S. Larribe, A. L’Hostis, M. Mayaud and K. Serrihini © Copyright 1997

Figure 12.22. Tree structure of the paths from Perpignan to the rest of France [CHA 97]
 Figure 13.2. Crossing the dune area: car parks are represented by a small blue line, tourist
paths by a blue line and the position of tourists on the beach by a blue dot; the sea at low tide
is in light blue, and the colors on the dune refer to the coefficient of practicability
ranging from easy in white to impossible (barbed wire) in red

Figure 13.3. Crossing the dune area according to two different

hypothesis regarding the practicability of paths
Figure 13.4. Simulations of the percolative diffusions of tourists according
to the digital model of the hill area, for a difficulty slope of 0%

Figure 13.5. Simulations of the percolative diffusions of tourists according

to the digital model of the hill area, for a difficulty slope of 14%

Figure 13.6. Simulations of the percolative diffusions of tourists according

to the digital model of the hill area, for a difficulty slope of 28%
 PART 1

Graph Theory and Network Modeling

 Chapter 1

The Space-time Variability of Road Base

Accessibility: Application to London

This chapter presents an attempt to integrate traffic conditions into an

accessibility calculation model associated with road networks at the regional scale.
For identical networks, congestion, which is closely linked to the use of
infrastructure, is responsible for variations of traffic speeds. In order to optimize
accessibility calculations it is of primary importance to obtain the “real” speed
section by section by taking into account the level of traffic. To that end we
endeavored to systemize the empirical approach initiated by the Transport Research
Board and developed in its Highway Capacity Manual [TRB 98]. This approach,
which is associated with a specific road network modeling method in the form of
valuated graphs, can indeed account for space and time variability of motor vehicle
traffic speed on roads.

1.1. Bases and principles of modeling

1.1.1. Modeling of the regional road network

Graph theory is one of the most effective means to model transport networks.
The vertices of a graph are associated with network nodes and the arcs are
associated with links. Along with the principle of mechanical graph description, this
type of modeling makes it possible to benefit directly from the calculating power
offered by computers.

Chapter written by Manuel APPERT and Laurent CHAPELON.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
4 Graphs and Networks

1.1.1.1. From network to graph

A priori no constraints restrict the choice of the graph vertices. The size of nodes
in terms of population, of economic or administrative functions, for example, may
be given prevalence. However, several factors must be taken into account in order to
improve the precision of the results.

First of all, the strategic role that certain nodes may play in the organization and
operation of the road network (crossroads, interchanges) makes it necessary to
compute them. Then, when the design features of the infrastructure or their
environment (urban/rural) change, it is advisable to create a new arc. The vertices
are placed consequently. Lastly, the graph can also adapt itself to the available
structure of traffic measurement. The limits separating two measurements series are
therefore likely to be regarded as vertices of the graph.

Optimally, an arc corresponds to a section of the road presenting relatively

homogenous design features and intensity of use over its entire length.

The arcs of the graph must render with precision the design features of the road
infrastructure, insofar as they directly influence the traffic speed. That relates
simultaneously to the structure and the quality of the roads. Among the elements that
may be noted let us cite the number, width, pattern, slope, sinuosity of lanes,
existence of a central reservation on dual carriageways, width of lateral clearances,
etc.

Within the framework of our applications these elements made it possible to

define 9 classes of infrastructure (Table 1.1). If the number of classes is not
restrictive, we postulate that the discrimination applied here is adequate to model
road networks on a regional scale with precision (Figure 1.3). Free speed is an
important input for the model. For a given vehicle type (here a car) this is the
maximum traffic speed technically allowed by each class of infrastructure within the
limits authorized by the Highway Code and without congestion. The variations
observed between French and British free speeds are due to differences in maximum
speeds authorized by the Highway Code and to differences in design features present
for the same infrastructure class. The geometry of roads in the French network often
has better quality than that of the British roads. Infrastructures are assigned to each
class by the operator on the basis of the technical information provided, in
particular, by roadmaps and road agencies.
 The Space-time Variability of Road Base Accessibility 5

Free speed

Class Design features of the infrastructure

Great
France
Britain

R1 Highway. 130 km/h 113 km/h

Intercity expressway: dual carriageway of the

R2 highway type and principal or regional 110 km/h 113 km/h
connection road with a separate roadway.

Urban expressway: highway or dual

R3 carriageway of the highway type with a 80 km/h 80 km/h
reduced shoulder (< 60 cm).

Trunk or regional connection road with 3 or 4

R4 90 km/h 96 km/h
lanes or 2 broad lanes, width: [7 m-9 m].

Trunk or regional connection road with 2

R5 80 km/h 80 km/h
“standard” lanes, width: [5 m-7 m].

Trunk or regional connection road and local

R6 service road with 1 or 2 narrow lanes, 70 km/h 49 km/h
width: ≤ 5 m.

Urban artery: penetrating or structuring by-

R7 50 km/h 49 km/h
pass of the urban road network.

R8 Urban boulevard: principal urban access road. 35 km/h 35 km/h

R9 Street: local urban access road. 35 km/h 35 km/h

Table 1.1. Infrastructure classes

6 Graphs and Networks

1.1.1.2. From graph to machine

Two types of files enable the electronic alphanumeric description of graphs.
Node files contain the name, code, population and geographical coordinates of
vertices. Arc files contain the codes of the vertices of origin, destination, class of
infrastructures to which the arc is attached, the length of the arc in kilometers and
the occupancy rate of the road lane or the hourly (or working day) rate needed when
taking congestion into account.

The principle of modeling makes it possible to work with orientated and two-
ways arcs simultaneously. The orientation applies to infrastructure classes. Since the
classes have at least one orientated arc (single direction of traffic, traffic data
differentiated by direction) all of their orientated arcs are in the database. In this case
a one-way road is described by an arc (i, j) and a two-way road by two distinct arcs
(i, j) and (j, i). The arcs of other classes are automatically regarded as orientated in
both directions and therefore they do not have to be computed twice in the database.

1.1.2. Congestion or suboptimal accessibility

In the case of small scale road modeling, the functioning of the system, which is
an essential factor in determining the accessibility levels, is often undifferentiated in
time as well as in space.

In this case traffic speeds are fixed regardless of the geographical location and
the time of day; we do not consider the effective operation of the infrastructure and,
thus, the possible local failures of the system to satisfy the mobility of people and
goods in the modeled zone. Taking into account the operating conditions of the road
network, i.e. congestion phenomena, appears less relevant. The use of average
speeds applied to the length of arcs in kilometers is an alternative that can be
adapted for the calculation of the durations of journeys, since for each pair “vehicle
type/infrastructure class” they are gauged with precision [CHA 97, CHA 98, CHA
00].

On the other hand, as the geographical scale increases, the omission of traffic
conditions becomes more problematic when road modeling is sought. For regional
applications, it is indeed necessary to evaluate real traffic speeds with more
precision, arc by arc, in close relation with localized system operation failures.
Finally, on a very large scale the analysis of vehicles movements at intersections
uses another category of methods known as microscopic.

Aiming our observations at a regional scale we have to measure the accessibility

of sites according to their geographical position, structure and quality of the
 The Space-time Variability of Road Base Accessibility 7

network, traffic codes and rules and vehicle types, as well as the congestion
phenomena, which, all things being equal, may imply a significant decrease in
accessibility.

Thus, the gains in accessibility that could be achieved by developing a fast

regional network may be partly cancelled out by congestion on its links or at its
intersections. We will therefore highlight situations of suboptimal accessibility, for
which taking into account real speeds by road segment modifies the accessibility,
which would be calculated on the basis of fixed average speeds.

If network failures were only accidental, taking them into account would no
longer be essential, since these congestion phenomena would not correspond to
network structural logics that could be subjected to modeling.

On the other hand, if the regional area under research manifests recurring traffic
problems with respect to space and time, we may seek to find out their impact on
accessibility. It would then be a question of modeling a progressive phenomenon, a
function of daily and hourly variations of network use (holiday periods, start of the
working week, rush hour during working days, etc.). We will thus admit the
relevance of a congestion function.

A short-term constant, i.e. the infrastructure capacity, is subjected to modulated

demand according to the stretch of road and time slots. The supply and demand ratio
measures the intensity of use of the infrastructure and, thus, its level of congestion.
To this end we have two types of demand variables: one measured in time (flow)
and the other in space (occupancy rate) [ABO 85], whose theoretical characteristics
and possible applications are different (Figure 1.1).

We will see how to introduce each type of variable into a specific congestion
modeling method adapted to accessibility calculations in a valuated graph.

Flow q is defined as the number of vehicles passing through the counting point
per unit of time. By extension, the obtained results are assigned to the road segment
delimited by two intersections.

To formalize: “let us consider a point x on a road and count the vehicles that
circulate there in one or both directions during the interval [t, t + τ∆]. Let us note by
n[t, t + τ∆] the number obtained. The flow at point x during the period [t, t + τ∆] is
by definition the number of vehicles passing by per unit of time, that is [LEU 93]:

q[t, t+∆t] = n[t, t+∆t] / ∆t” [1.1]

8 Graphs and Networks

Figure 1.1. Calculation of driving times

This type of data, which is currently used, poses a theoretical problem limiting
its use. The flow-speed curve (Figure 1.2) allows the same flow for two different
traffic speeds. A low flow may be interpreted either as a low traffic level (high
speed), or over saturation characterized by an extremely reduced flow (low speed).
 The Space-time Variability of Road Base Accessibility 9

Figure 1.2 shows in this respect that speed is a decreasing function of flow to the
critical point where capacity is reached. From this point the two variables decrease
simultaneously [COH 90]. Considered up to the critical point indicating
infrastructure saturation, the flow-speed ratio can be used both for affected and
observed flows.

Speed

Maximum
flow

Flow

Figure 1.2. Flow-speed curve

Once we have flows affected by a “traditional” traffic model [CER 90] as

emme/21 on a very complex graph, over saturated traffic states are avoided. Indeed,
in these sequential models with four stages (generation, distribution, modal split and
application) the last phase that consists of applying modeled traffic demand to the
routes proceeds by iterations of applications to the quickest paths between the
starting point and the destination. Consequently, once axis saturation is attained, the
ways not yet affected are used as alternative routes. The traffic conditions will thus
be saturated at most and the use of flows remains relevant.

On the other hand, when we have effective (observed) flows for a given time slot
it is impossible, as emphasized previously, to distinguish low traffic levels from over
saturation. In this case, good knowledge of traffic states of the analyzed network
makes it possible to locate the sections where a risk of over saturation exists and to
consequently adapt the flows. The ideal is, obviously, to get as much information
related to concentration (occupancy rate) as possible.

The occupancy rate of a road is also a specific measure, which can however be
easily linked to concentration, which is a variable of space. It is the part of the

1 Sequential traffic modeling software used, in particular, by highways agencies.

10 Graphs and Networks

measurement period during which a vehicle occupies the measurement point. More
formally: “at a moment t let us consider a stretch [x–∆x, x] of a road and count the
vehicles that are there. Let us note by n[x–∆x, x, t] the number obtained. The
concentration k[x–∆x, x, t] at the moment t on the segment [x–∆x, x] is by definition
the number of vehicles per unit of length [LEU 93], that is:

k[x–∆x, x, t] = n[x–∆x, x, t]/∆x” [1.2]

“Experiments with homogenous traffics show (…) that the speed of the flow
decreases as concentration grows. The assumption of the fundamental diagram
(statistical relation between flow and speed) is the postulate that there exists a
functional dependence u = u(k) between the flow speed and concentration” [LEU
93].

In this sense speed, which is the opposite function of occupancy rate, can be
calculated regardless of the present level of traffic, as opposed to what the use of
observed flow enables us. The use of occupancy rate is more relevant theoretically
and makes it possible to measure with precision road congestion and, in particular,
accessibility in cases of over saturation.

However, the method that we will endeavor to clarify hereafter allows the use of
two types of input variables, i.e. flow or occupancy rate, since the latter is seldom
available. This method is inspired by the Highway Capacity Manual2.

1.2. Integration of road congestion into accessibility calculations

1.2.1. Time slots

The amount of traffic in a given axis varies according to time. To evaluate the
real performance of a network we generally look at the period of its most intensive
use. Observable traffic points my arise by accident when they are the result of
particular occurrences or can be recurring when they are allotted to regular mass
movements, such as daily commutes (to or from employment centers), mass weekly
departures (during the weekend at the outskirts of large agglomerations) and mass
seasonal travel (in the outskirts of coastal towns and on intercity corridors).

2 The Highway Capacity Manual presents a method of road infrastructure capacity and
service level estimation. It does not propose legal standards, but rather represents the
commonly accepted American points of view summarized by the Transport Research Board
of Washington.
 The Space-time Variability of Road Base Accessibility 11

In urban or outlying environment it is common to carry out observations during

rush hour in the morning or in the evening, that is between 8 and 9 am and 5 and 6
pm respectively, even though variations may be recorded depending on the areas
studied [EPD 97]. Therefore, the occupancy rates and flows observed depend on the
choice of a particular time slot.

1.2.2. Evaluation of demand by occupancy rate

The estimation of traffic demand must take into account the various vehicle
types. When the occupancy rate is used this is automatically enabled “by an
electromagnetic loop placed under the road surface which makes it possible to
measure the occupancy rate, by recording the duration of electromagnetic contacts,
from which the duration of the real contacts is deduced, by taking into account the
influence of the circuit length” [LEU 93]. The number and length of vehicles are
thus associated to provide a reliable picture of spacing between vehicles responsible
for variation in traffic speeds.

Occupancy rates are measured over short time periods: from a few seconds for
networks equipped with dynamic regulation modes (real-time traffic management,
such as the Gertrude3 and Petrarque systems) to several minutes for networks
controlled by static devices. They can then be aggregated over an hour.

We then obtain the average percentage of the road occupied by the vehicles for a
given section of the road and a given time. 0% corresponds to zero traffic and 100%
to maximum occupation, literally bumper to bumper. Thus, occupancy rates
accurately account for the real level of traffic stretch by stretch and make it possible
to better take into account the localized problems.

Expressed as rates, they free the users from having to evaluate road capacity,
since they represent a supply and demand ratio. The measurement techniques used
provide information for each traffic direction, implying the use of an orientated
graph.

However, it is advisable to ensure an appropriate location of electromagnetic

sensors under the road. Indeed, placed near a vehicles stoppage area (ie: controlled
or uncontrolled intersection), these sensors may provide information unsuited to the
method used here. It is generally preferable to use several measurement points on
the stretch and to calculate an average or, failing that, a single judiciously positioned
measurement point [ABO 85, LES 92].

3 http://www.gertrude.fr.
12 Graphs and Networks

1.2.3. Evaluation of demand by flows

1.2.3.1. Flows
Flows measured or approached using a “traditional” traffic model are generally
expressed in vehicles per unit of time. However, finer models can differentiate
between cars, motorcycles and trucks. Moreover, if we consider their technical
differences (size, flexibility), the flow measurement would have to be homogenized.

Thus, the larger trucks occupy a greater area of the road. Moreover, their length
makes overtaking them more difficult. Taking into account their weight and the
speed limits imposed on them, these vehicles circulate at a lower speed, causing
blockages in the traffic flow. These become stronger if the road is sloped and
overtaking possibilities are limited. Even when a lane is available for overtaking, the
speed differential between the overtaking vehicle and the one being overtaken
causes a loss of space that could have been occupied in front of the vehicle being
overtaken. Finally, in case of jams4, trucks react more slowly to mode changes. The
same applies to buses, which may, moreover, slow down traffic during stops.

Since cars constitute the majority in the distribution of traffic volume, they are
often selected as standard. Consequently, the volume of trucks is represented in
automobile equivalent, such as 1 truck equals 2 cars, which is an average. Variations
are observable in reality, in particular, according to the gradient of the slope5.
Contrarily, motorcycles are considered less important than other vehicles. It is
generally retained that 1 motorcycle corresponds to 0.4 cars.

When these two standardized vehicles classes (trucks and motorcycles) are
added to the volume of cars we obtain for each section the total volume expressed in
car units (CU), which is then divided by the number of lanes and a unit of time
(CU/lane/h).

Depending on the availability of information, the flow may be differentiated

according to traffic directions, implying that the graph is directed and the value is
divided by the number of lanes in the direction considered. If it is not differentiated
(unorientation graph) we divide the value by the total number of lanes (both
directions together).

If the data consists of average annual daily flows (in CU/day), it is advisable to
apply the percentage of daily traffic circulating during the time period that we wish to
study to the value. This percentage may vary depending on the class of infrastructure

4 At traffic lights, roundabouts, stop or give way signs.

5 In this respect, a finer method is presented in the Transport Research Board, Highway
Capacity Manual.
 The Space-time Variability of Road Base Accessibility 13

to which the stretch belongs, which, in this particular case, is in fact an additional input
into the model. This calculation makes it possible to obtain flows in CU/h.

1.2.3.2. Capacity determining factors

First of all, it is necessary to isolate exogenous factors that are not taken into
account as they are not particularly explanatory of structural congestion problems.
For example, weather conditions can be responsible for variations [MAY 89].
Among the endogenous factors we may cite the frequency of bus stops and
intersections with and without traffic lights.

If the intersections are controlled by traffic lights, the fact of allocating a fraction
of time per movement phase (passage) involves considerable capacity variations,
especially as a safety lag must be observed between two phases and the reaction of
users is not instantaneous.

Moreover, certain endogenous factors may prove to be not easily quantifiable, on

the one hand, because they are very subjective, such as the differences in behavior
between commuters and occasional road users and on the other hand because they
refer to incidental occurrences, such as traffic accidents or roadworks.

Two major factors determine capacity. Firstly, the maximum speed authorized
by the Highway Code in the sense that for identical roads a higher authorized speed
can enable greater traffic flow. In addition, the number of lanes has the same effect
because, all things being equal, it can be estimated that the more the number of lanes
increases, the more the infrastructure capacity increases. Indeed, the multiplication
of lanes makes it possible for several vehicles to overtake simultaneously at the
same point.

In addition to legal speed limits and the number of lanes, technical features of the
infrastructure are major elements for an efficient capacity evaluation. Indeed, the
capacity of two roads with the same number of lanes may differ, since they exhibit
different cross-sections (width of lanes, existence of a central dual carriageway
separator, distances between the lanes and obstacles on the sides, etc.) Let us
consider, for example, a reduction of lateral space. It is generally noted that users
tend to compensate for this reduction by increasing the distance with the car in front,
which contributes to reducing supply capacity [CHAU 97].

Taking these design features into account in the evaluation of theoretical

capacities can be done in two manners: either by applying reduction coefficients
specific to the optimal theoretical capacity of the infrastructure category considered
[TRB 98], which are coefficients that take into account technical, weather or
behavioral factors, or, as we described previously, by adapting the principles of
14 Graphs and Networks

graph modeling to a fine description of the road network in order to distinguish the
various types of roads and the various traffic conditions that can be encountered.

At this time in the development of the method it is advisable to take into account
the scale, on which the analysis is carried out. Indeed, on a local scale, the most
exhaustive possible consideration of microgeographical elements that have just been
mentioned proves essential but within the framework of an analysis carried out on a
regional scale it is different. In the latter case only the most important elements have
to be retained for the obvious reasons of cost-efficiency-precision of modeling.

1.2.3.3. Capacity
Capacity may be interpreted as the maximum observed flow. We will discuss
theoretical capacity, which is intrinsic to supply (number and width of lanes, linear
or singular installations, access conditions), corresponding to optimum usage
conditions. We will also suppose that flows indicate “downstream fluid flow, a
sufficient upstream demand to approach capacity [and] vehicles sharing space
homogenously” [LEU 93].

Capacity in
Infrastructure class
CU/lane/h

Highway (R1)

Intercity expressway (R2) 1,800

Urban expressway (R3)

Trunk or regional connection road with 3 or 4 lanes or 2 broad lanes

(R4)
1,250
Urban artery (R7) and urban boulevard (R8)

Principal or regional connection road with 2 standard lanes (R5)

Principal or regional connection road and local service road with 1

1,000
or 2 narrow lanes (R6)

Street (R9)

Table 1.2. Capacity of the road infrastructures

 The Space-time Variability of Road Base Accessibility 15

Thus, each of the 9 road classes used for the road modeling network is associated
with a capacity provided by the infrastructure engineering departments (Table 1.2).
In cities the capacity may be evaluated in two ways: either by using an “intersection
by intersection” modeling logic, which requires integrating average waiting times at
each crossroad into the calculation of movement time, or by considering the most
important road sections (from 1 to 3 km) in order to obtain an average capacity, by
taking into account crossings of intersections controlled or uncontrolled by traffic
lights.

The temporal constraint of the latter will thus be expressed by a loss of capacity
leading to reduced speeds and therefore driving times. The meso-geographical
approach that we choose encourages us to adopt the second approach, the first being
preferentially applied on a microgeographical scale.

1.2.3.4. The flow/capacity ratio

Having established the flow and the capacity (in CU/lane/h) for each section of
the road network for a given time slot, it is then advisable to calculate the
flow/capacity ratio necessary for the section by section evaluation of traffic
conditions.

1.2.4. Calculation of driving times

To evaluate traffic conditions we use the classification by level of service (LOS)

developed in the Highway Capacity Manual on the basis of flow or occupancy rate
criteria (Table 1.3) [TRB 98, LEU 93]. For each stretch the level of service is given
by the occupancy rate value or by the flow/capacity ratio value (Tables 1.4 to 1.6).
The correspondence between the level of service and occupancy rate or
flow/capacity ratio varies according to the infrastructure class to which the stretch
belongs. A speed reduction coefficient defined as the ratio between the sought real
traffic speed and free speed (Tables 1.4 to 1.6) corresponds to each LOS and each
infrastructure class. Thus, for each stretch the real speed is obtained by multiplying
this coefficient by the free speed of the infrastructure class to which the stretch
belongs:

Real speed = free speed × speed reduction coefficient

Then, by dividing the length of the stretch in kilometers by the real speed we
obtain the time needed to traverse it:

Driving time = length of the section/real speed

16 Graphs and Networks

The components of the method presented above were integrated into the NOD6
software which was conceived to evaluate the projects aiming to modify transport
supply.

LOS Traffic Description of traffic conditions

Traffic conditions are ideal. Users tend to move at

LOS A
optimal speed without inconveniencing each other.

Fluid
Traffic is fluid. Overtaking, however, may prove
LOS B difficult and the observed speed is close to the authorized
limit.

Traffic is busy, maneuvers become difficult. Speed starts

LOS C Busy
to drop.

Traffic becomes denser. Lane changes are very difficult

LOS D
and cause considerable deceleration.

Dense
Traffic is strongly slowed down, flow becomes saturated.
LOS E Traffic conditions are very unstable, even a minor
incident leads to LOS F.

LOS F Saturated
Capacity is insufficient, backlog phenomena occur.
Traffic is slowed down or even stationary. The flow is no
Over longer continuous.
LOS G
saturated

Table 1.3. Description of the level of service

6 Chapelon L., L’Hostis A., Mathis P., 1993-2002.

 The Space-time Variability of Road Base Accessibility 17

Levels of service for highways (R1), intercity expressways (R2)

and urban expressways (R3)

Occupancy rate Coefficient of free

Flow/capacity
(%) speed reduction

LOS A [0 – 4] [0 – 0.318] 0.86

LOS B [4 – 7] [0.318 – 0.509] 0.81

LOS C [7 – 11] [0.509 – 0.747] 0.77

LOS D [11 – 16] [0.747 – 0.916] 0.66

LOS E [16 – 25] [0.916 – 1] 0.43

LOS F [25 – 40] 0.33

LOS G > 40 0.21

Table 1.4. Table of the level of service for highways and expressways

Levels of service for roads (R4), (R5) and (R6)

Occupancy rate Coefficient of free

Flow/capacity
(%) speed reduction

LOS A [0 – 4] [0 – 0.25] 0.92

LOS B [4 – 7] [0.25 – 0.4] 0.83

LOS C [7 – 11] [0.4 – 0.6] 0.75

LOS D [11 – 16] [0.6 – 0.8] 0.67

LOS E [16 – 27] [0.8 – 1] 0.54

LOS F [27 – 40] 0.33

LOS G > 40 0.21

Table 1.5. Table of the level of service for interurban roads

18 Graphs and Networks

Levels of service for urban arteries (R7), urban boulevards (R8)

and streets (R9)

Occupancy rate Coefficient of free

Flow/capacity
(%) speed reduction

LOS A [0 – 4] [0 – 0.25] 0.9

LOS B [4 – 7] [0.25 – 0.49] 0.7

LOS C [7 – 12] [0.49 – 0.6] 0.5

LOS D [12 – 20] [0.6 – 0.8] 0.4

LOS E [20 – 33] [0.8 – 1] 0.3

LOS F [33 – 50] 0.24

LOS G > 50 0.15

Table 1.6. Table of the level of service for streets

Examples of applications
Use of the occupancy rates
Let us take the example of a 5 km long stretch of an interurban highway with 3
lanes each way. This stretch belongs to the R1 class (highway). In France the free
speed is 130 km/h. What is the real speed between 5 and 6 pm?
– determination of the free speed reduction coefficient: the information
contained in the data file used in model input provides us with the average
occupancy rate of each stretch by time period, distinguishing between traffic
directions (directed graph). For the stretch that is of interests, i.e. between 5 and 6
pm, this rate is 15% in the north-south direction and 10% in the south-north
direction. With an occupancy rate of 15% the north-south direction forms part of the
[11 – 16] class of the table of level of service for highway infrastructure (R1). The
stretch exhibits dense traffic conditions (LOS D), with which a free speed reduction
coefficient of 0.66 is associated. The occupancy rate of the south-north direction
corresponds to the [7 – 11] class in this same table, that is, busy traffic (LOS C). In
this direction the free speed reduction coefficient equals 0.77;
– calculation of the real traffic speed: the real traffic speed is obtained by
multiplying the free speed by the reduction coefficient calculated above. The real
 The Space-time Variability of Road Base Accessibility 19

peed is thus 86 km/h in the north-south direction (130 × 0.66) and 100 km/h in the
south-north direction (130 × 0.77);
– calculation of driving time for the stretch: the time needed to drive the entire
length of the stretch is about 3 and a half minutes in the north-south direction (5 ×
60/86) and 3 minutes in the south-north direction (5 × 60/100).

Use of flows
Let us take the example of a 4 km long highway stretch with 3 lanes in both
directions located in a residential area with a lateral clearance less than 60 cm. This
stretch belongs to the R3 class (urban expressway). In France, as in Great Britain,
free speed is 80 km/h and the average annual daily traffic is of 100,000 vehicles
(including 10% of trucks) in both traffic directions. What is the time needed to
traverse the stretch between 5 and 6 pm?
– differentiation and homogenization of traffic: (100,000 × 10)/100 = 10,000
trucks circulate. Moreover, we consider 1 truck ⇔ 2 cars. Upon standardization this
represents 20,000 trucks (in equivalent cars or CU) and 90,000 cars (100,000 –
10,000). In total 110,000 CU circulate, that is, 18,333 CU/lane/day (110,000/6);
– calculation of hourly rate: with respect to urban expressways we could observe
that 9.1% of the total daily volume circulate during the height of the evening rush
hour, namely 5-6 pm. The theoretical flow per lane for this time slot is thus 1,668
CU (18,333 × 9.1/100);
– determination of the free speed reduction coefficient: a theoretical capacity of
1,800 CU/lane/h is associated with “urban expressway” infrastructure class.
Consequently, the flow/capacity ratio is equal to 0.93 (1,668/1,800). The
expressway is nearly saturated and therefore tends to operate in LOS E. From Table
1.4 we deduce that the free speed reduction coefficient is equal to 0.43;
– determination of the real traffic speed: the real traffic speed is thus
34 km/h (80 × 0.43). This speed applies to both traffic directions, since the initial
data does not make it possible to make a distinction;
– calculation of the driving time for the stretch: the time necessary to drive the
entire length of the stretch is about 7 minutes (4 × 60/34) in each direction.

1.3. Accessibility in the Thames estuary

London is the centre of the most populated urban area in Europe, i.e. the South-
East of England. The geographical area studied here relates more particularly to the
enlarged East Thames Corridor which is the eastern half of this area. The latter
includes the zones bordering the Thames estuary, namely, Essex, Kent, Greater
20 Graphs and Networks

London, but also Cambridgeshire, Hertfordshire, Sussex and Surrey. These counties
have been affected by profound spatial changes that have upset the distribution of
settlements during the last 50 years.

The processes of suburbanization and counterurbanization, which are at work

throughout the entire East Thames Corridor since the interwar period, have indeed
resulted in strong space dilution of activities and residences beyond the green belt
surrounding London in a radius of 30 to 40 km from the city centre [HALL 89]. The
large number of commuters and the lengthening of daily travel distances lead to
heavy traffic regardless of the distance from the centre of London. An analysis of
the distribution of average annual daily flows shows peak traffic on the M25
(200,000 vehicles/day), which is the ring road that encircles London at a distance of
25 km from the centre.

The road network, although improved over the last 20 years (M20, M25), is not
always able to support such amounts of traffic. This brings about high levels of
congestion. In addition to intensity, congestion is characterized by its large extent,
since recurrent traffic difficulties are observable within distances of more than 50
km from central London [IAU 98].

The graph of East London conceived by using MAP©7 software comprises 891
vertices and 1,321 arcs (Figure 1.3 in the color plates section). It corresponds to the
state of the road network in 1998 as described on the Michelin map number 404.
The graph contains the first eight infrastructure classes described in Table 1.1. The
free speeds used are presented in the same figure.

The arcs of the graph are not orientated since traffic measurements provided by
the Department of the Environment, Transport and the Regions (DETR) correspond
to average annual daily flows from 1997.

Hourly rates are obtained by using load diagrams of DETR [DEP 97]. The
proportions of daily traffic circulating during the two time periods used in this study
are summarized (Table 1.7).

7 Alain L’Hostis, Philippe Mathis, 1993-2002.

 The Space-time Variability of Road Base Accessibility 21

Infrastructure classes/time slots 1-2 pm 5-6 pm

Motorway (R1)
5.6% 8.9%
Intercity expressway (R2)

Urban expressway (R3)

Urban artery (R7) 5.7% 8.5%
Urban boulevard (R8)

Trunk or regional road with 3 or 4 lanes or 2

broad lanes (R4)
Trunk or regional road with 2 standard lanes
5.8% 8.6%
(R5)
Trunk or regional road and local service road
with 1 or 2 narrow lanes (R6)

Table 1.7. Proportion of daily traffic during the 1-2 pm and 5-6 pm intervals

The flows in CU/h obtained using these percentages are then divided by the total
number of lanes in the section (both directions together) in order to obtain traffic
data in CU/lane/h.

The Thames ferry crossing between Woolwich and Silvertown (Greater London)
is added to the description of the road network. The crossing’s modeling used
different principles than those described previously. Indeed, congestion does not
directly influence the crossing time (four minutes). On the other hand, the time spent
in waiting areas on both banks can vary strongly depending on the time of day. We
estimated this time at 15 minutes for the 1-2 pm period and 30 minutes for the 5-6
pm period. The duration of the journey (wait + crossing) is directly associated with
the corresponding arc.

1.3.1. Overall accessibility during the evening rush hour (5-6 pm)

Among accessibility indicators the sum of minimal driving times between each
node and all the others is one of the most synthetic:

n
t r (i ) = ∑ t r (i, j ) [1.3]
j =1
22 Graphs and Networks

where:
– i is the index of the node of origin;
– j is the index of the node of destination;
– tr(i, j) is the minimal driving time through the network between i and j;
– tr(i) is the sum of minimal driving times through the network between i and
every other city.

The interpretation of results must take into account the four principal explanatory
factors that affect this indicator simultaneously, namely: geographical position of the
nodes, road network structure, infrastructure quality and congestion.

This indicator highlights the accessibility structure for the selected time slot (5-6
pm). The central part of the estuary and notably the right bank benefit from the best
accessibility (Figure 1.4 in the color plates section). The centre-periphery ratio is
high, ever so slightly higher than 2.5. The A2/M25 interchange is the central point
of the structure and on the contrary Frinton, in Essex, is the least accessible place.

The class structure of the most accessible nodes, around the A2/M25
interchange, is characterized by a diffusion extended to the east. On the other hand,
it is strongly limited to the west by the congestion in the densely built-up area. The
relatively good functioning of the fast network on the right bank makes it possible to
maintain the same structure as far as central Kent. Conversely, the traffic difficulties
in London and on the M25 (Dartford) tend to sharply reduce accessibility to the west
and, to a lesser extent, in the left bank corridors. More generally, the progressive
character of the diffusion structures in the east tends to be replaced in the west, at
least in the agglomeration, by a juxtaposition of highly spatially restricted classes.
Moreover, accessibility is diffused more homogenously in the semi-rural territories
of Essex than in Sussex, since the latter has more irregular topography (inside of the
North and South Downs cuestas), which often leads to a more sinuous infrastructure.

The weight of the geographical position of the nodes is the dominant explanatory
factor of the accessibility levels observed in Figure 1.4 (see color plates section).
The choice of the space studied is not neutral. This is particularly clear for London
and the coastal cities. This edge effect precludes a good legibility of the structure, of
the quality and operation of the axes connecting the peripheral nodes. It is possible
to eliminate it and to isolate the sole network performance by calculating a specific
accessibility indicator.
 The Space-time Variability of Road Base Accessibility 23

1.3.2. Performance of the road network between 1 and 2 pm and 5 and 6 pm

The network performance indicator for each network node is the ratio between
the sum of the direct distances to all of the other nodes and the sum of minimal
driving times to these nodes:

n
∑ le ( i, j )
j =1
ve ( i ) = n × 60 [1.4]
∑ tr ( i , j )
j =1

where:
– i is the index of the node of origin;
– j is the index of the node of destination;
– le(i, j) is the Euclidean length separating i from j;
– tr(i, j) is the minimal driving time through the network between i and j;
– ve(i) expressed in km/h is an indicator representing the influence of network
performance on the accessibility of node i.

The value thus obtained places all the nodes on an equal footing from the point
of view of their geographical localization. The minimal driving time present in the
denominator is a function of the geographical position of the cities, structure, quality
and operation of the network. As a result, the Euclidean length in the numerator
erases the influence of the geographical position without deteriorating the impact of
network performance.

1.3.3. Network performance between 1 and 2 pm

The general structure of network performance during off-peak hours is clearly of

the “corridor” type. The best performances follow the layout of the expressways
(Figure 1.5 in the color plates section). The variations in accessibility exhibit a ratio
of 1.7 between the two extremes. The M11 interchanges show the best
performances, whereas the nodes of central London, such as Hyde Park Corner,
suffer from the worst performances in the network.

The London agglomeration encircled by the M25 ring-road presents the most
heterogenous picture. The strong proportion of urban boulevards, which are
24 Graphs and Networks

generally saturated, reduces average velocities. This is not only true for central
London, but also for the south of the agglomeration.

The borough of Croydon is thus exceptionally inaccessible by comparison with

others boroughs of Outer London. The A23 that connects it to London and the urban
roads (A232, A212) that connect it to other urban centers in the southern suburbs are
of poor quality and present a saturated operation.

The western end of the estuary, in turn, benefits from a much better accessibility.
This is even better to the south of the Thames where the A2 and A20 expressways
that are lightly congested during the off-peak hours maintain a “corridor” effect. The
Blackwall tunnel, even when congested, makes it possible to maintain relatively
good accessibility to the Greenwich and Stratford boroughs. The A406 inner ring
also makes it possible to diffuse the general network performances in a narrow
corridor in the north-east of the agglomeration.

Finally, the M25 does not seem to play a major part in diffusing accessibility in
the agglomeration. Its good performances are limited by the congestion of the urban
road connectors leading to it. On the other hand, it is on the external side where its
performances and its relative good functioning are truly diffused through the radials
and thus benefit a larger number of cities [LIN 91]. The nodes located at both ends
of the M25 should be more affected mechanically than those positioned halfway
through its course. This is verified for interchanges 6 and 8. On the other hand, the
interchanges 25, 26 and 27 suffer from the effects of congestion of the north-eastern
quadrant of the M25.

In the rest of the metropolitan area, “rural” East Sussex and the estuary
peninsulas (Essex) remain disadvantaged by the lack of fast infrastructures. For
Southend-on-Sea, which in a “dead-end” position in the network, the problem is
exacerbated by the congestion of the axes connecting it to the main network.

This indicator also proves relevant to demonstrate the strengthening of “tunnel

effects”. While the Eastern Kent network is performing well and is only lightly
congested maintaining good general accessibility, the towns of Rochester and
Maidstone suffer from the congestion of the axes connecting them to the main
network (terminal connections) that is still doing well. Finally, the interchanges of
the axes connecting the exterior of the agglomeration maintain a relatively good
performance, in particular for those furthest away from London and the M25.
 The Space-time Variability of Road Base Accessibility 25

1.3.4. Network performance between 5 and 6 pm

The overall accessibility structure for this time slot is more difficult to detect
than for the previous time slot (Figure 1.6 in the color plates section). The highways
and expressways that are, generally, the principal structuring elements no longer
exhibit the distribution of the accessibility levels so clearly.

The distribution of the latter yields a ratio of 1.9 between the most advantaged
node (Dover) and the worst off node (Hyde Park Corner). The London
agglomeration brings together the strongest contrasts. The saturation of the urban
boulevards which prevails in the city centre is responsible for the poor performances
detected of the central London nodes. This is in no way exceptional, nor
preoccupying, since, actually, only 14% of journeys in this area are taken by cars.
However, it is rare to find these low accessibility levels 18 km away from the city
centre. Thus, Croydon and Brixton have a score identical to Westminster. The
saturation of the South London axes, urban arteries and urban boulevards implies the
constitution of a corridor with extremely poor accessibility.

This is reinforced by the saturation of the M25, which cannot offer a proper level
of functionality, in particular, for routes undertaken from the south of the
agglomeration, as opposed to what is observed during off-peak hours. It is also the
case for nodes located in North London. Indeed, the congestion of the M25 and the
urban sections of the M11 deteriorates the performances of the Northern corridor,
which is, however, less congested. Nonetheless, since only the eastern part of the
London area has been adopted as the subject of this study, we have disadvantaged
the accessibility of this Northern corridor. Should we consider that the M25 is
saturated over its entire length, then the longer the stretch driven by the motorist, the
poorer the calculated performance will be and vice versa. We may thus explain the
relatively good performances of the nodes located along the M25 from Dartford to
the M25/M20 interchange. Indeed, within the selected area the nodes in question are
in the middle of the course of the M25. Its use will thus not be as systematic as for
the routes stemming from the north or south corridors.

The nodes located near the A12 and A10 owe their relatively poor accessibility
to the combination of delays on these two axes, as well as on the M25. Moreover,
the poor performance of the nodes located on the estuary peninsulas is explained by
the congestion of the Essex axes and by the sinuosity of the infrastructure used to
reach it. Southend-on-Sea is the one town, aside from London, which benefits the
least from network performance. The rural areas of inland Sussex are also at a
disadvantage due to the sinuosity and the spatial patterns of the routes, which lead to
long detours of the optimal paths, taking them away from the path as the crow flies.
26 Graphs and Networks

The eastern end of Kent appears to be an exception. Not only accessibility is

greater there, but it is still structured in corridors. The highest values follow the
course of the A299 (Faversham-Margate), A2 (Canterbury-Dover) and A20/M20
(Ashford-Folkestone). The remote location and the isolated nature of the county also
limit the traffic volumes in this part of England.

By comparing the two time slots (1-2 pm and 5-6 pm), we realize that the most
significant losses are found in, or in the vicinity, of the axes of the intercity London
suburb. This is verified inside as well as outside of the M25. The latter, whose
interchanges suffer from heavy losses, are once again poised to play a major part.
Nodes that are most likely to be used are the main victims. This is all the more true
as the axes that connect the M25 are already congested. It is the case, in particular,
in the North-East and the South. Conversely, the nodes of the agglomeration are
definitely less concerned. Since traffic Speeds are already low during off-peak
hours, the fall is somewhat attenuated during peak hours by comparison with the rest
of the metropolitan area. Similarly, the rural zones, which are often more remote, are
barely affected. It is particularly clear for South Sussex, East Kent and the cities and
villages at the “dead-end” of the estuary.

Taking into account the flows during peak hours highlights real accessibility
levels when network operations are the least effective. We note, by proceeding in a
similar manner, that it is primarily the inter-urban roads that suffers most from
accessibility losses, which is not exceptional. On the other hand, the scale and the
location of these losses are particular to London. Indeed, not all cities can complain
about accessibility reductions of more than 30% in a band ranging from 25 to 65 km
from the main urban centre!

1.3.5. Evolution of network performances related to the Lower Thames Crossing

(LTC) project

Figure 1.7 (see color plates section) highlights the importance and location of the
performance gains related to the construction of the Lower Thames Crossing. The
graph comprises 892 vertices and 1,327 arcs. The gains are obtained by absolute
difference between the nodal results with respect to two distinct states of transport
supply. On the one hand, the initial network dated in 2005, when the recalibrated
A13 and A2 will be operational and, on the other hand, the final network including
the Lower Thames Crossing (LTC) envisaged by 2015 [GLA 01].

This new Thames crossing comprises a 1.5 km bridge connected to the A13 to
the North and the A2/M2 to the South by expressway stretches (R2) with 2 × 2 lanes
for a total of 14 kilometers. Two arcs are added to the initial graph: one between the
M2 J1 interchange and Shorne and the other between Shorne and the A13/A1044
 The Space-time Variability of Road Base Accessibility 27

interchange. The traffic at the two sections should reach 1,500 CU/h in both
directions, that is, 375 CU/lane/h [DEP 93]. For the remainder of the network we
preserve the 1997 flows for lack of available reliable data of prospective nature. The
bridge has a triple objective. First of all, it is intended to improve connections
between the two economic centers of the estuary, namely, the Southend-Basildon-
Wickford conurbation in the North and the Medway Towns in the South. It is also an
issue of reducing the isolation of the peninsulas in the North of the estuary and
seeking an alternative route to the congested M25 motorway especially around the
Dartford area. According to the simulation, the gains observed overall are high and
diffused. This is explained by the relative position of the bridge in the estuary. Built
10 km away from to the East of Dartford it would be the first fixed link bridging the
two banks of the Thames. Moreover, the performance of the fast networks in Essex
(A12, A127, A13) and Kent (A2/m2, M20, A229) would make it possible to reach a
vast hinterland. Lastly, these gains are also the result of a decrease in detours. The
obligatory passage by Dartford, which is imposed by the configuration of the
network, would thus be cancelled for the North-East to South-East routes.

The distribution of gains is paradoxically asymmetric. The left bank benefits

more from the new infrastructure. The Basildon area, in particular, benefits from
particularly improved performances, whereas there is no equivalent on the right
bank. This is explained by the difference in the development of routes between the
two banks. The northern bank, and more particularly the area of Southend, has
suffered more from detours imposed by the network structure, since the axes
connecting it and thus the minimal paths following them currently tend to veer away
more from straight lines. This is in particular the case of the routes bound for the
South of the estuary. On the other hand, by comparison, on the southern bank the
orientation pattern of the A2/M2 and the M20 limits the detours for the routes
toward the northern half and thus, the impact is less strong.

The routes between Kent and Essex could thus be partly taken over by the Lower
Thames Crossing. Similarly, the relatively high gains recorded in the north-eastern
quarter of the London agglomeration show that the A406 and A13 corridors are
likely to drain optimal routes to or from Kent via the Lower Thames Crossing
instead of using the M25 from Dartford. The stretch of the A13 between the M25
and the LTC, which is less congested than its counterparts on the right bank, the
A2/M2, could perhaps support this concentration. More certainly, the M25 would be
relieved at the Dartford passage. Another consequence of the LTC is the
improvement of the link between the two main urban centers located on either side
of the bridge; Rochester and Southend would benefit from an improvement of 13
and 25% of their accessibility.
28 Graphs and Networks

The space-time evaluation of networks benefit from a new dimension when the
real infrastructure operation is taken into account. Speed losses are not distributed
homogenously and congestion seems to differentiate network space. We could thus
represent the temporal dilution of urban and intercity areas of London. In this
context even a minor modification of road supply may result in a proportionally
more significant impact on the network performance. Thus, a simple infrastructure
modification may involve an upheaval of the optimal routes because of the
heterogenous distribution of congestion levels. However, planners need to know the
consequences of the investments which they prepare to carry out, because the
functioning and the balance of the whole network are concerned. The principles of
modeling and the tools presented in this chapter clearly fit into this perspective of
assisting decision-making in regional planning.

1.4. Bibliography

[ABO 85] ABOURS S., Estimation des temps de parcours sur un axe urbain à partir de taux
d’occupation, no. 76, Arcueil INRETS, 1985.
[APP 99] APPERT M., Intégration de la congestion dans l’évaluation spatio-temporelle d’un
réseau routier régional: le cas de l’estuaire de la Tamise, Thesis, Montpellier 3, 1999.
[APP 02] APPERT M., CHAPELON L., Planification des transports régionaux en
Languedoc-Roussillon et Nord-Pas-de-Calais: évaluation de la concurrence rail-route,
Research Program INRETS/GRRT, Maison de la Géographie, Montpellier 2002.
[AUR 94] AURAY J.-P., MATHIS P., “Analyse spatiale et théorie des graphes”,
Encyclopédie d’économie spatiale, Economica, Paris, 1994.
[BER 83] BERGE C., Graphes, Gauthier-Villars, Paris, 1983.
[CER 90] CERTU, Les études de prévision de trafic en milieu urbain: guide technique,
CERTU, Lyon, 1990.
[CHA 97] CHAPELON L., Offre de transport et aménagement du territoire: évaluation
spatio-temporelle des projets de modification de l’offre par modélisation multi-échelles
des systèmes de transport, PhD Thesis, Tours University, 1997.
[CHA 98] CHAPELON L., “Evaluation des projets autoroutiers: vers une plus grande
complémentarité des indicateurs d’accessibilité”, Les cahiers scientifiques du transport,
no. 33, p. 11-40, 1998.
[CHA 00a] CHAPELON L., “Accessibilité routière et périphéricité des villes atlantiques”,
Les cités atlantiques, Publisud, Paris, p. 139-170, 2000.
[CHA 00b] CHAPELON L., Transports et énergie, Montpellier – Paris, RECLUS, La
Documentation Française, Atlas de France Collection, 2000.
[CHAU 97] CHAUVIN et al., “Une expérimentation de réduction de profil en travers sur
autoroute urbaine”, RGRA, no. 756, 1997.
 The Space-time Variability of Road Base Accessibility 29

[COH 90] COHEN S., Ingénierie du trafic routier, Presses de l’ENPC, Paris, 1990.
[DEP 93] DEPARTMENT OF THE ENVIRONMENT, Transport and the Regions, River
crossings to the east of Tower Bridge, Report of the British Ministry of the Environment,
Transport and the Regions, 1993.
[DEP 97] DEPARTMENT OF THE ENVIRONMENT, Transport and the Regions, London
traffic monitoring report, Report of the British Ministry of the Environment, Transport
and the Regions, 1997.
[DUP 96] DUPUY G., Les territoires de l’automobile, Economica, Paris, 1996.
[DUP 99] DUPUY G., La dépendance automobile, Economica, Paris, 1999.
[FAY 96] FAYARD A., “La congestion du réseau routier de rase campagne”, Revue
d’économie régionale et urbaine, no. 1, p. 41-52, 1996.
[GLA 01] GREATER LONDON AUTHORITY, Transport strategy (transport development
plan for the city of London), 2001.
[HAL 89] HALL P., London 2001, Unwin Hayman, London, 1989.
[IAU 98] IAURIF, Paris, London, New York, Tokyo: comparaison des systèmes de transport
de quatre métropoles, IAURIF, Paris, 1998.
[LES 92] LESORT J.-B., Étude d’indicateurs de circulation en milieu urbain, INRETS report,
no. 142, INRETS, Arcueil 1992.
[LES 93] LESORT J.-B., “Théorie simplifiée des flots de trafic”, Modélisation du trafic,
INRETS, Arcueil, p. 141-163, 1993.
[LEU 93] LEURENT F., Proposition d’indicateurs macroéconomiques des conditions de
circulation sur un réseau routier, INRETS report, Arcueil, 1993.
[LIN 91] LINNEKER B.J., SPENCE N.A., “An accessibility analysis of the impact of the
M25 London orbital motorway on Britain”, Regional studies, vol. 26, p. 31-47, 1991.
[MAT 93] MATHIS P., L’effet paradoxal des transports à grande vitesse, DATAR/Editions
de l’Aube, Paris, 1993.
[MAY 89] MAY A.D., BONSALL P.W., MARLER N.W., Travel time variability of a group
of car commuters in North London, Working papers, 1989.
[MER 84] MERLIN P., La planification des transports urbains: enjeux et méthodes, Masson,
Paris, 1984.
[MOG 90] MOGRIDGE M., Travel in towns: jam yesterday, jam today and jam tomorrow?,
MacMillan, London 1990.
[TRB 98] TRANSPORT RESEARCH BOARD, Highway capacity manual, 2nd ed., TRB,
Washington, 1998.
 Chapter 2

Journey Simulation of a Movement

on a Double Scale

An increasing number of urban dwellers choose to spend a few hours of

relaxation in the countryside. Over the past years we could observe an increase in
leisure and recreation activities around cities [DEW 90, DEW 92, DEW 93]. This
phenomenon of visiting natural spaces is scarcely studied, in the sense that there are
certain difficulties in determining the movement conditions of these visitors [TOU
98]. Indeed, if certain sites are frequented more than others, then necessarily they are
more attractive and able to captivate more visitors than nearby sites. This
phenomenon thus poses questions to the researcher on the nature of the
attractiveness of a site. These questions are central to our work: should it be
attributed to the activities on offer, to its generally remote location in a region, to its
natural environment or its fame?

If there is movement, there is a goal. This goal is to take a walk in “nature”, to

relax in the spare time. By recreational visits to a natural environment we mean a
process comprising two distinct displacements: access to the site and pedestrian
diffusion in the environment with the aim to enjoy the entertaining nature of the
activities that the visitor seeks at the site. Consequently, there arises the central
question of taking into account the space and the choices of the visiting individuals
related to the network, i.e. the network of sites structuring the area is likely to
determine induced forms of distribution and diffusion in the natural environments
which are open to public visitation.

Chapter written by Fabrice DECOUPIGNY.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
32 Graphs and Networks

2.1. Visitors and natural environments: multiscale movement

2.1.1. Leisure and consumption of natural environments

For many places of interest a large part of the visits arises from local
entertainment usage [CUV 98, IFE 00]. So-called tourist infrastructures are
increasingly used by the local population and in particular by urban dwellers [PL
98]. The study of visits carried out in the Hautes-Vosges [DEC 97] has revealed
large numbers of local or regional visitors1 with rates that can attain 75% on some
Sundays. In short, we noted that visits double on Sundays over the entire massif2
and that they correspond to rush hour phenomena during the day, with the peak of
visits occurring between 4 and 5 pm. Therefore, throughout this chapter, we prefer
the term visitor to this of tourist3 or excursionist. As a more neutral concept, first of
all, it enables us not to presuppose a segmentation of the visitor population that
could introduce a bias into the analysis by taking into account that a tourist a priori
exhibits different on-site behavior than an excursionist. We will therefore assume
that there is a visitor population that is homogenous not in its social practices
(choice of types of natural environments – sea, mountains, countryside – and
recreational activities), but in its behavior in discovering a natural environment (type
of movement and diffusion over the area), moving away from a place of residence
towards one or more sites offering natural environments.

Before carrying on, it is necessary to identify the two constitutive stages of

movement: choice and visit. The former involves socio-cultural conditions: why do
people choose the mountain over the sea, one activity rather than another? Why do
people leave for a period between one and four weeks, a weekend or just one
Sunday [CAZ 92]? The practices related to leisure involve a decision taken
individually. Although it is not our intention to analyze social practices, we believe
that it is important to determine from them the observable behaviors (the manner in
which the visitor populations are diffused). The very first observable behavior
manifests itself through the use of cars as movement vectors between the place of
residence and natural environments [CHA 99].

1 Bordering departments and/or regions.

2 These observations are also verified in other areas (Ecrins, Pyrenees, Pont du Gard, PNR
Brotonne).
3 For the OMT a tourist is a person who travels for leisure and spends at least a night at the
site.
 Journey Simulation of a Movement on a Double Scale 33

2.1.2. Double movement on two distinct scales

Behaviors may thus appear varied in the choice of visited natural sites, but they
are always made according to three principal characteristics of the natural
environment: protected nature that offers relaxation, managed nature that has been
made safe and pedestrian accessibility to natural attractions from the parking spots
[ENV 92]. The main difficulty in evaluating visitor displacements consists of
integrating a double scale of movements according to the behaviors linked to the
three factors of site visiting (safety, accessibility, protection). This double scale
depends on two movement processes: one or more displacements by car through a
network followed by one or more pedestrian displacements at the natural sites.

2.1.3. Movement by car

The first movement scale corresponds to the first process – movement by car –
that takes place in space and calls upon the morphology of networks and the
accessibility of sites at the entrance points, which we will refer to as “gateway
cities”4.

Natural recreational areas consist mainly of natural sites that do not all have the
same value in the eyes of the visitors. Visitors are likely to come to several of them
during a displacement and they make choices to visit certain sites on the basis of the
latter’s better or worse infrastructure, beauty, accessibility from a car park or
proximity to places of residence, etc. It is then possible to model this area by using a
network of sites (a reception network) which is characterized by a hierarchical
structure based around one or several main sites that condenses the flows and
redistributes them to secondary sites during the second displacement. One of the
recurring characteristics on these networks is the existence of more important sites
which will act as “attractors” for the nearby sites. The visitor who comes by car will
be partly attracted to one or more sites that he wishes to discover. Once the visit is
over, it may be that he decides to see another site in the vicinity, thus carrying out a
tourist tour. The very structure of the network, which is generally hierarchical,
induces forms of recreational consumption of natural environments. It is important
to know if the very structure of a reception network or, more precisely, its
morphology and hierarchy, which comprises various sites connected to each other
by a road network, is likely to direct or to transform certain types of movements.

4 These terms are generally used by regional natural reserves for the principal cities with
direct access to the parks and we will use them in turn to describe the cities emitting the main
incoming visitor flows.
34 Graphs and Networks

2.1.4. Pedestrian movement

The second movement scale is related to the way in which individuals use the
area once they leave their car – pedestrian movement, i.e. the way in which visitors
diffuse through the area. The study of visits to the Hautes-Vosges5 highlighted a
radio-concentric diffusion of pedestrian flows6 around access points, which are the
car parks. Whether a German tourist, a Parisian or a local resident of a higher socio-
professional category or not, all these individuals engage in pedestrian movement
around the parking, by doing either a loop circuit or an outward and return journey.
We could determine three types of visitors according to the pedestrian diffusion
areas that the visitors were likely to attend: “simple visitors”, walkers and hikers
(Table 2.1). Based on concepts of demographic and ecological load capacity7 of a
natural environment this classification makes it possible to quantify the human
pressures and to obtain information that is essential in developing the sites’ potential
to accommodate visitors.

Type of visitors Diffusion area Length of displacement

“Simple visitors” R < 500 meters < 1 km in an outward and return journey

Walkers – “simple R < 500 meters 2 to 4 km in a loop

visitors”

Walkers – “simple R < 1,000 meters 2 to 4 km in an outward and return journey

visitors”

Walkers R < 1,000 meters 6 and 8 km in a loop

Walkers R < 3,000 meters 6 to 8 km in an outward and return journey

Hikers R < 3,000 meters 10 to 20 km in a loop

Hikers R > 3,000 meters 20 km and more

Table 2.1. Typology of visitors and radio-concentric areas of pedestrian diffusion

5 Observations that have been verified by several studies: Pont du Gard in 1997 by the Nimes
CCI, Pyrenees national parks [DOC 94] and Ecrins [DOC 92].
6 During the study of the visits to the Hautes-Vosges it was extremely interesting to note that
some paths were used only for descending or climbing.
7 The demographic load capacity is the maximum quantity of visitors that a site can
accommodate, the ecological load capacity is the maximum quantity of visitors that a natural
environment can accommodate without degradation.
 Journey Simulation of a Movement on a Double Scale 35

The classification of visitor populations according to the types of diffusion areas

in natural environments enabled us to obtain groups of individuals segmented
according to behaviors that appear common to all visitor types: pedestrian
movement. This typology makes it possible to regard visitor groups, when
formalizing our assumptions, as spatial consumption agents of recreational natural
environments.

Interactions between network morphology and path behaviors manifest

themselves in variations linked to visit times and the number of sites attended in one
day. These variations induce a fluctuation of visitor distribution through natural
environments, by creating a greater or smaller number of visits of the attended sites
in time and quantity. Low movement times between sites may multiply the number
of visited sites and decrease the attendance times at each site concentrating the
visitors and the impacts on an area of 500 meters, whereas a sparse sites network
minimizes the number of visited sites but increases visiting times, which imply visits
and impacts that are more diffused but have a greater impact on the natural
environments.

2.2. The FRED model

2.2.1. Problems

On the basis of the main observations made about the extent of the tourist
infrastructure of an area and the types of recreational consumption of the natural
sites, the aim of constructing the FRED model [DEC 00] was to resolve a scale
problem in order to integrate the two movement forms. Let us recall that one of the
main difficulties is to define the variables and to determine the parameters in order
to model a multiscale process: kilometric scale for car movement and hectometric
scale for pedestrian movement.

According to the assumption that space predetermines or directs a certain

number of space consumption behaviors we suppose that there exists an area with
flows of visitors that are homogenous not in their behaviors but in the type of
activities that they have come to engage in (indeed, it seems obvious that people do
not come to Ecrins to practice water sports!). Thus, we suppose that there are areas
offering a set of sites that are homogenous with respect to the panel of proposed
activities (“spatial offer”). That makes these spaces more structured and attractive by
forming geographical entities for the discovery of natural environments.

The aim of formalizing is to obtain a relative value of the area according to the
type of recreational consumption represented by a distribution behavior. Thus, we
consider that the value of the natural environment differs according to the activity
36 Graphs and Networks

that the visitors engage in. Whether the visitor is a “simple visitor”, a walker or a
hiker, the attractiveness to the site is determined by the types of recreational
activities for the visitors, a “practical value” of the site. That is, the attractiveness of
a site depends not only on the structure of the area, but also of the type of visitor’s
recreational behavior – a relative attractiveness for a visitor type. All the
assumptions exposed below and the originality of the model are based on the fact
that there is no absolute value of space. Space is subjected to filters that are different
according to the types of practices of recreational space consumption. An area is not
seen the same way by “simple visitors”, walkers or hikers.

2.2.2. Structure of the FRED model

To this end, we exploit two calculation modules. The FRED model is broken up
into two parts and meets precise aims corresponding to specific modeling methods.

2.2.2.1. The calculation module of visitor access to the sites

The first models processes of spatial interactions linked to our three previously
defined spatial constraints: spatial offer, absolute attractions and accessibility of
sites. It models the distribution of visitors at the sites following car movement
initiated at the place of residence (or gateway cities) and ending at the car park with
access to natural sites. The simulation model is based on the architecture of a
gravitation model [HUF 62], which defines for each visitor type the attractions of
the various car parks with access to natural sites. These attractions are calculated
according to three spatial constraints.

On the basis of the various observed elements we have constructed a simulation

model which consists of calculating the probability that a car park has to fixate
tourist flows at certain entrance points determined by taking into account:
– the accessibility of the places of interest: “distance time” between the entrance
and the various car parks of the tourist area;
– the absolute attractions of each site in the network according to a recreational
practice of space consumption of the tourists (diffusion area);
– the existing spatial offer around a car park (level of infrastructure and
accessibility to natural attractions: mountain tops, panoramic viewpoints, lakes).

On the basis of accessibilities to a movement starting point (a “gateway city”)

the FRED model then calculates the probability that a car park can fixate
differentiated visitor quantities (“simple visitors”, walkers and hikers). By using the
graph theory and the former work by the CESA laboratory on the use of this theory
with respect to transport [AUR 94, HO 97] we have modeled a tourist reception
 Journey Simulation of a Movement on a Double Scale 37

network in a graph which is defined by arcs that symbolize the road connections in a
regional area and by nodes, connected by the arcs, that represent car parks with
access to natural environments. The modeling principle consists of calculating the
distribution probabilities for each visitor type – “simple visitors”, walkers and hikers
– on the graph.

2.2.2.2. Diffusion calculation module

The second module models visitors’ routes through the natural environments. In
this case it is a calculation of pedestrian movements of the individuals during their
walks through the natural sites on the basis of cellular automata programming. The
aim of this programming is to provide a tool enabling us to evaluate the potential
pressures of visitors’ movement by simulating the routes in the natural
environments.

On the basis of a maximum movement time corresponding to the distances of

500, 1,000 and 3,000 meters diffusion areas we calculate the number of times that an
arc of the path graph is used during a displacement with a car park (or point of
entry) as origin and a “stop” point (either a natural attraction or a natural site) as
destination. Movement is taken into account, if and only if the site can be reached
within the allotted time. The assumption here again is that movement in a diffusion
area necessarily implies a minimum time of presence at the natural site.

2.3. Part played by the network structure

Our prior work [DEC 00] clearly highlighted that the three spatial constraints –
accessibility, vicinity and spatial offer of the site – determined the distribution of
visitors on car parks with access to natural environments. We could therefore
identify several movement types subjected to threshold phenomena.

A first access movement oscillates between 45 minutes and an hour, then a

second movement (from 10 to 15 minutes) of redistribution on the reception network
(or circulation between sites). The results are interesting, since they made it possible
to determine spatial movement indicators which demonstrate very selective
recreational practices in the natural environments.

These results give us implicit information about the attendance time of the site
visitors. Indeed, if the visitors move in a network with poorly accessible car parks,
they visit few sites, perhaps only one. We may then suppose that walking time is
spent at one site only. In this case pedestrian diffusion is likely to be more
penetrative of the natural environment, i.e. movements tend more to be within the
1,000 and 3,000 meters diffusion areas around the car parks. Otherwise, a network
38 Graphs and Networks

which makes it possible to stop at several sites in one day implies that walking time
is divided between several sites. Pedestrian diffusion then tends to be only within the
500 meters zone from the car park and leads to a concentration of impacts around
the car parks.

As mentioned in the beginning with regard to territorial management it is

interesting to know the factors of visitor flow distribution in a site network. We
assume that network structure has an influence on the distribution forms. If the
accessibility of sites to “gateway cities” appears to be a factor of visitor distribution,
the density and the connexity of the relations between car parks with access to
natural sites seem determinant in visitor redistribution to the nearby sites.

Road network coverage, the distribution of sites through the mountain and
spatial characteristics associated with the graph (access distances of the sites from
the gateway cities and distances of mutual site interaction) reveal a tendency to
homogenize the distributions throughout the territory by assigning visiting
probabilities to all sites. However, heterogenous space distribution of the natural
environment offer of the sites (taking into account site hierarchy) would then
polarize the visits to clearly defined areas. The various sites of these sectors will
have an attraction effect by concentrating flows in their area, like first order places
of interest that redistribute flows to the surrounding sites. This phenomenon is all the
more amplified by the road network morphology.

Visitor distribution probabilities through the Hautes-Vosges car parks

0 5 10 km 0 5 10 km

Graphe a graphe b

Distribution of car parks starting at the gateway Cumulative distribution of visitors on

cities after a 45-minute movement car parks after a three-stop tour
Cartography using the FRED model. Fabrice Decoupigny, CESA Laboratory, 2000

Figure 2.1. Distribution of visitor flows throughout the Hautes-Vosges massif

 Journey Simulation of a Movement on a Double Scale 39

The proximity effect would thus reveal geographical aggregates of very

attractive sites (Figure 2.1). Indeed, these sites are then able to mutually exchange
visitors, thus increasing their attractiveness by comparison to sites that are generally
isolated in the network. It then becomes interesting to evaluate the influence of the
network structure (location of car parks and the road graph morphology) in the
forms of visitor distribution throughout the reception area.

2.4. Effects of the network on pedestrian diffusion

2.4.1. Determination of the potential path graph: a model of cellular automata

First of all, we calculate the practicable area, i.e. the area where it is possible for
an individual to move normally. This practicable area is calculated by introducing
the constraint of a limit slope. Any arc with a slope over 50% cannot be used as a
path arc and therefore cannot be taken by an individual. In a way the program
determines a potential area of practicable diffusion. This operation consists of
tracing on the digital terrain model (DTM) graph from any remarkable point to
every other all the outward and return routes that can exist between the specific
natural environment nodes (car parks and natural attractions). The path laying
process is conducted node by node following an 8-connected diffusion. Once the
diffusion is over we create a file of the nodes and arcs of the graph of the potential
routes obtained so as to simulate the routes and impacts according to a calculation of
optimization of the minimal paths in the second phase of the simulation.

The aim of the diffusion is to simulate movement from the departure point to an
arrival point passing by various DTM nodes supposing that the individuals have no
knowledge of the environment but see the node that has to be reached. The
assumption is very different from that of minimal path optimization. First of all, the
automaton defines all the potential routes in an area and only afterwards the course
through the paths is optimized. Moreover, this assumption also uses another element
that comes into play during the choice of routes: visibility. The cellular automaton
representing the walker describes routes without anticipating the total difficulty of
the path, the only information that it has is the location of the arrival point. It
optimizes movement only on sections with connected nodes. In this manner we can
take into account a very common type of visit for walkers: movement at sight.
Indeed, the assumption, as it is defined, supposes that the arrival point site is known
or seen; we thus suppose that the visitor knows the general direction of the node to
be reached.
40 Graphs and Networks

2.4.2. Two constraints of diffusion

The observed impacts on natural environments do not make it possible, a priori,

to evaluate movement types because the multiplicity of the impact types (shortcuts
through sharp bends, spontaneous routes, widening of the footpaths) in the natural
environments do not enable us to decide on the type of movements that generate the
impacts. We wish to test two types of movements in the environment (Figure 2.2 in
the color plates section). The question that we pose is to find out if the visitors move
in natural environments using the shortest distance path (called Shortest) or the path
that requires the least effort (called Easiest).

In the first case, the assumption of the model concerning the choices of paths
includes the general hypothesis of “the law of least effort”. It consists of stating that
people try to reach a point with a minimum of effort. A person will choose the
connected node according to the variation in altitudes making it possible to associate
minimal difficulty to the easiest path. In the second case, the choice of the related
node is made by minimizing the Euclidean distance without involving difficulty.
However, the program stores in memory the difficulty of the traversed arcs. Since, if
the visitor chooses the shortest path, he remains subjected to the constraints
generated by slopes and will not spend the least time to reach the target, it may even
be that this type of path is the longest in time.

Thus, it becomes possible for us to differentiate between outward and return

routes. Indeed, various observations have demonstrated that the routes between two
points may be different depending on the direction of walking. The outward and
return routes are not symmetrical.

Taking into consideration the results obtained by the cellular automaton, we have
a potential path laying area with no preferential routes. It is thus necessary to
reprocess the information obtained and to establish among these routes or portions
of routes those which are more likely to be used than others.

Indeed, we may admit that people have a certain knowledge of the areas either
because they have already visited, or because they use maps, or quite simply follow
a footpath and then retrace their steps. The objective of the model is to find the
factors governing movements in a natural environment and the induced impacts on
the basis of a simulation of the genesis of a network of paths.

Therefore, contrary to the assumption put forward previously regarding the

forms of movement, we suppose that people move from one point to another
optimizing the path in a potential network (Easiest or Shortest). Thus, we introduce
the assumption of preferential choice of path made according to an optimization of
the minimal path under the movement type constraint. This assumption of the
 Journey Simulation of a Movement on a Double Scale 41

optimization of the minimal paths is admissible insofar as we can suppose that when
the visitor leaves his vehicle he wishes to go and reach the natural site as soon as
possible. The path graph will be used as a basis to calculate the impacts of the
movements of the various types of visitors on the natural sites. For the two path
graphs (Shortest and Easiest) we will apply “Floyd’s” algorithm in order to search
the minimal paths. The fact of having differentiated outward and return movements
and the difficulty of the arcs makes it possible to have a dissymmetrical return path
layout compared to the outward one. The goal of the calculation of the optimal paths
on the path graphs consists of calculating origin-destination movements between the
various points of interest (car parks, natural sites) and distinguishing the arcs which
are likely to be the most used to get from one node to another under the constraint of
minimization of the distance covered according to the two types of movement
behavior in the natural environment, either the Shortest or the Easiest path.

Three types of routes are considered: access movements from car parks to the
natural sites (or outward movements); walking movements (or circulation) between
the natural sites, return movements from the natural sites to the car parks. The two
movement assumptions (Easiest and Shortest) associated with the movement
thresholds in the diffusion areas enable us to put forth assumptions about the types
of visitor movement through the natural environments. We will thus be able to see
whether the impacts generated by movement correspond to there and back walks or
loops. The initial conditions of the simulations observe movement times
corresponding to the diffusion areas of the various visitor types (Table 2.2).

For a walking speed of 3 km/h

“Simple visitors” Walkers and

Types of visitor “Simple visitors”
and walkers hikers

Walking time
10 minutes 20 minutes 60 minutes
less than

Maximum
500 m 1,000 m 3,000 m
distance covered

Table 2.2. Typology of movement spatial behavior

The impacts of the routes are yielded by the frequencies of use of the graph arcs
each time there is movement from one point to another. If the probabilities of
fixating on car parks are integrated in the form of the weight of pedestrian flows
entering through one, we do not proceed to the calculation of probability of path
42 Graphs and Networks

usage. The results give only potential impacts on the area regardless of the choices
of direction that may take place in the natural environment (Figure 2.3). That is,
starting at a car park we calculate only the number of times that an arc is used once
all the itineraries have been calculated.

3 3
1 1
2
2
4 4

6 5
7
8
9
"Shortest"
"Easiest"

Point 1: Car park at the nature reserve of Tanet - Gazon du Faing

Point 2: Panoramic table
Point 3: Soultzereneck
Point 4: Gazon du Faing
Point 5: Dreieck car park (or Collet du Lac Vert)
Point 6: Lac Vert
Point 7: Lac Vert car park
Point 8: Tanet car park

Figure 2.3. Comparison of a simulation of routes for a 20-minute movement in a 1,000-meter

diffusion area (nature reserve of Tanet and Gazon de Faing – the Hautes-Vosges massif)

2.4.3. Verification of the model in a theoretical area

We have simulated movements in a theoretical space under the same conditions

as at the Vosges. The aim of this comparison is to verify if the results obtained for
the Tanet area reveal specific characteristics due to local conditions
(geomorphology, location of the car parks, etc.) or to conditions of a more general
nature. To this end we have created a hilly theoretical area modeling the mountain
relief with altitudes oscillating between 600 and 1,400 meters and two potential path
graphs: Shortest and Easiest (Figure 2.4 in the color plates section).
 Journey Simulation of a Movement on a Double Scale 43

We can note that movement simulations yield different results than those
obtained at the Vosges (Figure 2.5 in the color plates section). The results are less
pronounced in the theoretical area. According to the type of movement constraint
(Easiest and Shortest) the impacts are of a different nature: either linear pressure or a
multiplication of the paths (routes generating impacts such as shortcuts, path
branching, etc.) The main difference stems from simulations of the movement
hypothesis corresponding to the Easiest constraint. At the Vosges we have
movements concentrating on routes along footpaths, whereas in the theoretical area
we see extensive movements multiplying the paths (see Figure 2.5). The differences
between the two movement assumptions are much more marked in the theoretical
area. Contrary to the Vosges the induced impacts are reversed: for the “Easiest”
movements we have impact types that would tend to multiply the footpaths
spontaneously, whereas for the “Shortest” movements the impact types would tend
to widen the footpath.

What do these two contradictory results mean? At first we may think that the
geomorphology of the site leads to different movements. However, simulations in a
theoretical area should have yielded results similar to these obtained for the Vosges,
since we took the same physical characteristics. It is therefore necessary to seek the
answer elsewhere. If it is not just the relief that induces impact types, we may
consider that the reception facilities may play a part in the diffusion forms in natural
environments.

In order to verify this assumption, we will again address the specificities of the
Tanet area (Figure 2.6 in the color plates section) in a theoretical area. To that end
we have kept only three car parks at mid-slope and two at the bottom of the slope to
obtain a structured area similar to the Tanet nature reserve (with the same number of
natural sites and car parks). We then obtain completely different results compared to
the first simulations (Figure 2.6) corresponding to those obtained for the Vosges. For
the latter we note that the number of routes is denser when there is a certain number
of natural sites and network entrance points.

Although theoretical space is denser at natural sites we observe the same types of
routes as these at the Vosges, with a linear concentration of routes on the Easiest
path graph and a multiplication of the routes accompanied by a north-south axis of
pedestrian circulation on the slopes of the Shortest path graph. This axis also appears
at the Vosges. These simulations show that areas are differentiated not only
according to the types of spatial usage and site but also according to the reception
facilities and accessibility of the massif.

These results are fundamental. We recall that one of the observations made
during the study of visits to the Hautes-Vosges [DEC 97] tended to say that visits to
a massif depended on the number of car parks and not on the number of places. That
44 Graphs and Networks

is, the pressure of visitor diffusion on natural environments depended more on the
number of access points than on the load capacity of the car parks.

The model shows that the impacts to a large extent depend on the reception
facilities in natural environments. The number of car parks and their location can
play a determinant part in the types of impacts of anthropogenic pressure.
Depending on the network structure, movement type and geomorphology of the
area, anthropogenic pressures can take different forms. Not only does the network
structure play a dominating part in car movements and the choice of the site to visit,
but it also seems to be the case for pedestrian movements within the natural
environments.

As regards the planning and protection of natural environments, network

management may differ from one area to another if we wish to concentrate or
diffuse the visits to certain places of interest. To this end tourist visits must be
evaluated by taking into account three interacting elements: the site, the network and
the visitor. This amounts to saying that the management of visitor flows and their
impacts on natural environments are apprehended simultaneously on two distinct
scales: car movement associated with a type of pedestrian movement.

2.5. Bibliography

[AUR 94] AURAY J-P., MATHIS P., “Analyse spatiale et théorie des graphes”, in J.P.
Auray, A. Bailly, P.H. Derycke, J.M. Huriot (ed.), Encyclopédie d’économie spatiale:
concepts, comportements, organisations, Economica, Paris, p. 81-88, 1994.
[CAZ 92] CAZES G., Fondements pour une géographie du tourisme et des loisirs, Bréal,
Paris, 1992.
[CHA 99] CHARDONNEL S., Emplois du temps et de l’espace. Pratiques des populations
d’une station touristique de montagne, PhD Thesis, Grenoble Joseph Fourier University,
1999.
[CUV 98] CUVELIER P., Anciennes et nouvelles formes de tourisme: une approche socio-
économique, L’Harmattan, Paris, 1998.
[DEC 00] DECOUPIGNY F., Accès et diffusion des visiteurs sur les espaces naturels.
Modélisation et simulations prospectives, PhD Thesis, Tours University, 2000.
[DEC 97] DECOUPIGNY F., Etude fréquentation Hautes-Vosges, Tours, 1997.
[DEW 90] DEWAILLY J.-M., Tourisme et aménagement en Europe du Nord, Masson, Paris,
1990.
[DEW 92] DEWAILLY J.-M., “Les citadins et les loisirs de nature”, in Tourisme et
environnement, La Documentation Française, Paris, p. 30-32, 1992.
 Journey Simulation of a Movement on a Double Scale 45

[DEW 93] DEWAILLY J.-M., FLAMENT E., Géographie du tourisme et des loisirs,
SEDES, Paris, 1993.
[DOC 92] “La fréquentation touristique”, Documents scientifiques du Parc National des
Ecrins, no. 4, 1992.
[DOC 94] “La fréquentation touristique”, Documents scientifiques du Parc National des
Pyrénées, no. 28, 1994.
[ENV 92] MINISTERE DE L’ENVIRONNEMENT, MINISTERE DU TOURISME,
Tourisme et environnement, La Documentation Française, Paris, 1992.
[FER 95] FERBER J., Les systèmes multi-agents, vers une intelligence collective, Paris,
InterEdition, 1995.
[HUF 62] HUFF D.L., Determination of intra-urban retail trade areas, University of
California, Los Angeles, 1962.
[IFE 00] INSTITUT FRANÇAIS DE L’ENVIRONNEMENT, Les indicateurs: tourisme,
environnement, territoires, Orléans, 2000.
[LHO 97] L’HOSTIS A., Image de synthèse pour l’aménagement du territoire: la déformation
de l’espace par les réseaux de transport rapide, PhD Thesis, Tours, 1997.
[PLA 98] Premier Ministre, Commissariat Général au Plan, Secrétariat au Tourisme,
Direction du Tourisme, Réinventer les vacances. La nouvelle galaxie du tourisme, La
Documentation Française, Paris, 1998.
[TOU 98] SECRETARIAT D’ETAT AU TOURISME OBSERVATOIRE NATIONAL DU
TOURISME, L’évaluation par les flux: outil d’analyse touristique territoriale,
Observatoire national du tourisme, Paris, 1998.
 Chapter 3

Determination of Optimal Paths

in a Time-delay Graph

3.1. Introduction

In urban planning, the applications of graph theory have demonstrated their

effectiveness in modeling transport networks and evaluating their impacts on
topological space through the contraction of space-time and its representation in the
form of Chronomaps [LHO 97], on accessibility using NOD software [CHA 97], on
the operation of the network, i.e. the organization of potential movement flows, or
on its evolution considering a dynamic network evolving over time [BAP 99].

First of all, these approaches distinguish the “containing” part (linear and nodal
transport infrastructure) from the “content” part (means of transport); both of these
are nevertheless joined together in the form of functional binomials associating a
certain infrastructure with a certain means of transport. Each one is distinguished
from the other according to the announced aim: analysis and evaluation of supply
modifications, taking demand into account, integrating the temporal factor. Using
the latter objective we can classify the types of graph necessary for simulations into
three categories.

The graph referred to as “traditional” distinguishes between arcs according to

their physical characteristics (rail, road, number and width of lanes, driven by
electricity or not, etc.) refined by topography and sinuosity constraints, the valuation
of these arcs in duration or cost resulting from all the above criteria. The means of

Chapter written by Hervé BAPTISTE.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
48 Graphs and Networks

transport are considered jointly leading to the development of binomials with

permanent functionality. Anyone in a particular vehicle can use a road at any
moment, unconstrained by time, if we agree to neglect climatic risks (shutting-down
of snow-covered sections, for example), works on roadways or structures. This type
of graph paves the way for diachronic analyses by comparing the performances of
two graphs according to a distinct transport supply.

The adaptive graph is, in fact, a graph of the same type (described by permanent
functional binomials, a valuation of the arcs of the same nature), but considered
from a temporally dynamic, long-term point of view. On the basis of an initial
graph, the improvement of a transportation axis frequently and durably exceeding its
maximum capacity leads, following a cyclic process, to the plotting of a series of
new graphs, which are described by arcs with different characteristics (widening of a
road followed by doubling of the number of lanes or pure and simple creation of a
new highway arc, for example).

On the other hand, the time-delay graph adds another constraint that is not only
infrastructural, but rather more functional, thus by definition discontinuous in time
for a mass transport network. This constraint is necessary when infrastructure
evaluation stems from supply quality analysis. Indeed, the existence of a TGV train
station in a town is only of limited interest for an individual if the number of
effective stops is restricted. The main requirement was thus to consider binomials
with temporary functionality via a coupling of permanent infrastructures (rails, roads
even air corridors, etc.) with means of transport functioning punctually throughout
the duration (trains, buses, planes, etc.), hence the name of a time-delay graph with
functional temporization.

To this end an algorithm, based on Floyd’s algorithm, was developed and aimed
to determine the optimal paths between all the nodes of a time-delay graph, thus
integrating departure and arrival timetable constraints of trains and other mass
transport vehicles, leading to simulations of the type: which cities can I reach in less
than an hour, departing from my origin at a certain time? From which city can I
leave in order to arrive at the destination before a certain time? Or more generally,
what timetables do I have at my disposal to go from any station to another?

We will not develop the founding model, i.e. Floyd’s algorithm [BAR 86], nor its
modifications directed more specifically towards urban planning simulations and
more precisely towards access problems in the transport domain, as it has already
clarified in [MAT 78]. We will only point out its main characteristics, namely, the
determination of the optimal path (in time, in cost, etc.) from any point to any other
point by enumerating the series of arcs and we will use an example in order to
describe its operation.
 Determination of Optimal Paths in a Time-delay Graph 49

On the other hand, the evaluation of the quality of service being the principle
basis for the chosen modeling, we will develop the concepts of “door to door”
movement and interconnection constraints (time needed for a journey on foot, using
public transport, etc., number of changes beyond which the path is abandoned,
whether in favor of a possible alternative or not, etc.).

The “traditional” and “timetable” versions of the developed Floyd’s algorithm

are distinguished above all by the structure of the descriptive matrix of the arcs, by
introducing the third dimension needed to account for temporal discretization of the
relations (a train that left five minutes ago has lost its functionality for the user who
arrived late at the platform, although the infrastructure remains). In the matrix, all
nodes taken two by two yield a number of relations, defined by departure and arrival
times, existing for a given period (a day, a week, etc.), to which relations with
permanent functionality can be added.

Initially neglecting, in order not to confuse matters, the connection constraints

specific to towns of different sizes or with several train stations and induced by the
time wasted to get off a train and get on another one, the set of paths is determined
by successive loops. For a given origin and a destination, for each departure during
the initially chosen period and regardless of the intermediate destination, the
algorithm will construct, if it exists, the series of arcs that respond to the constraint
of the time of arrival at an intermediate node before the time of departure from this
same node for the rest of the journey.

Formalization itself will be clarified by choosing an example of several nodes

connected by several arcs described in the form of a timetable, which will be
progressively detailed in frames as the calculation is developed.

3.2. Floyd’s algorithm for arcs with permanent functionality

Before describing Floyd’s algorithm that is specifically developed to integrate a

transport timetable offer we begin by briefly pointing out the operating mode of the
primitive algorithm, with complexity of the 3rd order which is based on triple
looping through n nodes (n3). For didactic reasons we take an example limited to
four nodes A, B, C and D and three arcs connecting A and B, B and C then C and D
(Figure 3.1) by chronologically exposing the computing process in a formal manner
and referring the reader for the algorithmic description to [BAP 99, CHA 97].
50 Graphs and Networks

D
B

C
A

Figure 3.1. Example of four stations connected by a railway

For each pair of nodes AC, AD, BD (by considering undirected relations and
therefore neglecting symmetric relations) which are connected by paths1 (in the
opposite case, i.e. a pair of nodes connected by an arc, the calculation continues
without processing the relation for this pair) each node is regarded as an
intermediate transit node. If the graph were directed, 64 relations would be possible
including loops and counting outward and return journeys along an arc or a path
separately. Relations AB, BC and CD are also processed on the assumption, which
is invalid in this particular example, of a shorter duration when passing by one or
more intermediate nodes.

Thus, after processing the intermediate node A, which is done quickly since it is
located at the end of the network, the intermediate node B is taken for the pair AC
by connection of the AB and BC arcs. For the AD pair, node B is an intermediary
but no path is determined because the path between B and D has not been
constructed yet. Then starting with the intermediate node C, relation AD is
determined as associating the AC path (built previously) and the CD arc, as well as
the BD path associating the BC and CD arcs. Finally, since node D is in the same
situation as node A, the calculation is completed by having indexed all of the
possible relations (paths and arcs) between each pair of nodes. We have
purposefully omitted symmetric relations, which are of course determined during
double looping at the origin and then at the destination for each intermediate node
(that is, the CA, DA and DB paths).

The brief presentation of this algorithm shows its interesting adaptation to

determining the paths with permanent functionality, where the selection of the arcs
and the shortest paths (in time, in cost, etc.) constitutes a first extension of the
primitive algorithm. For this latter aspect a preliminary selection of the optimal arcs
between each pair of nodes is carried out when it is connected by several arcs of
different nature (traditional in the transportation field where a local or regional road
is doubled by a trunk road with greater throughput). This procedure initiates the
calculation of minimum paths by being positioned upstream of the main algorithm.

1 Understood as a juxtaposition of at least two arcs.

 Determination of Optimal Paths in a Time-delay Graph 51

In addition, the functionality of the primitive algorithm that enables determining

“antecedents”, i.e. for each node the immediately preceding node in a route with
multiple transit nodes, is preserved. In particular, it makes it possible to determine
the principal corridors which are identified when a significant number of paths
follow the same arcs.

The description that has just been made deserves a comment on its possible
application to a network that is completely or even partially comprised of public
transport connections. The permanent functionality of road transport networks
allows interconnection without load rupture, in other words, immediate passing from
one arc to another (by neglecting the time spent at a node, which in addition is
solved by multiscale modeling and is applicable to Floyd’s algorithm by simply
considering “interurban” and “urban” functional binomials, the latter connecting
complementary intra-urban nodes [CHA 97]).

3.3. Floyd’s algorithm for arcs with permanent and temporary functionality

3.3.1. Principle

The particular case of a network consisting of binomials with temporary

functionality, to which a user is constrained by the times of departure and arrival,
has imposed a major modification of Floyd’s algorithm, which requires to take into
account the load rupture times and an adequate distribution of the timetables in time
during connections2.

For this reason several versions were developed according to the needs of studies
and research [COL 96, BAP 02]. On the basis of an algorithm making it possible to
determine the set of “timetable” paths we have successively developed new
functionalities, such as the need for arriving at a destination before a certain
schedule (arrival time constraint), or for leaving an origin only after a certain time
(departure time constraint). Moreover, according to the space scale considered and
the mass transport means (rail, air, mixed), specific connection constraints have been
worked out particularly in individualized processing of paths requiring a connection
with a change of station in the same city (Paris, etc.), which is described by only one
node but by several stations, as well as constraints on terminal stops in order to
consider door to door path laying.

2 This transformation of the algorithm was imposed when it became necessary to pass from a
simple logic of transport supply (infrastructural) to a quality of service logic, within the
framework of research for the DTT and the OEST (ex-DAEI).
52 Graphs and Networks

We will develop hereafter the most general version determining the set of paths
within a time range defined by the user.

3.3.2. Description

Similarly as for the traditional version of Floyd’s algorithm we will expose its
operation by using an example limited to four nodes, which are connected
successively by three links: each proposing two departures during a certain
considered period. By neglecting the extremities of the line we have certainly
chosen a less probable example because the return is impossible (a case that is,
nevertheless, not useful here, since its study concerns the same method (Figure 3.2)).
We will follow the algorithmic steps chronologically by presenting the concerned
procedures successively.

D
B C
A 11:00 am
9:00 am 9:05 am 10:00 am 9:59 am
8:00 am 11:30 am
9:45 am 9:40 am 10:20 am 10:25 am
8:30 am

Figure 3.2. Four stations connected by rail with timetables

3.3.2.1. Creation of the minimum arcs matrix

In the first instance, the arcs matrix is processed in order to check the potential
existence of arcs with permanent functionality in addition to the timetable arcs
(Algorithm 3.1). According to this hypothesis, if two relations with permanent
functionality exist for the same arc, the fastest is preserved (example of a relation by
walking and a faster one by car). Similarly, if there are also relations with fixed
timetables, they will only be preserved if they are faster than a relation with
permanent functionality. Flexibility is retained here to privilege an unconstrained
timetable relation if it performs better with respect to duration (a relation of the same
duration will also lead to privileging that without timetable constrains). At this level
nothing would prevent implementing the opposite choice resulting from taking into
account unmotorized travelers and privileging in the long term the mass transport
route constrained by timetables.

Moreover, the maximum number of timetables for a given arc (in other words
the maximum matrix thickness) is determined; the timetable arcs are placed starting
 Determination of Optimal Paths in a Time-delay Graph 53

from the second layer, whereas the best performing permanent arc with
functionality, if it exists, is placed in the first layer.

CREATEMATARCSMINTIMETABLE
Knowing the set of arcs referenced by the iOri indices of the origin node and the iDes indices of the destination nodes, their means of transportations,
traveling time taken from the corresponding input file,
For each arc k = 1, 2,…, n (of iOri origin and iDes destination indices):
If the arc k has permanent functionality:
If the minimum arc (iOri, iDes) does not yet exist (value equals zero),
or if the arc k is shorter than the minimum arc, (iOri, iDes) already stored:
* The journey duration and means of transportation are stored in the “ArcsMin” matrix of the minimum arcs with indices iOri, iDes.
* The new minimum arc (iOri, iDes) is affected to the depth of 1
(layer allocated to the permanent functionality relation that is the most performing in terms of duration).
End If
Otherwise, if the arc k has temporary functionality:
If the minimum arc (iOri, iDes) does not yet exist (value equals zero):
If its depth is zero (no arc with permanent functionality):
* A depth of 2 is allocated (to position the timetables starting from the second layer)
Otherwise:
* An additional layer is allocated.
End If

If the depth obtained is greater than the maximum depth until now:
* Maximum depth is adjusted to this new value.
End If
Otherwise, if the arc k is faster or has the same duration as the minimum arc (iOri, iDes):
An additional layer is allocated,
If the depth obtained is greater than the maximum depth until now:
* Maximum depth is adjusted to this new value.
End If
End If
End If
Next arc k

Algorithm 3.1. Creation of the minimum arcs matrix

3.3.2.2. Determination of the range of timetable processing

Secondly, the user defines the time range within which the search for the
minimum path must be carried out, which is limited by processing start and end
times. The start time is immediately assigned to a “reference time”, which will
progressively increase during processing to determine the valid paths
chronologically.

3.3.2.3. Partial and provisional filling-in of the minimum paths matrix

As a precondition to the main calculation to determine the paths, a matrix of
precedents must be partially filled (Algorithm 3.2): this matrix makes it possible, if
necessary, to recompose the path by using intermediate nodes, as was underlined
previously.
54 Graphs and Networks

FLOYDPREPARATION
For each node i = 1, 2,…, n:
For each node j=1, 2,…, n:
If the minimum arc (i, j) exists:
* The cell (i, j) of the matrix of precedents recovers the index i and the value of the minimum arc (i, j).
End If
Next node j
Next node i

Algorithm 3.2. Preparation of the calculation of the minimum paths

3.3.2.4. Processing of the minimum arcs

The following procedure still concerns the preparation of the calculation of the
optimal paths themselves, specifying, nonetheless, elements depending on input data
(Algorithm 3.3). In fact, the specific case of trains leaving one day and arriving the
following day requires particular processing in order to have an arc in the matrix that
is coherent with respect to the operation during a single day. Indeed, if the aim is to
determine the set of routes and timetables possible during one working day, 24 hours
are added to the time of arrival, thus meaning arrival is later than departure3.

Moreover, if the true paths, i.e. cumulating at least two minimum arcs, are only
created afterwards, the minimum paths matrix during this procedure will be partially
filled using the already stored minimum arcs, with neither connection, nor stop
required.

In order to optimize the timetable search procedure adapted to work out the
shortest possible path, the timetable permitting, for a given arc, to arrive at the
earliest is placed in the first layer4.

3 A specific version of Floyd’s algorithm was developed to determine minimum paths in the
form of timetables and is identical to that presented here, but considering days as well as
hours. The addition of 24 hours is no longer necessary, since it is substituted by checking the
day.
4 If arcs with permanent functionality are mixed with “timetable” arcs, the latter are placed in
the first layer, because they are always adapted to supplement a path.
 Determination of Optimal Paths in a Time-delay Graph 55

PREPCALCPPATHTIMETABLE
For each node i = 1, 2,…, n:
For each node j=1, 2,…, n:
If the thickness value of the minimum arc (i, j) exceeds the maximum thickness for all arcs:
# Maximum thickness is assigned to this value.
End If
Next node j
Next node i

For each node i = 1, 2,…, n:

For each node j = 1, 2,…, n:
If i=j:
# The (i, j) path is assigned the value of 32,000 (“infinite”)
Otherwise:
If the value of the minimum arc (i, j) is zero (no permanent functionality means):
If the depth of the minimum arcs matrix for (i, j) is zero (no timetable either):
# The (i, j) path is assigned a value of 32,000.
Otherwise (the arc is only described by timetables):
# The (i, j) path is set to 32,000.
For each layer k = 2, 3,…, n’ of the minimum arc (i, j)
(sweeping the entire depth of the arc apart from the first):
If the (i, j) arc in layer k has an arrival time earlier than the departure time
(change of day):
# The arrival time is modified by adding 24 hours.
End If

If the departure time of the (i, j) arc in the layer k is before the reference time:
# We add 24 hours to both departure and arrival times.
End If
# Filling in the paths matrix (value, departure and arrival times, means of transportation,
number of the start and finish line, index of the nodes i and j).
* If the (i, j) arc in layer k also makes it possible to arrive before the time is stored in
the first layer (reserved for the arc that arrives the first for a given relation):
# It is placed in the paths matrix in the first layer with all the characteristics
(value, departure and arrival times, means of transportation, number of the start
and finish line, index of the nodes i and j).
End If
Next layer k
End If
Otherwise (the arc is described by a permanent functionality relation):
** The minimal arc (i, j) is placed in the paths matrix storing the departure time as reference time,
the arrival time as this same time plus the duration, as well as the means, its value and the indices i and j of the nodes.
If the arc is also described by timetables:
For each layer k = 2, 3,…, n’ of the minimum arc (i, j):
If the (i, j) arc in layer k has an arrival time before the departure time (change of day):
# The arrival time is modified by adding 24 hours.
End If

If the departure time of the (i, j) arc in layer k is later than the reference time:
# It is placed in the paths matrix with all its characteristics (value, departure and arrival times,
means of transportation, number of the start and finish line, index of the nodes i and j ).
End If
Next layer k
End If
End If
End If
Next node j
Next node i

Algorithm 3.3. Processing of minimum arcs

56 Graphs and Networks

3.3.2.5. Determination of paths

Once these first initialization procedures have been implemented the calculation
itself to determine all the “timetable” path can begin. Triple looping on k, i, j
implemented for the “traditional” version of the algorithm is kept because the
innovations relate to taking on board the potential multiplicity of timetables for a
given arc.

As in the traditional version of Floyd’s preparation, we will not consider the

cases where A and D are intermediate nodes, which are not of interest here since
they do not enable the development of a path.

For the intermediary B, the first origin A is initially tested to verify the existence
of an arc or path between A and B, in which case the processing with the following
destination point C is useful. In the opposite case, the following origin is considered,
since AB does not exist, the interest to calculate BC is zero and unnecessarily
expensive in terms of time to find a path AC by B.

For the BC timetable, a first comparison of the arrival time is carried out with the
arrival time of the AC path (if there is any) in order to preserve the shortest
timetable of arrival at C and at the same time minimize the number of connections.

In our example there is no AC path and a new test is carried out to verify the
adequacy of the time of arrival at B from A and the departure time of B to C. The
former must logically be earlier than the latter in order to enable load unloading at
B5. Information over the connection time at B is thus found using the difference in
the two timetables above.

In practice, for the relation AC with B as intermediate point, the two timetables
(8:00-9:00 am) between A and B, and (9:05-10:00 am) between B and C thus
constitutes a first path, which did not exist until now. The next timetable (9:40-10:20
am) between B and C does not replace the previous one since the arrival time is
later.

Still with B as intermediate point, the relation AD cannot be determined, since

the BD path has not yet been found. It will be determined now by considering the
intermediate point C. Let us immediately observe that the second timetable (8:30-
9:45 am) between A and B is considered only later on when the reference timetable
supplants the current one (that is, at the first passage, the initial processing
timetable).

5 A connection is accepted by default which is a minimum time of two minutes, or when it

indicates a simple stop at a station (which is identified when the line number does not
change).
 Determination of Optimal Paths in a Time-delay Graph 57

According to point C the AD relation will be on the basis of the already existing
path AC with the timetable (8:00-10:00 am) and passing by point B (with five
minutes connection time) and the arc CD. The possibility (9:59-11:00 am) is quickly
discarded, since it does not enable the connection at C from A to D. It is thus the
latter which will be retained, leaving at 10:25 am from C and arriving at 11:30 am.
The complete path thus includes two connections at B at C, leaving A at 8:00 am
and arriving at D at 11:30 am.

From B to D the path found starts at 9:05 am, arrives for the connection at C at
10:00 am and again necessarily follows the timetable (10:25-11:30 am) between C
and D. This result is stored at this stage or only written down if the following
reference hour is later than the departure of BD.

FLOYDTIMETABLE
If at least one path is modified:
For each node k=1, 2,…, n:
For each node i=1, 2,…, n:
If a path exists between i and k:
For each node j=1, 2,…, n:
For each layer m=1, 2,…, n’ of the (k, j) arc:
If the (k, j) arc has permanent functionality:
# Departure time from k to j is stored as arrival time at k from i.
# Arrival time at k from i is stored by adding to the arrival time at k from i the duration to
effectuate the relation from k to j.
Otherwise:
# Existing departure and arrival times for the relation from k to j are stored.
End If

If this arrival time precedes the arrival time of the (i, j) path:
If this departure time from k to j is later than the arrival time at k from i:
# The means of the relations (i, k) and (k, j) are recovered to test the possible connections
(when two trains are used with different line numbers, if the connection time is greater or equal to 2
minutes; no connection if one of the two means has permanent functionality).
If the connection is possible or if there is none:
# The sum of the connection times is recalculated.
If the path does not exist yet or if there already exists one that performs worse
# The table is filled with the duration, departure and arrival times, the means (41 for train,
4141 for a combination of trains), connection time, indices i and j of the line numbers of the
start and the end.
Otherwise:
# It is a diagonal element that is not processed.
End If
End If
End If
End If
Next layer m
Next node j
End If
Next node i
Next node k
Loop

Algorithm 3.4. Main calculation to determine the minimum paths

Processing continues in this manner with looping until no timetable change can
be obtained. This assumption can sometimes be verified, since the creation of a new
58 Graphs and Networks

path n may make it possible to find at the following iteration, for another relation, a
better performing path p which nevertheless takes the path n. Algorithm 3.4
describes in formal language the main procedure to determine optimal paths.

3.3.2.6. Determination of a new reference time

Following the previous main processing the new reference time is calculated by
considering all the departure times of all the relations, after the previous time of the
start of processing (Algorithm 3.5). The nearest time (by at least 1 minute) serves as
the new basis for the Floyd calculation. According to the following example, 8:30
am is the new reference time.

SEARCHNEXTDEPARTURE
# Reset “NextDep” to 32,000 (“infinite”)
For each node i=1, 2,…, n:
For each node j=1, 2,…, n:
If the depth of the the (i, j) arc exceeds 1 (more than a timetable):
For each timetable layer k=2, 3,…, n for the (i, j) arc:
If the departure time from i to j is later than the current reference time:
If the departure time from i to j is less than “NextDep”:
# “NextDep” takes the value of this departure time.
End If
End If
Next layer k
End If
Next node j
Next node i
# The new reference time is from now on equal to “NextDep”, which is at least one minute later than the
previous reference time.

Algorithm 3.5. Search for the next departure time

3.3.2.7. Validation of the minimum paths matrix

It is only at this stage that the paths which are included between the previous
reference time and the new time are really written (Algorithm 3.6). The paths
already found in this range are stored in order to avoid heavy repetitive calculations
in calculation time and written later when they are included in the range of reference
times.
 Determination of Optimal Paths in a Time-delay Graph 59

FILLPATHTABLE
For each node i=1, 2,…, n:
For each node j=1, 2,…, n:
If the departure time of the (i, j) path is earlier or equal to the new reference time:
If the (i, j) path exists:
If the distance (duration) is less than 32,000:
If the path has temporary functionality (described by timetables):
# Filling in of the definitive paths matrix (departure and arrival times, means,
duration, connection times, i and j indices, start and end line numbers),
by replacing or creating, if the path has not yet been stored in this matrix.
End If
End If
End If
End If
Next node j
Next node i

Algorithm 3.6. Filling in of the minimum paths matrix

By recalling the example with the new reference time equal to 8:30 am and using
the intermediate point B no new relation AC is stored, insofar as the timetables
(8:30-9:45 am) from A to B and (9:40-10:20) from B to C are not compatible for a
connection.

Now by considering C, the AD link is not possible, insofar as AC is incompatible

as shown above. On the other hand, BD can be constructed with the timetables
(9:40-10:20 am) then (10:25-11:30).

By following the process of minimum path determination we have voluntarily

refrained the symmetric relations from D to A from the demonstration, since this is
incompatible with the creation of new paths. However, since the algorithm loops all
the points, these symmetric relations are also analyzed.

3.3.2.8. Process looping

The process then continues as long as the reference time calculated afresh
precedes the processing end time given by the user at the start.

The procedure of preparation of the minimum paths matrix (Algorithm 3.2) is

carried out again, then Floyd’s algorithm itself (Algorithm 3.4) and, finally, the
search for the new reference time and the filling in of the final paths matrix
(Algorithms 3.5 and 3.6).
60 Graphs and Networks

However, an additional procedure is inserted between the preparation of the

paths matrix and Floyd’s algorithm, thus substituting the preparation procedure of
the calculation of the timetable paths (Algorithm 3.3). This is done in order to avoid
recalculating the already determined paths that have not been stored in the final
paths matrix.

3.3.2.9. Creation of standardized files

At the end of the process a file of standard format is produced, in which each
column of fixed width refers to an element of the path description: names of the
stations of origin and destination, departure and arrival times, means of
transportation (single train or series of several trains), length of trip, connection
time, numbers of the start and end lines, codes of departure and arrival stations
specific for each one.

3.4. Conclusion: other developments of Floyd’s timetable algorithm

3.4.1. Determination of the complete movement chain

Before describing the timetable development of Floyd’s algorithm we have

raised the specific case of connection constraints as well as the necessary addition of
pre- and post-routing durations to the timetable matrix in order to refine the quality
of service concept up to a door to door processing. We will now describe these
initial developments, which have been operational since 1996.

To describe the algorithmic developments specific to connection constraints we

will recall a study carried out in 1996 on behalf of the DTT and the OEST, which
was tasked with defining a level of service grid for the SNCF and interior aerial
transportation system within the framework of evaluating the governing principles
of the transport infrastructure. In the general algorithm no constraints are applied to
the number of connections and the maximum waiting time; only the minimum
duration must be longer than two minutes. At national level it was necessary to
substitute these considerations by stricter constraints. For interior aerial transport
only one connection was accepted, with a maximum connection time of two hours
when there was no change of airport, or ranging between one and three hours on the
assumption of a connection with change of airport.

For rail transport two connections were accepted with methods exploiting a
variable duration depending on whether a city like Paris or a provincial town was
considered. In Table 3.1 we indicate the assumptions envisaged and introduced into
Floyd’s algorithm regarding the allowed connection times.
 Determination of Optimal Paths in a Time-delay Graph 61

In addition, the individualized knowledge for each graph node of the average
duration observed to reach a public transport station from home, or conversely to
reach home after leaving the station would yield a first response to evaluating the
quality of service over the entire movement chain. The lack of this type of data for
all the nodes, nevertheless, led us to consider approximate times which, however,
corresponded to a certain observable reality and were variable according to the size
of the cities and the means of public transport used for the main journey.

Time Time Total time

Connection 1 Connection 2
(in hours) (in hours) (in hours)

G6 ≤2 – – –

Gp1–Gp17 ≤2 – – –

Gp1–Gp28 1≤≤3 – – –

G – G – ≤2

Gp1–Gp1 – G – ≤2

Gp1–Gp2 1≤≤3 G ≤1 –

G – Gp1–Gp1 – ≤2

G ≤1 Gp1–Gp2 1≤≤3 –

Table 3.1. Connection constraints

Therefore, the initial travel times to reach an airport are traditionally longer than
those to reach a railway station. According to this principle and with the help of a
matrix of the nodes valuated by duration of the two final routes, the optimal paths
matrix is modified by bringing forward the departure times to integrate the duration
of preliminary routing and moving back the arrival times to take into account the
final leg of the journey. In addition, a time known as safety departure is added to
account for the “anxious behavior” of a user not wishing to miss his train or plane.
Two other developments stem from the above considerations.

6 G: connection at a provincial station.

7 Gp1-Gp1: connection in Paris without change of station.
8 Gp1-Gp2: connection in Paris with change of station.
62 Graphs and Networks

3.4.2. Overview of all the means of mass transport

First of all, Floyd’s algorithm, which has been specifically obtained to provide
an account of a means of transport with temporary functionality, is also adapted for
railways as for other means such as aerial, maritime, river transport or on another
spatial scale for transport in a city bus, tram, etc. The availability of timetables and
constitution of the corresponding matrices are the only constraints, which when
exceeded become open to possible comparative analyses of the efficiency of each
means of individual or public transport. This method of sector by sector analysis of
the means of transport does not at all overshadow the possibility of evaluating a
multimodal service incorporating all these means, in particular, to account for the
hub and spokes operation of aerial transport, frequently requiring pre-routing by
TGV, which in turn implies an initial journey on foot or by taxi from home.

3.4.3. Combination of means with permanent and temporary functionality

Secondly, the combined processing of means with temporary and permanent

functionality, which is possible by using the algorithm presented, can solve the
problem of approximating the lengths of the final journey. According to the nodal
zoom method [CHA 97] the creation of new intra-urban nodes (stations, airports,
sites attracting employment, housing, etc.) and of new arcs connecting these stations
or airports to workplaces or homes by using such means as walking, bus, urban
methods of transport, etc., considerably refines the movement chain by
individualizing each path determined by the algorithm. To take but one example, the
departure time from home to an SNCF railway station can thus be calculated on the
basis of the departure time of the main route, which is brought forward by a length
of time determined for an initial pedestrian arc. Other urban means of transport are
recognized in Floyd’s algorithm. Associated with various types of infrastructure
(generally extensive public roads system, at a site of generally high urban density, at
a usual site or a site with exclusive right of way), this prospect leads to
distinguishing three functional binomials for the bus (8, 15 and 25 km/h depending
on traffic density or the width of lanes), two binomials for the tram (18 and 45 km/h
depending on the type of material), one for the subway (24 km/h), or two more for
walking (3 and 5 km/h). The list is naturally dependent on the needs that have arisen
up until now, but remains perfectly open to other means (roller skates, etc.) or finer
differentiations (pedestrian on a conveyor belt, on a narrow or a broad sidewalk,
etc.).

Graph theory and Floyd’s algorithm offer a significant development potential,

which is limited only by the model creator’s preoccupation with precision. Thus,
why not imagine the processing of the performance of river and maritime transport
networks by considering the constraints of current, regulation speeds, starting with
 Determination of Optimal Paths in a Time-delay Graph 63

directed arcs with permanent functionality, or even more exactly using a series of
arcs with temporary functionality required by the effect of tides and closings
associated with locks and able to repeatedly prevent a ship from moving upstream?

3.4.4. The evaluation of a timetable offer under the constraint of departure or

arrival times

Another extension related more to an algorithmic modification than to a graph

problem still has to be specified here. In the introduction we have already posed the
problem of simulations using an additional constraint of departure or arrival times;
i.e. which means of transport should I choose if I wish to travel from Lyon to
Marseilles and be there before 8:00 am? The performances of the line assigned to
TGV at first glance appear as the most adequate alternative. Nonetheless,
introducing a constraint on the time of arrival at destination may lead to prefer the
use of an individual vehicle for the flexibility that this means of transport allows
while being free from departure time constraints.

The general algorithm already enables the determination of the set of paths for a
given time range. The remaining developments stem from these particular problems
of arbitration between several means of transport, depending on the imposed arrival
or departure timetables. They were implemented by Alain Hostis and Christophe
Decoupigny [LHO 00].

If a chronological presentation of the algorithmic development for a user is

preserved, the procedure of creating the minimum arcs matrix (Algorithm 3.1)
remains unchanged.

Then, depending on the imposed constraint of arrival time, the user may input
the reference arrival time, after which no path will be determined, since they become
of no interest for a user who does not wish to be late for a meeting or work, for
example. Let us recall that in the general case the user delimited the processing time
range using both departure and arrival times. Except for this modification, the partial
and provisional filling of the minimum path matrix (Algorithm 3.2) remains
unchanged.

The true change is observed in the main processing of the minimum arcs
(Algorithm 3.3).

If there is no arc valuated by a means of transport with permanent functionality,

it is the IF/End If loop indicated with an asterisk in the general procedure that is
modified. For a given arc (origin, destination, time) this is stored in the first layer of
the paths matrix only if the arrival time is earlier or the same as the reference time
64 Graphs and Networks

and if the departure time is later than that already found (meaning the possibility of
leaving the origin later, while minimizing the total duration of the journey).

If there is an arc valuated by a mode with permanent functionality, the

modification of the algorithm is found at the level of the double asterisk in
Algorithm 3.3. The duration of the arc with permanent functionality added to the
reference time in order to provide the new arrival time is now deduced from the
reference time to provide the departure time. This minor change has little
importance insofar as it is the durations that are really processed thereafter.
Nevertheless, in the actual calculation to determine the paths (Algorithm 3.4), it
makes it possible not to generate paths that can arrive after the arrival time
constraint when a permanent arc is added to a time arc.

Next, the determination of the minimum paths is carried out using a procedure
similar to Algorithm 3.4 except for the fact that for the general algorithm we have
looked for the optimal paths between i and j, for a given relation ik (k being the
intermediate node) by sweeping all of the relations kj. With the constraint on the
time of arrival for each given relation kj we work out the paths by sweeping the
relations ik, so as not to exceed the arrival time at j from k, which has been imposed
at the start.

In the following procedure (Algorithm 3.6), which is specific to the arrival

constraint, we gradually fill in the matrix when the arrival time of a path precedes
the reference time (whereas in the general case the departure time of a path had to be
earlier than the reference time, but the arrival time could be later). Moreover, the
path being processed is preserved only if, at the time of arrival equal to another path
for the same origin/destination, it leaves later, thus minimizing the total duration of
the path, while respecting the arrival constraint.

Lastly, we find the same looping already clarified for the general algorithm
which makes it possible to look for the reference time of the next departures, and so
on until no longer entering new minimum paths.

On the contrary, the constraint on the departure time is a simple alternative to the
determination of the minimum paths within a time range which is limited by
processing start and end times. The procedures of partial and provisional filling of
the minimum paths matrix (Algorithm 3.2), of processing the minimum arcs
(Algorithm 3.3) and of determining the paths (Algorithm 3.4) are carried out without
modification. However, the looping leading to the search for the next departure
within the time range and so on for all the ranges is not carried out. Indeed, at the
end of the first iteration all the paths that leave starting from the time indicated as
departure constraint are given.
 Determination of Optimal Paths in a Time-delay Graph 65

The possible applications for arrival and departure constraints are most
frequently based on the comparison between, for example, individual transport and
the railway transport. The arrival time being known, it is enough to compare the
possible departure time by car (by deducting from the imposed arrival time the
minimum transport duration according to its type) with that determined for public
transport (i.e. the latest departure time to arrive at the nearest possible time to the
one imposed).

3.4.5. Application of Floyd’s algorithm to graph properties

The description of the developments of Floyd’s algorithm or, more generally, of

the methods implemented to specify multimodal routing of the door to door type
would be incomplete without introducing the latest publications relating to the use of
graph properties to evaluate a transport timetable offer.

The quality of service of public transport is traditionally evaluated, as we have

demonstrated, by explicitly characterizing the relations between one or several
origins and one or several destinations. This approach is naturally relevant with a
view to timetable optimization, better spatial distribution of the relations, or
measuring the possibility of effectively and quickly linking the possible destinations
with a given origin, and thus evaluating the spatial diversity of the offer.

However, the study of graph properties, such as external and internal half-
degrees, enables to provide additional information by evaluating the importance of a
station, an airport, etc., according to the number of time relations carried out by
train, by plane, etc., arriving or departing from a city. Such calculations are
implemented on the basis of the matrix of minimum paths which are determined
beforehand using Floyd’s algorithm. This application is thus strictly speaking not a
development of the algorithm, but rather a new analysis of the results emanating
from it by partially answering to the deficiencies of graph theory to describe and
characterize the nodes.

In practice, for a rail transport network, for example, the sum of the relations
with a given city as destination (thus the number of arrival times) provides an
interesting indication about the convergent traffic (the internal half-degree) and from
that about the importance of the hierarchical position of a city in an urban network9.

9 Despite everything the results must take into account a certain observation, in the sense that
it is the number of arriving times from all the origins at a given destination that is measured,
and not the number of trains or planes really arriving at a station or at the airport. By way of
illustration, a train may very well leave Bordeaux, stop at Poitiers, then arrive at Tours, thus
accounting for two relations on arrival at the last station, but for only one train. Consequently,
66 Graphs and Networks

This indicator may be supplemented by opposite information, the external half-

degree, by summing the number of departure times from the same city, the
observation for its interpretation being in the same veins as for the internal half-
degree.

3.5. Bibliography

[BAP 99] BAPTISTE H., Interactions entre le système de transport et les systèmes de villes:
perspective historique pour une modélisation dynamique spatialisée, Thesis, Tours, 1999.
[BAP 02] BAPTISTE H., L’HOSTIS A., Analyse de la qualité de service du réseau de
transport collectif en région Nord-Pas-de-Calais et Languedoc-Roussillon, recherche
menée pour le compte du Conseil Régional de la région Nord-Pas-de-Calais, CESA,
Tours/INRETS, Villeneuve d’Ascq, 2002.
[BAR 86] BARTNIK G., MINOUX M., Graphe, algorithmes, logiciels, Bordas, Paris, 1986.
[CHA 97] CHAPELON L., Offre de transport et aménagement du territoire: évaluation
spatio-temporelle des projets de modification de l’offre par modélisation multi-échelles
des systèmes de transport, Thesis, Tours, 1997.
[COL 96] BAPTISTE H., LARRIBE S., MATHIS P. (ed.), Grille de niveau de service,
recherche menée pour le compte de la DTT/OEST, Rapport de recherche, Tours, 1996.
[LHO 97] L’HOSTIS A., Images de synthèse pour l’Aménagement du territoire: la
déformation de l’espace par les réseaux de transport rapide, Thesis, Tours, 1997.
[LHO 00] L’HOSTIS A., DECOUPIGNY C., MENERAULT P., MORICE N., Cadencement
et intermodalité de l’offre en transport collectif en Nord-Pas-de-Calais: analyse et
propositions d’amélioration, INRETS, Villeneuve d’Ascq, 2000.
[MAT 78] MATHIS P., Economie urbaine et Théorie des systèmes, Thesis, Tours, 1978.

the result must be interpreted as a measurement of the richness of the material offer coupled
with the diversity of possible origins.
 Chapter 4

Modeling the Evolution of a Transport System

and its Impacts on a French Urban System

4.1. Introduction

A quick analysis of the present French urban network shows a rather good
adequacy between the size of cities and the quality of transport infrastructure
servicing them. Taking into account the advanced state of the implementation of
various road and railway projects, the largest agglomerations will all be connected
by TGV connections and/or highway type links within a few decades [REY 93].

This development, whose foundations were laid around the 15th century with the
centralization policy implemented by Louis XI, accelerated appreciably since the
18th century with the development of postal roads [ARB 92, STU 95]. The
improvement of transport networks to the benefit of the largest cities made it
possible to set up an urban network of a higher level, to some extent providing the
latter with an advantage compared to their respective countryside, thus forming a
network among networks. These mechanisms initially found their origin in the
strong political will to control the territories which led to imposing an “invisible
hand” stretching from Paris towards the main French ports through ease of
communication [BRA 90b].

Today the various laws on regional planning still encourage greater accessibility
of the worst connected territories, still and always to the benefit of the capital or the
largest agglomerations [JOU 95]. However, a dominant evaluation method intended
to estimate the needs and choices with respect to the construction of transport

Chapter written by Hervé BAPTISTE.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
68 Graphs and Networks

infrastructure was added to the political aspect. Indeed, it is shown that

costs/benefits analysis leads to a pattern according to which the demand for transport
generates the resulting supply and tends to reinforce the existing tendencies with
respect to movements, as well as increase the polarization of territories (see [BAP
99], p. 56). Nonetheless, this mechanism supports the concentration of population on
a regional or national scale, which also results in the opposite in more restricted
areas where the increasingly lengthening intra-urbanization follows the
improvement of transport networks and the development of mobility.1

The main objective here is to estimate the long-term consequences of the

evolution of departmental transport networks on the spatial distribution of
alternating migrations and then on definitive migrations, but also its opposite, whose
processes tend to connect the best performing transport networks with the most
important cities. In other words, it is a matter of specifying, of dynamically
measuring the interactions between a departmental system of cities and the transport
system.

Modeling by using graph theory is an effective response to this type of question.

With respect to the multiplicity of interrelationships between several hundreds of
cities which are, in particular, manifested through a significant number of transport
infrastructure installations, the description of area by nodes and arcs offer invaluable
assistance in the understanding of such mechanisms. While by necessity simplifying
reality, the data processing tool in this context remains the most appropriate means
of analysis by simultaneously making it possible to take into account tens of
thousands of spatial and quantified data inputs.

4.2. Methodology: RES and RES-DYNAM models

4.2.1. Modeling of the interactions: procedure and hypotheses

Two types of models are used for this purpose: the first (RES) aims to create
theoretical areas from any part of a graph responding to many constraints in order to
reveal similarities with the areas observed. The second (RES-DYNAM) is a
calculation model intended to exploit the previously created graphs by implementing
processes of transport network modification, generation of commuting and definitive
migration.2

1 See [PUI 94] for a development of the polarization concept, its definition, manifestations
and causes.
2 These two models are more broadly clarified in [BAP 99].
 Modeling the Evolution of a Transport System 69

Modeling by using RES-DYNAM is based on two functionally and temporally

dependent processes (see Figure 4.1): on the basis of any graph (of theoretical or
real area) the optimal routes are determined following a temporal measurement of
the transport network (1)3. A gravity model4 is used at this level (2) to determine the
mutual influences exerted by the cities and the potential movement flows (in
particular the home-work commute), the latter being finally assigned to the various
arcs of the graph (3). The spatial configuration of the transport network, the quality
of the infrastructure play a major part in the distribution of movement, since space
acts as a brake for various flows.

This module qualified as static supposes an immediate determination of the

attraction of each city and an immediate adaptation of movement flows to the
transport supply. This philosophy of simulation, restrictive as it is, is, however,
widespread in this field because it makes it possible to model the observed flows
with unquestionable efficiency.

A dynamic module is then exploited leading to a redistribution of a fraction of

the population by definitive migration, according to a global mobility rate selected
for the area used (4). The decision to migrate definitively and especially the
destinations of each individual are determined according to the generally attractive
nature of each city, which is itself derived from the results of the gravity model.

However, a factor that proves attractive to one person may be repulsive for
another. For example, the attraction of a level of services, employment or important
infrastructure can lead to a migration towards a city of considerable size (the
correlation between size and the functional level of a city being significant), land
costs or quality of life are then not perceived as determinant. However, on the
contrary, a stronger attachment to the latter criteria can lead to a migration towards
smaller towns, which are close to a large employment pool, and reveal a diagram of
migration by centrifugal redistribution from large cities. In other words, these two
“residential strategies” are just one example among all the modeled motivations
integrated into the RES-DYNAM model. At this level, a looping is carried out (5):
the new settlement pattern of each city is placed at the origin of a new gravity
calculation which determines the new influences exerted by cities. Loop after loop
the hierarchy of city sizes changes, each one of these iterations being comparable to
a year when the selected global definitive mobility rate is coherent with real life
observations.

3 For a clarification of the model used based on Floyd’s algorithm see [CHA 97].
4 The model used here is derived from the model of Mathis [MAT 78] which is based on the
population and temporal distance measured in the network, respecting the hypotheses of
mutual influences between cities (attraction of cities taken two by two and being then
relativized with respect to all other cities) and of dissymmetry of flows.
70 Graphs and Networks

Until now, the hypothesis of temporal modification of the transport network

characteristics has not been used. Its integration is based on the existence of
saturation thresholds of the transport axes: after gravity calculation, determination of
potential movement flows and new partial population redistribution, the crossing of
the threshold would cause an infrastructure modification (6) using methods clarified
later on5. Once the transport network configuration has been modified the optimal
routes have to be calculated afresh during the following iteration (7), since the new
journey times are integrated in the gravity calculation. These durations of journey, as
well as the population, will necessarily intervene in distributing the influence
between cities and, in the long term, new population localizations. Thus, there is
clearly a dynamic interaction between the system of cities and the transport system.

Graphs STATIC MODULE

of
theoretical
areas Gravity model
Optimal routes
determination model (determination of interurban influences
of observed
and movement flows)
areas

Assignment of flows

Transport network modification model Definite migration

(modification or creation of additional arc) generation model

DYNAMIC MODULE

Figure 4.1. Schematic modeling diagram

The results are interpreted using a comparative method: two simulations are
carried out, one using a temporally invariable transport network, the other
considering the transport network evolution through the improvement of
performance along certain axes. The analysis by differences leads to a conclusion
regarding the role of transport networks in the processes of commuting and
definitive migration, the dynamic process of population redistribution being in turn
rigorously identical, just as the parameter values and the calculation hypotheses.

5 If the saturation threshold is not reached, the iterative process continues without
modification of the transport network (5).
 Modeling the Evolution of a Transport System 71

4.2.2. The area of reference

The graph of theoretical area used below, which is generated by the RES model,
is constructed on the basis of simple assumptions to observe a certain similarity with
the area of the Indre-et-Loire department.

The theoretical area of reference

a seedbad of 304 nodes and a partial
road network of 50 km/h roads

35,051 132,000 inha.

Partial network
CESA Laboratory, 1998
50 km/h road Created by: H. Baptiste, Ph. Mathis
With the collaboration of S. Larribe,
0 20 km A. L'Hostis, M. Mayaud, K. Serhini

Figure 4.2. Initial theoretical area

The theoretical area extends over 100 km from East to West and 110 km from
North to South and is described on the basis of 300 towns, which are located
randomly but, nevertheless, respect a minimal spacing of 5 km, which is the average
observed distance (Figure 4.2). Four towns were added afterwards in order to benefit
from a certain openness of the system of cities located between 80 and 100 km from
72 Graphs and Networks

the center of the theoretical area, which is a distance compatible with the spacing
observed between two prefectures of bordering departments6.

The populations were distributed according to the rank-size law, following a

form that makes it possible to preserve common points with the Indre-et-Loire
department: the total population of the theoretical area reaches 520,000 inhabitants
(not including the four peripheral towns), the main town – approximated to the
department prefecture – is assigned a population of 130,000 inhabitants, as well as
the four peripheral towns, thus making it the equivalent of the prefectures of the
bordering departments. The location of the second town by size is selected near the
first city, similarly to Joue-le-Tours spatially which is contiguous with Tours.

All the towns are connected by a partial infrastructure network7 of “50 km/h
roads”, which can be assimilated with respect to the maps published by Michelin to
roads captioned as: “regional or release roads” presenting strong sinuosity, or
“surfaced or unsurfaced road or road with bad viability”. These are two-lane axes,
whose width is most frequently three meters in each direction, where the average
speed is approximately 50 km/h. In addition, a planarity principle is adopted, such as
can be generally observed in real areas of this type with this kind of road network.

The interest of the presented theoretical graph then lies in enabling a certain
generalization of the conclusions obtained during simulations, while preserving a
degree of reality compatible with interpretations that could be made regarding the
departmental area of Indre-et-Loire alone.

Moreover, the process of transport network evolution involves additional partial

networks: the “70 km/h road” is captioned in roadmaps as a “main road” since the
speed selected here is an average value taking into account, in particular, stops,
crossroads and successive decelerations. The “60 km/h road” refers to the same
roadmap legend, but with stronger sinuosity and inducing a more limited average
circulation speed. Finally, the expressways and the highways are considered
separately according to the respective average speeds of 90 and 110 km/h.

6 Let us note, however, that the RES model enables the representation of more atypical
situations, like that of Alençon in the Orne department, which is located at the periphery of
the department, at a distance of about 50 kilometers from Le Mans.
7 A partial infrastructure network can be defined as the entirety of the infrastructures of a
given means of transport and presenting a certain homogenity particularly in terms of average
circulation speed (see [LHO 97], p. 60).
 Modeling the Evolution of a Transport System 73

4.2.3. Initial parameters

4.2.3.1. Static modeling

According to the description of the transport network given above, each arc and
each path from any node to every other node is assigned a minimum time value
measuring the optimal duration to reach all the nodes in the graph.

Specifically for average movement durations within the same town we choose a
time equal to 15 minutes, according to the results obtained beforehand during the
model calibration phase using alternating commutes in Indre-et-Loire in 1990 (see
[BAP 99], p. 306). The movements during simulations retained hereafter will thus
relate only to the home-work commutes. Only the active population with an
employment will be considered and determined on the basis of the population of
each town according to an average ratio of a number of people for each employee.

4.2.3.2. Dynamic modeling of population evolution

The dynamic process of evolution of populations in town follows the
characteristics observed between 1968 and 1990 in the Indre-et-Loire department
only for definitive internal migrations. However, the values of attraction indicators
were modified slightly to simulate an evolution towards a more marked area
polarization, in order to make the final results more perceptible. In other words, we
accentuated the attraction of the towns in terms of employment.

The total rate of final mobility is chosen to be equal to 5% per iteration8 for all
simulations. Each one of these iterations is thus – by the selected parameter values –
about one year long and the simulations relate to a period of 100 years.

4.2.3.3. Modeling of transport network evolution

The dynamic process of transport network improvement implemented for one of
the two following simulations works according to a “following” transport policy, in
the sense that it meets existing needs, represented by saturation of road axes, but
with a chronological delay between the initial circulation difficulties and the
implementation of correction measures.

In fact, we considered a period between 10 and 20 years depending on the case,

during which the infrastructure capacity was regularly exceeded before an
improvement could be offered to the users (in a network that he could use, the

8 That is, a probable value signifying a sufficiently strong mobility compared to the average
annual rate of 3.6% which was observed on an inter-commune scale in France from 1982 to
1990.
74 Graphs and Networks

construction period was included). The very basis of transport network evolution
thus has to be sought among the capacity values of each transport axis.

The capacity of road infrastructure corresponds to the maximum flow of vehicles

that it can support. On the one hand, it depends on the geometrical characteristics:
the number and width of lanes, the number of lines of traffic circulating in the same
direction, slope, visibility range and curvature radius. In addition, it is dependent on
the circulating vehicles, with respect to their types (cars, trucks with and without
trailers, etc.) and speed of circulation.

The practical capacity of an infrastructure is a measurement of the maximum

flow that still preserves good circulation conditions, in terms of speed, safety or
overtaking maneuvers. At this level of traffic the circulation starts to be disturbed,
the average speed of the vehicles decreases, the first indicators of obstruction
appear, but still under conditions which are considered bearable (see [COQ 72], pp.
140 and 146). This value thus corresponds to the flow limit, up to which the
infrastructure is sufficient. It corresponds to the saturation threshold, beyond which
stops in the traffic flow are more and more prolonged; the difficulties of circulation
increase requiring frequent braking and acceleration, which leads to reducing the
average speed of the vehicles.

The “possible capacity”, also called physical capacity or saturation flow, is the
most intense theoretical flow that the infrastructure can cope with, where the
vehicles are in a situation of “driving in a procession” ([COQ 72], p. 146): they
follow each other at intervals equal to the minimum safety distance, at reduced
speed. At this level of traffic any circulation incident (breakdown, accident,
crossroads, etc.) results in the full halt of the entire line of vehicles concerned.

The capacity is generally expressed as the number of vehicles per hour observed
at a point, the result then being extrapolated to a portion of the infrastructure
presenting equivalent characteristics. From this unit the capacity can then be
converted into the average number of vehicles circulating daily, by considering that
the normal rate during rush hour (repeated periodically during a short period of time,
such as during commuting migrations, for example) is equivalent to a sixth of the
average daily flow calculated over a year [COQ 72, p. 36]. The values of
infrastructure capacity mentioned further on are therefore not determined on the
basis of the largest flow values, if these are specific for a time of year (in particular,
holiday departures). The traffic observed during these exceptional rushes cannot
indeed be retained as a reference value, since economic conditions cannot
reasonably lead to over-develop the road networks to absorb certain exceptional
traffic jams.
 Modeling the Evolution of a Transport System 75

In addition, if the concept of average is recurrent in the determination of

infrastructure capacities, it is because the latter are implanted in geographical space.
Thus, topographic constraints (slopes, sinuosity minimizing visibility, etc.) make it
is extremely difficult to adopt the same value of capacity for the entire
infrastructures of the same partial infrastructure network. Moreover, many other
factors intervene to modify the capacity of the same road: the existence or absence
of obstacles at points of access (low walls, bridge railings), the percentage of trucks
or periodic traffic fluctuations.

For reasons of feasibility we, nevertheless, chose a single value for each of these
partial networks, each one of those having been determined to group infrastructures
with similar characteristics (in terms of width, number of lanes, sinuosity and slope).
Consequently, the bias is relatively weak.

Finally, the concept of economic capacity is used to determine a maximum

economically sustainable flow. Thus, a higher value in an urban environment is
acceptable, in the sense that the distances covered by users are small, urban
constraints are stronger that the maximum traffic is observed during a short period
of time during a day, all this leading to a raised saturation threshold. On the other
hand, the threshold is fixed at a lower value for interurban environments where
temporally durable saturation is no longer acceptable for long distances. Since the
arcs of the graphs connect different towns or cities, we then consider values related
to circulation in an interurban environment.

In any case, hereafter we have retained values relating to saturation thresholds

and not the “possible” capacity, insofar as we saw that the traffic flow using the
latter concept is extremely unstable, and unsustainable in an interurban network,
where the inconvenience caused to vehicles is no longer acceptable for longer
distances.

Finally, the criteria used to determine the saturation threshold of each partial
infrastructure network are:
– interurban circulation only;
– total value by type of partial network;
– value expressed as an average daily number of vehicles, since the traffic
figures assigned to arcs are not segmented by time;
– balanced circulation in both directions for infrastructures with undivided
highways, i.e. symmetric distribution of traffic in each direction. However, the
traffic seldom obeys this situation (in particular for the alternating commutes),
which leads us to overestimate the true capacity of these types of roads;
76 Graphs and Networks

– saturation thresholds determined for circulation include 10% of commercial

vehicles, which once again leads us to overestimate the real capacity of the
infrastructure, insofar as the distribution of road traffic in France by type of vehicle
shows that trucks, coaches and light commercial vehicles represented approximately
20% of the total traffic in 1996. However, we minimize the share of these categories
of vehicles because light commercial vehicles have a less considerable impact on
capacity reduction than trucks or coaches;
– value provided without crossroads, or any other device involving difficulties
for the traffic flow. Once again the indicated capacity will therefore tend to
overestimate the real figures.

Number of Number of
vehicles vehicles with
Partial
without 10% of Observations
network
commercial commercial
vehicles vehicles

Good geometrical characteristics,

70 km/h road 2,700 2,400 visibility more than 450 meters
at any point

Higher sinuosity (visibility lower

60 km/h road 2,400 2,100 than 450 meters for approximately
30% of the length of the road)

Reduction of the width

50 km/h road 2,100 1,800
of the lanes to 3 meters

Expressway 11,125 10,000 Twice two ways

Highway 14,400 13,000 Twice two ways

Table 4.1. Saturation thresholds for one direction9, average number of vehicles per day

Overestimating the real capacities is justified by the fact that the improvement of
transport networks or the construction of a new axis generally takes a relatively long
period of time, during which the true saturation threshold is exceeded. A higher
value then makes it possible to take this inertia into account in the overall response

9 Assumption of balanced circulation in interurban environment without taking crossroads

into account.
 Modeling the Evolution of a Transport System 77

time. On this basis the capacities retained for each partial network integrated into the
process of transport network evolution are determined (Table 4.1).

For the “70 km/h road” partial network the lanes are, with rare exceptions, broad
(3.5 meters) and come in twos (one in each direction). Roads with three lanes exist
in a minority and we have adopted a saturation threshold valid for a seven meters
wide road with good geometrical characteristics (visibility over 450 meters at any
point of the infrastructure, flat ground and weak slopes).

The capacity of such infrastructure is 900 vehicles/hour in both directions if we

consider that only cars are in circulation, in the absence of crossroads and in
interurban environment. By considering 10% of commercial vehicles the decrease of
the above capacity by extrapolation also amounts to approximately 10% of an
interurban road (see [COQ 72], p. 145). It is then equivalent to approximately 800
vehicles/hour in both directions, that is, an average of 4,800 vehicles per day and,
thus, 2,400 vehicles in each direction (considering balanced circulation).

For the “60 km/h road” partial network the increased sinuosity compared to the
previous partial network cuts down visibility values. Roger Coquand estimates that
when 20% of the stretch of road studied has a visibility below 450 meters, its
capacity decreases from 900 to 840 v/h; from 900 to 760 v/h for 40% of the length
of this road (see [COQ 72], p. 150). In view of these observations we choose a
capacity of 800 vehicles/hour (both directions, without commercial vehicles), i.e.
approximately 2,100 vehicles per day on average (with 10% of commercial vehicles
in each direction).

Roger Coquand determines the reduction in capacity between a type “70 km/h
road” infrastructure and a three meters per lane road, that is, a “50 km/h road”, at
23% (see [COQ 72], p. 149), which yields an approximate capacity of 1,800
vehicles per day and direction on average.

Infrastructures with a central separator have a higher capacity by lane in the

sense that flow limitation no longer depends on increased or decreased difficulty of
overtaking maneuvers. Nonetheless, traditional crossroads (left turns, etc.) may still
exist on expressways, whereas only interchanges exist on highways. However, if
transverse circulation is considerable, the traffic flow on the main roads may be
affected by it, thus decreasing the saturation threshold (see [COQ 72], p. 272).

The saturation threshold for a highway lane is fixed at 1,200 v/h. A two-lanes
highway track thus has an approximate capacity of 14,400 vehicles per day and
direction on average. Considering that expressways are an intermediate type of
infrastructure between highways and divided highways with four lanes without a
central reservation (capacity of 7,500 vehicles per day and direction [COQ 72, p.
78 Graphs and Networks

159]), we arbitrarily adopt for them a saturation threshold equal to the average
value, that is, 11,125 vehicles per day and direction.

The values indicated above apply to circulation made up exclusively by cars. By

extrapolation, according to methods presented previously, the capacities obtained
with 10% of commercial vehicles are decreased to obtain the following figures
(expressed as number of vehicles per day and direction on average, for four-lane
divided highways):
– expressway: 10,000;
– highway: 13,000.

The simulation of the transport network evolution during 100 iterations for a
theoretical area is obtained on the basis of capacity figures integrating 10% of
commercial vehicles.

The transformation itself of partial networks road (from 50 km/h to expressways

assigned an average speed of 90 km/h) is carried out by widening existing lanes,
thus by substitution of the successive networks, whereas a highway is added to the
existing networks. We have, nonetheless, rendered the mechanism more complex
according to the following assumptions:

The successive transformation of a “50 km/h” partial network into an

expressway is carried out in 10 year periods upon exceeding capacity; on the other
hand, the construction of a highway requires an interval of 20 years before it is
implemented. Then, by supposing an already stated construction schedule, the later
sections are placed at the disposal of users after 10 years.

This expresses quite well the idea, according to which “in general, a generation
needs to pass between the moment when the idea to implement new transport
facilities for a given service emerges and when this idea becomes popularized and is
ready to become operational” (see [MER 94], p. 6).

The durations taken into account here are, however, lower, insofar as the times
chosen for this application stem from a transport policy which is based more on
permanent adjustment to the movement demand – but shifted temporally – than on
real long-term planning. However, we roughly respected the financial constraints
related to the differentials of capital costs according to infrastructure types, leading
to higher time intervals before the installation of a highway stretch10.

10 Merlin thus evaluates approximately the capital costs (in francs, in 1990) at 1.5 million
francs per kilometer for a road in town (the equivalent of the “50 km/h road” partial
network), 3 million for a secondary road with 2 lanes (“60 km/h road”), 8 million for a trunk
road with three lanes, 18 million for an expressway with 2×2 lanes, and above all 25 million
 Modeling the Evolution of a Transport System 79

Table 4.2 shows in detail and in chronological order the various terms, during
which infrastructure capacity is exceeded, before any improvement or construction.

Necessary Type of
Initial network Resulting network
duration (years) modification

50 km/h 60 km/h 10 improvement

60 km/h 70 km/h 10 improvement

70 km/h 90 km/h (expressway) 10 improvement

90 km/h (expressway) 110 km/h (highway) 20 construction

then 90 km/h
110 km/h (highway) 10 construction
(expressway)

Table 4.2. Transport network modification assumptions

4.3. Analysis and interpretation of the results

4.3.1. Demographic impacts

First of all, we show the final configurations obtained for the system of cities
under consideration before clarifying its generating mechanisms.

Demographic trends which were only attributed to the inequalities in spatial

distribution of definitive migration at the end of 100 years show an increase in urban
concentration regardless of the hypothesis, but following different patterns (Figure
4.3):
– by densification of the main town following the hypothesis of a transport
network invariable in time (this town increasing from 132,000 inhabitants to
181,986 inhabitants11);

for a highway (see [SEA 94], p. 227). Since in the application the expressway substitutes
existing roads by simple improvement, we considered a 10-year modification just as for
networks of lower hierarchical levels despite the higher construction costs.
11 If we continue the simulation for an additional 50 iterations, the population of the main
town stagnates, to the benefit of its primary crown, which is explained by a saturation
threshold caused by an elasticity lower than 1 of one of the attraction indicators; nevertheless,
peripheral diffusion remains very limited beyond the primary crown.
80 Graphs and Networks

– by peripheral diffusion of housing in the second simulation (see right-hand side

of Figure 4.3), with decreasing intensity as the distance from the city-center
increases, which is hereafter compared to the prefecture of the theoretical area (its
population decreasing from 132,000 to 129,632 inhabitants).

Simulation following an invariable transport network Simulation according to a dynamic transport network

Partial network
110 km/h highways
70 km/h road
19,094 181,986 inha.
60 km/h road 9,123 30,791 129,632 inha.
50 km/h road
CESA Laboratory
Laboratory, 1998 0 20 km
Created by H. Baptiste, Ph. Mathis
With the collaboration
With the collaboration of
of S.
S. Larribe,
Larribe,
A. L'Hostis,
A. M. Mayaud
L’Hotis, M. Mayaud and
and K.
K. Serrhini
Serhini

Figure 4.3. Populations after 100 iterations

Nevertheless, for both hypotheses the peripheral zones of the area suffer from
demographic loss due to the effect of definitive migration in the direction of the
large employment pool. on the other hand, on the basis of Figures 4.4 and 4.5, we
note that regardless of the chosen simulation hypothesis the hierarchy of the sizes
for the 300 towns (not including the four bordering prefectures), according to the
rank-size law representation, is generally respected over time.
 Modeling the Evolution of a Transport System 81

Hierarchy of town sizes after 100 iterations

(simulation with an invariable transport network configuration)
1,000,000
Town population (logarithmic scale)

100,000
Town population
Linear regression straight

10,000

1,000

100

10

1
1 10 100 1,000
Town ranking (logarithmic scale)

Figure 4.4. Rank-size distribution of towns with an invariable network

Hierarchy of town sizes after 100 iterations

(simulation with an invariable transport network configuration)
1,000,000
Town population (logarithmic scale)

100,000
Town population
Linear regression straight

10,000

1,000

100

10

1
1 10 100 1,000
Town ranking (logarithmic scale)

Figure 4.5. Rank-size distribution of towns with a dynamic network

From a negative slope initially equal to 1.2, the slopes are respectively reduced
to 0.95 for an invariable configuration of the transport network and 1.15 by
considering the dynamics of the transport networks. The primacy of the main city is
82 Graphs and Networks

especially clear in the first case, whereas the intermediate levels are much better
represented in the second.

4.3.2. Alternating migrations revealing demographic trends

In order to clarify the mechanisms leading to the trends above, at this stage we
detail the orientations of alternating morning migration flows, i.e. the proportion of
active population with employment moving from one town to another following the
indicated arcs, and the proportion of active population remaining in their town of
origin, according to the color of the nodes (Figure 4.6). The initial theoretical area
shows a strong polarization of the main town and its primary crown, which is more
marked than for the Indre-et-Loire department. Some other towns remain attractive,
but flows between peripheral towns are very weak. The towns sending more than
40% of their active citizens outside their territory are few in number; only the
prefecture and a town located in the east of the area fall into this profile12.

Alternating migrations in the initial Alternating migrations after 100 iterations.

theoretical area (iteration 0) Simulation using an invariable transport network

Alternating migrations after 100 iterations.

Simulation using a dynamic transport network
Propor tion of active popu lation w ith employment
and working i n anot her tow n (in % )

between 2 and 10 more than 25 to 40

more than 10 to 25 more than 40 to 100

Propor tion of active popu lation w ith employment

and working i n their tow n of residence (in % )

between 2 and 10 more than 25 to 40

more than 10 to 25 more than 40 to 100

0 20 km
CESA LAboratory, 1998.
Created by: H. Baptiste, Ph. Mathis

Figure 4.6. Flow of alternating migrations which are not assigned on the transport network

12 However, we did not represent the directions of morning flows for map legibility purposes.
 Modeling the Evolution of a Transport System 83

After 100 iterations the structures of attraction are appreciably different

according to the two hypotheses. The polarization increases in both cases, which is
visible through the strong reduction in the number of towns at the periphery of the
area offering employment to more than 25% of their active population (black
circles), and the increase in the number of towns sending more than 40% of their
active population to the center of area (black thick arcs) thus indicating a major
reduction in their own attraction. However, the distance at which this polarization is
felt varies quite clearly depending on the hypothesis.

For an invariable transport network configuration, for which the single average
speed of circulation is 50 km/h, the towns sending more than 40% of their active
population to the prefecture or its close periphery are located within a perimeter of
approximately 10 to 20 km around the center, as opposed to 20 to 30 km for a
transport network undergoing successive improvements.

In addition, despite the greater area polarization and regardless of the hypothesis,
after 100 iterations we observe the conservation of certain spatial diversity of
employment for a dynamic transport network in towns located at the periphery of
the area: those offering employment to 25% of their active population, or even up to
40%, are still rather numerous compared to the initial area. On the other hand, for an
invariable transport network almost all of these towns provide a little less than 10%
(a significant number preserving even less than 2%, which is visible on the map by
the absence of a circle). This is explained by a smaller number of definitive
migrations originating at these small peripheral towns for a dynamic transport
network and could mean the conservation of a certain number of day-to-day
services, this being a point which we will reconsider later on.

On the other hand, the decreased influence of peripheral towns during the first
simulation with invariable transport network is explained as follows: since the
distance resistance remains strong in the first area where the spatial concentration of
employment increases, the residents of towns far away from the employment pool
migrate definitively to the close periphery in order to go to work there. In addition,
exchanges between the towns of this area and the four other bordering prefectures
are negligible (less than 2% of active population sent by each of the 300 towns,
including the prefecture, and most often less than 0.5%).

The closest towns (located at about 50 km) are mainly attracted towards the
prefecture of the studied area, which is better described and takes into account a
wider employment pool. On the other hand, the inter-departmental exchanges with
the main city of the area require approximately and 1.5 hour-long journey in the
morning, which plays here a strongly discouraging role; with the improvement of
the transport network under the second hypothesis – decreasing this journey time to
84 Graphs and Networks

less than 45 minutes – this type of exchanges becomes largely favored, as we will
see in the map of flows assigned to transport networks13.

4.3.3. Evolution of the transport network configuration

Up to now we have limited the analysis to the clarification of modifications with

respect to spatial orientation of alternating migration flows between cities within the
area of reference and those obtained after 100 iterations following the two
assumptions regarding transport networks. The results reflect the way in which the
definitive migrations take place, being strongly dependent on this scale on the sites
of employment, which, finally, contribute only “to emptying” the peripheral zones
of the theoretical areas of their inhabitants to a larger or smaller extent.

After 20 iterations After 50 iterations

After 80 iterations After 100 iterations

Partial networks
110 km/h highway
90 km/h expressway
70 km/h road
60 km/h road
50 km/h road
0 20 km CESA Laboratory, 1998.
Created by: H. Baptiste, Ph. Mathis

Figure 4.7. Evolution of the transport network

13 Indeed, no commune of the area taken individually sends 2% or more of its active populace
to the bordering prefectures, regardless of the hypothesis upon 100 iterations. On the other
hand, the number of communes nonetheless participating in these exchanges is much larger
for the hypothesis of a dynamic transport network. For this reason only assigning flows to
arcs makes it possible to visualize it.
 Modeling the Evolution of a Transport System 85

It remains to demonstrate the evolution of the transport system during the 100
iterations of the dynamic simulation, in order to compare side by side the two
evolutions of the system of cities observed. We present here the transport network
configurations for several milestone dates (Figure 4.7). Since political decisions
regarding transport are taken according to the intensity of circulation flows, the
situation after 20 iterations is logical; since considerable flows of alternating
migrations converge towards the prefecture of the theoretical area, the capacity of
the corresponding axes is systematically exceeded in the mornings, which otherwise
constitutes the maximum rush hour of a weekday.

The 60 km/h road partial network prolongs the star diagram of 70 km/h roads. At
this level, if the definitive migration flows converge towards the center of the area –
not being able to easily carry out a daily commute towards the employment pool
using the 50 km/h roads transport network – on the other hand, the towns located at
the periphery of the prefecture, which were recently connected to a network of 70
km/h roads, can from now on accommodate definitive migrants searching for a more
pleasant lifestyle, more affordable land or attracted by the possibility of home
ownership.

This pattern makes the intercity towns more attractive and continues to spatially
widen according to the movement possibilities provided by the quality of the
existing infrastructure.

The connection between the bordering prefectures and the center of the area is
more visible after 80 iterations, except for the southern part, in the form of
successive sections of yet unequal quality due to the unfinished construction of the
highway between these large cities. At the end of the simulation they are fully
connected by the highway and other sections converge towards the prefecture of the
studied area from the initially largest towns in a radius of approximately 30 to 40
kilometers. We have to point out that the improvement or creation of an
infrastructure is final for the duration of the simulation; in other words, a
hypothetical reduction in the population of the connected towns may lead in the long
term to the network being under-used.

Thus, transport infrastructures are built for the long term; “those that will be
decided upon and financed at the end of the 20th century, will have to be used for
generations, even centuries” (see [SEA 94], p. 208). It is thus from this point of view
that the evolution of the transport system during the 100 iterations had been
“creative”, in the sense that the infrastructure improvement or construction projects
were implemented irreversibly.

In any case, the distribution of the town sizes at the end of this simulation is fully
explained by the increase in the polarization of the area around the prefecture; but
86 Graphs and Networks

the transport networks contribute to spatially widen this perimeter, by means of the
concentration process, i.e. suburbanization. The improvement of these networks thus
facilitates maintaining certain population diffusion around the main pole.

By using only the mechanisms of population dynamization – generation of

definitive migrations – and transport networks – delayed temporal adjustment
shifted from supply to demand – we quantify the role of transport networks
configuration for the spatial extension of the city (in its broader meaning of an
agglomeration) and the greater population diffusion within the theoretical area under
consideration.

We will not draw any conclusions here on the realism of the consequences
highlighted by comparing the two areas, with and without modification of road
networks. Indeed, the increasingly marked peripheral extension of cities on areas
that had previously been agricultural or presented natural or historical value, clearly
poses the problem of deterioration and induced effects, such as the costs of
decontamination, in particular.

On the other hand, the participation of the towns affected by this sort of
demographic trends towards the periphery can be a contributing element to ensure
the conservation of certain day-to-day services and thus of employment, in an area
larger than just the agglomeration, if the PLU (a French official urban planning
document) or, failing that, the MARNU are coherent with a controlled urbanization
and an active management of natural or historical areas.

4.4. Conclusion

Finally, we may also draw certain conclusions with respect to several other
elements which are particularly visible in Figure 4.8. Apart from the already
mentioned strong convergence of the alternating migration flows towards the
prefecture of the studied area, we also observe a fairly clear pattern of directed flows
following a circular configuration of the periphery/periphery type, which indicates
the importance in terms of employment of zones external to the city center and the
exchanges occurring between them14. In addition, the considerable size on the map
of the upper semicircles (sum of incoming flows) for nodes located in the primary
crown around the main town also proves their economic importance, which is also
an observable pattern in real areas, such as the towns of Joue-les-Tours or
Chambray-les-Tours located in the immediate periphery of Tours in Indre-et-Loire.

14 We recall that on the map considered, the direction of home-work commute flows follows
the rule of right-hand circulation; in this context, bold lines represented on the “right lane” of
certain arcs connected to the city center indicates a strong convergence of morning commutes
towards it.
 Modeling the Evolution of a Transport System 87

Flow of alternating commutes from the home to the workplace

assigned to the transport network after 100 iterations.
Simulation with a dynamic transport network

Flow (number of movements)

1,000 500 100

The directions of the flows are determined
between two nodes by the line: in all cases
the flows "drive" on the right (only flows
of >= 20 movements are represented)

Sum of the incoming flows (maximum: 39,610)

Sum of the outward flows (maximum: 44,391)
0 20 km

CESA Laboratory, 1998.

Created by: H. Baptiste, Ph. Mathis

Figure 4.8. Alternating migration flows assigned to the transport network

A transport policy based exclusively on a short-term vision of adjustment to the

transport demand really helps to explain the elements stated in the introduction with
respect to the evolution of the transport system configuration: from this point of
view and excluding any political will controlling the territory, the city “copes” well
with transport, the city models the spatial configuration according to its
requirements – strong demand – manifested through the continuity of the highway
network of the prefecture under consideration towards the four bordering cities of
the same size and radially towards its “hinterland”. Besides, we note in Figure 4.8
the attraction exerted by these cities on the studied area thanks to the boldness of the
lines converging towards them.

Although the number of daily incoming and outward commuters of the towns
connected to a highway interchange is quite considerable during the morning
journey, we note a real variety of situations in terms of balances15: regardless of the
highway networks considered (connecting the prefecture of the area to each of the
four cities in the bordering departments) the migratory balances are sometimes
positive, zero or negative. No regularity can really be discerned thus verifying the
assumption, according to which linear causality between infrastructure and

15 Let us specify that each commune located along the highway has an entry and exit
interchange.
88 Graphs and Networks

economic development must be rejected and no automatic reaction can be

declared16. The highway can successively generate development (highlighted by the
attraction of migrants, or a strong positive balance), provoke a significant number of
alternating migrant departures to make up the deficit by reciprocal flows (reflecting
a supply related to weak employment and therefore a low number of activities).

However, the balances of alternating migrations, whether they are positive or

negative, are increasingly large on this scale in towns with a highway interchange
than in the interstitial zones located between two of these axes. Moreover, these
towns are slightly larger in size, which is explained by the fact that their populations
may either work there when the creation of an interchange resulted in development
or reach one of the prefectures with greater ease every day. From this point of view,
there clearly is an amplification of the transport network impacts, which are either
positive or negative, resulting in population gains or losses, more intensive
alternating movements for the towns connected by a rapid transport infrastructure, as
successfully demonstrated by Plassard [PLA 77].

4.5. Bibliography

[ARB 92] ARBELLOT G., Autour des routes de poste: les premières cartes routières de la
France, XVIIe-XIXe siècle, Bibliothèque Nationale/Musée de la Poste, 1992.
[BAP 99] BAPTISTE H., Interactions entre le système de transport et les systèmes de villes:
perspective historique pour une modélisation dynamique spatialisée, Thesis, Tours, 1999.
[BRA 90a] BRAUDEL F., L’identité de la France: Les Hommes et les Choses, volume 1,
Flammarion, 1990.
[BRA 90b] BRAUDEL F., L’identité de la France: Les Hommes et les Choses, volume 2,
Flammarion, 1990.
[BRE 98] BRETAGNOLLE A., PUMAIN D., ROZENBLAT C., “Space-time Contraction
and the Dynamics of Urban Systems”, Cybergeo, no. 61, 1998.
[CHA 97] CHAPELON L., Offre de transport et aménagement du territoire: évaluation
spatio-temporelle des projets de modification de l’offre par modélisation multiéchelles des
systèmes de transport, Thesis, Tours, 1997.
[COQ 72] COQUAND R., Routes: circulation, tracé, construction, Livre 1: circulation, tracé,
Eyrolles, 1972.

16 This aspect of automation, of simple causality between infrastructure level and socio-
economic development was the object of numerous publication, such as: [VOR 84, PLA 92],
sometimes purely and simply calling into question the very concept of effect: [OFF 96] p. 52,
where Offner speaks of congruence by considering that the infrastructure arrives at the
moment when the socio-economic context of the newly connected city generates needs that
cannot be satisfied that by this infrastructure.
 Modeling the Evolution of a Transport System 89

[DOC 69] DOCKES P., L’espace dans la pensée économique: du XVIe au XVIIIe siècle,
Flammarion, 1969.
[FIE 86] FIERRO-DOMENECH A., Le Pré Carré: Géographie historique de la France,
Robert Laffont, 1986.
[FLA 89] FLAMANT M., Dynamique économique de l’histoire: deux siècles de progrès,
éditions Montchrestien/E.J.A., 1989.
[JOU 95] JOURNAL OFFICIEL DE LA REPUBLIQUE FRANÇAISE, lay no. 95-115, 4
February 1995 d’orientation pour l’aménagement et le développement du territoire, 5
February 1995.
[LHO 97] L’HOSTIS A., Images de synthesis pour l’aménagement du territoire: la
déformation de l’espace par les réseaux de transport rapide, Thesis, Tours, 1997.
[MAT 78] MATHIS P., Economie urbaine et Théorie des systèmes, Thesis, Tours, 1978.
[MER 91] MERLIN P., Géographie, économie et planification des transports, PUF, 1991.
[MER 94] MERLIN P., Les transports en France, La documentation Française, 1994.
[NAV 84] NAVARRE F., PRUD’HOMME R., “Le rôle des infrastructures dans le
développement régional”, RERU, no. 1, p. 5-22, 1984.
[OFF 96] OFFNER J.M., PUMAIN D. (ed.), Réseaux et territoires: significations croisées,
éditions de l’aube, 1996.
[PLA 77] PLASSARD F., Les Autoroutes et le Développement Régional, PUL/Economica,
1977.
[PLA 92] PLASSARD F, “L’impact territorial des transports à grande vitesse”, Derycke P.H.
(ed.), Espace et dynamiques territoriales, Economica, p. 243-262, 1992.
[PUI 94] PUIG J.P., THISSE J.F., JAYET H., Enjeux économiques de l’organisation de
l’espace français: 1. Polarisation et concentration, CESURE, 1994.
[REY 93] REYNAUD C., OLLIVIER-TRIGALO M., Tendances du transport européen et
besoins en infrastructures, INRETS, 1993.
[STU 95] STUDENY C., L’invention de la vitesse: France, XVIIIe-XXe siècle, Gallimard,
1995.
 PART 2

Graph Theory and Network Representation

 Chapter 5

Dynamic Simulation of Urban

Reorganization of the City of Tours

In Tours, at the city’s “Maison des sciences”, there is a group of researchers

comprising archaeologists, geographers and urban developers, whose objective is to
compare the points of view, the contributions of various disciplines on the same
subject in a very free, exploratory and possibly iconoclastic way. This group chose
the evolution of the city of Tours as a theme, on the one hand, because the known
data is immediately available and, on the other hand, because the subject concerns
all three disciplines. The selected area is illustrated by the plans of 1619, the map of
1770 (Figure 5.1) and the enlargement of the city of Tours on the same plan. These
three plans demonstrate the evolution of the network, but also the stability of
urbanized space.

The general hypothesis of this work is as follows: it is possible to model the

changes of the urbanization axis of the city of Tours. The traditional East-West axis
developing along the Loire is gradually being supplanted by the new North-South
axis during the second half of 19th century, whereas the Cluzel opening exists since
the 18th century.

That implies taking into account:

– construction of the road of le Pont de Pierre and of the Grammont dam as a
precondition for the considered reorganization of space, as shown by the map of
1770, at which date the city still had its former structure;

Chapter written by Philippe MATHIS.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
94 Graphs and Networks

– evolution of road and river transits, as well as a population increase;

– a very detailed description of space on the basis of circulation networks and
localization of populations.

Figure 5.1. 1619 and 1770 plans of the city of Tours

 Dynamic Simulation of Urban Reorganization of the City of Tours 95

Figure 5.2. Land parcels in the city of Tours, Napoleonic land register of 1836

In short, it is necessary to answer the question: how do we go from what we call

state 1 of the system, which is described above by the Napoleonic land register
(Figure 5.2), to state 2 of the system reflected in the drawing from 1855 below,
where we already see the buildings on the present day Jean Jaurès square, which is
to become the center of the city of Tours, the unloading dock or first Tours train
station built outside the city, at the edge of the mall that replaced the fortifications.
However, this urban expansion remains very feeble and we may note that many
intra-mural spaces are still free of constructions.

More precisely, the general hypothesis involves the following specific question:
does merely taking into account flow modifications, populations and their locations
make it possible to account for the phenomenon by means of a simple dynamic
simulation model?

We then transform this specific question into a work hypothesis enabling a

rationalization by means of land values, which enables one of the possible
formalizations of the problem.

Figure 5.3. 1855 drawing: oblique wiew of Tours from a tethered balloon
96 Graphs and Networks

The precise work hypothesis will then be the following: a differential evolution
of transit flows involves an evolution of the land price patterns, which in turn
induces a modification of population localization. This localization modification in
turn involves an evolution of local flows, which reinforces the effect of the
differential evolution of transit flows and so on; the very type of circular causality of
explosive nature described by Denise Pumain.

Technically these various modifications are relatively easy to include in

D.LOCA.T1, which is the simulation model used here. However, that raises several
problems related to the data:
– the precise localization of the population, the distribution of census data;
– the problem of lands whose values we do not have immediately; it would be
possible to collect them by using notary records, but that is an arduous job for a
historian that does not fall within our abilities;
– the problem of traffic which again points to the lack of immediately usable
data.

We will thus deal with the problem of data in the first part and in the second we
will deal with the model and its transformations, so as to expose in the third part the
most interesting results of our simulations.

5.1. Simulations data

As far as the population is concerned, we have used the results of censuses

published by the INED2. We have the censuses from Year II to our days and prior to
that data concerning fires.

This data is naturally global. To distribute it spatially we have the Napoleonic

land register of 1832 and the works of the UMR Archeology determining the
coefficients of land use on the basis of this diagram (Figure 5.4).

1 D.LOCA.T Dynamic of Territorial Localizations, Mathis, patented software, is the

prototype of a certain number of models of the CESA laboratory. It originates in various
GAST models [MAT 78].
2 INED population census.
 Dynamic Simulation of Urban Reorganization of the City of Tours 97

0.14 and less

0.15 – 0.29
0.3 – 0.44
0.45 – 0.29
0.6 – 0.74
0.75 – 0.89
0.9 and more

Figure 5.4. Coefficients of land use established by the UMR

Archeology under the direction of Galinié

Thus, upon analysis with our archeologist colleagues, there was a possibility of
decentralized population distribution in the urban area whose approximation
would be relatively operational. The hypothesis on which this spatial distribution
of the population of the city of Tours rests is as follows: the population is
proportional to the coefficient of land use according to the Napoleonic land
register. This, of course, remains a simplification that does not take into account a
certain number of factors, such as the difference in density related to incomes and
the status of the people or households, the existence of mixed surfaces, housing
and activity etc. However, albeit simplified, this spatialization seems acceptable as
a basis for a simulation, all the more since we had no other possibility in the short
or medium term.

Regarding transport, we knew of the existence of several networks according to

the old plans, in particular, the map of 1770 for the paths outside the city and the
land parcel chart of 1836, which was corroborated by the drawing of 1855. The
present network has already been modeled by the CESA Laboratory3.

We were then faced with the difficulty of choosing a network: should we take
into account only the network in existing 1850 or could we retain the current
network? In the second case we tend to induce the result that we wished to simulate,
whereas in the first case we complicate the simulation problem by adding another

3 EA 1376 then EA 211; see Chapter 13.

98 Graphs and Networks

dimension: that of simulating network generation4 with all the possible delay or
anticipation problems that this implies.

Initially we adopted the second solution. Indeed, the current network is the
product of evolution and if this introduces temporal auto-correlation, then we avoid
distortion and complication. The primary object of the model is to simulate the
evolution of population localizations according to the differential evolution of the
two largest means of transport and exogenous interventions. If we add the network
development simulation, which was carried out afterwards5, the problem clearly
becomes more complicated and we avoid doing that at first, taking into account
possible substitutions to measure the importance of a particular phenomenon.

The populations were localized at network nodes. The population has thus been
distributed through the current network according to the land usage coefficients
determined on the basis of the 1836 Napoleonic land register. This does not
introduce practically any bias for the old town, which has remained identical until
the rebuilding after the Second World War. In fact, the nodes punctually localize a
population that actually occupies a certain area, whose intake capacity can be known
and which can be determined by taking the dual of the graph comprising the road
network6.

Duality makes it possible to pass from the network to the surface, when the
holding capacity of a surface (here a set of pieces of land) is saturated. We slip from
a transport problem to a percolation problem, with stock taking increasing
importance in terms of significance, at the expense of the flows that become of
secondary importance, which is normal when the observation period is extended.

Once the hypotheses have been specified, the origin of the simulations was fixed
by using the 1855 drawing7 reproduced above: the city of Tours seen from a
balloon. In this document we can note that the evolution of the city has just started at
this date, whereas the unloading dock, i.e. the old Tours train station, is already
constructed, as well as the railway connection to Paris.

4 The problem of network extension by auto-similarity had not yet been solved at the time of
this simulation.
5 See Hervé Baptiste above for the development in capacity and speed, and Philippe Mathis
for the growth of networks by using fractal properties: see below in Part 3.
6 The other solution would have been to localize the populations at the nodes of the dual
graph of the road network, which would have corresponded to localize them at the “centroid”
of the surfaces of the road network. This solution would have been preferable, since the
inhabitants are localized in the small islands and not in the networks, but the algorithmic
solution has not been found for this case yet either. Actually, these are, among others, the
difficulties involving the more theoretical aspects exposed in Part 3.
7 1855 drawing.
 Dynamic Simulation of Urban Reorganization of the City of Tours 99

The stability of the urban structure until this time encouraged us to modify our
specific work hypothesis and to simulate the evolution of the city only by starting
from 1851, since from this date we have networks, localization of populations and,
most importantly, three means of transport instead of two: the river, the road and the
railway.

Moreover, transport has led us to partially reconsider our point of view. Indeed,
we have a good indicator of population localization and thus of an estimate of land
values but, on the other hand, we do not have at our immediate disposal precise data
on traffic volumes and especially on their differential evolutions. To take this fact
into account we reversed the problem: instead of simulating the effect of differential
transport evolutions on urbanization, we sought the hypotheses to keep in order for
the model to really simulate urban evolution, such as it was retraced by the evolution
of the population. With the work hypothesis specified, in the following section we
quickly expose the model used for this simulation and its adaptations.

5.2. The model and its adaptations

For this new type of simulations we have used the D.LOCA.T model, which was
appropriately modified for this experiment.

5.2.1. D.LOCA.T model

The D.LOCA.T model developed by the author is the predecessor of the

laboratory models, since its premises date back to the 1970s. Written in Fortran it
was, on the one hand, modified and adapted to the changes in the capacity of the
following computer generations and to the considerable modifications in the
programming language from Fortran IV to 90 and, most of all, it was used
systematically to demonstrate the possibility of simulating a particular type of
problems. However, quite unlike its efficiency in calculation, its lack of ease of use
caused the development of models in Visual Basic, which are better adapted to the
current standard with a strong graphic dimension.

At the start of this research this model comprises six stages: a space description
module using graph theory and taking into account the multimodal aspect of the
transport system, a module of calculation of the “shortest” paths in terms of
distance, journey duration, cost or a combination of these characteristics, a gravity
module generating the origin/destination matrix, a modal assignment module, a
space assignment module and a module of evaluation of certain consequences8.

8 See Chapter 8.
100 Graphs and Networks

Using dynamic allocation the model adapts to any type of space: from the city
center to Europe. It is dynamic, simulating by successive iterations from period to
period. Its present limits are, on the one hand, the capacity of the computer used and,
on the other hand, the absence of graphic aspects. The latter point is mitigated by the
use of Laboratory models and, in particular, of MAP9. D.LOCA.T was used for
calculations and, in particular, successive iterations, where its speed of calculation
could be used to the full. After each iteration it wrote on disk the files with the
results in the standardized laboratory format; files that were read and transformed
into maps by MAP. Thus, we did not mitigate the absence of graph capacity in
Fortran 90, but rather a greater difficulty to use them as compared to the highly user-
friendly capabilities of Visual Basic in this area. Both types of models were then
fully and logically used.

The description method of the area is multimodal and multiscale. It is based on

the zoom method with quasi-fractality10. In the present research the possibility of
multiscale justifies the general hypothesis: we can effectively simulate the two
levels simultaneously, which implies, in fact, studying a certain number of
consequences, such as the auto-transformation of flow variables, as well as the “state
variables”11.

With respect to the module of calculation of the “shortest paths”, which is based
on Floyd’s algorithm, the present version takes into account primarily road traffic at
a uniform intra-urban speed, which corresponds to the reality of the time: horse-
drawn or pedestrian circulation at practically identical speeds.

The gravitating model is a model of general access, an extension of the Reilly

model used since 1973, whose efficiency has been recently confirmed by tests
conducted by Chapelon [CHA 97] and Baptiste [BAP 99].

Spatial assignment is easy thanks to Floyd’s algorithm that simultaneously

provides the shortest path and the node preceding the destination node for all pairs
of nodes. From there on the procedure is very simple: from one previous node to the
other we reach the origin.

The modules of analysis of the consequences were not taken into account for this
application, since the important consequences that we wish to simulate, the
modification and extension of the residential zones, were not yet tackled then.

9 Alain L’Hostis.
10 Laurent Chapelon. See also in Part 3.
11 In the sense of J.W. Forrester, 1969, Urban Dynamics, MIT.
 Dynamic Simulation of Urban Reorganization of the City of Tours 101

5.2.2. Opening of the model and its modifications

Firstly, the population develops considerably during the 1851-1936 simulation

period: it goes from 30,000 to 150,000, which corresponds to an average annual rate
of increase of 2% in the city of Tours. The use of an average annual rate is a
simplification which can be avoided, because the model can run with an automatic
annual increase or a rate that can be annually set by the user, just as the majority of
other characteristics.

River, as well as road or rail traffic is taken into account, primarily, by creating a
flow vector for transit as well as for movement originating at or bound for Tours.
Firstly, we measure its hypothetical effect on the city, due to the inversion of the
problem and, secondly, it can be integrated as data or calculated at the higher level
of the system of the cities where it is endogenous, as interurban and city-countryside
relations.

The hypothesis of constancy of flows does not hold in our case and especially for
this period. Thus, we integrated a differential evolution coefficient of modal flows:
river circulation decreases since 1850 until it disappears before the First World War,
whereas road traffic grows in a relatively constant way. Moreover, the railroad
appears at the beginning of this period. The Tours train station has already been built
in 1855 and develops exponentially until the end of the century, but it generates
intra-urban or extra-urban traffic from the freight station. These flows therefore
cross the city only partially, but create or consolidate the development pole to the
south of the old town.

In our model land values depend on accessibility and this simple hypothesis was
corroborated in our former work. Accessibility itself depends on distances, modal
flows, the evolution of the population and its location.

Lastly, the possibility to construct in the area, which depends on how easily a
certain section in this zone can be flooded naturally or artificially, the political will
to manage city planning, the existence of various road networks or land retention
were taken into account by means of an urbanization vector whose value is equal to
zero when, at a determined point, construction is impossible, equal to 1 when it is
neutral and more than 1 when there is a will to manage the development and to
construct in this area. This aspect has to be developed.

These modifications are neither truly important, nor heavy in the model, but they
lead, as we shall see, to rather radical changes in the basic mathematical tool.
102 Graphs and Networks

5.2.3. Extension of the theoretical base of the model

Indeed, the nature of data like that of the followed objective makes the
simulation develop from a problem of flows solved by using graphs to a problem of
percolation where the objective becomes the analysis of the modifications of space
localizations: variables of state.

Firstly, it should be stressed that the description of space by graphs is merely

partial: only the locations of poles and the relations that exist between them are
described. The quasi-fractal method of space description developed at CESA very
clearly develops it using the zoom system.

There is an internal homothety whose principles are constant and apply to all the
levels of modeling of transport networks. When applied to the city this enables the
transition of the taking into account of isolated poles to quasi-continuous spaces.

There is a semantic and theoretical glide: nodes represent a set split into zones.
The dual graph defined on the basis of the primal that constitutes the road network
becomes a percolation grid whose primal graph describes the passage from one cell
to another and the problem of quadriconnectivity or octoconnectivity no longer
arises in this case. Space then becomes quasi-continuous on a specific scale.

What becomes essential to follow are no longer flows in the primary sense, but
rather the dynamic differential evolution of populations localizations and, in
particular, localizations of population increases, knowing that for these periods
densification is limited by the construction method and the available space.

To take into account the physical impossibility of too high a concentration at a

point in space we find a population maximum per node. If this maximum is reached,
the node in question is no longer taken into account during the process of relocation
of the populations moving at each iteration.

The map of population locations reveals the networks: the primal graph and the
delimitation of zones around each pole and the dual graph. The latter becomes a
percolation grid whose graph describes the possible passage from one grid to
another. This passage is not necessarily performed contiguously, with order 1
connectivity, but, on the contrary, may be carried out in jumps. The model is capable
to precisely retrace these various movements for each cell where there are
simultaneous population arrivals and departures.

Thus, we can, on the one hand, spatially analyze the total population movement
and, on the other hand, follow it punctually which, once we segment the population
in the next version, will make it possible to follow each category or population
 Dynamic Simulation of Urban Reorganization of the City of Tours 103

group specifically. This aspect presents particular interest in monitoring the effect of
urban development measures and the influence of amenities.

5.3. Application to Tours

Applying the model to the case of Tours obliged us to confront certain

difficulties during simulations before leading to convincing results.

5.3.1. Specific difficulties during simulations

In fact, these difficulties were, a priori, foreseeable. There are two aspects linked
to the spread of a gravity model. Indeed, even if the elasticity assigned to the
population is equal to one, the simple fact of spatially spreading the population has
two consequences. Firstly is that the result, i.e. the total attraction of the spatially
spread population, is not equal to the attraction of the same population concentrated
in a same point because internal distance is introduced and this internal distance is
necessarily equal to at least 1, otherwise an undesirable multiplicative effect is
obtained. Thus, the distance between two connected points of the second order (with
an intermediate point) will be equal to at least 2 on a grid, by simple respect of
triangular inequality, otherwise it is no longer a distance. The simple fact of initially
distributing a population therefore has a considerable effect which decreases its total
influence with respect to the value obtained when it is concentrated in a single point.
This is a utopian situation in the full sense of the expression, even if this latest
hypothesis is very often used for simplification.

A second phenomenon comes into play at this level: the influence of inequality
of graph faces. If, indeed, the localized populations are punctually different for each
node of the primal or each face of the dual, the simple mechanics of the dynamic
model means that after a certain number of iterations the imbalance between
populations by sector increases very strongly. It is therefore necessary to ensure a
relatively homogenous or relatively “normal” distribution over all of the
contiguously inhabited zones. The possible maximum population per point makes it
possible to solve this problem, since when a location becomes very attractive, its
population grows and when it reaches the maximum, the attraction benefits the
connected zones.

This difficulty is found on the scale of the town, i.e. between Tours and the
peripheral towns. Indeed, if we spatially disaggregate the main city while
considering the populations of the peripheral towns which are concentrated in a
single point of their territories, even if these populations are small, they are then
often clearly greater than a single peripheral sector of Tours. The same phenomenon
104 Graphs and Networks

of growing imbalance is reproduced to the benefit of external centers in the latter

case. Thus, we are obliged to either spatialize the peripheral towns to a non-
negligible distance or simulate only the city of Tours itself.

For the simulated period we chose to focus solely on the city of Tours given that
the growth of the peripheral towns started later, primarily after the 1950s. For the
period considered (1850-1950) the influence of the peripheral towns remains fairly
negligible in terms of attraction and possible relocations, and thus our simplifying
choice is acceptable.

5.3.2. First results of simulation

The results presented here are among the very first obtained, the model not
having yet been subject to systematic digital exploration. However, two main
simulations have been carried out: on the one hand an ad hoc evolution regarding the
transport flows for each mode as constant and on the other hand, an evolution
integrating a differential variation of modal flows (decline of river transport and
growth of the road transport).

The first simulation describes the evolution dynamics for the stable case, all else
also being equal, i.e. without changes at the infrastructure level or technical progress
of any kind. This simulation serves as a point of comparison for the other series of
hypotheses.

Figure 5.5. Pattern of settlement in 1851 according to the coefficient

of land usage (Napoleonic land register, 1836)
 Dynamic Simulation of Urban Reorganization of the City of Tours 105

The second simulation integrates structural modifications and very fast technical
progress for road and, especially, rail transport, which will quickly supplant river
transport and make it disappear within half a century.

The population is unequally distributed inside the city walls, it is concentrated

mainly in the west and, more particularly, in the Saint Martin’s district where the
density is highest (Figure 5.5).

In both cases considered the annual population growth rate is 2% with respect to
the population in 1851, which corresponds to the average growth rate between the
1851 and 1936 censuses, a period during which the city increases from over 30,000
to more than 150,000 inhabitants.

5.3.2.1. Evolution simulated with constant importance of river and road traffic
The simulation proceeds without opening of additional locations, considering
that the intake capacity of intra-mural Tours is largely sufficient as shown by the
drawing from 1855 where many unbuilt areas may be seen. The following map
(Figure 5.6) shows the absolute differential evolution: populations localized at the
beginning of the period (shown in gray) are superimposed on populations localized
at the end of the period (shown in black).

We note very clearly, first of all, the very strong influence of the proximity of
large transport arteries: the river and rue Royale12. The influence of the river
remains largely dominant in this simulation and the traditional East-West axis is
extended and reinforced in a district which was traditionally oriented towards river-
related activities: the old port of Saint-Pierre-de-Corps is in the immediate vicinity
of the Loire-Cher canal which is located between the towns of Tours and Saint-
Pierre-de-Corps. This canal replaces ruau Saint-Anne, which previously ensured
communication between the two rivers, as shown in the maps of 1619 and 1770, and
gradually became silted up.

The second observation is a tendency for rebalancing to the east and very slightly
towards the south. Indeed, as mentioned above, we introduced a population
maximum per node to avoid local concentration that would exceed the reception
capacities of the frame. This maximum by node was fixed at the level of the
population located between Saint-Martin and the Loire, considering that it
corresponded to the maximum coefficient of land usage in the Napoleonic land
register.

12 Rue Royale is the old name of the present-day rue Nationale.

106 Graphs and Networks

Figure 5.6. Development over 10 years without additional locations

(influence of river and road traffic axes, population growth of 2%)

However, this hypothesis does not correspond to the reality of location and
relocation movement at the time.

5.3.2.2. Evolution simulated with “differential growth” of means of transport

Under this hypothesis we kept the above coefficients with two important
changes: the influence of river and road means of transport is no longer constant,
therefore river transport decreases by 5% per annum and road traffic, on the
contrary, intensifies at 5% per annum. The trends of this differential development
conform to reality, since the river traffic disappeared at the beginning of the 20th
century and road traffic shows a strong growth on the Paris-Chartres-Tours-Poitiers-
Bordeaux axis.

The map of population growth below (Figure 5.7) showing, as previously, the
superimposition of populations at the end and at the beginning of the period enables
us to immediately note a very weak relocation close to the river, except in the east.
However, there again it is definitely less strong than under the previous hypothesis.
On the other hand, we note a much stronger growth on the North-South axis of
Tours, the street which became rue Nationale.

It is clear that taking a differential rate into account, i.e. growing for roads and
decreasing for the river, modifies the relocations or new localizations.
 Dynamic Simulation of Urban Reorganization of the City of Tours 107

The phenomenon appears much more clearly when we take into account the
absolute variation of populations by site, which is illustrated in the map below
(Figure 5.8). We can observe that all considerable absolute variations take place
along two axes: decrease at the river, which loses its attractiveness, and at the same
time growth along rue Royale. Over the period considered the population growth is
not strong enough to spatially diffuse outside the limits of the old ramparts and the
dominant phenomenon are still quite clearly relocations.

Figure 5.7. Development over 10 years without additional locations (increasing influence of
the road to Bordeaux and decreasing influence of the Loire, annual population growth of 2%)

Figure 5.8. Absolute differences between the hypotheses with differential rates of
development of the means of transport: urban inversion emerges
108 Graphs and Networks

However, this makes it possible to note new tendencies which are masked by
absolute values: relative growth in space, albeit still discrete, and, in particular, in
the south and the west, as if the population started a process of diffusion or rather of
rebalancing from the densest zones at the beginning of the period. It is a relative
increase in density in the southern zones close to the ramparts or of reallocation of
areas formerly used as parks or gardens by the religious congregations, which were
established at the periphery of the old town, as may be seen from the 1855 drawing.

However, intra-mural space is still far from being saturated! The negative values
appear as lighter circles and the positive values as darker circles. The figure of
relative growth (Figure 5.9) enables us to specify the phenomenon: it confirms the
preceding observation regarding the importance of movement along the two
transportation axes.

These initial simulations thus tend to corroborate the modified work hypothesis
and the general hypothesis: it appears that, indeed, it seems possible to model the
phenomenon of axis change in the city of Tours, and it is not enough to merely take
into account the attractiveness of transport routes to arrive at the observed
development, it is also necessary to take into account the differential development of
the two types of traffics corresponding to the historical reality.

Figure 5.9. Relative differences between the hypotheses with and

without differential development of the means of transport
(positive evolution darker and negative evolution lighter)
 Dynamic Simulation of Urban Reorganization of the City of Tours 109

5.4. Conclusion

We are aware that it is only a very first attempt at clarifying the problem and that
the model is still neither completely explored, nor even complete for this specific
type of simulations. However, we may note as of now that the opening of the model,
possibly with respect to two levels, and the extension of its capacities through the
introduction of minor modifications, two vectors and three constants tend to change
its nature at the level of the theoretical base of the tool by gliding from one concept
to another, from graph to percolation. In terms of methodology, it appears that the
inversion of the problem that we were obliged to perform due to the unavailability of
data proves to be effective.

With respect to the field of application we pass from a praxeologic model to a

cognitive historical model. From a certain point of view, following the example
experimental archeology, we introduce experimental history by modeling.

A result obtained indirectly is the existence and temporal localization of a

change in the spatial evolution of the city and this is shown, if not demonstrated.

A general result of the theory of systems is corroborated with regard to

reproduction: by using the model we show that the system can reproduce itself
differently, at least with respect to its spatial form, although the reproduction
mechanism remains identical.

Lastly, in terms of “explanation”, we go from “everything happens like this” to

“what is needed within the framework of the logic adopted for everything to happen
like this”. In this case the model produces second degree knowledge!

5.5. Bibliography

[BAP 99] BAPTISTE H., Interactions entre le système de transport et les systèmes de villes:
perspective historique pour une modélisation dynamique spatialisée, Thesis, Tours, 1999.
[CHA 97] CHAPELON L., Offre de transport et aménagement du territoire: évaluation
spatio-temporelle des projets de modification de l’offre par modélisation multi-échelles
des systèmes de transport, Thesis, Tours, 1997.
[MAT 78] MATHIS P., Economie urbaine et Théorie des systèmes, Thesis, Tours, 1978.
 Chapter 6

From Social Networks to the Sociograph

for the Analysis of the Actors’ Games

As has been illustrated up to now, it is rather common to use graphs to deal with
transport and some other technical networks. We will now attempt to approach the
social sphere by means of social networks. Although, usually it is a field of interest
for certain sociologists, we will see how they can specify such concepts as “system
of actors”, “game of actors”, “network of actors” or that of governance, which are
often present in contemporary debate and publications on urban planning. However,
unlike these paradigmatic concepts that are often used to mask the thorny issue of
power and politics in our discipline, we will choose a more operational point of
view, which, in a certain manner, conforms fairly well to the general spirit of those
who use graphs.

“A social network is by construction a set of relations between a set of actors”

[FOR 97]. Behind this use of “social network” there is a paradigmatic model, which
poses that the behavior of actors is determined by their relations and their attributes.
The term “actors” is generally used to designate individuals, but it can also refer to
households, associations or any other group form. They are mutually engaged in
varied relations of power, friendship, distrust. The set of these relations is properly
known as the social network. Finally, actors are distinguished from each other by
attributes, i.e. the values of variables such as age, resources, socio-professional
category, religion, which are supposed to generally directly explain their behaviors.

Chapter written by Sébastien LARRIBE.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
112 Graphs and Networks

Moreover, a “system of actors” has to be understood as a set of explicit or

implicit rules that organize an activity (in fact, involving actors in it). The “game of
actors”, in turn, is the manner, in which the actors take on these rules, interpret them,
constrain themselves accordingly or circumvent them in order to achieve their
respective goals and carry out their projects [MER 92]. From this point of view, the
social network is therefore a means for actors to play inside a system.

Sticking with the game metaphor, we can say that when there are only two
players involved (duel situation), the situation is perceived by each of them directly,
since it is clear whether the other player has adopted a cooperative or an
uncooperative attitude (in other words, if he is an adversary or an ally).

However, when there are three or more players everything changes. Cooperation
attitudes and generally lasting alliances and coalitions may appear and render the
tactics less direct, more ambiguous and fluctuating. Consequently, the entire game
changes because everyone may be obliged to establish a “shifted” relationship with
the others, which is a parallel relationship at the margins of the “real game”. To
some extent a “meta-game” with vague rules is created. For some the true interest of
the game lies therein. For others, on the contrary, these attitudes are comparable to
cheating, which may lead to situations of rupture, exclusion, or appointment of a
referee.

If we wish to study them, it is convenient to represent these game and

cooperation-conflict situations using the formality of graphs. We obtain figures
comprised of nodes representing individuals and edges or arcs between two nodes
which represent the relations, if they exist, between two individuals.

Here we propose to analyze precisely this exercise. Passing by sociometry and

social networks analysis, we will dedicate greater attention to sociography, whose
viewpoint largely exceeds the connotation or even “illustrative” usage which is
sometimes too attached to graphs.

6.1. The legacy of graphs

When only two actors or players and the fact that there is or there is not a
relationship between them are considered, the situation is common and the graph
that can be plotted is no less so: the two individuals are either connected or not by an
edge. However, still with just one type of relation possible and three non-
interchangeable individuals, there are already eight possible situations. If we then
consider the case where two relations are used to describe the operation of this
tripod and that these two relations can exist simultaneously between all three pairs
of players, 64 possible arrangements are obtained (Figure 6.1).
 From Social Networks to the Sociograph 113

Figure 6.1. 64 organizations of a tripod with 2 possibly probable

cumulative (but not directed) relations

From this figure we can deduce the potential role of an actor relative to his
position in the structure (it is exactly the sense, which is usually given to
“hierarchical position”, for example). It is clear that this role results at the same time
from the relations that an actor does or does not maintain with the others and,
simultaneously, from the relations that the others do or do not maintain between
them. This observation does not, however, come without a cost.
114 Graphs and Networks

On the one hand, it necessarily implies the construction of relatively broad and
global figures and views to represent the analyzed situations. Moreover, it leads us
to think that the reading process needed for the analysis of these graphs has to be
multiscale, thus leading us to gradually decipher the operation of binomials, tripods
and sub-groups of any other size. Later on we will return to consider these
characteristics, but suffice it to say for now that they lie at the heart sociography.

Let us suppose that in Figure 6.1 the gray edges symbolize a hierarchical relation
between A, B and C and the black edges symbolize a friendly relation. Figures only
containing black edges are traditional organizational diagrams. We also talk of arcs
and not of edges by considering that hierarchical relations are not symmetric.

When we transfer our attention to the figures comprising gray and black edges
simultaneously, we understand quite quickly how the gray edges “disturb” this
hierarchical organization. In other words, the gray subgraph superimposed on the
black subgraph, obviously, considerably modifies the interpretation that can be made
of the graph on the whole.

It is also necessary to pay attention to the absence of edges, which does not
necessarily mean absence of relationship. We are talking here of the property of
transitivity. For example, in all the configurations that contain two edges, gray and
black, we may think that there is no relation between two of the three nodes.
However, if A and C are friends (connected by a gray edge) and A is the superior of
B then, consequently, and although the relation between B and C is not visible, we
can suppose that, functionally, C has a “means” (A) of influencing B.

In all this, the question of power in general, in such dimensions as authority,

potentiality and, especially, influence, is central. The general aim of the exercise is
to know how the sociogram (informal relations) of an organization or a company
disrupts the structure of the organizational diagram (formal relations), how it makes
it possible for certain actors to perform “hierarchical jumps” of several levels, how
actors are not, in reality, “in the place” which we could have believed to be theirs. It
is, of course, in this sense that the term “network” recovers its somewhat negative
connotation. It conveys the idea of avoidance, bias, maneuvering that would enable
one individual to “reach” another by means of intermediaries or by playing on
another relational register, whereas structurally and formally he has not been given
this possibility.

We owe the first analyses based on data collected about interpersonal relations to
Moréno [MOR 70]1. It gave rise to sociometry defined as “the science of measuring

1 “Sociometry has the aim of studying the psychological properties of populations

mathematically; to this end it implements an experimental technique based on quantitative
 From Social Networks to the Sociograph 115

relations between humans”. Among the tools introduced and used by Moréno for
data processing we are only directly interested here in the sociogram. It is a graph
constructed exclusively on the basis of emotional relations which are recorded
between individuals and considered by the author as the real maps of “psychological
geography” [MOR 70]2.

The structural method3 introduced by Moréno has seen many developments, in

particular, stemming from anthropology and field of organizational research. In a
certain manner, the organizational chart is for Fordism what the sociogram is for the
current of Elton Mayo’s human relations, and it is perhaps to Michel Crozier that we
owe one of the first attempts to bring these two movements closer. Although
schematic, since the beginning of the 1980s this constant evolution constitutes and
organizes a specific field dedicated to the analysis of social networks.

This latter type of analysis is no longer limited to the psychological level and,
therefore no longer limits itself to the sole study of emotional relations between the
members of a group. In principle, any relation between members can be processed,
even though, in general, we limit ourselves to taking into account the relations of
collaboration, support, advice, influence or control.

As for the processing, it may be a question of producing individual scores

(scores of centrality, prestige, etc.), of determining the make-up of sub-groups in
order to have a structure emerge in a vast network of individuals, of highlighting
particular structural configurations (determining the equivalent positions in the
structure), or revealing the indirect relations between the sub-units of a group.

From this point of view and in general, the method of behavior analysis
following the use of the conceptual “network” model is organized in two stages. The
first consists of determining a structure in the set of data about “actors” and
“relations”. To this end we primarily use sociometric indicators (density, centrality,
prestige, etc.) and methods of determination of cohesive sub-groups (i.e. where the
members maintain more relations between them than with members of another
group).

methods and exposes the results obtained through the application of these methods. Thus, it
performs a methodical investigation into the development and organization of groups and into
the position of individuals within groups…” [MOR 70], p. 26 and XLI.
2 “By psychological geography we mean the graphic representation of the interrelationships
that link the members and groups of a same community, a) with respect to their topographic
location and b) to the psychological currents that circulate between them” [MOR 70], p. 26
and XLI.
3 Sometimes referred to as “neo-structural” to take into account the “structural” approach of
Claude Levi-Strauss [LAZ 98], p. 3.
116 Graphs and Networks

Actually, all the formalism of network make-up and of the tools of graph theory
may only be applied here on the condition that the result can yield a social
interpretation4. For example, the degree of a node becomes “centrality” in the
sociometric context, whereas the internal half-degree may be compared to
“prestige”. Using the same idea, a complete maximum subgraph becomes a clique
(or 1-clique), i.e. all the actors are directly connected thus indicating a very cohesive
group. If a subgraph admits n as the maximum length of the paths connecting all the
nodes, then we speak of an n-clique (n becomes the number of intermediaries). If a
node in a subgraph does not have relations with at most k other nodes, it is indicated
as k-plex. Conversely, in a k-core each node is attached to k different nodes. A
clique (A, 1-2-3), a 2-clique (A, 1-2-3-4-5, the path between 4 and 5 passes by 6), a
1-clan (A, 2-3-4-5-6), a 3-plex (B, at least one node does not reach three others
directly) or a 2-plex (C) (Figure 6.2).

2 3

4 5

A B C
Figure 6.2. Graphs or analysis of social networks?

The second part of the analysis consists of trying to connect this structure with
the observed behavioral phenomena, possibly by using information relating to the
attributes. Strictly speaking, we qualify the first stage as network analysis and the
second as structural analysis. As far as we are concerned they are not dissociable
and that confers all its specificity to the procedure.

Most of all, we should not lose sight of the fact that in this approach the network
analysis serves structural analysis. The first objective is clearly to understand,
analyze and even anticipate the behavior of an actor. By behavior we understand
here the exercise of power, i.e. again, the faculty or tendency to be able to make and
impose a decision.

This type of graph has two limits: the organizational diagram and the sociogram.
The organizational diagram reflects only the hierarchical relations, i.e. the power

4 Which is what has been carried out, at least partly, in [FLA 68].
 From Social Networks to the Sociograph 117

distribution in the organization. As for the sociogram, it reflects only the emotional
relationships of selection or rejection between the members of a group. The aim of
strategic analysis is then found admirably transcribed in these two graphs: start
knowing the relations of power (the organizational diagram), consider the informal
relations (in a certain manner, the sociogram) and, thus, reveal a specific action
context.

6.2. Analysis of social networks

Strictly speaking, in order to carry out a structural network study, we require,

first of all, to collect three types of information. Firstly, we have to identify the
individuals and their relations. That can be done by means of sociological scale
surveys, by determining a type of relation which is supposed to characterize the
analyzed system (we speak of a “relational variable”, i.e. a criterion that makes it
possible to connect the actors of the system), or by using “name generators”5. The
second type of information consists of actors’ “attributes”. These are variables likely
to suggest why an actor would be in a relation to a particular person (age, gender,
social status, etc.). Finally, the last type of information can be collected regarding
the behaviors of actors that are thought to be connected, through analysis, to the
position of the individual in the network (mobilization, decisions, performances,
etc.).

Cohesion Equivalence Hierarchy Sub-groups

Figure 6.3. Reference structures for the analysis of the social networks

5 Name generators are questionnaires prepared with the goal of recreating a social network. It
may be simply a question of the “with whom do you perform a particular activity?” type. In
these proceedings we only avoid formulations that may cause various interpretations by the
questioned individuals, such as the “who are your friends?” kind, since the definition of
friendship may be rather variable. Questions about activities do not have this disadvantage,
since it is the investigator who determines the associated relation type.
118 Graphs and Networks

However, when we consider their implementation, these methodological

elements have a direct feedback effect on the sense and even the use of the analysis.
As an analysis tool, the graph is at their disposal to suggest a group functioning
mode which comes in all the nuances ranging from a clique to the hierarchical tree
structure (Figure 6.3).

However, we can also imagine an “in play” tool of network analysis, in other
words, information about the network which is used and exploited by a particular
actor. The point of view is strategic here because the actor in question may seek to
work out his own game tactic by this means, for example, by wanting to know if
there is a path between this actor and another, “how to reach so-and-so” and which
intermediaries “to use” if this relation is not direct.

Under this new light, and if we wish to preserve all the sense of the game as
such, it is not possible that a player sets up systematic protocols of data collection
envisaged by strategic analysis. Because, of course, and yet again, to preserve all the
interest of the game, some of this data is not intended to be revealed and may even
stem from very personal appreciations or even indiscretions or other confidences
collected under quite various circumstances. Would that be reason enough not to
take an interest in it? Within the framework of a scientific work, perhaps. Within the
framework of personal tactic and according to the teachings of Sun Tzu or
Klausewitz, certainly not.

Thus, if we wish to preserve the possibility of network analysis in this particular

context, it is quite necessary for us to admit:
– that it is up to each player to get information about the others and the relations
that may exist between them without this being able to be done in an official and
systematic manner;
– that information processed will be thus, by construction, biased if not partial
and can only correspond to his own point of view at a given moment;
– that taking into account the unquestionable inaccuracy of this data and its
approximate nature, it would seem inappropriate, at least, to use too fastidious a
formalization (k-plex, k-core and others) to finally describe the game in which he
participates.

Under these conditions, it is not truly social network analysis that each player
engages in, but an analysis of his own representation of the network in which he is
inserted6. And since we deliberately do not place ourselves in a deductive process,

6 However, it is not a question of analyzing his “personal network” because in the analysis of
social networks the formula draws a radiating relationship diagram centered on a subject (the
ends of the rays are not connected between them).
 From Social Networks to the Sociograph 119

the construction of such a representation may only proceed by successive iterations.

Indeed, it is quite necessary to imagine that, when confronted with the first
impression, the player-analyst will quite quickly have additions to bring, individuals
and relations to add. In this sense network analysis will be for him a means of
identifying his lack of knowledge about the system and of mitigating it by means of
many return trips between his periodic obligation to play and the preparations that
he may make for the following step by taking into account what cold have occurred
in this interval of time.

However, if we now extend the concept of a player to the more general one of an
actor, we can undoubtedly consider this process of objectification of reality on other
scales or in other contexts that hardly dissimulate the game metaphor. On the basis
of elements that may sometimes be relatively uncertain or emotional, an actor could
be brought, from result to result and from assumption to assumption, to change a
certain number of his opinions, impressions and prejudices. By these means, which
are altogether rather simple, there is a possibility of estimating or simulating the
consequences of the existence, transformation, or discovery of new relations
between two or several players. Being confronted with a representation of his
knowledge, it finally becomes possible for him to reveal, formalize and specify his
ignorance of the reality, in order to better fill in the gaps. And, at the end of the day,
in this particular meaning of the “networks of actors”, the determination of the
concrete context of action of an actor is at stake.

By starting from the analysis of social networks we clearly see how this
approach preserves certain aspects and how much it deviates in some others. On the
one hand, the structural perspective seems to us particularly adapted to the
description of a true game in progress and well beyond what the rules would enable
us to understand. On the other hand, the conditions of data access and the ambitions
of the analysis are specific to us. Everything that we propose to name as
“sociography” is based on this distinction. Its methods and tools enable us, perhaps,
to propose an original interpretation of graphs.

6.3. The sociograph and sociographies

To us “sociography” thus indicates this reflection method which is applied more

specifically to relational data. We say “graphy” because we believe that graphs in
general, as visual representations, at any rate at our level and with a little practice,
can render a sufficiently good account of the operation of a network without it being
necessarily useful to resort to overly elaborate mathematical singularities.

The project seems to us all the more accessible, since Bertin’s graphic semiology
already deal with networks [BER 68, p. 263-283] and since in graphs we can find all
120 Graphs and Networks

the principles and the advantages attached to representation and information

communication systems [BON 88], to images and graphics [BON 75].

Is it not the case that a graph enable us to grasp very intuitively the specificities
of a clique (Figure 6.2, A, 1-2-3), a 2-clique (Figure 6.2, A, 1-2-3-4-5, the path
between 4 and 5 passes by 6), a 1-clan (Figure 6.2, A, 2-3-4-5-6), a 3-plex (Figure
6.2, B, at least one node is not directly connected to three others) or a 2-plex (Figure
6.3, C)? Following this design or scheme we clearly see the difference between
organizational structures and we already “sense” that they cannot function in the
same way. By a notable kind of shortcut: “you can see it”.

We say “socio”-graphy to recall that the data which are handled simultaneously
concerns formal and informal relations. “Network images” or the graphs that we
obtain are thus a mixture of “shares” of organizational diagrams and sociograms,
which is different from case to case.

If a camera makes it possible to take photographs, a sociograph is a “machine to

take sociographs”. It is thus more of a tool than a diagram. It must enable a
representation of a database containing individuals and the relations that they can
have between them. It must help the diagnostic of actors, but certainly not replace it.
For that it must be relatively interactive and flexible in use because it is used in an
iterative trial and error process, which seems to us to be the most effective means to
approximate our reality. If graphs must appear – and they undoubtedly do – they are
“exploratory graphs”. However, we may also imagine that it can represent the
attributes of actors in order to facilitate behavioral interpretation. Thus, we obtain a
doubly “strategic” image. Strategic, because that is what the position of an
individual in a network inevitably is; strategic, because this position really only
becomes that when coupled with resources, at least potential ones (attributes).

We do not forget either that the network is still considered from the point of view
of a given actor, in consideration of which the “sociographies” are by nature “self-
centered”. It is not, however, an “ego-network” as understood by sociometricians or
analysts of social networks, i.e. a network of relations of an individual that could be
represented in the shape of a star.

From the moment when edges can connect two points, one of which is not the
center of the circle, the personal network becomes only a particular case of
sociography, more precisely a subgraph. However, let us come now to a more
concrete presentation of our project.

Thanks to computers, we can have a tool for automatic design and interactive
exploration of graphs, in other words, our sociograph. Relational information can be
stored in any kind of spreadsheet or database management system. The functional
 From Social Networks to the Sociograph 121

basic unit comprises binomials, for which two actors and a possibly valuated
relation are indicated (Table 6.1). By convention, we can state that when the relation
is directed, the direction is indicated by the order of appearance of the actors in the
same line.

Actor Actor Relation Valuation

A B R1 1

A C R1 1

C D R2 2

C B R3 1

… … … …

Table 6.1. Relational binomials

As far as the attributes are concerned, the input of information is not more
complicated. It is possible to specify a parameter list for each actor supposed to
qualify him or to single him out. However, it is also possible and, perhaps, more
useful to consider from the start the attributes which could possibly become
exploitable later (Table 6.2). Starting from this stage, which is already partially
carried out, the graph can be plotted. At the start, it is a global graph therefore
representing all the existing relations. Since network analysis privileges structural
data with respect to category data, the attributes are not represented (Figure 6.5a).

Actor Studies Age Marital status Chaser …

A F1 24 Single Not …

B F2 37 Married Yes …

C F1 21 Single Not …

D F2 57 Divorced Yes …

… … … … … …

Table 6.2. The classification of individuals according to their attributes

122 Graphs and Networks

A B

C D
Figure 6.5a. Generating a graph with relational networks

This graph is the starting point for the exploration, which consists of moving the
nodes to limit the tangles as much as possible and to reveal a readable structure. In
other words, it is a question of approaching a planar graph as well as possible. Or
even to locate sub-groups which are made up of nodes having more relations
between them than with any others. There are automatic placement algorithms that
may facilitate this operation but we have not yet felt the need to implement them.

This structural exploration may be made easier by successively representing each

possible type of relations (Figure 6.5b). With the addition of a new relational type,
the user observes new permutations and repositions the nodes accordingly. It also
happens that we observe that this addition does not change anything in the
representation. That is, two modes can be redundant to some extent, in which case
either one of them is removed, or a “composition” of them both is performed.

Here we do not deal with partial graphs. Indeed, the fact of always representing
all the nodes makes it possible to also locate those that “do not belong to the
network” in a given relational mode.

Lastly, once a structure has been highlighted, displaying one or more attributes
may make it possible to discuss the reasons that could have led to the observed
forms. That can also lead to reposition certain nodes, if it is thought that the
similarity of profile may favor a certain “bringing together”.
 From Social Networks to the Sociograph 123

A B

C D

Formation Age Marital status Hunter

F1 F2 <25 >50 Sing. Mar. Div. no/yes

Figure 6.5b. Plotting of the graph of relational networks according to attributes

These two figures illustrate the guiding principle of sociography. The first is
made up only on the basis of relations that are, nonetheless, distinguishable. It is
complemented in the second figure by a possible representation of the attributes that
highlights “resemblances” between nodes. Everyone is then free to interpret these
similarities in terms of potential cooperation-opposition relations.

Krackplot 3.0 is a software which makes it possible to plot graphs by giving the
possibility of distinguishing nodes by text color and geometrical forms, that is, two
attributes7 (Figure 6.4 in the color plates section).

This completes the first iteration of the process. The highlighted graph or
sociography makes it possible to return to the database in order to supplement or
correct the initially entered information in terms of binomials. From there a new
graph can be generated on the basis or not of the preceding configuration. With time,
as alliances and dissensions are formed, the process of real-time objectification of
the game of actors is engaged.

7 Source: David Krackhardt. http://www.heinz.cmu.edu/~krack.

124 Graphs and Networks

Thus, this is what we indicate as sociography. The exercise is altogether rather

simple, which, in our opinion, does nothing to diminish its operational character.
Even though as this explains that. Since, indeed, let us not forget that it is a work
that must be able to be undertaken quickly, in real-time, since the game is in
progress. Simplicity and speed in this context are thus, undoubtedly, assets that are
at the very least non-negligible.

However, from here on we can, perhaps, consider a less game-related and more
concrete implementation than the one that we have just described. Indeed, the
readers would have all clearly understood that the term “game” is used as a
metaphor to describe situations of actors in very diverse contexts. In particular, that
has already for some time been the case in the field of environment and this
application has spread more recently to the disciplines of urban planning and
development. It is thus, of course, in a more specific situation of this type that we
will try to develop our approach later on.

However, before proceeding further, we clearly need to specify that our aim and
our analysis are predictive and not prescriptive. Unlike game theory and some other
attempts, our ambition is neither to determine nor to even approximate an optimal
organizational solution for a group of actors placed in front of a given problem.
Several authors have already underlined the vacuity of such an ambition in the field
of urban planning and development and we will not return on that subject there. Our
own ambition is rather to try to specify the decisional context of a project by
rendering more complex the organizational framework that it seems to have. In this
sense, it is a question of limiting as much as possible the set of possible or
imaginable decisions by restricting it in a certain manner within the mesh of a graph.
As we can transmit onto a map the geographical limits of a project, so we try to
transmit onto a sociography the contours of the decisional space that has to lead to
its implementation.

Here is the type of figure (Figure 6.6 in the color plates section), from which
sociographic exploration may begin. All the relations between individuals are
represented without distinction. The placement in circle is arbitrary but rather
convenient for a first attempt. Here, the star shape suggests that the operator was
integrated in the relational base, but the observable relations in the lower right
quarter show than it went beyond the ego-network. Since each edge represents a
relational mode whose legend is found in the left-hand column, the graph is at most
a 12-graph.
 From Social Networks to the Sociograph 125

It is there that we finally see the great interest of the structural approach and
social networks analysis for our problem, since the method “aims to systematically
contextualize the behavior of actors” (see [LAZ 98], p. 7). We do not specifically
seek to understand the sense of an individual’s action, but we try specifically to
position an actor in his network in order to estimate his potential of action, his
opportunities, as well as his constraints. The method aims at the predicable not the
predictable. In other words, the idea of this approach is to try to represent the
“playing field” where we watch our actors move. In a certain manner it is, first of
all, about strategy rather than about trying to clearly define the game environment.

If we have thus somewhat given up on the idea of really being able to understand
the strategy of an actor, we think that the only alternative is to encourage reflection,
even imagination, by trying to determine all of his resources or attributes. It can be a
question of financial capacity, time, access to information, membership in various
associations, i.e. a whole set of parameters constituting a potential for action.
Obviously, interpersonal relations form part of this potential and, for this reason, the
distinction from “attributes” is sometimes thin. It is primarily due to the potential
character of the resource. Let us say that the relation is really observed in practice,
whereas the attribute is a capacity that is considered to be available for mobilization.

We try to represent these two dimensions in the sociograph. In its usual form the
graph represents the relational capacity of both. By means of various formalizations
we try to represent the resources-attributes.

This figure (Figure 6.7 in the color plates section) was obtained by “purifying”
the previous figure. All the relations in the modes that do not make it possible to
distinguish between nodes have been eliminated. The nodes in Figure 6.7 have been
moved by the user to obtain the present figure (Figure 6.8 in the color plates section)
without calling upon a systematic process or an optimization algorithm. However,
we observe that the nodes connected by gray edges form a clique, that the exchanges
are also very intense between the individuals in the “red group”, or that individual
number 3 has a pivotal role between these two sub-sets.

To this end, it is necessary, first of all, to represent the resources upon which, at
least partly, the action can be based. Then we have to render this representation
dynamic, since, by nature, such dissemination of the resources is not immutable.
126 Graphs and Networks

Figure 6.9. Requests for advice and attributes of lawyers in a firm

In Figure 6.9 we finally illustrate the difficulty that may exist in plotting a graph
comprising a significant number of nodes and a significant number of attributes. The
nodes are placed on a normalized reference scale whose origin is located in the top
left. The x-axis represents the advice given: the more an individual is located
towards the bottom, the more advice he gives. The y-axis represents the requests for
opinion: the more an individual is located towards the bottom, the more he requests
advise from others. The circle slices represent the possible attributes of the
individuals. They highlight a certain similarity between the three individuals in the
upper right quarter (who give out a lot of advice but hardly receive any). However,
they also make it possible to observe individual singularities, such as node 16, which
is in the same zone with an “attributes profile” that is extremely different from the
three others or, on the contrary, node 0, which, despite its attributes, does not have
the position that we would have believed.
 From Social Networks to the Sociograph 127

6.4. System of information representation

Consequently, and in spite of criticisms that we could formulate in its respect, it

is obvious that sociography is quite close to the theoretical framework of “limited”
or “procedural” rationality. It is different, however, in the sense that while focusing
on the dynamics of information representation-interpretation, it tries to avoid as
much as possible any consideration of the actual motivation of individual action.

To solve such difficulties, that is, to contribute to decision making in this

environment, three large logic types can be developed: that of multicriterion, a more
critical approach of rationality in the organization and, finally, “information
representation systems”, among which we place our sociographs.

Another more recent attempt to sweep the field of the possible games of actors is
to build a multiagent system (MAS). By “agent” we understand the active elements
that are like our actors, but also passive agents that can be perceived, created,
modified or destroyed by the former. In principle, it “is enough” to formalize the
rules of behavior of each of these agents and then to let data processing “do the job”
in order to observe what occurs when several agents are found in the same
environment. To date MAS were used experimentally as an aid to public decision-
making or for problems of territorial management [FER 98]. In our case we would
have to create as many agents as actors with their own rules of behavior in order to
observe the result of their interactions and we would have to restart the process as
many times on the basis of assumptions about the individual strategies of the agents.

By adopting a more descriptive point of view, other authors on organizational

sciences criticize the very principle of rationality of the decision. We group a little
hastily in this set the work of the human relations school, the group dynamics of
Kurt Lewin, those on organized collective action, i.e. studies that showed how the
human element came “to disturb” the mechanical operation of industrial companies.
The model of the “concrete system of action” (CSA) by Crozier and Friedberg is one
of most famous on the subject [CRO 77, FRI 93]. It states that an organization
creates a context of action by imposing formal relations of power between
individuals, but that the latter change its composition through their informal
relations, which finally yields a CSA. In this sense, however, it should be clearly
understood that the CSA is not meant to generate situations of conflict and
opposition between actors, but enables them, on the contrary, to solve a sum of
concrete, daily and detailed problems, so that the organization may really engage in
its functions. In this system individuals act to preserve an influence or minimal
power and this influence in turn depends on the ambiguity of the role that they have
in the organization, or that they believe to have. However, all these power games
may on certain occasions be opposed to the principle of rationality, which is really
why it is described overall as “limited” or sometimes “strategic”.
128 Graphs and Networks

The third possible way that we wished to explore is that of the “interactive data
visualization” tools. Here, the focus is on the acquisition and exploitation of
information, since these two stages should be used as a pivot and to stimulate
reflection. In this case, it is a question only of shaping (often graphically) this data in
order to enable the user to detect in this heterogenous information mass the
singularities, which he would then be able to analyze more thoroughly.

Web Path is an interactive data visualization tool. It is a file manager, also called
Visidrive, which shows a tree structure of data in 3D (Figure 6.10 in the color plates
section). There are several types. Here, the file labels appear in a sort of a tube, then
on a wall in perspective8.

From this point of view, strictly speaking, it is not a model of analysis, but at
most a set of tools or means. However, these are tools that do not yield direct and
clear results. Contrary to many tools, data input into the sociograph is not
homogenous. Consequently, it is not possible to reduce it to some kind of indicator
in the hope of then determining the optimal solution to our problem.

On the other hand, by using the sociograph, i.e. by giving this problem the shape
of a graph, it is, perhaps, indeed possible to initiate an analysis. Presented in the
form of a relational database or various tables containing the attributes of the actors
and their relations, the decision-making problem cannot be perceived in all its
dimensions and therefore cannot be solved. We substitute this with just one image, a
sociography, which is able to open perspectives for discussion and thus, possibly,
(but that almost exceeds our ambition) of resolution.

For this reason a sociography is clearly a graph with all its operational logic.
Indeed, the graph could not solve the Koenigsberg problem anymore than it could
directly solve the problem of the shortest path. On the other hand, it is, indeed, from
the moment when the problem is presented in this form of arcs and nodes that
solutions to these questions can be conceived and then generalized. Sociography is
no more ambitious when it proposes to apprehend a decisional problem.

Our sociograph is an attempt to take on directly these “additional” questions,

these small “asides” that are still regarded by many as disturbing elements of a
robust and reasonable practice, but which for us are a specific and irreducible
dimension of the praxis of urban developers. We were largely inspired for its
implementation by the data processing tools used by social networks analysts.
However, we associated with it a certain number of specificities which are better
suited to make it usable in situ by urban planners or developers. It is still not certain

8 http://davinci.informatik.uni-kl.de/vis98/archive/tp/papers/webpath.html.
 From Social Networks to the Sociograph 129

whether the former and the latter have a real will to apprehend this dimension of the
praxis.

To conclude, it is necessary to consider that a sociography is altogether quite a

commonplace object: a non-planar p-graph enriched at the nodes by some figures
symbolizing the attributes. As far as we are concerned we have not yet found more
effective means of accounting for these relations and attributes. Having said that, it
is perhaps this simplicity or this minimalism wherein lies the power of images in
general and, thus, of graphs in particular.

6.5. Bibliography

[BER 68] BERTIN J., Sémiologie graphique, Gauthier-Villars, Paris, 1968.

[BON 75] BONIN S., Initiation à la graphique, EPI, Paris, 1975.
[BON 88] BONIN S., “La cartographie, processus de communication: méconnaissance et
sous-utilisation des images graphiques dans l’enseignement et dans la presse”, Bulletin du
comité français de cartographie, no. 1, issue no. 115, 1988, p. 19-31; and the special
edition of the Comité Français de Cartographie, “Théorie de l’expression et de la
représentation cartographiques et exercices pratiques”, Bulletin du comité français de
cartographie, issue no. 117-118, p. 19-31, 1988.
[CRO 77] CROZIER M., FRIEDBERG E., L’acteur et le système, Seuil, Paris, 1977.
[FER 98] FERRAND N., DEFFUANT G., “Trois apports potentiels des approches ‘multi-
agents’ pour l’aide à la décision publique”, p. 373-385, CEMAGREF-ENGREF, Colloque
Gestion des territoires ruraux: connaissances et méthodes pour la décision publique,
Clermont-Ferrand, 27-28 April 1998.
[FLA 68] FLAMENT C., Théorie des graphes et structures sociales, Mouton/Gauthiers-
Villars, 166 p., Paris, 1968.
[FOR 97] FORSÉ M., LANGLOIS S., “Réseaux, structures et rationalité”, L’année
sociologique, 1997, 47, no. 1, p. 27-35.
[FRI 93] FRIEDBERG E., Le pouvoir et la règle: dynamique de l’action organisée, Seuil,
1997, Paris.
[LAR 99] LARRIBE S., Représentations auto-centrées et interactives d’un réseau d’acteurs en
aménagement, PhD Thesis, Tours, 1999.
[LAZ 98] LAZEGA E., Réseaux sociaux et structures relationnelles, PUF, Paris, 1998.
[MER 92] MERMET L., Stratégies pour la gestion de l’environnement: la nature comme jeu
de société, L’Harmattan, Paris, 1992.
[MOR 70] MORENO J.-L., Fondements de la sociométrie, PUF, Paris 1970.
 Chapter 7

RESCOM: Towards Multiagent Modeling

of Urban Communication Spaces

7.1. Introduction

In the present chapter we aim at developing a data processing tool within a

systemic individually-centered approach dedicated to the study of the spatial
structure of urban communications. The utility, which we would like to attach to this
tool, consists, first of all, of setting up a method of graphic representation of urban
communication spaces, so as to deduce several explanatory indicators with the
ultimate goal of assisting decision-making in urban development.

To arrive at our goal, we employ a theoretical framework enabling us to study

the city as a socio-cultural communication system where individuals exchange
information in specific spaces whose design can create favorable conditions for the
initiation and maintenance of the communication flow between them. These spaces
are described as phatic.

However, we have observed that to simulate a permanent social process

integrating multiple behavior modes, the multiagent system (MAS) approach offers
the methods necessary to achieve our objective. The MAS consists of representing
in computerized form the behavior of entities (commonly referred to as agents) that
act in the world with the objective of making a phenomenon emerge on the basis of
their interactions. Each one of these interactions develops according to rules,
conventions, etc. which are determined by a certain “interaction protocol”. The

Chapter written by Ossama KHADDOUR.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
132 Graphs and Networks

definition and installation of this protocol bring into play a great number of
mechanisms and properties which relate not only to the protocol itself, but also,
more specifically, to the communication medium as well as the regulation
mechanisms that may exist at the overall system level. In our work concerning an
urban system, agents are individuals with the desire, the need, the obligation or
prohibition (according to a socio-cultural code) to use phatic spaces as
communication (interaction) media.

Called RESCOM, our data processing tool has been developed in a CAD
environment by using a functional programming language: Lisp. The development
of this tool enabled us to represent graphically on two- and/or three-dimensional
maps a set of indices which look further into our knowledge of phatic spaces in the
city. We will describe, in the following pages the main functions of RESCOM
which show the utility that this tool could have both on the epistemological and
operational level of the urban development.

It should be noted that all data that we will use to illustrate the iconographic
RESCOM products are the result of in situ work near the inhabitants of the town
center of Latakia in Syria.

7.2. Quantity of information contained in phatic spaces

To understand what a quantity of information may represent it is necessary, first

of all, to recall that a piece of information indicates, by definition, one or more
events among a finite set of possible events. Thus, there is an inversely proportional
relationship between the frequency of appearance of an event and the information
which it brings.

In a practical sense we may use the following example: if when looking for a type
of building in a city containing a set of buildings (N), we specify that this building is a
“mosque”, we provide information that reduces the search time proportionately to the
number of mosques (n) in the city. This information is all the more interesting, since it
reduces the number of later possibilities, in other words, it reduces uncertainty. We are
thus led to define the quantity of information as a growing function of N/n: the smaller
the number of mosques, the higher the ratio N/n and, thus, the higher the quantity of
information carried by the statement “it is a mosque”. According to Shannon, we can
express this definition mathematically:

QI = –log pi (1
1)
 RESCOM: Towards Multiagent Modeling of Urban Communication Spaces 133

where “QI” is the quantity of information carried by the statement and pi is the
probability of event i.

On the basis of the mathematical Shannon’s communication theory we propose

to apply this formulation of quantity of information to a set of phatic spaces
(e1, e2, ...en) by basing ourselves on descriptive criteria (category, type, time, etc.) of
each space. Let us take the example of a city holding six phatic spaces whose
descriptive criteria are established in the following table.

Space Category Type Time

e1 religious mosque 1940

e2 religious mosque 1980

e3 religious church 1940

e4 religious church 1920

e5 religious synagogue 1920

e6 public market 1900

Thus, space e1 has a probability of appearance of 5/6 in its category (five

religious spaces among six spaces studied), 2/6 for the type (mosque) and 2/6 for the
time (1940), thus a quantity of information equal to –log(5/6*2/6*2/6) = 3.4.
According to the same calculation, space e6 represents a quantity of information
equal to 7.7.
134 Graphs and Networks

On the basis of relation (1

1) and by using the table above we may calculate the
quantity of urban information contained in each of the considered phatic spaces.

Probability
Space Category Type Time Quantity of information

Cat Type Time

e1 religious mosque 1940 5/6 2/6 2/6 –log (5/6*2/6*2/6) = 3.4

e2 religious mosque 1980 5/6 2/6 1/6 –log (5/6*2/6*1/6) = 4.4

e3 religious church 1940 5/6 2/6 2/6 –log (5/6*2/6*2/6) = 3.4

e4 religious church 1920 5/6 2/6 2/6 -log (5/6*2/6*2/6) = 3.4

e5 religious synagogue 1920 5/6 1/6 2/6 –log (5/6*1/6*2/6) = 4.4

e6 public market 1900 1/6 1/6 1/6 –log (1/6*1/6*1/6) = 7.7

In terms of data processing the algorithm associated with this function can
generate the quantity of information following one, several or all the descriptive
criteria adopted for the spaces. In terms of programming, this algorithm is inbuilt
into an interactive instruction that makes it possible to choose the criterion or
criteria, according to which the user wishes to calculate and then to represent the
quantity of information contained in each studied phatic space.
 RESCOM: Towards Multiagent Modeling of Urban Communication Spaces 135

On the right: map of the quantity of

information contained in the phatic
spaces mentioned by the inhabitants of
the town center of Latakia. On the
right: various views in perspective
projection with a rendering showing
the town center and the quantity of
information in the phatic spaces in
Latakia. The values are represented by
cylinders. Everything is represented
against the cartographic background
of topographic services.

Design and plotting: O. Khaddour, Ph.

Mathis. CESA, 2003.

Figure 7.1. Various representations of the attractiveness of phatic

spaces according to Reilly’s model (see color section)
136 Graphs and Networks

The quantity of information contained by each phatic space constitutes a

component of its semantic contents and will thus form part of the graphic entity (the
point, for example), which represents it. This enabled us to use it in other
explanatory indices of the socio-cultural communication network in the city. Thus,
in RESCOM we have developed functions that calculate and represent phatic
centrality according to Levaitt and the degree of cognitive dissonance according to
Festinger [KHA 04], which we will not be described here in order to leave enough
space to demonstrate the usefulness of RESCOM on the operational level of the
urban development.

7.3. Prospective modeling in RESCOM

We tried to concretize the prospective dimension of RESCOM by developing the

modules concerning the representation of phatic attraction surfaces, on the basis of
which we proposed a “game” simulating the behavior of the individuals looking for
sources of information and communication in an urban environment. From the same
point of view, we have assigned to RESCOM a functionality which enables the
development of requests in pseudo-SQL relating to data used in RESCOM and,
finally, an inference module which helps observing the changes in spatial practices
after the establishment of a new phatic space.

7.3.1. Phatic attraction surfaces

Other possibilities to study phatic spaces on a cellular basis are also possible in
RESCOM, in particular with respect to graphic analysis of spatial data. Within this
framework we have tried to explore the potential of the gravity model to limit the
attraction surfaces of each phatic space. The interest of the gravitational model for
the estimation of the attraction surface had been pointed out in the 1920s by the
economist Reilly.

Reilly’s objective was to determine how the unsatisfied demand of an

intermediate place (m) located between the two larger cities (a and b) would be
distributed. Thus, he starts with two poles considered without dimension and studies
their influence on space [MAT 71a].

Va /Vb = (Pa /Pb)* (Dam²/Dbm²)

where Va and Vb are the sales of a and b in m, Pa and Pb the “Weight” or

population of a and b, and Dam and Dbm the distances from a and b to m. By
 RESCOM: Towards Multiagent Modeling of Urban Communication Spaces 137

generalizing Reilly’s statement we can calculate the attractions of all the points a,
b, … n at the point m.

Aij = Pi/(Distanceij)2 (2
2)

We then probabilize linearly by dividing each attraction by the sum of the line, in
order to pass from a ratio of attraction between two points to a generalization
between n points, which enables us to find Huff’s formulation

P{Arij} = Pi Dij–2 /Σk (Pi/Dij–2 )

In the case of phatic spaces where the weight is expressed in terms of quantity of
information we can, in all spatial points j, look for the phatic space that exerts the
strongest attraction on this point and thus trace the spatial extension of the maximum
influence areas of each space. As a result, the attraction surface of a geographical
site (phatic space in our model) draws a hyperbolic cone, which when projected onto
a map is represented by concentric circles of decreasing intensity around the site
concerned [PUM 01].

The use of cellular space and thus of the attributes of each of its cells led us
towards a specific formulation based on the observations of Philippe Mathis in
connection with “the fields’ equation” [MAT 71b], whose simulations were limited
to two-dimensional projections (the computer equipment in 1971 did not make it
possible to plot three-dimensional representations). Along the same lines, by using a
CAD system we set up a technique, which made it possible to generate three-
dimensional representations of attractiveness in a cellular space.

The first results obtained from this simulation do not coincide with the networks
obtained following the investigations. Thus, the most characteristic spaces in the
Latakia area, i.e. the University (in the right corner) and the Corniche Sud (in the
bottom corner), seem to become less attractive (since they are at the periphery of the
city) with respect to other spaces that are geographically more accessible in terms of
distance.

In fact, the use of quantity of information as “mass” revealed to us the need for a
specific development of the distance covered in search of this rather particular
“good”, since it is clear that for certain people neither the distance nor its cost play
the decisive part in their choice of phatic spaces.
138 Graphs and Networks

We have thus set up a prospective module in the form of a game aiming at

simulating the behavior of an individual who is subjected to the constraints of
belonging to a socio-cultural category (gender, age, profession, etc.) in his choice of
reference phatic space. Below we will explain the functioning of this module, as
well as its ramifications.

7.3.2. Game of choice

In parallel to the thermodynamic laws of Boltzmann and Shannon’s formulation,

we will suppose that a blank cell contains informational potential if only by its
“unique” localization in the grid constituting the geographical space of the city. That
is, in a grid comprising 12 cells where any cell contains a quantity of information
equivalent to: –log 1/12 = 0.2.

Consequently, in the case of movement within a grid of “n” isomorphic cells,

each cell that we pass through brings a quantity of information equal to (log 1/n).
The usefulness of this formulation is proven when the cell corresponds to a distance
so that the distance covered in the grid is expressed in cellular unit (cellular
distance). The equivalent of this distance in terms of information will be:

QId = –log (dc/ N)

where (dc) is the cellular distance and N is the total number of cells. By applying this
concept of cellular distance to a space containing several sources of information (p1,
p2, ..., pn), which correspond to various quantities of information (QI1, QI2, ..., QIn),
we can simulate the behavior of an individual traversing this space by consciously or
unconsciously looking for a maximum of information with a minimum effort and
cost.

For our game we suppose that the individual placed in this space must evaluate
all the available sources of information in order to choose one of them. However, the
fact of moving to get to the node corresponding to the individual’s search adds a
quantity of information generated by the distance covered.

QIt = QIn + QId

where “QIt” is the quantity of information that the agent could acquire during
movement, “QIn” is the quantity of nodal information that a source of information
 RESCOM: Towards Multiagent Modeling of Urban Communication Spaces 139

contains. Finally, “QId” is the quantity of information generated by distance. On the

basis of this formulation we generated the second simulation which shows the
difference between the effective choice of actors as revealed by surveys and the
choice of actors according to the opportunity to choose a space ensuring an optimum
quantity of information. The following figure shows, in particular, the great
variation of distances covered by individuals to get to various phatic spaces of their
environment.

Figure 7.2. Comparison between the network according to surveys (left)

and the network of the simulation according to a game of choice (right)

If we take a closer look at the statistical indicators of these two maps, we will
clearly see that, on the one hand, the distance covered is greater in reality than
according to the simulation and on the other hand, the scattering of values is also
more considerable in reality than in the model.

The summary table provided below therefore only represents the “temporary”
results showing the need for additional adjustment of the algorithms integrated into
RESCOM.
140 Graphs and Networks

Indicator According to surveys According to RESCOM

Maximum distance 3391.72 m 875.85 m

Minimum distance 29.09 m 29.09 m

Average distance 628.1 m 240.53 m

Standard deviation 906.3 197.92

By analyzing the link between distance and the quantity of information, which
we calculated on the basis of data received from the inhabitants of the town center of
Latakia, we noticed that the quantity of information increases with distance. The
logarithmic function in the algorithm specifying the calculation of the quantity of
information generated by distance can partially explain this phenomenon.
Otherwise, we see in this link an echo of Webber’s observation on the relationship
between the quantity of information received per hour and per person, the level of
specialization and the distance separating the urban domains [WEB 94].
Quantity of information (bit)

Figure 7.3. The link between distance and the quantity of information
for the inhabitants of the town center of Latakia

However, the following graph poses several questions about the nature of
thresholds where the quantity of information seems invariable despite the increase in
the distance covered (between 700 m and 3,500 m). However, this threshold groups
 RESCOM: Towards Multiagent Modeling of Urban Communication Spaces 141

together the majority of the choices expressed by individuals questioned during the
surveys. Within the framework of this chapter we can only propose partial
explanations by leaving the door open to future projects tackling this question.

Below, we will test several hypotheses that may lead to simulations that are
closer to the “limited rationality” of individuals in their choice of phatic spaces.

Let us recall that an individual with a perfect knowledge of his environment

really does not exist. For this reason we propose to incorporate in our simulation a
set of constraints that will be imposed on the mobile agents: proximity of a central
point, agent-specific resources, maximum distances to cover. We can also introduce
the relation to individuals “residing” in the neighboring cells, so that, if “the
majority of my neighbors go to a certain communication space, I go there too”.

QIt = QIn + QId + Qic

where
– “QIt” is the total quantity of information;
– “QIn” is the quantity of nodal information contained in a phatic space;
– “QId” is the quantity of information generated by distance where the distance
covered must be equal or lower than a maximum distance;
– “QIc” is the quantity of information representing the number of individuals
who choose a communication space. (Qic = log n/N, where n is the number of
individuals who choose a specific space and N is the total number of individuals
simulated.)

Lastly, we can add a multiplicative parameter of the binary order α – called

dummy variable by economists. This parameter makes the frequenting of a given
space by a certain category of individual either possible (α=1) or impossible (α=0).

QIt = (QIn + QId + QIc)*α (4)

The binary parameter works in situations where a certain socio-cultural code

almost completely reduces to zero the possibility that a Muslim man goes to a
church or that a Christian woman goes to a mosque. Consequently, we can simulate
situations where girls from districts known as “traditional” cannot go to nearby
phatic spaces and prefer to go further away in order to escape certain codes of
conduct (the prohibition of the mixing of genders).
142 Graphs and Networks

Once the quantity of information has been calculated according to the criteria
mentioned above, and in order to put into action a few hypotheses, we can simulate
the possible behavior of agents in search of sources of information and
communication in the urban environment. To this end we use a variation of Reilly’s
law, which has been widely used in geomarketing.

7.4. Huff’s approach

Huff’s model proposes taking into account the uncertainty regarding the exact
localization of the borders of areas of influence by defining a relational probability,
which is the relationship between the influence of a site and the influence of all the
sites. Huff’s problem can be formulated as follows:
– In a set of sites k, each site with a mass M and a location j represents, for the
individual located at a point i, an opportunity Oij that can be evaluated using Reilly’s
formula: Oij = Mj/(Dij)2.
– The probability of choosing a destination is equal to the opportunity of this
destination divided by the total sum of destination opportunities: Pij = Oij/Σk Oik.

By substituting mass with quantity of information we propose to simulate the

behaviors of individuals involved in a search for information and communication in
the city, by using the following formula:

Pij = [QIj/(Dij)2]/[Σk QIk/(Dik)2] (5)

where (QI) is the quantity of information as previously defined.

By limiting ourselves to the geometrical criteria of individuals (location in the

geographical space) we implemented the algorithm which is necessary to test the
formulation suggested above. The results obtained using this algorithm reveal a
considerable attenuation of the decisive character of Reilly’s law and reveal
behaviors that hardly fit a homo economicus.

Apart from the representation of the potential attraction of phatic spaces, this
simulation revealed to us other aspects of the socio-cultural network of
communication in the city. From the phenomenological point of view, by comparing
networks of individuals and their phatic spaces on several scales we discover auto-
similarity [MAT 03]. This leaves the door open to future research aimed at
thoroughly investigating the modeling of socio-cultural communication networks in
the city by following the principles of fractal geometry.
 RESCOM: Towards Multiagent Modeling of Urban Communication Spaces 143

Figure 7.4. 1. The attraction of phatic spaces according to a variation of Huff’s model,
2. zoom on the network showing the planar “aspect” of the graph,
3. cartography of the global communication network

Up until now we showed the possibilities that RESCOM possesses to develop and
help to interpret the spatial structure of communication. However, we marked out the
limits of this chapter by aiming primarily at representing socio-cultural communication
networks in the city, and secondly at the possibility of intervening as a designer and/or
a decision maker in the operation and the organization of these networks. This leads us
to the “inference” module that we have integrated in RESCOM.

7.5. Inference

We prefer to use the term inference, although it is not modern, to indicate

(according to the French “Petit Robert” dictionary) a “logical operation, by which one
admits a proposal in virtue of its connection to other proposals already held as true”.

By doing this it is possible to simulate the behavior of individuals in the event of

the opening of one or more new phatic spaces. The characteristics of this space,
144 Graphs and Networks

geometrical (location) as well as semantic (descriptive criteria), will play an

important part in its potential to attract individuals searching for an information and
communication space.

By keeping the example of the town of Latakia, we simulated a scenario of the

opening of several shopping malls trying to respect a spatial distribution covering
four sectors of the city: the city center (C2), the northern periphery (C3), the eastern
periphery (C4) and the unused port grounds in the western part of the city (C1).

The following illustration clearly shows that, in fact, the shopping malls located
in the city center sector would be more attractive than those located in the periphery.
It should be noted that the shopping mall that we proposed on the unused grounds of
the port (C1) would be more attractive compared to that located to the east of the
town center (C2).

C3

Corniche Ouest

Rue Mutanabi

Rue Bagdad

C2

C1
Place Al Yaman

Universitˇ

C4
Corniche SUD

Figure 7.5. Illustration: Simulation of the attractiveness of the new phatic spaces suggested
for a project. Design and drawing: O. Khaddour, Ph. MATHIS. CESA, 2003. Urban planning

On the other hand, the introduction of these new phatic spaces did not reduce the
performance of the already existing spaces, in particular, Rue Baghdad and rue
Mutanabî, which coincides with the results obtained by using Huff’s approach. This
leads us to believe that these spaces deserve careful consideration, first of all, to
reveal their architectural and urban characteristics, but also in order to propose the
necessary interventions in terms of urban development operations which are aimed
to improve their communication nature.
 RESCOM: Towards Multiagent Modeling of Urban Communication Spaces 145

7.6. Conclusion

We believe that the quantity of information, such as we have formulated it, is

still far from having revealed its entire potential, due to the fact that we could not
make a comparative study between several cities, because the mathematical
formulation of this function leads to a specific notion of entropy which makes it
possible to define the specificity of each type of city. Consequently, the quantity of
information would differ for each city depending on its entropy.

We also believe that it is possible to use the model of indifference of Nicos

Devletoglou to reveal more characteristics of the areas of influence of phatic spaces
in the city. Set in a duopoly, this model allows the consumer to make choices
according to the distance to cover. If the difference in distance between two poles
decreases (while remaining equal or larger than a certain value), there will arrive a
moment when the “customer” regards them as equidistant from his residence: the
difference in distance then becomes imperceptible. That leads Devletgolou [DEV
65] to propose the concept of perceived minimal difference which corresponds to a
zone covering the areas of influence of two poles, inside which individuals have no
spatial preferences, but are rather submitted to a kind of mode effect pushing them
to follow mass behavior in a larger zone [BER 71]. We generalize by calculating a
zone of indifference between two dominant attractions which enable the
cartography.

In fact, if we generalize Reilly, we obtain Huff’s formula, but each point of space
is still subjected to a dominant attraction except in the case of triangular or square
distribution of strictly identical poles or in extremely improbable cases where the
difference in weight would be strictly compensated by a difference of the square of
the distances. In reality a point is always located in a single dominant zone or at the
border or the intersection of two borders, although then it is necessary to take into
account “the thickness” of the border. The generalization of Develtoglou introduced
by Mathis [MAT 71a] (generalization to any exponent of the distance n and to poles
of different weights) made it possible to reveal indifference zones which have very
different or even dissymmetric form.

Thus, we finish these few illustrations on the manner in which we chose to use
RESCOM. Let us note that a developer, a town planner or an architect continuously
propose virtual, hypothetical places that might well be accepted by reality as well as
refused or neglected. Consequently, all information drawn from simulations can be
validated or rejected only through experience which generates a new perspective.
RESCOM is restricted to the role of a “machine of questions” striving unceasingly
to acquire new data and generate knowledge by means of new descriptive criteria of
new phatic spaces and/or their users.
146 Graphs and Networks

7.8. Bibliography

[BER 71] BERRY B., 1971: Géographie des marchés et du commerce du detail, Paris:
Armand Colin, pp. 153-154.
[DEV 65] DEVLETGOLOU N., 1965: “A Dissenting View of Duopoly and Spatial
Competition”. Economica. Vol. 32. pp. 140-160.
[KHA 04] KHADDOUR O., 2004: Modélisation individu-centré des espaces de
communication dans la ville. Vers une modernisation de la représentation de
l’environnement urbain au Levant, PhD Thesis, University of Tours.
[MAT 71a] MATHIS P., 1971: Introduction à une théorie unitaire des implantations
commerciales. PhD Thesis, Paris I University, p. 234.
[MAT 71b] MATHIS P., 1971: op. cit. pp. 290-291.
[MAT 03] MATHIS P., 2003: “Une méthode de simulation et de visualisation multiniveaux”
in MATHIS P. (ed.): Graphes et réseaux, modélisation multiniveaux, Paris: Hermes,
p. 286.
[PUM 01] PUMAIN D., SAINT-JULIEN T., 2001: Les interactions spatiales. Paris: Armand
Colin, p. 37.
[WEB 94] WEBBER M. M., 1994: L’urbain sans lieu ni bornes. Paris: Aube, p. 82.
 Chapter 8

Traffic Lanes and Emissions of Pollutants

8.1. Graphs and pollutants emission by trucks

The formulae used to calculate the emissions of vehicles (regardless of the

category) take into account the type of vehicle (catalytic or not, diesel, petrol, etc.),
automobile park, weight (cars, utility vehicles or trucks), highway type associated
with characteristic driving cycles (urban, rural and road). It appears that a vehicle’s
load has little influence on the emissions of cars as opposed to trucks, more so since
it varies between load when empty and fully loaded vehicle. There is, therefore, a
strong link between the load of a vehicle, the consumption of energy and emissions.
The larger the load (increase in resistance to the movement of the truck) the more
work is required to drive that mass at a constant speed (70 km/h, for example). This
resistance depends on the load of the truck, but also on the gradient of the road,
since the work required from the engine is not the same during ascent as during
descent. Consumption as well as the fuel breakdown products are then greater or
smaller depending on the needs.

We easily calculate (i.e., obtain a satisfactory evaluation of the emissions) the

annual emissions of light vehicles according to vehicle, type of driving (urban rural,
etc.) and annual mileage. However, this method is not easily applicable to the trucks
(effect of mass), if the gradient of the road is considered. It is, thus, of primary
importance to consider the routes (origin, destination, gradient, etc.) in addition to
the method used for passenger vehicles.

Chapter written by Christophe DECOUPIGNY.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
148 Graphs and Networks

We will show in several stages how the use of the graph theory may or may not
be justified: what does it bring in the evaluation of heavy vehicles emissions? We
will also tackle the problem of the integration of the road gradient into a hitherto
two-dimensional graph.

For that we will first present the method used, in order to offer a better
understanding of our process and results. As the aim is not to comment on the
method of calculating the emissions, we will limit ourselves to evaluating carbon
monoxide (CO) and nitrogen oxides (NOx) for trucks lighter than 7.5 tons.
Emissions will be calculated by using a programming protocol in Visual BASIC,
which is added to the NOD model by Laurent Chapelon1.

We will work in two stages. The first consists of observing the behavior of
emissions depending on speed, slope and load on a section of the A28 highway. The
second will enable us to extend the emissions calculation formulae to a road
transport system graph.

Figure 8.1 shows the section of A28 developed on the basis of a digital terrain
model [SER 00], whereas Figure 8.2 represents the graph of the French road
network. This graph is based on a file with 300 points representing the largest
French cities, either by population or by importance in the road network, and a file
with 912 undirected arcs (or 1,824 directed arcs). This file is composed by using a
national transportation system logic. There are five hierarchical levels of roads, to
which a mean velocity for each type of vehicle is associated (that is, five functional
binomials2 for trucks lighter than 7.5 tons):
– highway: 110 km/h for cars and 70 km/h for trucks;
– expressways: 90 km/h and 60 km/h for the latter;
– roads: 70 - 70 km/h and 55 for trucks (in the majority they correspond to trunk
roads);
– roads: 60 and 50 (40 and 30 km/h for trucks).

Several types of arcs may connect two cities. The graph of the transport system
is a p-graph where width or colors correspond to the various partial infrastructure
networks. The partial graphs defining the infrastructures are directed, since the
nodes have an altitude, in order to calculate an average slope for the arcs.

1 “Software for space-time evaluation of transport supply modification projects”, design and
realization: Laurent Chapelon, Philippe Mathis, CESA laboratory version 4.2, August 1998.
2 See Chapelon, ibid.
 Traffic Lanes and Emissions of Pollutants 149

A28 SEQUENCE AROUND “CHANCEAUX – SUR – CHOISILLE”

North

Section of the A28

PROFILE

Profile and absolute value of the slopes

Ch. DECOUPIGNY, K. SERRHINI, CESA April 2000

Figure 8.1. Profile of a section of A28

Road network
Ch. Decoupigny, K. Serrhini, CESA Laboratory, April 2000
Highways and
expressways
70 roads
60 and 50 roads

Figure 8.2. Graph of the road network

150 Graphs and Networks

8.1.1. Calculation of emissions

We use formulae of the hot emissions for heavy-duty vehicles that take into
account the weight of the empty vehicle, the road gradient and the load state of the
vehicles [HIC 99]. They are derived using data from engine test-bed measurements
according to the standardized driving cycles performed on vehicles that are
representative of the vehicle fleet. They are expressed in grams per kilometer
according to speed, the road gradient, the load and type of the vehicle.

8.1.1.1. According to speed

There are several cases depending on the pollutant considered and the speed of
the trucks.

For carbon monoxide:

– speed ranging between 5 and 100 km/h

HotCO,j,k = 37.28 Vk-0. 6945 [8.1]

where:
– hotCO,j,k is the hot emission in g/km of vehicle j over an arc k;
– Vk is the average speed over the arc k.

For nitrogen oxides:

– speed ranging between 5 and 50 km/h

HotNOx,j,k = 50.305 Vk-0. 7708 [8.2]

– speed ranging between 50 and 100 km/h

HotNOx,j,k = 0.0014 Vk2 – 0.1737 Vk + 7.5506 [8.3]

We then have the data for five pollutants i, carbon mono and dioxide (CO and
CO2), nitrogen oxides (NOx), particles (PM), and volatile organic compounds
(VOC). There are four classes of vehicles j and five classes of arcs k, to which the
mean velocities Vk are associated.
 Traffic Lanes and Emissions of Pollutants 151

We then obtain the unitary emission in grams per kilometer for a pollutant and a
type of empty vehicle for each arc of the graph. After that the slope factor can be
introduced into the calculations.

8.1.1.2. According to the slope

Emissions according to the slope are determined by multiplying the emissions of
an empty vehicle without slope by a correction factor [HAS 97]:

ASi,j,k = A6,i,j,k Vk6 + A5,I,j,k Vk5 + A4,I,j,k Vk4 + A3,I,j,k

Vk3 + A2,I,j,k Vk2 + A1,I,j,k Vk + A0 [8.4]

where:
– ASi,j,k is the correction factor for a pollutant i of a vehicle j over an arc k with
a slope ranging between – 6% and 6%;
– Vk is the average velocity over an arc k;
– A6, . . . . A0 are constants for each pollutant i of a vehicle j.

The emissions according to the slope are then given by the relation:

EChoti,j,k = ASi,j,k * Ehoti,j,k [8.5]

8.1.1.3. According to the load

In the case that interests us the load of the truck has great importance. Indeed,
fuel consumption and, thus, emissions vary with the load of the vehicle. As well as
taking slope into account, there exists a correction factor function for emissions
according to the load [HIC 97]:

Φ (γ,v) = K + n*γ + p*γ2 + q*γ3 + r*v + s*v2 + t*v3 + u/v [8.6]

where:
– K is a constant;
– n - u are coefficients;
– γ is the gradient in %;
– v is the average speed.
152 Graphs and Networks

The emissions are then given by the relation:

ε = εu * Φ (γ, v) [8.7]

where:
– ε is the emissions rate corrected for the load and the gradient in g/km;
– εu is the base emissions factor;
– Φ (γ, v) is the load correction function;
– γ is the slope in %;
– v is the average speed.

We have the emissions in grams per kilometer for our two pollutants, according
to the load of the vehicle and the slope for each arc of our graph. From here on each
arc of the graph is defined by average slope, distance, speed of circulation and
unitary emission in g/km for each pollutant.

8.1.2. Calculation of the minimum paths

For the calculation we use Floyd’s algorithm. This latter is carried out in three
stages: the first is the creation of the matrix of minimum arcs, followed by Floyd’s
preparation and, finally, the algorithm itself.

8.1.2.1. The matrix of minimum arcs

For any arc connecting two nodes i and j we determine the one with the smallest
value of the selected character. It may be the shortest distance or driving time, etc.

8.1.2.2. Floyd’s preparation

This is a simple but important stage because it enables the transition from the
concept of an arc to that of a path. For any pair of items i and j, if there is a
minimum arc connecting these two nodes, then the path between i and j is
implemented with the value of this arc, otherwise we assign the value 32,000 to the
path, given that regardless of the pair of points, the value of the path is less than
32,000.

Thus, we have the matrix of paths which are not yet minimum, where all the
values of the paths between two items i and j are defined, but not determined. This
simply means that when the path between two nodes is composed of only one arc, it
is at the same time defined and minimum. On the contrary, if it is composed of
 Traffic Lanes and Emissions of Pollutants 153

several arcs, its value is automatically fixed at that of the largest integer, waiting for
a lower value that would replace it, if this exists after Floyd calculates them.

8.1.2.3. Floyd’s algorithm

This is carried out in three stages:
– calculation of a path between i and j;
– comparison with the already defined value;
– replacement if the newly calculated value is less than that already existing.

The matrix obtained is called matrix of precedents, since we have not only the
values of the minimum paths between all pairs of nodes i and j, but also of the index
of the previous node, which is the destination node of the path, thus making it
possible to find the routes.

We note by PathMin (…) the shortest time value, for example, to connect city i
to city j.

We then have, according to the parameters of minimum arcs selection, four

matrices of precedents:
– according to time: PathMin (i, j, time);
– according to the distance in kilometers: PathMin (i,j, distance);
– according to the minimum emission of the pollutant considered: PathMin (i, j,
emission);
– according to the percentage of emission (emission indicator): PathMin (i, j, %
emission).

Once we know for each arc the distance in kilometers, the time of driving
through the arc according to the speed and the unitary emissions, it is easy to
determine for all the minimum paths, regardless of the selection parameter, the
driving time, the distance covered and the emissions of all the paths.

We can thus compare the elements of the same matrix, but also the various
matrices two by two. We have various methods of comparison:
– monopolar: we study, for example, the accessibility from all the cities to city i.
We can also isolate two cities to study their relations;
– multipolar: we study, for example, the accessibility from any point to all the
others.
154 Graphs and Networks

8.1.3. Analysis of subsets

We have developed this method to study Programs one European Spatial

Planning (SPESP)3. It made it possible to fill in the absence of the traditional nodal
approach (monopolar or multipolar) by offering a correct definition of the corridors
between large geographical areas (such as the Ruhr, the plain of the river Po or Ile-
de-France), on the one hand, and the variation indicators of these corridors, on the
other hand. It makes it possible to compare territorial subsets between them or with
the rest of the area studied by proposing their specificities, their resemblances or
dissimilarities.

8.1.3.1. Implementation of the method

Each node is defined by a name, a code, coordinates and a number of row i in the
file:

number of row
Node(77).x = 537400
Node(77).y = 2383299 coordinates
Node(77).z = 142
Node(77).name = Chartres
Node(77).code = 2800 code of the node corresponding to the department, here
28.

The fact that a node belongs to a subset is fixed by an index. By default, the
absence of an index (or 0 index) means that the nodes are elements of our space of
origin, but do not belong to any subset. Hereunder we define the nodes belonging to
subsets that have to be studied by indices other than zero. In the example treated
later on in this chapter we chose to select two sets: the cities of the Nord-Pas-de-
Calais (index 1) for the first, and index 2 for the cities in departments on the
Mediterranean coast.

3 Study Programs One, European Spatial Planning, Working Group 1.1 Geographical
Position, December 1999.
 Traffic Lanes and Emissions of Pollutants 155

EXAMPLE.–
Node(5,2).x = 846400.
Node(5,2).y = 1815199.
Node(5,2).z = 48.
Node(5,2).name = Marseille.
Node(5,2).code = 1300.
Node(50,1).x = 650099.
Node(50,1).y = 2626199.
Node(50,1).z = 51.
Node(50,1).nom = Lille.
Node(50,1).code = 5900.

Index of the subset

The number of subsets is limited to n-2 (n being the number of nodes). Indeed, if
this number is equal to n or n-1, all the nodes are element sets, bringing us back to
monopolar or multipolar nodal analysis. Similarly, the number of nodes included in
a set is limited. Two cases are possible:
– if there is only one subset indexed 1, then the number of elements of this subset
must be strictly larger than 1 and less than or equal to n-2;
– if there are several subsets (at least two indexed 1, 2), then each set must
comprise at least one element and no more than n-nx elements (where nx is the
number of elements already belonging to a subset).

Let us note that we exclude the possibility of an element belonging to two

subsets.
– formal statement:
- let there be a graph G = [X,U],
- X set of vertices or nodes numbered i = 1, 2, 3, etc., N,
- U set u(i, j) of ordered pairs of nodes or arcs with i ∈ X and j ∈ X.

We define a subset as a subgraph GA = [Xa, Ua] such that GA ∈ G, with the

following.
156 Graphs and Networks

1,1 2,1 3 4,2 5,2 N Σ

Val (1,1;2,1)
1,1 0 Σ [1,1;N]
P (1,1;2,1)

Val (2,1;1,1)
2,1 0 Σ [2,1;N]
P (2,1;1,1)

Val (3;1,1) Val (3;2,1)

3 0 Σ [3;N]
P (3;1,1) P (3;2,1)

Val (4,2;1,1) Val (4,2;2,1)

4,2 0 Σ [3,1;N]
P (4,2;1,1) P (4,2;2,1)

5,2 0 Σ [5,1;N]

6 Σ [6;N]

N
R
0 Σ [N;N]

Σ Σ [N;1,1] Σ [N;2,1] Σ
Σ [N;3] Σ [N;4,2] Σ [N;5,2] [N;N] Σ Σ
[N;N]

Σ [N, n1] Σ [N, N2]

where Val (etc.) is the value of the minimum path and P (etc.) is the preceding node

Table 8.1. Matrix of precedents and subset

 Traffic Lanes and Emissions of Pollutants 157

where the index of the subset numbered A = 1, 2, 3, etc., is M with 1 ≤ M ≤ N – 2 if

G1 ∩ G2 ∩ G3 ∩, .... ∩ GM = ∅4.

And Xa ∈ X the set of nodes of the subgraph numbered a = 1, 2, 3, etc., n, with:

If A = 1, then 1 < a ≤ N – 2.

If A > 1, then 1 ≤ a ≤ (N) – ∑ X a.

where ∑ Xa is the sum of the elements already belonging to a subset.

Each subgraph defined in this way is included in the graph of the original
transport network with the same typology of infrastructure. The analysis of several
subgraphs highlights relations included in the original graph.

We thus have our various territorial subsets of our original space, upon which we
employ Floyd’s algorithm to obtain the following matrix of minimum paths (see
Table 8.1). In this matrix we observe the minimum paths from a point to any other
point, which are organized in three groups: from light gray to dark gray:
– the first: relations of set 1 that are either at the destination or at the origin of all
the other nodes;
– the second: relations of set 2 that are either at the destination or at the origin of
all the other nodes;
– the third: relations of set 1 that are either at the destination or at the origin of
all the nodes of set 2.

In the case of pollutants emissions, ∑[n,n1] and ∑[n,n2] are the sums of the
emissions of each path bound for the nodes of the sets 1 and 2, ∑[n1, n] and ∑[n2, n]
are the sums of the emissions of the paths originating at the nodes of sets 1 and 2
and bound for all the nodes of the graph. In our example we defined two subsets but
more can be created. We may choose to analyze the relations between a subset and
the rest of the space (analogy with the method of traditional monopolar analysis) or
of several subsets, either between them or with respect to the whole of the graph.

The relations between the various elements of our graph can be defined
according to four strategies of movement between two nodes.

4 In the example developed below we decided to prohibit the intersection of subsets.

However, it is possible for an element to belong to two subsets. This possibility has the
advantage of enabling us to widen the type of analysis suggested by comparing the relevance
of an element – or a group of elements – belonging to a particular subset, for example, from
the point of view of accessibility.
158 Graphs and Networks

8.1.3.2. Indicators used or choice of the selection parameter

The objective being to quantify the emissions for the arcs, on the one hand, and
for the paths, on the other hand, we chose to select:
– minimization of driving time;
– minimization of the network distance in kilometers;
– minimization of the emission for a given pollutant and type of vehicle (unitary
emission in g/km multiplied by the distance covered);
– minimization of the emission of all the pollutant together for a given type of
vehicle = synthetic indicator.

8.1.3.3. Why a synthetic indicator?

As we will see later on, pollutants do not exhibit the same behavior with respect
to slope, load or speed. The least polluting path for carbon monoxide is not
necessarily the least polluting path for nitrogen oxide. Therefore, how do we
determine, if a path is more polluting overall than another? It is not enough to add
up the quantities, since CO or NOx emissions are to the order of grams per
kilometer, whereas the emissions of CO2, for example, are in the order of hundreds
of g/km. For this reason the following synthetic indicator is proposed:

For each pollutant we have the cumulated emission in grams per arc (unitary
emission in g/km multiplied by the network distance of the arc):
– let Ei, j, k be the emission of pollutant i of the vehicle j over the arc k in grams;
– regardless of k → number of arc, ∃ at least an arc with the strongest emission
of the pollutant noted by Emax i, j;
– let PEi, j, k be the proportion of the emission of pollutant i of the vehicle j over
an arc k with respect to Emax i, j such that:
- regardless of k → number of arc, PEi, j, k = Ei,j,k / Emax i,j,
- we note by %Ej, k for an arc k and a vehicle j the percentage of emission of
all the pollutants together such that:
- %Ej, k = (PEco,j,k + PEnox,j,k) * 100 / Nbi, where Nbi is the number of
pollutants considered.

We then obtain for all the arcs of our graph the percentage of emission of all the
pollutants together with respect to the maximum emissions of each pollutant without
taking into account the unit of mass. Consequently, we can determine the paths with
the minimum emission of all the pollutants together.
 Traffic Lanes and Emissions of Pollutants 159

8.2. Results

We will start with simulations on the section of the A28 to highlight the relations
between emissions, speed, slope of the arcs and load of the trucks. It will then be
easier to identify the links between graph and emissions, first of all, for movement
between two cities and then for relations between two subsets.

8.2.1. Section of the A28

The calculation of emissions requires certain modifications of the profile of the

A28. The values of slopes were limited so as to lie between – 4% and + 4%, in order
for the network speed defined beforehand to agree with the speed limits adopted in
the formulae. From the digital terrain model (Figure 8.2) we work out the profile of
the section, which is composed of 40 arcs defined by two nodes, a slope and a
length. We then calculated the emissions for each arc (for various initial conditions).

The results presented below represent the emissions for the section in its totality
according to the direction of movement – direction 1: movement from left to right,
direction 2: movement from right to left– of the slope of each arc, speed (constant
for all the arcs) and of the load.

Tables 8.2 and 8.3 recapitulate the cumulative emissions on the section of 2.9 km
for three speeds.

Emission of co of an empty vehicle and 1.95 g/km

without slope at 70 km/h 5.67 g

Emission of co of an empty vehicle and 2.305 g/km

without slope at 55 km/h 6.7 g

Emission of co of an empty vehicle and 2.876 g/km

without slope at 40 km/h 8.36 g

Emission of NOx of an empty vehicle and 2.251 g/km

without slope at 70 km/h 6.55 g

Emission of NOx of an empty vehicle and 2.291 g/km

without slope at 55 km/h 6.66 g

Emission of NOx of an empty vehicle and 2.929 g/km

without slope at 40 km/h 8.52 g

Table 8.2. Emissions of empty vehicles and without slope

160 Graphs and Networks

We observe in these tables that the effects of the variables measured on CO or

NOx emissions are the same:
– regardless of slope or load, the more the speed increases, the more the emission
decreases;
– there is over-emission or under-emission with respect to the emission of empty
vehicles and without slope when we take slope and load into account;
– the emissions are different according to the direction of circulation.

On the other hand, the impacts of speed, slope or load do not have the same
intensity. Table 8.4 shows the variations of emissions according to speed.

Function of
Function of Function of Function of
slope and
slope: slope: slope and load:
load: direction
direction 1 direction 2 direction 1
2

Emission of
5.124 6.398 5.337 6.871
CO at 70 km/h

Emission of
6.397 7.76 6.813 8.526
CO at 55 km/h

Emission of
8.376 9.906 9.028 11.02
CO at 40 km/h

Emission of
NOx at 70 5.715 8.352 6.77 10.26
km/h

Emission of
NOx at 55 6.199 8.909 7.5 11.19
km/h

Emission of
NOx at 40 7.981 10.844 9.685 13.725
km/h

Table 8.3. Emissions (in grams) according to slope and load

 Traffic Lanes and Emissions of Pollutants 161

CO Zero slope Negative slope Positive slope

Difference between 55 and
+ 18 + 30 + 19
70 km/h in %
Difference between 40 and
+ 19 + 35 + 26
55 km/h in %
NOx
Difference between 55 and
+ 1.7 + 13 +6
70 km/h in %
Difference between 40 and
+ 27.5 + 41 + 15
55 km/h in %

Table 8.4. Variation of emission depending on the slope, in %

This table indicates that for the same slope the effect of the reduction of the
speed on the emissions is more marked during movement down a negative slope.
However, the emission in grams per kilometer is greater for positive slopes than for
negative ones. In addition, we can see that CO seems more sensitive to speed,
whereas NOx seem more sensitive to slope (Table 8.5).

For CO, the emissions for a negative slope are lower than those for a zero slope,
even more so when speed increases. Conversely, the emissions increase for positive
slopes – compared to zero slopes – as speed decreases. This behavior is easily
understood through the study of forces applied to a mass moving on a tilted plane.
The CO emissions are proportional to fuel consumption. On the other hand, the
behavior of NOx is more ambiguous. However, we can say that the slope of the road
affect the emissions of NOx more than those of CO.

CO 70 km/h 55 km/h 40 km/h

Difference between negative
– 28 – 20.8 – 13.9
slope and zero slope
Difference between positive
+ 31 + 33 + 35
slope and zero slope
NOx
Difference between negative
– 45 – 39 – 32
slope and zero slope
Difference between positive
+ 70 + 68 + 56
slope and zero slope

Table 8.5. Variation of emissions according to slope, in %

162 Graphs and Networks

Hereafter the slope – and thus the profile – developed here diverges during
calculation of the emissions on the graph. Indeed, the slope associated with an arc
will be calculated on the basis of altitudes of the nodes of the arc, and we will
associate the path between an origin and a destination to a profile. For this reason
Table 8.6 provides, on the one hand, the emissions according to an average slope
calculated for the first and the last node of the profile and, on the other hand, the
average emission – value in brackets – calculated for the emissions of each arc.

CO At 70 km/h At 55 km/h At 40 km/h

Direction 1: slope of
1.4 (1.76) 1.82 (2.2) 2.47 (2.88)
– 0.2 %

Direction 2: slope of 0.2% 2.56 (2.2) 3.06 (2.67) 3.86 (3.41)

Table 8.6. Variation of emissions with load

When we calculate the emissions for an average slope we observe in our profile
that emissions are underestimated in the descending direction of the profile and
overestimated in the ascending direction. Indeed, during calculation with the average
slope we consider a constant slope for the entire profile without taking into account
the rises and descents that may follow one another.

Thus, the behaviors of CO and NOx are different according to the pair (speed,
slope). Moreover, it appears that CO is less sensitive to the slope than NOx. This
explains the difference in emissions between direction 1 and direction 2. As a result,
for our profile a movement from left to right (direction 1) can be described as
downward movement, whereas from right to left (direction 2) we have an ascending
movement.
 Traffic Lanes and Emissions of Pollutants 163

at 70 km/h CO NOx

p=0 + 3.5% + 14%

p<0 + 3.5% + 13%

p>0 + 18% + 41%

at 55 km/h CO NOx

p=0 + 6.5% + 17%

p<0 + 4.3% + 13.6%

p>0 + 21% + 44%

at 40 km/h CO NOx

p=0 + 8% + 19%

p<0 + 5.6% + 15.3%

p>0 + 22.7% + 45.8%

where p is the slope

Table 8.7. Variation of emissions according to the slope in %

CO 70 km/h 55 km/h 40 km/h

Difference between negative slope and zero

– 28 – 22 – 15.8
slope

Difference between positive slope and zero

+ 50 + 51 + 52
slope

NOx

Difference between negative slope and zero

– 45 – 41 – 35
slope

Difference between positive slope and zero

+ 99 + 107 + 91
slope

Table 8.8. Variation of emissions according to the slope in %

164 Graphs and Networks

For both pollutants there is over-emission at full load (it is stronger for NOx),
which is amplified for a positive slope (Table 8.7). The difference between vehicles
with or without load on the same slopes increases slightly with the reduction of the
speed. We also notice that the percentage of the slope accentuates the effect of load
on the emissions (Table 8.8), especially in the case of NOx. The latter are more
sensitive to the load of the vehicle than CO.

Taking into account the load of the vehicle does not modify the behavior of the
pollutants, however, it intensifies the behaviors of the pollutants.

With this first approach we show that emissions are specific to the profile,
according to the origin and destination of the movement and depending on the
pollutants considered. We have also shown the sensitivities of the pollutants with
respect to the initial conditions (speed, slope and load). We will now consider the
case of graphs. However, before proceeding it is necessary to reflect on the relations
of the trinomial composed of relief, infrastructure and the graph used.

We saw previously that the method used below underestimates emissions for
negative slopes and overestimates emissions for positive slopes. Consequently, our
approach can appear satisfactory in areas that are not too uneven (Rhone valley,
Landes, etc.) and much less so in more hilly areas.

In addition, maximum slopes vary depending on the road infrastructure used. For
highway structures, they are minimized with respect to the relief. Moreover, the
kinematics of driving are very different for different types of infrastructure; on
highways speeds are very close to the average speed, as opposed to trunk roads, for
example, where speeds fluctuate more (entering an agglomeration, passing through a
village, etc.). We deduce from it that emissions are better estimated for highway
structures, especially if the profile is not very uneven. It is important to dissociate
relief and profile in the case considered (the estimate of the emissions is finer for
alpine highways which follow the bottoms of valleys than for hilly areas where
descents and ascents follow one another). Consequently, the graph used is
dominating. A simple but paramount question needs to be posed: how has the graph
been elaborated? This question does not refer to techniques of digitalization or
modeling, but rather to its later use: will we calculate accessibilities, plot
Chronomaps [LHO 97] or estimate emissions? It should be specified that the graph
used here was not developed to calculate emissions of pollutants from trucks, but for
calculations of driving time. It was slightly modified for this new use.
 Traffic Lanes and Emissions of Pollutants 165

8.2.2. French graph

Having studied the behaviors of pollutants for a profile we will generalize on a

graph in two stages. The first consists of studying the Paris-Marseille and Paris-
Toulouse routes, whereas the second is between two geographical subsets.

We do not have the profile along each arc and, nevertheless, we associate a path
(or a route) to a profile.

Table 8.9 below shows the emissions of CO and NOx for the return journey
according to three criteria of minimum paths selection.

Paris-Toulouse Paris-Marseille Toulouse-Paris Marseille-

Paris

CO NOx CO NOx CO NOx CO NOx

time 1.39 1.63 1.51 1.81 1.47 1.81 1.76 2

mileage 1.88 1.82 1.54 1.66 1.51 1.83 1.76 2.11

emissions 1.3 1.55 1.46 1.55 1.45 1.5 1.55 1.65

Table 8.9. Cumulated emissions of CO and NOx in kilograms

Figures 8.3 and 8.4 present the minimum paths bound for Toulouse and
Marseille and bound for Paris, as well as the carbon monoxide emitted over each arc
by a vehicle. Several comments can be made concerning carbon monoxide:
– the traffic corridors used are naturally the same (with different means) except
for the journey from Marseille to Paris according to the emission;
– the south-north direction is more polluting;
– generally speaking, finding the optimal routes according to the minimum
emission amounts to optimizing the driving time except for the journey from
Marseille to Paris;
– the emission according to mileage is the most polluting.

These observations are easily understood. According to time we optimize the

route according to the pair speed-distance, whereas according to mileage we
optimize only the distance and for the emission we try to optimize the trio speed-
distance-slope.
166 Graphs and Networks

Emission according to time Emission according to emission

Emission according to distance
minimum in g/km minimum in g/km
minimum in g/km
2.66 – 3.07 2.66 – 3.07 2.66 – 3.07
2.24 – 2.66 2.24 – 2.66 2.24 – 2.66
1.82 – 2.24 1.82 – 2.24 1.82 – 2.24
1.4 – 1.82 1.4 – 1.82 1.4 – 1.82

Road network Ch. DECOUPIGNY, CESA JANUARY 2001

Highways and
expressways
70 roads
60 and 50 roads

Figure 8.3. Emission of CO per arc, of trucks lighter than 7.5 t

and minimum paths, north-south direction

Emission according to time Emission according to distance Emission according to emission

minimum in g/km minimum in g/km minimum in g/km
2.66 – 3.07 2.66 – 3.07 2.66 – 3.07
2.24 – 2.66 2.24 – 2.66 2.24 – 2.66
1.82 – 2.24 1.82 – 2.24 1.82 – 2.24
1.4 – 1.82 1.4 – 1.82 1.4 – 1.82

Road network Ch. DECOUPIGNY, CESA JANUARY 2001

Highways and
expressways
70 roads
60 and 50 roads

Figure 8.4. Emission of CO per arc, of trucks lighter than 7.5 t

and minimum paths, south-north direction
 Traffic Lanes and Emissions of Pollutants 167

Figures 8.5 and 8.6 show the minimum paths for the return journey, as well as
the emissions of nitrogen oxides per arc. We observe that:
– the traffic corridors used are the same, as expected, for different means in both
directions except in the North to South direction according to emission;
– contrary to CO, the difference between paths according to minimum emission
and to time is a lot larger;
– the North-South direction is more polluting;
– the emission according to mileage is the most polluting.

There is greater variability of routing for NOx than for CO. We saw above that
NOx is more sensitive to the slope than CO. We realize in this example that NOx is
more sensitive to the trio speed-distance-slope than CO. NOx is thus more sensitive
to the structure of the network and the initial conditions than CO.

By observing the unitary and cumulative emissions values for each path it
appears that there exists little difference between an ecological approach seeking to
minimize emissions and a transport approach seeking to minimize travel time (in
particular for CO). We find here in a general manner the results presented for the
profile of the A28.

Emission according to emission

Emission according to time Emission according to distance
minimum in g/km
minimum in g/km minimum in g/km
3.76 – 4.55
3.7 – 4.5 3.79 – 4.6 2.18 – 2.97
1.3 – 2.1 1.39 – 2.19 1.39 – 2.18

Road network
Ch. DECOUPIGNY, CESA JANUARY 2001
Highways and
Expressways
70 roads
60 and 50 roads

Figure 8.5. Emission of NOx per arc of trucks lighter than 7.5 t
and minimum paths, north-south direction
168 Graphs and Networks

Emission according to time Emission according to distance Emission according to emission

minimum in g/km minimum in g/km minimum in g/km

3.75 – 4.55 3.83 – 4.64

3.74 – 4.53 2.19 – 3.01
1.37 – 2.16 1.37 – 2.17 1.38 – 2.19

Road network Ch. DECOUPIGNY, CESA JANUARY 2001

Highways and
Expressways
70 roads
60 and 50 roads

Figure 8.6. Emission of NOx per arc, of trucks lighter than 7.5 t
and minimum paths, south-north direction

The contribution of graphs to this case makes it possible to locate the most
polluting arcs and to highlight the heterogenity of emissions with respect to the
network. According to the latter, the origin and the destination of displacement we
obtain different emissions. It is on a local scale that the contribution of graphs will
be the most relevant. This local scale may be integrated and thus analyzed, in
particular, due to nodal zoom, which makes it possible to include a much more
precise level n-1 graph into a level n graph [CHA 97].

8.2.3. Subset

Figures 8.7 and 8.8 show the relationships between the cities of the Nord-Pas-de-
Calais and those on the Mediterranean coast. Four pieces of information are illustrated:
– the number of minimum paths per arc used;
– the quantity of pollutants (in kg) emitted per arc by all the vehicles using this arc;
– the unitary emission (in g/km) is the average of the pollutant emitted by a
vehicle along the route starting at the elements of a subset and bound for an element
of the second subset;
– the cumulated emission (in kg) is the sum of the emissions (in g) of CO or
NOx of a vehicle along the route starting at the elements of a subset and bound for
an element of the second subset.
 Traffic Lanes and Emissions of Pollutants 169

We realize that paths with destination or origin in the south via Paris are
identical to the paths found in the previous case; this shows that the study of subsets
fits in well with the whole of the graph.

If we compare, for example, the CO emitted over the territory with the
destination of Alès and Perpignan according to time, the cumulated emissions with
the destination of Perpignan (28 to 30 kilograms) are higher than those for Alès (26
to 28 kilograms), since the kilometric accessibility of Ales is better. However, it is
more polluting to go to Alès than to go to Perpignan. Indeed, taking into
consideration unitary emissions, a vehicle will discharge on average two grams of
monoxide per kilometer during its journey between a northern city and Alès,
whereas this emission is only 1.8 grams per kilometer for Perpignan. This
observation is valid regardless of the direction of movement, on the one hand, and
the criterion of minimum paths selection, on the other hand. Therefore we demonstrate
the smoothness of the connection between network quality and emission. The interest
here is not to realize that going to a point A is more polluting than going to a point B,
but rather to visualize using routes where the emissions are likely to be the highest.
We, thus, clearly see the importance of the network for emissions.

In the case of a temporal approach to the minimum paths, there exists a main
traffic corridor circumventing the Paris area via the A26 and the A31 and going
through the Rhone valley. According to minimum CO emission, in the north-south
direction there is a main corridor passing by Paris and following the Rhone valley
and two secondary corridors, the first skirting around Paris, as previously, and
joining the main corridor in Beaune and the other passing by Limoges. In the south-
north direction there are also one main and two secondary corridors. The latter are
very different from those found in the north-south direction. Moreover, we observe
that according to driving time, the connections originating at or bound for Toulouse
are achieved exclusively via Limoges, whereas other routes pass through the Rhone
valley. This exclusiveness is not found in routes laid according to minimum
emission. In the case of a bilateral connection, by considering emissions and the
frequency of use of the arcs, it appears that with a temporal approach the emissions
concentrate on only one main traffic axis, whereas there is a distribution of
emissions over three large circulation axes according to the minimum emission.

Finally, the calculation of minimum paths according to time or minimum

emission gives very similar emissions. On the other hand, the relationships between
the two subsets resulting from the two approaches are very different, with very
marked local consequences. We show therefore that these emissions have a strong
sensitivity, and in particular those of NOx, to the initial conditions (speed, slope,
distance), i.e. to the graph.
170 Graphs and Networks

This traditional approach of graph theory by calculating emissions according to

the slope between two nodes does not take profiles into account. The main problem
is linked to the lack of reference to the profile of the roads. However, how can we
take into account the profiles for a graph of national or international transport
network without weighing down the representation or calculations of accessibility,
for example?

In a graph each node is defined by coordinates (x, y, z) and a code. We propose

to introduce the degree of the node (the degree of a node is the number of graph
edges incidental to the node). From the profile of a road (or a digital terrain model)
it is possible to create a node of the second degree at each point of inflection of the
profile; these nodes are neither crossroads nor cities (which need to have a degree
equal to or higher than 3). The aim is to transform the arcs into 3D polylines with
ends at nodes of a degree equal to or higher than 2 and with nodes of the second
degree as intermediate or inflection points. These polylines make it possible to take
into account the profile of a road by calculating the absolute values of the slopes for
each segment of a polyline.
According to minimum driving time According to minimum emission time

Number of minimum
paths Number of minimum
150-295 paths
50-150 150-295
0-50 50-150
0-50

Ch. Decoupigny, CESA, January 2001 Ch. Decoupigny, CESA, January 2001
Emission per Cumulative emission Emission per
Unitary emission Cumulative emission Unitary emission arc in kg ***
per city in g/km * per city in g/km ** arc in kg *** per city in g/km * per city in g/km **
34.8-57.2 34.8-57.2
2.034-2.066 30.14-32.07 23.2-34.8 1.98-2.034 30.14-32.07 23.2-34.8
1.98-2.034 11.6-23.2 1.926-1.98 28.19-30.14
11.6-23.2
28.19-30.14 0-11.6
1.926-1.98 0-11.6 1.872-1.926
1.848-1.926 26.24-28.19 1.818-1.872 26.24-28.19
24.29-26.24 24.29-26.24

* it is the average of emissions in g/km emitted over a territory with an origin at every northern
city and bound for the city considered
** it is the sum of emissions in g (emission in g/km * distance to cover) emitted over a territory with
an origin at every northern city and bound for the city considered
*** it is the product for each arc of emission in g/km * the length of the arc in km *
number of minimum paths per arc

Figure 8.7. Emission of CO and minimum paths starting at Nord-Pas-de-Calais

and bound for cities in the South
 Traffic Lanes and Emissions of Pollutants 171

According to minimum driving time According to minimum emission

Number
of minimum paths

Ch. DECOUPIGNY, CESA JANUARY 2001 Ch. DECOUPIGNY, CESA JANUARY 2001

Emission Cumulative emission Unitary emission Emission Cumulative emission Unitary emission
Per arc in kg*** per city in kg** per city in kg* Per arc in kg*** per city in kg** per city in kg*
42.6 – 56.9 50.2 – 66.7
48 – 53 49.1 – 50.22 2.14 – 2.222
28.4 – 42.6 2.198 – 2.316 33.5 – 50.2
44 – 48 2.059 – 2.14
14.2 – 28.4 2.08 – 2.198 16.8 – 33.5 48.13 – 49.1 1.978 – 2.059
0 – 14.2 36 – 40 1.962 – 2.08 0.1 – 16.8 47.08 – 48.13 1.897 – 1.978
46.03 – 47.08

Figure 8.8. Emission of CO and minimum paths starting at

the southern cities and bound for Nord-Pas-de-Calais

One of the advantages of introducing the degree of nodes is to be able to simplify

the graph. During modeling it is easy to select only the nodes of a degree equal to or
higher than 3 so as not to weigh down certain calculations.

Thus, we justify the advantage of using graph theory in the study of the
emissions of the pollutants discharged from transport, in particular, from the
prospective point of view. It is all the more justified when we introduce the slope
factor into the calculation of emissions, which then prohibits the traditional
evaluation of emissions (emission in grams per kilometer multiplied by the annual
mileage of a vehicle). Indeed, the raw data on emissions between several cities, for
example, does not make it possible to localize emissions and does not provide us
with information concerning the spatial heterogenity of emissions.
172 Graphs and Networks

The definition of subsets makes it possible to determine traffic corridors and

emission corridors which are specific to the analyzed subsets. It also enables a better
understanding of the modifications of the relations resulting from the change of the
initial conditions, which are impossible to highlight by using a traditional monopolar
or multipolar analysis (interspecific relations). Moreover, it makes it possible to
shed light on the specificities of the elements belonging to the same subset (intra-
specific relation).

Therefore, the use of graph theory in the calculation of emissions has an

advantage in terms of modeling, on the one hand, (calculation of emissions for all
the arcs of the graph regardless of the initial conditions or the study area) and in
terms of representation, on the other hand since it makes it possible to spatialize
emissions.

The cold emissions were not covered in this chapter; the data on this subject is
scarce, although just as for cars there is over-emission for cold trucks. If we consider
the cycles of driving imposed on the driver, it becomes clear that after nine hours
rest there is cold over-emission. It is possible to locate the arcs where there is a stop
(for the rest hour) and to evaluate the cold over-emission resulting from it.

Nevertheless, the method used has its limitations. They relate, first of all, to the
graph and, secondly, to emissions formulae. For the graph a directed arc is defined
by the nodes origin and destination, which is a means that is associated with a
constant average speed over the arc according to a vehicle type, a distance and an
average slope defined by the altitude of the origin and destination nodes. A route
along the arcs does not correspond to the real profile (however, we saw that this
problem could be solved by integrating polylines between two nodes of a degree
which is equal to or higher than 3). There can also be variations in speed at the
access to large urban conglomerations or cities (which can be approximated using
nodal zoom). Indeed, we work with arcs by considering nodes only as origin and
destination of the journey or as intermediate cities. However, there is a great
diversity of relations between a node (a city) and the transport network, such as, for
example, the existence or absence of a ring road, which introduces large differences
at the local level. The sinuosity of roads has also not been taken into account, in
particular, for minor roads.

The second problem relates to emission formulae and the lack of data on trucks.
Most of the data is obtained for a test bed with standardized cycles and not in situ.
However, this method has the merit of offering a first approximation by yielding
minimum emissions according to the quality of the network (Figure 8.9 in the color
plates section).
 Traffic Lanes and Emissions of Pollutants 173

8.3. Bibliography

[CHA 97] CHAPELON L., Offre de transport et aménagement du territoire: évaluation

spatio-temporelle des projets de modification de l’offre par modélisation multiéchelles des
systèmes de transport, Thesis, Tours, 1997.
[HAS 97] HASSEL D., WEBER F.J., “Gradient influence on emission and consumption
behaviour of light and heavy duty vehicles”, delivery 9 of the MEET project, 3 TUV
Rheinland, Cologne, Germany, 1997.
[HIC 97] HICKMAN A.J., “Emission functions for heavy duty vehicles”, delivery 10 of the
MEET project, Project report SE/289/97, Transport Research Laboratory, Crowthorne,
United Kingdom, 1997.
[HIC 99] HICKMAN J., HASSEL D., JOUMARD R., SAMARAS Z., SORENSON S.,
“Methodology for calculating transport emissions and energy consumption”, Report
SE/491/98, Transport Research Laboratory, Crowthorne, UK, 1999.
[LHO 97] L’HOSTIS A., Image de synthèse pour l’aménagement du territoire: la déformation
de l’espace par les réseaux de transport rapide, PhD Thesis, Tours, 1997.
[SER 00] SERRHINI K., Evaluation spatiale de la covisibilité d’un aménagement, sémiologie
graphique expérimentale et modélisation quantitative, PhD Thesis, Tours, 2000.
 PART 3

Towards Multilevel Graph Theory

 Chapter 9

Graph Theory and Representation of Distances:

Chronomaps and Other Representations

9.1. Introduction

Born from the resolution of practical questions which cannot be easily solved by
using graphs (such as the problem of crossing the bridges of Königsberg (Figure
9.1), which was solved by Euler, or the problem of the four colors, which was
introduced for the first time by the cartographer Guthrie in 1852), graph theory
constitutes a mathematical framework that makes it possible to tackle problems in a
very vast field.

Figure 9.1. Bridges of Königsberg

Chapter written by Alain L’HOSTIS.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
178 Graphs and Networks

In the domains of geography, urban development and spatial planning, graph

theory is used to tackle questions arising in the field of networks (whether they are
transport and communication or even social networks). The general procedure is
based on the translation of these problems into the mathematical language of graph
theory so as to be able to treat them using, for example, the properties of
connectivity or by exploring minimum paths. For this reason, in its applications to
the social sciences, graph theory deals mainly with mathematical properties and
much less with the graphic representation of nodes and arcs1. By refocusing on the
graphical dimension, from which it was born, here we wish to develop the aspect of
graph theory that relates to representation (plotting of graphs) in its applications to
problems of spatial analysis.

Since graph theory is particularly adapted to the modeling of transport networks,

the representation of the network graph may constitute an investigative direction that
can answer the difficult question of the representation of distances.

A classical illustration of the difficult problem of the representation of distances

is provided by Müller [MUL 79, p. 216]. In Figure 9.2 we laid out three cities noted
by p, q and r. There is a highway connecting the cities q and r, and another between
q and p, whereas r and p are connected by a small road. The shortest path by
duration between the two latter cities passes by the city q and follows the highway.
The question raised by Müller is as follows: “How is it possible to determine, on a
map with a time scale, the position of points p, q and r when the shortest path from
one point to another is no longer a straight line?” [MUL 79, p. 215].

Autoroute
Highway r

Petite road
Small route
Highway
Autoroute

Figure 9.2. Representation of the shortest paths

1 When we study graph theory, one of the few properties that relates to their representation –
their plotting – is the concept of a planar graph. A graph is known as planar if it can be drawn
on a plane so that the nodes are distinct points, the edges are simple curves and no two edges
meet apart from at their ends.
 Graph Theory and Representation of Distances 179

The straight line almost never corresponds to the general form of movement, i.e.,
to the general form of geographical distance. Consequently, as Angel and Hyman
put it: “if we wish to study transport within a geometrical framework, we must
develop non-Euclidean geographical theories” [ANG 72, p. 366]. In this spirit we
will start by establishing a general form of distance on a graph on the basis of the
basic definitions of graph theory. The goal is to give a mathematical form to a
geographical measurement: the distance built in this fashion will make it possible to
find support in mathematics to better envisage the spatial dimension associated with
transport and movement and, finally, to question the cartographic representations
used in geography.

To deal with the domain of representations we will create a method for reading
distances on a map. The comparison of these measurements which are read with
those provided by the geographical data will make it possible to characterize the
various representations according to their capacity to represent geographical
distances.

9.2. A distance on the graph

Mathematical distances are measurements between objects that respect the

properties of positivity, uniqueness, separation and triangular inequality. From the
mathematical point of view as well as from the point of view of geographical
interpretation, one of the most fundamental properties of distances is that of
triangular inequality because it introduces the idea of the minimum [LHO 97, p.
102 and 113]. The distance between two objects is associated with the
measurement of the smallest gap that can be found. To illustrate this principle, on
Müller’s diagram the distance between points p and r can have two forms: if we
measure it in kilometers, the distance is given by the small road, whereas, if we
calculate it in duration of transport, the distance follows the highway route passing
through point q.

In graph theory we have a particular application which associates a value to the

shortest path: if a and b are two nodes of a graph G, the measure e (a, b) is given by
the length of the shortest path between a and b.

If we consider the case of a valuated graph2, the length of a path corresponds to

the sum of the values associated with the arcs, of which it consists. With the
measure e, we construct an application that associates a value in R to a pair of
elements of the set of nodes S. It should be noted that in this definition there is no

2 In a valuation graph a numerical value is associated with each arc.

180 Graphs and Networks

specification of the mathematical properties of the application. In particular, it is not

specified, if the measure e is metric, i.e., if it respects the properties of distances.

In the general case of a non-valuated graph Flament indicates the construction of

a metric on the basis of the measure e [FLA 68, p. 39]. The only condition for the
graph is that the value associated with a link between two nodes is identical in each
direction. The graph must therefore be symmetric. The measure between one point
and another is considered zero. Defined on a symmetric non-valuated graph, with
conventions for infinite and zero paths, the margin is metric.

On a valuated graph in order for the measure e to be a distance in the

mathematical sense, we must verify one by one the metric properties (symmetry,
positivity, separation and triangular inequality) for any pair of nodes. That leads to
two conditions, which make it possible to establish a graph distance [LHO 97, p.
137]: the graph must be symmetric and for a valuated graph the values associated
with the arcs must be strictly positive.

The distance criterion of minimum paths calculation can integrate the duration of
transport, the cost, the length of the journeys, but also the various modes of transport
(with a quality of connection that may or may not be homogenous) or even the
nature of what is transported. If the graph considered is not symmetric, we can
construct only a non-symmetric metric, a structure that belongs to the more general
field of impoverished metrics3. Although non-symmetry is the general case for
geographical distances (in time or cost) we will limit ourselves here to the analysis
of symmetric distances by considering a simplification of the collected data.

The distance defined in this fashion is purely mathematical: it is only the

connection between pairs of places with a measurement that takes the form of a
measure e. It is a mathematical object that characterizes geographical data, but a
priori does not have immediate and univocal translation in the sphere of
cartographic representations.

9.3. A distance on the map

Although a geographical map primarily provides information about the location

of places, it also informs us about the relations between these locations. In
particular, reading a roadmap makes it possible to work out routes and determine the
distances associated with them.

3 For more on these structures and their use in spatial analysis see [HUR 90, p. 225].
 Graph Theory and Representation of Distances 181

To define the function of distance that we have in a map Müller [MUL 79, p.
224] introduced the concept of “graphic distance”. This distance is defined in a
Euclidean form: we consider that the length of a path between two points on the map
is directly proportional to the length of the rectilinear segment which connects them.
However, movements carried out in the geographical space use transport
infrastructures that have very little chance to follow a straight line between two
places chosen at random. Thus, in order to measure a distance in kilometers on a
roadmap the reader must transform the graphic distance mentally to obtain a useful
measurement: the sinuosity of the road and the extent of detours along the way make
it possible to correct, by scaling up, the Euclidean length of a route. In this context
the Euclidean distance offers a first approximation of measurement. It is then
adjusted by information about the infrastructure used. In fact, the concept of distance
“as the crow flies”, on which Müller’s graphic distance is based, cannot be used to
read a cartographic representation if we wish to understand the logic of movements.

For this reason it is necessary to introduce a different concept, that of “visual

length” [LHO 97, p. 137], which is derived from Müller’s graphic distance. The
visual length of a connection between two points is equal to the length of the
corresponding path. If, for example, the path takes the form of several segments
placed end-to-end, the visual length corresponds to the sum of the lengths of these
segments. Visual length is, thus, not a Euclidean concept. It should be noted that,
according to this definition, visual length does not exist between two points unless
there is a path. Visual length makes it possible, in particular, to provide a
measurement of the routes and paths on the basis of a graph plot, i.e., on the basis of
the map of a transport network.

By abandoning graphic distance in favor of visual length we lose the reference to

the metric. The motivation of this choice becomes apparent when we refer to
Müller’s diagram (Figure 9.2). The visual length of the connection between p and q
corresponds to the distance between p and q, if it is measured in kilometers.
However, that is not true if the distance is calculated in journey duration. In this case
the visual length read on the map along the small road is not a distance since it does
not correspond to the minimum possible measurement in space-time.

By uncoupling the concepts of distance (geographical concept) and visual length

(cartographic concept) we pave the way for a better understanding of the
representations of distance.

Thus defined, the concept of visual length calls into question the usual use of
map scale. Indeed, on a traditional geographical map the scale has the function to
provide measurements of distance in kilometers as the crow flies: it is clearly a
Euclidean notion of distance. This use of scale implies the postulate of verifying the
182 Graphs and Networks

isotropy and the uniformity of the areas represented: two properties that cannot be
regarded as given when geographical spaces are dealt with.

Visual length establishes a measurement which, converted by using the scale of

representation, produces a geographical distance. It is a non-Euclidean use of the
map scale. A cartographic representation of distances is considered to be
satisfactory, if the distances read are coherent with the initial data, that is, with
geographical distances.

For Bunge there are two manners of representing distances: we may show on a
traditional map the paths that can be sinuous, or we may represent distances in a
simplified form but on a deformed map [BUN 66]. The second approach refers to
the anamorphoses introduced by Tobler [TOB 63, p. 59-78] and consists of moving
the locations of places according to distances4. This second approach is the one that
interests us because it directly involves the representation of the transport network.

9.4. Spring maps

To illustrate the impact of the modes of transport on spatial relations Tobler

considers the representation of distances in a mountainous zone [TOB 97]. The
initial data set consists of measurements of real distances in kilometers that often
follow sinuous routes. The degree of circuity of the associated graph (which is
defined as the measurement of the difference between the distances in kilometers
and the distances as the crow flies [KAN 63, p. 93-121]) is thus very high5. The
spring map (Figure 9.3) is constructed on the basis of a set of places which are
positioned at their real location [CAU 84, p. 40], i.e., their position on the plane of
the map is not modified by anamorphoses.

The principle of spring representation is to depict the link existing between two
points in space in the shape of a spring whose length is proportional to the length in
kilometers. By reference to the Euclidean distance, connections which are longer
than those “as the crow flies” are represented in the shape of a generally compressed
spring. This graphic principle makes it possible to represent paths which are longer
than the straight line without modifying the location of the ends.

To read such a map, it is necessary to get rid of our Euclidean reflexes because
the distance between places is indicated by a path which is not a straight line. This
distance can be understood only by using visual length because the graphic distance
defined by Müller results in Euclidean measurements. The map is read by evaluating

4 On this subject see [CAU 96] (www.cybergeo.presse.fr).

5 For developments and analyses of the degree of circuity see [CHA 97, p. 330].
 Graph Theory and Representation of Distances 183

the intensity of the compression of the springs: the more compressed the spring is,
the more difficult the connection (in other words it contains more turns and slopes
due to the mountainous terrain, which lowers the average speed) and the more it
moves away from the straight line. The representation is built around a direct
reference to the Euclidean distance, but a reference that is implicit. Here we find the
status of the Euclidean space as proposed by Cauvin, that is, as a reference space
[CAU 84, p. 72].

Figure 9.3. Spring graph of the road network of Western Colorado

The spring graph may be seen as the plotting of a graph. The graph is that of the
transport network considered, which is valuated by the measurements taken for the
network, and its plotting undergoes three constraints:
– the nodes of the graph are placed at the geographical position of the nodes,
which they model;
– the length of the arcs of the graph is proportional to the associated value;
– the arcs of the graph are drawn in the shape of springs which ensure the
proportionality of the lengths.

The distances in kilometers are considered here to be symmetric and they define
a graph distance. The construction principle of the spring map makes it possible to
establish a representation which is coherent with the associated mathematical
structure. Therein lies the main merit of this type of map, which is to offer a
representation of distances that remains coherent with real measurements, although
they elude the Euclidean domain.
184 Graphs and Networks

On the basis of the measurements represented on the spring map, Tobler then
proposes three-dimensional anamorphoses. The adopted principle is to slide the
location of the places on a vertical axis to stretch the distances “as the crow flies” in
three dimensions. A regression using least squares makes it possible to find a
configuration which minimizes the cumulated error for all the links (Figure 9.4).

Figure 9.4. Road distances approximated by displacing the nodes in the third dimension

Then, a transport surface can be calculated by interpolation, by building a

continuum based on the previously positioned sites (Figure 9.5). It is possible to find
the distances between places by using a measurement along a two-dimensional
plane.

Figure 9.5. Transport surface constructed by interpolation

 Graph Theory and Representation of Distances 185

Let us imagine tracing the original transport network on this surface. This
plotting of the graph would make it possible to find a coherence between the real
distances and the represented distances, similarly to what is proposed by the spring
map. If we consider that the initial data makes it possible to construct a graph
distance (where positivity and symmetry are verified), the suggested map is coherent
with this data, as well as with the mathematical structure that they imply and
therefore also with the geographical reality.

The construction principle of the interpolated transport surface comprises,

however, two important differences with respect to the spring map representation
mode. Firstly, regression and interpolation introduce an error, which, even if
minimized, makes the representation enter the domain of approximation, since the
measurements taken have to integrate an associated error. Secondly, the use of the
third dimension implies a projection system onto a two-dimensional surface of a
paper sheet that carries the image. And there too we introduce an approximation,
though of a different nature, due to the projection angles. This approximation can be
largely reduced, if we help the reader to reconstruct the third dimension mentally. In
Figure 9.7 the deformation of the grid, the hidden surfaces (in the south of the
surface) and the three-dimensional coordinates system invite the reader to enter into
the third dimension of the map and help him to reconstruct the surface topology.

Mâ con
Bourg-e n-Bre ss e
Ge neva

Roanne
Annecy
Lyon

Cha mbéry

St Etie nne
Grenoble

Vale nce

Figure 9.6. “Spring” graph of the Lyon region

186 Graphs and Networks

The second application that we propose [PLA 87, p. 12] (Figure 9.6) shows the
quality of service of the rail network in the Lyon region. The value of the arcs in the
graph indicates the quality of service, which in turn is constructed on the basis of the
quality of service of every line.

Visually, on spring maps, it is difficult to directly perceive the exact length of the
arcs. We first see the generally “compressed” aspect of the spring which provides an
indication of its length. The notion of visual length, which is given by the overall
length of the spring, does not result here from a direct visual perception. In the case
of the spring graph we are closer to the idea of a mental transformation carried out
by the map reader than of a direct and proportional visual perception. However, does
this mental transformation not proceed from the learning inherent to any non-
traditional system of representation? As Tobler proposes, these representations may
seem strange at first, but this is largely due to our more favorable attitude towards
more traditional and also more familiar maps, which means that we tend to consider
only conventional maps as realistic and correct [TOB 61, p. 164].

To answer the problem of the representation of distances, the plotting of a three-

dimensional graph offers other latitudes and allows other modes of representation,
which are to some extent connected with Tobler’s explorations. We will now present
the space-time relief maps, from the viewpoint of coherence with the initial data, but
also by way of comparison with the systems presented up until now.

9.5. Chronomaps: space-time relief maps

Space-time relief maps, or Chronomaps, aim at representing a space of transport

deformed by the coexistence of modes of transport with different performances
[LHO 96, p. 37-43]. In the domain of passenger transport speed and cost per
kilometer vary very strongly according to the transport modes, but also within a
mode depending on the quality and type of the infrastructure used. An inhabitant of
Aurillac who wishes to go by car to the prefecture of the area, Clermont-Ferrand,
starts by driving on a trunk road with an average speed of 60 km/h until he reaches a
highway entry, from where he drives at 110 km/h on average. In this example, the
highway portion of the journey is covered almost twice as quickly as the portion on
the trunk road. The possibility of swapping between the sub-networks of the
transport system are very strongly conditioned by the nature of the networks and,
therefore, the cartography of such a transport space must take into account the form
of the networks and the heterogenous nature of the system. This difficult problem,
which in anamorphoses is dealt with by dilating the space in a non-homogenous
manner according to the quality of the connections, is solved on the relief map by
employing the third dimension. To represent these distances, we do not change the
location of the places, but the way of drawing the connections.
 Graph Theory and Representation of Distances 187

Chronomaps are constructed on the basis of a graph (Figure 9.7 in the color
plates section, stage b). The graph representing the transport network must be planar
and saturated in order to enable the construction of a three-dimensional surface
(Figure 9.7, stage g). To respect the heterogenity of the network the graph adopts the
shape of a p-graph6, in which each sub-network appears as a partial graph7. Thus,
between Nantes and Angers, the highway and the road are superimposed in the p-
graph and each infrastructural network is considered as a partial graph of the p-
graph. The nodes of the graph are again assigned to the main cities of the area
considered, but also to the singular places, whose consideration is necessary for a
thorough understanding of the transport network, such as the Essarts highway
interchange located between La Roche-sur-Yon and Cholet. The choice of the
density of nodes in the space considered is one of the determining elements of the
form of the relief and, thus, of the produced image and the message that it conveys.
Low density can make it possible to show the effect of networks with a broad grid
(TGV and highway) over the entire area, whereas greater density makes it possible
to express the complexity of local distortions.

Representing distances in a homogenous network with connections that are not

far from the straight line is easy: in the plot in the plan of the highway graph (stage
c) the length of the arcs plotted is directly proportional to the time-distance and the
associated errors are small. On the other hand, in a heterogenous network, where the
differential journey speeds (or costs) are considerable, the proportionality of the
lengths of the arcs cannot be respected if they are drawn in the form of segments in
the plan. In space-time relief maps the arcs with poorer performances are traced
under the plane of the nodes in the shape of two segments, in such a way that their
length is proportional to the journey durations. The worse a connection performs, the
more an arc becomes concave and moves away from the straight line on the plane
(stages d, e and f). During the last stage the coloring of the graph surfaces in
different shades of gray makes it possible to reconstruct the transport surface and
facilitates reading the relief.

The Chronomap is constructed as the plotting of a graph representing a

multimodal transport network. The principle of plotting follows three rules that can
be stated as follows:
– the position of the nodes in the graph corresponds to their real position;
– the length of the arcs in the graph is proportional to their effective length (be it
in kilometers, hours, etc.);
– to enable the proportionality of the lengths, the arcs are drawn in the third
dimension.

6 In a p-graph there can be no more than p distinct arcs connecting the same pair of nodes.
7 A partial graph G’ of a graph G contains all of the latter’s nodes, but only a part of its arcs.
188 Graphs and Networks

From the point of view of graph plotting, the relationship of this representation
principle with that of Tobler’s spring maps is very strong. The location of the nodes
is the same and the principle of proportionality for the length of the arcs is common.
The only difference lies in the manner of tracing the arcs.

Another interesting point of comparison of the two models is their way of

representing the continuum. According to Tobler, the passage from graph plotting to
the transport surface is achieved by using a regression along a vertical axis followed
by interpolation. In the space-time relief map the passage from graph plotting is
established by tracing the facets of the graph on the basis of the arcs. Each triangular
facet (the graph must be planar and saturated) is in turn divided into four triangles
which are traced from the middle of the facets. The surface constructed in this
manner then undergoes the deformation of graph arcs. The surface is associated with
the road graph and not to the highway graph because access to normal space is only
possible by using the road. This idea of universality of the road network [DUP 94, p.
146], defined by Dupuy, is expressed by Tobler by defining the border of the
network: the border of the road network is constituted by the edge of the
carriageway, whereas the border of the highway network is constituted by the access
points [TOB 61, p. 89]. On the relief map the slope of the arcs directly translates the
associated relative speed: the more marked the slope, the lower the connection speed
is with respect to the possible maximum speed appearing in the plan of the nodes
occupied by the highway network. By geometrical construction the slope of the
facets in the graph is sharper than that of the arcs upon which they are based. This
means that the performance of the interstitial sub-network that services the area of
the facet is poorer than that of the connections bordering it. This principle is
coherent with that of network hierarchy: in transport networks the performance
drops with the size of the grid [PLA 91, p. 22]. The representation of the continuum,
which is defined as the dual of the network, does not contradict the principle of
relief representation. However, it is necessary to draw attention to the fact that the
distances shown by the relief map are, first of all, those produced by the transport
network. Here, the first function of the continuum is graphic: the shades of gray of
the facets help the reader to imagine the third dimension, just as the deformed grid
used by Tobler did.

Chronomaps do not show space in the form of plane layers, but of superimposed
complex surfaces. The image in Figure 9.8 in the color plates section shows two
sub-networks of the road transport network which are arranged in superimposed
layers in the third dimension and coming into contact only at the sites of highway
interchanges. This is a direct illustration of the tunnel effect of fast infrastructures.

Each space is registered according to its own relief: the highway is in the plane
of the cities where the straight line draws the fastest path but where space is reduced
to a set of points (network space), whereas the other networks follow a relief which
 Graph Theory and Representation of Distances 189

becomes more marked as accessibility diminishes (banal space). The space of the
road is a continuous but irregular surface, less accessible, but where the concept of
proximity still operates. The map shows a space dualized by speed.

The large highway axes which serve the territory appear clearly inscribed in the
plane of the cities. However, in the interstices of this network the space-time relief
reflects the lack of accessibility due to the inferior quality of the infrastructure. The
network space of the Riviera, from Avignon to Menton, which is well connected
thanks to an extensive highway grid, strongly contrasts with the poorly accessible
space-time of Alpes-de-Haute-Provence and Hautes-Alpes.

The Chronomap of the road and the highway renders readable the minimum
duration transport routes by showing the degradation of the relative conditions of
circulation outside of the highway type network. Thus, the Grenoble-Nice
connection is ensured via the Rhone valley along a highway route that traces a vast
loop whose visual length remains lower than that of the route closer to the straight
line through Gap and Digne which passes through an irregular space-time relief.

In formalizing the construction principle of relief maps, the first constraint

relates to the length of the arcs without taking their visual length into consideration.
In this respect we may note a difference with the construction principle of “spring
maps”. In spring maps visual length is directly proportional to the effective length of
the arcs, even though the visual length of springs is more inferred than directly read.

Relief maps are particular views of a general three-dimensional model. The

length of the arcs of the construction principle is a constraint on this general model.
The images, whether projected onto a screen or onto a sheet of paper, are two-
dimensional. The concept of visual length may be used only with respect to the
images. An image is created by the transformation of a three-dimensional structure –
the model – into a two-dimensional form – the map. Just as on the interpolated
transport surface of Tobler, the complex play of three-dimensional structure
projections onto the plane of the map causes disturbances in the lengths of the arcs.
These disturbances justify the fact that we only have an “approximate scale” of
space-time in relief maps. The visual length which is available for images is not
exactly proportional to the effective length present in the data. Mathematically, this
results in the fact that a three-dimensional surface and the final image are not
isometric: there is a distortion due to the projection on the plane. For arcs belonging
to the third dimension, the distortion between visual lengths and effective lengths is
not homogenous. It is therefore difficult to evaluate visually: it is necessary to find
the position of the segments in three dimensions. It is the role of gray shaded facets,
of viewing angle indication, to help to mentally reconstruct the relief in order to
make the distortion understandable. Let us note that the general principle of relief
maps construction places no constraints on images, meaning that the adjustment of
190 Graphs and Networks

the image generation parameters arises from the choices made by the cartographer.
An infinite number of possible images correspond to the single model, according to
the choice of the viewing angle, or also of the point of view8.
Although the approximation due to the projection of the three-dimensional
structure is common to that of the interpolated transport surfaces of Tobler, quite
unlike the latter, from the point of view of the construction principle, space-time
relief maps address the exact domain. This means that the length of arcs of the three-
dimensional model is strictly equal to that coming from initial data.

9.6. Conclusion

Distance belongs to concepts applied in disciplines that are very far from each
other. In Tobler (for whom any map results from a mathematical transformation
whose parameters it suffices to modify), in Müller [MUL 79, p. 215-227], or in
Huriot and Perreur [HUR 90, p. 225] the issue of the representation of distances, in
geography as well as in other disciplines that concern space, insistently refers to
mathematics.

Data Mathematical
structures
- graphs
- metrics
Representation

Figure 9.9. The role of mathematical structures in the representation of distances

In this spirit, the representation procedures exposed here are all based on graphs.
The very object of this contribution (the issue of the representation of distances) is a
call for a positioning in the field of metrics. Graph distance has been defined in this
manner. The representations all are constructed on the basis of a graph plotting.
Reading the map requires measuring instruments (which are the scale of
representation and the graphic length), which make it possible to reconstruct metric
space and understand its distance.

8 For relief maps of the Atlantic coast according to an unusual point of view see [MAT 96, p.
97-111].
 Graph Theory and Representation of Distances 191

Whether implicit or explicit, the reference to mathematical structures is

necessary for the procedure of representing distances (Figure 9.9). The use of
mathematical forms enables the clarification of principles and ensures the freedom
of contradiction [BUN 66, p. 2].

The principles of representation clarified here show the essential contributions of

graphs use in the cartography of distances and the understanding of spatial relations.
All the avenues are far from having been explored and the models still hold
important development prospects. Indeed, if the relief maps presented show the
connections in the form of two broken segments, nothing prevents us from
imagining other configurations using curves, for example.

The cartographies of distance can hardly be understood without instructions (a

scale of correspondence and the way to use it), which sometimes disturb the
established order of cartographic conventions.

9.7. Bibliography

[ANG 72] ANGEL S., HYMAN G.M., “Urban spatial interaction”, Environment and
planning, GB, vol. 4, p. 350-367, 1975.
[BUN 66] BUNGE W., Theoretical geography (revised 2nd ed.), Gleerup (Lund studies in
Geography), Lund, Sweden, 1966 (1962).
[CAU 84] CAUVIN C., Espaces cognitifs and transformations cartographiques, Strasbourg,
Thesis, 1984.
[CAU 96] CAUVIN C., “Au sujet des transformations cartographiques de position”, Revue
électronique Cybergéo (www.cybergeo.presse.fr), 1996.
[CHA 97] CHAPELON L., Offre de transport et aménagement du territoire: évaluation
spatio-temporelle des projets de modification de l’offre par modélisation multiéchelles des
systèmes de transport, Thesis, Tours, 1997.
[DUP 94] DUPUY G., “Réseaux”, Encyclopédie d’économie spatiale: concepts -
comportements – organisations, J.-P. Auray, Antoine Bailly, P.-H. Derycke, J.-M.
Huriotb (eds.), Economica, p. 145-151, Paris, 1994.
[FLA 68] FLAMENT C., Théorie des graphes et structures sociales, Paris: Gauthier-
Villars/Mouton (Mathématiques et sciences de l’homme), 1968.
[HUR 90] HURIOT J.M., PERREUR J., “Distances, espaces et représentations”, RERU,
France, no. 2, p. 197-237, 1990.
[KAN 63] KANSKY K., “Structure of transport networks”, Chicago: Department of
Geography, Research Paper no. 84, 1963.
[LHO 96] L’HOSTIS A., “Transports et Aménagement du territoire: cartographie par images
de synthesis d’une métrique réseau”, Mappemonde, no. 3, p. 37-43, 1996.
 Chapter 10

Evaluation of Covisibility of Planning

and Housing Projects

10.1. Introduction

The aim of this chapter is to represent the visible landscape. The study of the
visibility of a linear transport infrastructure calls upon graph theory, which is a
modeling tool of the studied terrain. The latter is considered on a local or even
microlocal scale.

To achieve our goal, a quantitative decision-making aid tool specific to the

object of managed landscape has been developed. In this context we propose to
summarize the main functionalities of the computerized model: 3D-IMA. The latter
reverts to the traditional GIS architecture and uses the principles of digital image
synthesis.

On the basis of graph theory and within the framework of landscape law of
January 8, 1993, 3D-IMA proposes two main complementary functions:
– multiresolution topography;
– localized visibility or covisibility (determination of the area visible to an
observer from a single point; see below) and generalized visibility.

The first part of this chapter is dedicated to the method of multiresolution

topography and its usefulness with respect to the surface and/or voluminal

Chapter written by Kamal SERRHINI.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
194 Graphs and Networks

representation of a given space. In the first place we will briefly present the model of
visible landscape for planning (VLP), which is the true backbone of this chapter.

In the second part, we will explain in detail the principles and characteristics of
the method of measuring the visibility surface of an object (highway, building, etc.).
Some examples used for visibility results will illustrate the subject. We will then
study the contribution of graph theory to the determination of various variants of
covisibility.

After the thorough examination of the covisibility approach the last part is
dedicated to a comparative analysis between the existing method and our method of
measuring the visibility of an object in its spatial environment. Finally, we will
conclude by describing the usefulness to space management of creating specific
functions, such as covisibility and perspectives.

10.2. The representation of space and of the network: multiresolution

topography

The 3D-IMA1 tool is specialized in dealing with the thorny issue of quantitative
evaluation of the visual impacts of large landscape management projects. Indeed, the
implementation of an extensive project, such as a highway, a new TGV line, high or
very high voltage lines, etc. should be preceded by numerous feasibility studies,
comparisons with past situations, digital and/or modeled simulations, etc. as well as
the evaluation of the visual impacts of the project on the landscape (landscape law
1993). Our work forms part of this framework.

10.2.1. The VLP system

The VLP system (see Figure 10.1) is a systemic formalization of the problems of
this research: determining the landscapes which are most frequently visible from the
construction site and vise versa, estimating management sites that are generally
visible to an observer from the landscape.

We consider this diagram as the discussion thread, which will guide us

throughout this chapter. By using colored arrows with different forms or aspects we

1 The main characteristic of this tool is the implementation of thematic and spatial analysis
functions. This model bridges the gap between digital databases (vector data and/or rasters),
on the one hand, and CAD software (AutoCAD®) for the restitution of results, on the other
hand. Indeed, our aim is not to fully recreate a SIG, but rather to develop a specific model by
taking into account previous achievements in geographical data processing.
 Evaluation of Covisibility of Planning and Housing Projects 195

will distinguish amongst four families of interrelations, which are at the origin of the
regulation mechanisms of the VLP system:

Relations 1 and 3
These two relations symbolize the transition between two stages of different but
successive levels or ranks.

Relations 2 and 4
They symbolize interrelations or, to be more exact, loops which are either inside
the same subsystem or between two different subsystems. It is a process, which in
certain or favorable situations will take place a certain number of times: it is the
case, for example, of the determination of the generalized visibility of an
installation.

Relations 5 and 6
They symbolize return actions or feedback [BER 73, p. 42]. Both relations 5 and
6 are the feedback emitted by the results subsystem (covisibility and cartography) to
regulate either the simulation-modeling subsystem, or the natural and/or anthropic
subsystems.

Feedback 5 could induce modifications, adaptations or adjustments of the

modeling-simulation subsystem. It may act in order to modify certain constraints,
among which is the observer’s height, the simulation of various heights of forests (5,
10, 15 meters), etc.

Feedback 6, on the contrary, may act directly upon the natural and anthropic
systems to integrate new geographical information (a new installation, a future
ZPPAUP or ZNIEFF, etc.). This feedback may also act by removing existing
information: comparisons between situations with or without a given infrastructure.

Relation 7
It symbolizes the contribution of decision-making aid elements in landscape
planning by using the covisibility method. The decision of whether to set up or not
to set up the future project depends on many systems or criteria of various natures:
economic, sociological, legal, political, etc. and some remain still largely in the
subjective domain (that of relations, of the game of each actor or groups of actors
linked to the project in question). This latter idea, symbolized in relation 7 by a
green arrow with a multipolar base, reverts to the multicriteria methods of assistance
to decision-making.
196 Graphs and Networks

Action:
Actionplanning and housing
: aménagement

(7)

Covisibilité
Covisibility

Viability,
Specific cartography

:
Cartographie spécifique

dynamiques
Généralité
applicability
Simulations:
(3) (4)

Simulations
Visibilité
General
Dynamic
Simulations (3D-IMA) (5)
3D-IMA Simulations
Multi-resolution,
Multirésolution,
variantes,etc.
etc.

Modélisations : Théorie
(2) Models: Graph Theory

Cartographie de la covisibilité
ou feed-back)
variants,

des Graphes

Cartography of covisibility
(1)

Anthropic: NZEFFI, (feedback)

(Rétroactions
Anthropiques : ZNIEFF,ZIPPAUP
ZPPAUP,
populations,etc.
populations, etc.
naturel
subsystem

Biotic: forests,
Biotiques woods,
: forêts, etc.
bois, etc.
Ecosystem
Ecosystème
Sous-système

Abiotic: :topography,
Abiotiques topographiedistance,
, distance ,
Natural

hauteurs,
heights, etc. Géosystème
etc.:: Geosystem (6)

Figure 10.1. VLP system [SER 00b]

Figure 10.1 summarizes all of the subsystems (abiotic, topography, water, biotic,
forest, anthropic, ZNIEFF2, ZPPAUP, populations, flow of travelers, networks, etc.)
and their interactions, thus making it possible to obtain a precise answer to the
question of the estimate of the visual impact of an installation.

2 ZNIEFF and ZPPAUP are biotic spaces (forest) or abiotic (monument) which, however,
benefit from the statutory protection of human design, thus explaining their belonging to an
anthropic subsystem.
 Evaluation of Covisibility of Planning and Housing Projects 197

10.2.2. Acquiring geographical data: DMG and DMS

In order to measure certain environmental impacts of landscape planning we

need to collect data with respect to, on the one hand, the work itself (possible
locations, longitudinal profile, dimensions, etc.) and, in addition, to the space that
receives this project (DMG, existing transport system, forests, protected areas etc.)

10.2.3. The Conceptual Data Model (CDM) starting point of a graph

The CDM is one of the bases of the MERISE method [MAT 94]; it is used to
formalize information systems in general and the data stored in the GIS in particular.

Figure 10.2. CDM3 of the three main types of data4

This CDM was implemented in two steps, as follows:

– prior to this research there existed already5 various structured information lists.
They correspond to files: triangular nodes, arcs and facets. These pre-existing files
are dealt with as they are with possibilities for improvement (deductive step);

3 The CDM rests primarily on the three concepts of entity, relation and property.
4 This CDM was carried out by using Access for the graphic functionalities. The geographical
data itself is in text files (ASCII) according to an appropriate format.
5 Laboratoire du centre d’études supérieures d’aménagement EA1373, University of Tours.
198 Graphs and Networks

– then, for the purposes of this work, a new type of surface data was developed:
the square mesh (in vector and not in raster mode). The latter is necessary to restore,
among others, the topography of the entrance area of an installation (inductive step);
– in addition, while the typology of data (various files) is based on the nature of
the object considered (punctual, linear and zonal), each data type, on the other hand,
simultaneously groups semantic and geometrical information specific to this type.

Figure 10.2 summarizes only the three principal types of the processed
geographical data, their attributes and relations. In this context some clarifications
are essential:
– the various tools of the CESA laboratory (D.LOCAT.T, NOD, MAP, RES,
FRED, 3D-IMA) generate multiple processing files and intermediate or final results
that are not represented in order to avoid overloading Figure 10.2;
– through the codes of the various nodes of the linear or surface elements, the
two types of arcs and meshes maintain the relations with the type of nodes. For the
cardinalities of these relations we have two eventualities:
- either a cardinality (1, n): meaning that a given node (1) belonging to the
nodes type can contribute to one or more arcs or meshes (n). In other words, the
nodes of the meshes or the arcs necessarily appear in the nodes type. In other words,
it means that in the nodes type there are no isolated nodes,
- or a cardinality (1, 1): where only one specimen of a node of an arc or a mesh
exists in the nodes type.

Thus, the relational nature of a database management system stems from the
preceding relations provided with their cardinalities.

10.2.4. Principle of multiresolution topography (relations 1 and 2 of the VLP)

The nature, the quantity and the quality of data necessary depend mainly on the
project under study. Since the present work primarily addresses to big planning and
housing projects, which need many data, one of the functions of 3D-IMA is, thus,
the creation of specific DMG. Although it clearly is a function of geographical data
acquisition, we will see that it is also a function of analysis.

Figure 10.3 in the color plates section summarizes for us the principal stages of
creation of a DMG referred to as multiresolution (often referred to as the Regular
hierarchical terrain model, [FLO 99, p. 8]). According to this figure it is therefore
possible:
– to obtain a surface representation (2.5D modeling: a single z for every pair of
(x, y)) of the original terrain on the basis of a simple topographic map with a scale of
 Evaluation of Covisibility of Planning and Housing Projects 199

1/25,000 with the possibility of overlapping the various resolutions only locally,
from where the term of multiresolution originates;
– to create geographical data in various geometrical forms which are specific to
various uses: punctual data (a scatter graph or DMG), linear (in the form of
segments of straights or arcs) and zonal (data in the form of facets or meshes). All of
this information has three coordinates (x, y and z) leading to a better level of
abstraction than that of models and 2D GIS.

If the method of multiresolution topography forms part of data acquisition

functions, it also implements analysis and calculation functions, and, mainly,
interpolation methods [ARN 00]. Interpolation to the nearest neighbor [SER 00, p.
327] enables us to recreate heights on the basis of altitude curves in order to go from
a 2D scatter of points (with only two coordinates: x, y) to a 2.5D scatter (a z for per
(x, y) pair): DMG. Obtaining the 2.5D grid is the result of surface modeling
(DMS). Lastly, a true 3D representation implies implementing voluminal modeling
(several z for certain (x, y) pairs). A 3D vector modeling is used to restore the
voluminal aspect of particular geographical information, such as a building, a forest,
a listed monument, a nuclear power station, a dam, a viaduct, etc.

In this chapter we will eventually call upon voluminal 3D modeling to represent

noticeable landscape elements after successively implementing a 2D modeling (2D
scatter of points) and then a 2.5D modeling (DMG and surfaces) with or without
multiresolution.

10.2.5. Need for overlapping of several spatial resolutions (relation 2 of the VLP)

This function, which may appear traditional at first, is actually fundamental.

Indeed, instead of having a very high definition topography in a generalized manner
(a DMG with a step of 10 or 5 meters, for example), we can imbricate various
resolutions around or along the processed installation. Our objective is not to
introduce a hierarchy in the landscape according to its distance from the observer in
the form of plans whose scale is increasingly small as we move away from the latter,
as is the case in a recent article by Brossard and Joly [BRO 99]. A ZNIEFF6 may be
located at the edge of the studied terrain, whereas the installation itself is rather
more central in this space. In this case a multiresolution topography is implemented
simultaneously for the ZNIEFF and for the installation. The goal is to obtain a fine
spatial definition only where it is useful, without worrying about the position of the
observer or that of the project: “since not all tasks require the same level of detail,
the use of high resolution models may affect applications for which many of the

6 Natural zone of ecological, floral and faunal interest.

200 Graphs and Networks

details are not relevant” [FLO 99, p. 8]. This leads to an unquestionable gain in
terms of quantity of geographical data to process.

Step

Pas /22
Step
Vue de face
Frontal view

Vue de profil
Sideways view

Excédent
Excess Déblai
Minimum
excavation
de déblais excavation
minimal
Route
Road

Figure 10.4. Mounds, excavations and multiresolution

Which spatial resolutions should be chosen? (50, 25 m, etc.)? And where?

Finding exact answers to these questions remains very difficult since each
installation and landscape constitute particular cases. This technique is also essential
for embankment building and excavation needed to maintain the profile of the works
(see Figure 10.4).

10.2.6. Why a square grid?

If, potentially, there are various geometrical forms [CHA 91, p. 27] (triangle7,
square, hexagon) to restore any surface (including that of the original terrain), then
why did we prefer using the square mesh?

Various reasons may explain this choice.

7 We also have the possibility of plotting a graph containing triangles. This is the case for the
planar saturated graph of the center of Tours, where the nodes represent crossroads in the city
and the facets are the result of a triangulation using Delaunay’s method [SER 97].
 Evaluation of Covisibility of Planning and Housing Projects 201

On the basis of a regular DMG it is possible to model a surface containing square

meshes or triangular facets. In fact, each mesh may easily be transformed into two
isosceles triangles with two nodes and the arc connecting them in common.
Ultimately, in terms of representation in wire, the same “surface” can be described
either through four points connected to each other by four “arcs” (a mesh) or by four
points and five “arcs” (two triangles). Several consequences arise from such a
transformation:
– a surface composed of triangles (regular geometry) contains twice the number
of triangular elements than if it were composed of square meshes. This results in
weighing down any possible processing. Moreover, the approach of this research
setting often requires processing large quantities of geographical data and the use of
a geometry based on triangles does nothing but double the occurrences of possible
Triangle objects with respect to the Mesh objects used (see CDM below). For
example, performing a simple classification (bubble sort) would have to deal with
twice as many triangles as meshes, or even the restitution (the automation of the
drawing by AutoLisp) of a topography using AutoCAD® would take practically
double the time with triangles than it would with meshes;
– in the case of an DMS formed by triangles the centroid (the dual point) of each
triangle has only three close neighbors (the nodes of the considered triangle),
whereas linear interpolation usually requires the four closest neighbors, which a
mesh structure enables us to easily obtain.

A mesh covers a larger surface than that of the two triangles in this mesh. The
definition of topography by facets is unquestionably finer than that offered by
meshes. However, this definition is not only more difficult to manage (storage,
processing, display), but, moreover, it does not appear to be always useful. To obtain
a better spatial resolution, the topography method makes it possible to only overlap
various space resolutions where it is useful: an operational multiresolution.

Square mesh topography seems more interesting to us mentally than its

equivalent in triangles, since it is closer to the mental representation than we make
of the surrounding space (the pieces of land are more often rectangular or even
square than triangular and the same applies to houses, buildings and roads). If this
observation proves to be founded, then a meshed structure should have a better
visual impact on the reader of the cartographic document. Moreover, such
topography is graphically lighter (at least one arc per mesh) and, consequently, it
would be less complex, more esthetical and thus, most probably, attractive.
202 Graphs and Networks

10.2.7. Regular and irregular hierarchical tessellation: fractalization

In the hierarchical tessellation used for covisibility, when that is necessary to

define more precisely the space for which we wish to determine the properties of
visibility, a certain number of invariants remain: the form and the limits of the
geometric forms (triangle, square, etc.), nodes and arcs. It is, moreover, a whole
multiple in terms of surface of the new cube or, conversely, a whole number of new
cubes of the same form which occupy the surface of the cube of hierarchical level
immediately above. All the elements of auto-similarity and of internal homothety,
which is a fundamental property of fractals, are present. The phenomenon can be
generalized to statistical homotheties [THI 87].

10.3. Evaluation of the visual impact of an installation: covisibility

10.3.1. Definitions, properties, vocabulary and some results

This function, i.e. the evaluation of the visual impact of an installation,

corresponds to the determination of the zones of space which are visible from the
installation (punctual or linear) and, conversely, the zones of the project which are
visible from the surrounding space.

This function is implemented, in the present state of development of this work,

with the following elements:
– input data:
- topography of the studied terrain (DMG),
- geographical position and height of the present obstacles (forests and others),
- geographical position and height of the object (profile of installation:
embankment, trench, etc.) integrated into its environment;
– parameters of the model:
- geographic position(s) and height of the observer,
- maximum visibility distance,
- population per considered zone (mesh, etc.) of the space,
- Euclidean distance separating the observer (local resident) from the project,
- the direction of movement of the driver (case of a linear transport infrastructure),
- speed of travel of the driver.
 Evaluation of Covisibility of Planning and Housing Projects 203

Punctual housing
Target = Space project (like a Linear works
building)

Observer = punctual installation

Localized visibility = covisibility
or a single point in space…

Table 10.1. Localized visibility

We distinguish two types of visibility.

A localized visibility (Table 10.1). When we determine the visibility area on the
basis of a single point of the object (case of a punctual installation: building [SER
99]), “two points V and W on a topographic surface are said to be mutually visible if
and only if interior of the straight-line segment joining them lie strictly above the
terrain (…). Given a viewpoint V on a terrain, the viewshed of V is the set of points
of the surface which are visible from V” [FLO 99]. In this case the answer is of the
“all or nothing” type, i.e. visible or invisible (see Figure 10.5).

Case 1 of Figure 10.5 in the color plates section shows the red zone which is
visible to an observer from the green mesh of the works: localized visibility.
Moreover, this green mesh of the section of the A28 is visible from all the red
meshes in the area. The localized visibility is symmetric (the symmetry is absolute),
hence we have the term covisibility. Therefore this term supports two pieces of
information at the same time: on the one hand, we have localized visibility and, on
the other hand, it is symmetric. This fundamental property stems from the fact that
the localized visibility is of the centroid-centroid type (the observer and the target
are each reduced to a single point or centroid with three coordinates x, y and z)
[SER 00] (see Table 10.3): “any visibility function can be represented by a visibility
graph with arcs that link the nodes corresponding to intervisible entities. The
visibility graph for point-point visibility is straightforward because any given point
is either visible or invisible from any other point” [NAG 94, p. 764].

The term covisibility is new compared to the terms of visibility, intervisibility or

such expressions as “field of view”, “landscape opening” and “subject to viewing”.
Indeed, the existing terms and expressions specify neither the nature of the visibility
(localized or generalized) nor the property of the symmetry.

Cases 2 and 3 of Figure 10.5 in the color plates section correspond to taking or
not taking into account, in the calculation of visibility, a future viaduct which
crosses the Choisille valley and which the A28 could use. This possibility (bridge) is
presented only as illustration. Indeed, even though the unevenness of the valley is
204 Graphs and Networks

approximately 50 meters, the viaduct solution is not possible due to its cost and the
fact that it is already planned that the A28 crosses the valley atop an embankment.
Each of the five flagstones of the bridge is a simple three-dimensional parallelepiped
containing six facets and two more for the safety fences, so that it really is a three-
dimensional modeling. Thus, while the majority of the illustrations presented show
only the surface, the visibility model perfectly integrates voluminal information,
such as buildings or bridges.

In the two latter cases the implementation of voluminal modeling (3D) makes it
possible to take into account the visibility between the piles and under the arch of a
bridge or a viaduct: possibility of making the visibility radius “pass” under the
bridge and between its piles in the algorithm.

A generalized visibility. Here, we perform the calculation of the visibility area on

the basis of many points of the installation in order to simulate, for example, the
movement of a driver (case of a linear installation: highway) or of the space. The
result obtained is presented in the form of a color scale corresponding to classes of
visibility percentages which go from the most visible zones to the least visible or
even hidden zones (Figure 10.6 in the color plates section).

Linear planning
Target = Space
(like a road)

Observer = space or linear installation Generalized visibility

Table 10.2. Generalized visibility

Figure 10.6 in the color plates section shows six zones of visibility degrees along
the works. In order to immediately differentiate the zones that are most frequently
visible (generalized visibility) from the section of the A28, the number of classes
was limited to six: “in selective perception it is advisable not to exceed six to seven
shades – black and white included” [BER 67].

For a linear installation visibility is not symmetric. The use of the term
covisibility is no longer justified and the statistical results obtained require prudence
in their interpretation. For example, for the visibility class [0; 10%] we will say that
approximately 9% of the section of the installation is visible from certain centroids
(or certain “meshes”) of the zones of space belonging to this visibility class.
Conversely, we will say that certain centroids (or some “meshes”) of the zones of
this same class are visible from approximately 9% of the section of the installation.
 Evaluation of Covisibility of Planning and Housing Projects 205

Determining the identity (localization) of the centroids of the installation (i.e.

approximately 9% of the “meshes” of the installation) which are visible from one of
the centroids of the zone of space belonging to the visibility class [0; 10%] requires
memorizing all the intermediate results (who do I see? and who sees me? all in all,
who sees whom?) of the calculation of generalized visibility.

In conclusion, since for a moving driver the term covisibility is no longer

adequate, we will then speak of generalized visibility.

Mesh-mesh Mesh-Zone Zone-Mesh Zone-Zone

Pairs = (Centroid- (Centroid- (Centroids- (Centroids-
Centroid) Centroids) Centroid) Centroids)

Unsymmetric
Symmetry of Absolute Absolute Absolute
(generalized
visibility (covisibility) (covisibility) (covisibility)
visibility)

Table 10.3. Symmetry of visibility

10.3.2. Operating principles of the covisibility algorithm (relations 3 and 4 of the VLP)

Having carried out the various stages necessary to create the set of geographical
data, the visibility calculation algorithm can be brought into play. The operation of
this computer program may be described in three stages.

10.3.2.1. Loading of geographical data into memory (1st stage)

This first stage corresponds quite simply to the 3D-IMA model reading and
storing in the microcomputer memory of the geographical data files produced
previously. There are two main matrices:
– the data matrix concerning the topography of space, including all the
information acquired, such as natural obstacles (forests and others). It is “the
equivalent” of data that can be found and collected by digitalization and/or vectoring
on a topographic map (raster type document) but in digital form: ASCII8 files of
digital data. This data format lends itself well to the restitution of the result in vector
form;
– the matrix relating to the installation and its characteristics. This second matrix
is at the origin of the opposition between covisibility and generalized visibility.
Indeed, the geographical description of each installation consists of at least one
object of the zonal type (mesh, facet, etc.). Thus, thanks to the number of objects

8 American Standard Codes for Information Interchange.

206 Graphs and Networks

comprised in the works, the algorithm automatically determines if there is localized

visibility (a single zonal entity to describe the object: localized installation) or, on
the contrary, generalized visibility (generally linear works comprising many zonal
entities).

10.3.2.2. Preparation of the visibility calculation (2nd stage)

This last stage preceding the visibility calculation makes it possible to collect
interactively certain additional information which are essential for the operation of
the algorithm:
– height of the observer: the computer model requires the user to input the height
(size) of the observer considered as being located at the installation or in the area.
The position of the observer corresponds to the centroid of each facet of the
installation or of the space;
– movement speed (strictly positive and less than or equal to 130 km/h), if the
observer is a driver;
– maximum visibility distance: if close-up the human eye has a resolution power
of about 0.2 mm in diameter, in remote vision this power very quickly loses
significance and the distinction of remote objects becomes fuzzy. To integrate this
physiological constraint into the algorithm we chose the following solution.

The modeling of an installation is systematically centered on a square

topography with sides of approximately 5 to 10 km. In other words, the linear
installation will not be modeled as one entity, but section by section (10 km long at
the maximum). This gives to the covisibility algorithm its physiological sense with
respect to the maximum visibility distance. Indeed, it is estimated that beyond
approximately 2.5 km on both sides of the works, the eye distinguishes only the
general forms (a forest in the form of a potatoid, a building in the form of a cube,
etc.) and very few or no details. Moreover, this solution also makes it possible to
implicitly minimize the effect of the curvature of the Earth.

Thus, in the case of a linear installation which extends over hundreds of

kilometers it is useless to calculate the visibility for the totality of the works (it is as
if we asked a driver located on the A10 highway somewhere around the Parisian
region whether we could see Bordeaux located 600 km away).

10.3.2.3. Calculation of visibility itself (3rd stage)

At this development stage the model contains all the necessary information and
can start evaluating the visibility of the works. Calculating the visibility itself may in
turn be subdivided into three stages.
 Evaluation of Covisibility of Planning and Housing Projects 207

1st sub-stage: systematic calculation of topographic cross-sections

Knowing the location of the observer (centroid of the red mesh in Figure 10.7)
and the target point, for which we are trying to determine whether it is visible or
hidden (centroid of the green mesh of the same figure), the algorithm calculates at
first the equation of the straight line (line of sight) connecting the observer and the
target. Then, it determines all the segments of straight lines (arcs) that cut across this
line of sight; this is a limited topographic cut (Figure 10.7 in the color plates
section).

Figure 10.7 shows four examples of topographic cross-sections from a vertical

perspective (X, Y plane). The 3 topographic cross-sections located on the left
constitute three particular cases:
– a horizontal cross-section parallel to the x-axis;
– a vertical cross-section parallel to the y-axis;
– and a final diagonal cross-section (angle of 45° to the horizontal);
– the 4th and last cross-section (the one on the right, Figure 10.7) is a random cut.

To determine the areas of space that are most frequently visible to a driver from
the installation (Figure 10.7), the topographic cross-sections are calculated from all
the meshes of the works to all the meshes of space: we thus have a systematic
sweeping of all the potentially visible zones from all of the elements of the structure
considered.

Conversely, to determine the sections of the installation which are most

frequently visible to an outside observer, the topographic cross-sections are
calculated from all the meshes of space to all the meshes of the works by using
systematic sweeping. For both events we say that the aiming is multidirectional.

The total number of topographic cross-sections depends simultaneously on the

number of entities of the installation and the number of entities of space. Thus,
punctual works generate a number of topographic cross-sections which is identical
to the number of entities describing space. In the opposite case, the determined
number of topographic cross-sections is the product of the number of objects, of
which the section of the works is comprised, by the number of objects that form the
space of this installation.

2nd sub-stage: visible or hidden?

Once all the arcs cutting through the line of sight have been calculated, the
algorithm analyzes the current topographic cross-section. Each arc of the
topographic cross-section has a (2D) point in common with the line of sight: it is the
208 Graphs and Networks

intersection point. The latter can have two different altitudes according to whether it
is on the line of sight (in 3D) or on the arc of intersection (also in 3D).
Consequently, what remains to be done is to calculate (by simple interpolation
following the proportionality rule along a line – Thales’ theorem) these two possible
altitude values of the intersection point and to compare them. If the altitude of the
intersection point on the arc of the topographic cross-section is higher than or equal
to the attitude, which it has on the line of sight, then the algorithm deduces from this
information that the target is hidden for our observer. In order for the model to
consider the target as visible, it is necessary that the altitude of all the intersection
points on the arcs of the topographic cross-section is lower than that of these same
points on the line of sight.

In all cases the final objective is to obtain a visibility index per spatial entity.
Various types of visibility indices are possible.

Statistical visibility index VIs. This corresponds to the ratio between the number
of times that the spatial (or the works’) mesh appears visible from the installation’s
(or spatial) meshes and the total number of meshes of the installation (or space).
This index can be without a unit or expressed as a percentage;

V [10.1]
VIs =
N

where:
– V: is, for example, the number of times that the mesh of the space appeared
visible from the meshes of the installation.
– N: is the total number of meshes (facets) which constitute the works (N = 1:
punctual work; N > 1: linear work) or the space.

Consequently, with covisibility, a space mesh which has a VI = 1 (100%), is

covisible (absolute symmetry) from the centroid of the installation mesh. With
generalized visibility a mesh of the space which has a VI = 0.5 (50%), is covisible
(absolute symmetry) from the centroids of a set of meshes corresponding to half of
the section of the installation.

0 ≤ VI ≤ 1 [10.2]

The visibility index can be balanced by the Euclidean distance (d) separating the
observer (local resident) from the installation which we will call VId (see below).
 Evaluation of Covisibility of Planning and Housing Projects 209

The visibility index can be also expressed according to the crossing time of the
transport infrastructure (see section 10.3.2.5).

Lastly, the visibility index can also be expressed according to the population P,
which we will call VIp (see below).

According to the number of installation meshes we can summarize the various

scenarios of the visibility index values in Table 10.4.

Number of installation meshes: N N=1 N>1

Visibility index: VI 0 or 1 0 <= VI <= 1

Visibility type Covisibility Generalized visibility

Color scale
Cartography restitution 2 colors
(classes of generalized visibilities)

Table 10.4. Characterization of the visibility index

3rd sub-stage: balancing with the population

If we consider that the visibility of an installation from a sparsely populated or
unpopulated zone (a corn field, a forest, etc.) does not have the same repercussions
on planning as that from a populated zone (a borough, a tourist site, etc.), then it
seems interesting to estimate visibility according to the population variable.

This balancing is conditioned by the availability of the data regarding the

population on a very fine spatial scale (mesh with sides of 100, 50 or 10 meters).
Without this data the algorithm omits this last stage. In the opposite case, the
visibility index is the sum of the populations of all the meshes from which each
element of the installation can be seen.

n
VIp = ∑P
i =1 i
[10.3]

where Pi is the population of the ith space mesh that sees the target (installation).

We will distinguish two sub-types of visibility indices according to the

population (Figure 10.8):
210 Graphs and Networks

– a visibility index calculated only with respect to the real population of the
studied terrain that sees the installation; this is the effective population. This
effective population is the product of the number of houses per mesh (topo map with
a scale of 1/25,000) by the average number of people per household;
– a visibility index calculated simultaneously with respect to the effective and
potential population. The potential population relates to space meshes which do not
contain houses but from which the works can be seen. In this case the population of
these meshes is equal to 1.

In three scenarios of Figure 10.8 in the color plates section (height of the
observer: 1.75 m) we voluntarily eliminated the meshes from the topography for
better legibility.

The choice of a visibility index type depends on the objectives set, the spatial
analysis performed (study of landscape or human impact, planning integration in
landscape, etc.) and the availability of the data.

10.3.2.4. Generalized visibility balanced by the distance to the installation

Although it is definitely interesting to take into account the relative proximity of
the residents in the vicinity of the installation during the determination of the
scenarios of road planning, it seems possible to us, however, to go even further by
considering not only visibility and the potential and/or present population of the site,
but also the Euclidean distance.

The combination of these three constraints (visibility, population and distance

from the installation) makes it possible to take into account, for example, all the
people who are near, or even very near, to the works but who would not necessarily
be inconvenienced by them due to the quite simple fact of the presence of an
intermediate visual screen (landscape, woods, etc.) Thus, considering the distance to
the works in order to estimate the number of people who are potentially affected by
the arrival of a highway infrastructure seems necessary but insufficient.

Covisibility, generalized visibility and anthropic generalized visibility can be

balanced by the relative distance of the residents to the infrastructure. The closer the
observer is to the installation, the more the balancing of visibility by distance
justified. Similarly, the larger the modeled space (more than 10 kilometers long), the
more this balancing is essential, especially for taking into account the case(s) of
town(s) (concentration of houses, of activities, etc.) that are not in the immediate
vicinity of the installation (Figure 10.9 in the color plates section).

Figure 10.9 is an example where generalized visibility of the section of the A28
was balanced by the distance of each space mesh to each installation mesh.
 Evaluation of Covisibility of Planning and Housing Projects 211

The index of generalized visibility which was balanced by the distance

implemented to obtain this map has the following form:

n ⎛ vj ⎞ [10.4]
VI dij = ∑ ⎜⎜
j =1 d ij
∗k⎟
⎟
⎝ ⎠

where:
– vj is the result of visibility (according to the traditional topographic constraints)
from the ith space mesh to the jth target mesh (the works);
– dij is the Euclidean distance between the observer (space) and the target (the
works);
– k is a visibility adjustment coefficient. It corresponds to the minimal theoretical
distance which separates the target (installation) from the observer and for which
visibility remains clear.

The result of Figure 10.9 may be regarded as an index of the clarity of the
generalized visibility of the A28: the more we move away from the installation, the
more details are lost from the visibility. Figure 10.9 shows three zones along the
A28 (in the form of aureoles) of various colors. The more we move away from the
works, the less bright the color becomes: red and magenta are in direct contact with
the meshes of the A28, then there is the green, cyan and on the outside we have blue.
The latter corresponds to the most remote space meshes, from which parts of the
works can be seen. We note the concentric aspect of the space zones of various
colors around sections of the A28.

10.3.2.5. Dynamic generalized visibility

In addition to the set of constraints considered above (height of the observer,
populations, proximity of the residents in the vicinity of the works, etc.), it is also
possible to take into account in the 3D-IMA model the time and direction of
movement of a driver on a highway section: space-time directed visibility.

An example simultaneously illustrating directed static generalized visibility (see

case 1, Figure 10.10) and directed dynamic generalized visibility (see case 2, Figure
10.10) is presented in the color plates section.

From an algorithmic point of view, this stage (directed visibility) takes place
very early in the covisibility algorithm. Indeed, it is useless to determine the
visibility of a target mesh from the site of the observer while knowing that this mesh
targets is not in the observer’s field of vision (180° for the directed visibility).
212 Graphs and Networks

The interest of the directed visibility is certain. An observer located on the last
floor of the Montparnasse tower does not see the same landscape from its various
sides (decision-making aid with respect to the choices of orientation of future large
scale development projects). A Parisian elected official would like to know only the
landscapes most frequently seen by a driver on the A10 leaving Tours. Conversely,
an elected official or a decision maker from Tours would try to get an idea only of
the spaces which are generally seen by a tourist on the same A10 highway in the
opposite direction. The objective, obvious in terms of management, for a space
manager (urban developer, planner or elected official) corresponds to the will to
improve the image of a region by using an attractive esthetic framework (entrances
to cities or areas with improved visual comfort), thus making it more competitive.

With regard to dynamic visibility, knowing the movement speed of a driver (100
km/h for the example of Figure 10.10) and when the target mesh (pertaining to the
landscape) is considered visible by the algorithm (3D-IMA), the latter assigns to it
the crossing time of the observer’s mesh (located on highway A28). This time is
determined by considering a certain speed and network distance of each entity of the
installation.

Thus, case 2 of Figure 10.10 represents, according to a color scale, the maximum
time (in seconds) of visibility of each space mesh. For example, a magenta space
mesh is potentially visible between 53 and 75 seconds along the section and by
considering the southwest-northeast direction of movement.

10.3.3. Why a covisibility algorithm of the centroid-centroid type?

Having described the operation of the covisibility algorithm, we now will specify
the reasons for choosing an algorithm of the centroid-centroid type.

Let us recall that in the covisibility algorithm of the centroid-centroid type, a

target mesh is considered visible from the observer’s mesh, if and only if its centroid
is visible from the centroid of the observer’s mesh. Thus, we perform a double
generalization of the results of visibility obtained at the level of the two centroids to
two corresponding meshes (of space and installation): the calculations performed
are of the centroid(s)-centroid(s) type and the restored results are of the mesh(es)-
mesh(es) type.

With regard to graph theory the algorithm of the centroid-centroid type employs
the dual graph on the basis of the initial primal graph (DMS).
 Evaluation of Covisibility of Planning and Housing Projects 213

The generalization of the visibility result obtained at the level of the centroid
tested to the entire mesh, to which it belongs, may appear hazardous to some. What
are the elements leading us to such a choice?

Continuing on the basis of the point-point type algorithm we performed the

calculation of visibility. We did not consider the tested centroid mesh, but its various
nodes instead: 1, 2, 3 or 4 nodes of each target mesh. In this case the point-point
type visibility algorithm uses the primal graph (DMS).

Figure 10.11 in the color plates section exhibits five types of covisibilities:
– Case 1: it determines all of the space meshes visible (exactly 330 visible
meshes out of 1,024: 20%) from the green mesh of the section of the A28. Here, a
space mesh is considered visible, if at least one of its four nodes is visible from the
centroid of the green mesh of the installation;
– Case 2: it presents the space meshes, at least two nodes out of four of which
are visible (275 visible meshes: 17%) from the green mesh of the A28;
– Case 3: it shows us only 216 (13%) space meshes, where three nodes out of
four are visible from the green mesh of the A28;
– Case 4: is shows only 165 (10%) space meshes where four nodes are visible
from the green mesh of the future installation simultaneously;
– Case 5: it proposes a calculation of the visible field by not considering one,
two, three or four nodes of each space mesh, but its centroid. The height of the
centroid is calculated using the method of linear interpolation: centroid-centroid
localized visibility. Thus, on the whole we obtain 219 (13%) space meshes whose
centroid is visible from the green mesh of the A28.

The higher the number of nodes involved in determining the covisibility of each
space mesh (passing from case 1 to case 4 of Figure 10.11), the more the visible
zone is reduced: a very fine estimate of the landscape visible to the observer.

Thus, in comparison with the first four cases, the last covisibility situation (case
5 of Figure 10.11) seems to us to be a better compromise and a good indicator of the
extent of the field of vision.

For specific development ends and in order to better approach the calculation of
covisibility, instead of leaning towards the development of particular covisibility
calculation algorithms (see below), we chose a solution associating multiresolution
topography to a covisibility algorithm of the centroid-centroid type with a result
restored in 3D in the shape of a wire mesh (vector data) and with colors, whereas the
majority of publications on visibility use raster data.
214 Graphs and Networks

10.3.4. Comparisons between the method of covisibility and recent publications

After having studied in detail the entire method of covisibility, it becomes easier
for us to compare and discuss the specificities of this approach with respect to the
results of recent and pioneering research.

This comparison encounters two main difficulties:

– a very poor bibliography concerning algorithms of visibility surface calculation
of an observer, despite the multiplication of GIS that propose spatial analysis
functions, including visibility. “In most documentation of information systems (GIS)
it is very rare to find details of the algorithms used in the software” [FIS 93, p. 331].
This first handicap must be a consequence of the GIS market laws which oblige their
designers to respond to competition with secrecy surrounding the main analysis
functions which they propose. Consequently, the comparison is limited to research
publications other than commercial GIS. Moreover, the comparison of visibility
functions of certain GIS in the market, mainly through the results obtained, has
already been carried out relatively recently by Fisher [FIS 93]. The latter, in order to
undertake his comparative study of visibility, had the possibility of experimenting
with seven GIS (Idrisi, OSU MAP for PC, PC MAP, Map II, GRASS, Arc/Info and
EPPL7), which is not the aim of this chapter;
– besides the GIS there exist, most probably, models of visibility determination
which are specific to the requirements and objectives of organizations, such as the
military, EDF (EVELYNE software), etc. Unfortunately, we have no hard evidence
of this research. The secrecy would be even stronger around military research than
around civil research (EDF).

In these conditions the comparison is limited to the recent civil research on

visibility. The latter, which also does not clarify all the stages and specificities of the
created or exploited visibility algorithm, obliges us to base this comparison on
relatively accessible elements.
 Evaluation of Covisibility of Planning and Housing Projects 215

Algorithmic characteristics

Visibility
Data Constraints Sweeping Display
type

Rasters
[RAN 99],
[DU 93] Especially
and [BRO Localized views in the
Recent Mainly Partial aiming
84], rarely (rarely (2D) plane;
publications topological (rarely complete)
vectors generalized) possibility of
[FLO 99] overlay
and [NAG
94]

Different
Vector:
views from
(points, Systematic:
Topological the start
arcs and Localized or complete or
3D-IMA and according to
facets: generalized multidirectional
development angles and
graph aiming
3D points of
theory)
view

Table 10.5. Comparison between various recent publications on visibility

Thus, if we disregard the potential work done in this field by military

laboratories (American, European and others), research that became the subject of
publications is not plentiful. We can mainly quote the three most recent working
groups:
– English-speaking researchers: Randolph [RAN 94, RAN 99], Nagy [NAG 94]
working within the Rensselaer Polytechnic Institute (New York), Lee [LEE 92] of
the University of Kent (Ohio) and also Fisher [FIS 93] of the University of
Leicester;
– the works of Italian researchers: De Floriani and Magillo [FLO 99] of the
University of Genoa;
– those of French teams: University of Franche-Comté [BRO 94, BRO 99],
COSTEL laboratory of the University of Rennes [DU 93] and the CESA laboratory
of the University of Tours [SER 99, SER 00a, SER 00b].

The comparison between these various publications will be summarized as

presented in Table 10.5. This reiterates the essential of the characteristics that appear
most often in publications dealing with visibility.
216 Graphs and Networks

Before concluding this chapter we will distinguish in this comparison three

primary categories of different elements: data, algorithms and results.

10.3.4.1. Raster and/or vector data?

The majority of visibility algorithms consider the studied terrain via a raster grid
(pixel matrix coupled to that of corresponding altitudes). In this research we favored
the use of vector data created on the basis of an image raster (Figures 10.1 and 10.3
in the color plates section).

The use of the vector mode gives us, on the one hand, a facility in handling
graphic entities and, on the other hand, a high degree of accuracy. Thus, it is
possible to modify (either via the CAD software used or via the 3D-IMA model
itself) the altitude of a set of geometrical primitives (punctual, linear and/or surface
data) relating, for example, to a wood, a forest or something else: different scenarios
of forest height simulation (10, 15, 20 m, etc.) before the calculation of the visibility.

Contrary to the AMAP (CIRAD – Montpellier) model where the growth of flora
in a landscape is simulated according to its age and agronomic constraints, in this
research the landscape is modeled as a whole. We will not try, for example, to
identify each tree in a forest but rather to consider the forest as a mass: a volume of
an average height. Having said that, the use of the vector model enables us to make
the grid thicker at any moment (DMG and DMS) and at a very precise site in order
to integrate new information on volume(s), thus approaching as much as possible the
complete form of this information. This possible return journey in landscape
modeling (by addition, suppression or modification of information) is conditioned
by the availability of time, means and especially the objectives required. The larger
the development project, intended to spread for dozens of kilometers, with a lifespan
of around a century and involving colossal expenses, the more justifiable it is to
employ this kind of approach (as is done more and more often).

10.3.4.2. Characteristics of the covisibility algorithm

The main characteristics of the covisibility algorithm are as follows.

Complete or multidirectional aim. This complete aim is the consequence of the

overlap of two main loops in the algorithm. The first makes it possible to traverse
the entities corresponding to the set of the possible viewpoints (file of a highway
section described by many rectangular meshes). The second loop makes it possible
to sweep all the target meshes of the surrounding landscape (file of all the entities
describing the landscape around the installation). Thus, this overlap performs a
systematic sweep in a precise order of all the entities of the installation and its space
simultaneously.
 Evaluation of Covisibility of Planning and Housing Projects 217

The algorithm of visibility implemented here is of the centroid-centroid type. The

exploration is conducted on the basis of a dual observer or origin graph – towards a
dual target or destination graph (see above). This algorithm, for very sensible spaces
(ZNIEFF, etc.), can be easily made more exact through coupling with the
multiresolution topography method. For our purposes this coupling is scientifically
satisfactory in comparison with publications that advocate:
– the implementation of a probabilistic model to take into account altitude
errors [LEE 93]. “A probabilistic representation of the viewshed is more acceptable
than the usual binary product” [MADE 93, p. 331];
– the use of the vague set theory. Knowing that these are relatively complex
and sophisticated algorithms, researchers dedicated themselves to the development
of parallel visibility algorithms (multiprocessor machines): “because visibility
computations are so computer-intensive, they form has natural target for
parallelization” [NAG 94, p. 769].

Localized or generalized visibility. It is one of the characteristics of the 3D-IMA

model. The latter automatically determines the type of visibility in question. The
first (covisibility) is the most frequent, in particular, in commercial GIS [FIS 93].
The second, which is a generalization of the previous one, is much less common
because of the quantity of data and calculations that it involves, without forgetting
that the researched objectives are often far away from the installation. Figure 10.11
of the submission for review of the article of Brossard and Wieber [BRO 84, p. 10]
is the result of the intersection after superposition of various visibility cones
obtained with local photographs. It should be stressed that, even if this method was
limited to only a few points of view (around 10), the approach itself would remain
pioneering in the subject matter. It is one of the first forms of generalized visibility.

Integration of constraints specific to the field of landscape management. Among

these we have the maximum visibility distance which was integrated since the start
of the method by limiting the surface of the modeled space. In the case of a linear
installation, the visibility is studied section by section of approximately 5 to 10
kilometers with overlapping zones between the various sections. Each section
corresponds spatially to the segment of the works, which raises interest and concern,
first of all, where it crosses towns and affects local residents. The desire to finish a
future highway project by using, in particular section by section and not lump
visibility simulations could arise during:
– the determination of the various alternatives of the future project taking the
landscape [SER 01], as well as the other technical and economic criteria (slope,
shortest path, radius of curvature, etc.), into account;
– taking into account the opinion of the future residents within the framework of
improving a participative democracy;
218 Graphs and Networks

– the search for the most sensitive spaces (from a landscape and human point of
view: pressure of local residents interest groups) that have priority with respect to
“1% landscape and development”.

10.3.4.3. Graphic restitution of the results: color and shape representation

To discuss, accept or express reservation about a certain visibility calculation
algorithm requires either the study of the algorithm itself or the analysis of its results
(images, maps, statistics) obtained by taking the stated constraints into
consideration. Since the first alternative is generally unavailable, we are left with the
second possibility. “Unlike some other GIS functions the viewshed is not actually
verifiable in the field nor can it be logically validated, anything more than trivially,
by examination of test figures …” [FIS 93, p. 333].

During such a procedure the majority of results regarding visibility reach us in

the form of an image raster [BRO 84] with a view in the (2D) plan, even though at a
given time the calculations have to involve the matrix of the altitudes (DMG). This
form of image or cartographic restitution does not speak in favor of a better visual
perception and therefore the understanding or even the appropriation of the results is
obtained according to the landscape and the main obstacles. In raster mode it is
necessary to resort to the overlay technique to obtain 3D raster visualization. This
restitution, although interesting, remains despite everything a typical (2.5D) surface
representation.

With a model vector, the change and multiplication of the points and angles of
view are one of its intrinsic characteristics. This three-dimensional and colored
cartographic representation is more easily perceptible, aesthetically better and seems
to better convey the expected message. The user, the resident, the elected official or
the decision maker can visually and immediately interpret, at least partially, the
results obtained according to the landscape and the main obstacles at the site: the
phenomenon of results appropriation by the decision maker. Thus, we estimate that
we will obtain better information transmission in the hope of an optimal effect.

The mode of representation of the obtained results is mainly conditioned by the

nature of the data used. They are two different ways (raster, vector) to approach
modeling and, thus, the representation of a landscape.

10.4. Conclusion

Nowadays the GIS propose to the users multiple analysis functions that are very
attractive in many fields, on the other hand, in terms of:
– determining the best sites or routes of an installation;
 Evaluation of Covisibility of Planning and Housing Projects 219

– estimating the visual impacts of the future installation on the landscape for an
optimal and, most of all, equitable implementation of the equipment of “1%
landscape and development”.

The majority of GIS do not propose specific functions, such as covisibility and,
even less, generalized visibility (simulations of landscapes which are most
frequently seen by a driver).

Hence there is the need to have or create tools which are appropriate for the
questions and problems treated in landscape management, where the usefulness of
cartography and digital imagery is undeniable. Among these models we also find
3D-IMA.

Let us specify that although covisibility and its alternatives give a precise answer
to the question of visual impact and the insertion of an installation into the
landscape, on the other hand, this function cannot be considered individually when
deciding on the future of a potential installation: we need multicriteria approaches
[SER 01] to help manage decision-making.

Within the framework of implementing a future installation, the visibility

algorithm can be used at various stages:
– upstream of project realization: making it possible to determine zones that are
most exposed to the view by using calculations of generalized visibility of space to
space. Once detected, these zones could possibly be rejected during the search for
alternative sites;
– during the search for project variants: the use of visibility helps to evaluate the
visual pressure of each variant in order to implement sorting and selection, in
particular, in conjunction with other criteria (slope, minimal path, cost, presence or
proximity of protected zones, etc.) and taking into consideration the opinion of the
residents;
– downstream of the project: for a better implementation of the “1% landscape
and development”, the algorithm of visibility (covisibility) may also act very locally
to determine with precision the generally affected areas in each town.

The method of visibility, in its current state, integrates certain constraints that
make it more humane and operational. Thus, it is possible to determine the visible
landscape according to various assumptions on the height of the installation:
embankment, ditch or the same altitude as its topography. It is also possible to
calculate the visibility by taking into account the intermediate obstacles
(constructions, woods, etc.) and the characteristics of the population that sees the
installation.
220 Graphs and Networks

Finally, among possible future developments we wish, on the one hand, to widen
the field of constraints of visibility calculation (flow of travelers, effect of proximity
of a notable element, such as a monument or a natural attraction) and, on the other
hand, to test this tool on cases of installation projects in an urban environment. The
second prospect will most probably require significant adaptations of the model to
simultaneously respond to the characteristics of the urban environment (high density
of geographical information) and the requirements of urban dwellers in terms of
quality of life by means of an even partial control of the various sources of potential
disruptions (visual, sound, air pollution, etc.) that may be generated during the
realization of a new urban development.

10.5. Bibliography

[ARN 00] ARNAUD M., EMERY X., Estimation et interpolation spatiale, Hermes, p. 221,
2000.
[BER 67] BERTIN J., Sémiologie graphique, Paris-La Haye: Mouton, Paris: Gauthier-Villars,
1967.
[BER 73] BERTALANFFY L., Théorie générale des systèmes, Paris, Dunod, 1973.
[BRO 84] BROSSARD TH., WIEBER J-C., “Le paysage trois définitions, un mode d’analyse
et de cartographie”, L’espace Géographique, no. 1, Paris, Doin, p. 5-12, 1984.
[BRO 94] BROSSARD T., JOLY D., LAFFLY D., VUILLOD P., WIEBER J.-C., “Pratique
des systèmes d’information géographique et analyse des paysage”, Revue internationale
de Géomatique, vol. 4, no. 3-4, Paris, Hermès, p. 243-251, 1994.
[BRO 99] BROSSARD T., JOLY D., “Représentation du paysage visible et échelles spatiales
d’information”, Revue Internationale de Géomatique, vol. 9, no. 3, Paris, Hermes, p. 359-
375, 1999.
[CHA 91] CHASSERY J.-M., MONTANVERT A., Géométrie discrète en analyse d’images,
Paris, Hermès, p. 358, 1991.
[COL 92] COLLET C., Systèmes d’information géographique en mode image, Presses
polytechniques et universitaires romandes, collection Gérer l’environnement, p. 186,
1992.
[DU 93] LE DU L., GOUERY P., “Paysage littoral: cartographie des degrés de visibilité”,
Mappemonde 2/93, GIP-RECLUS, p. 9-11, 1993.
[FIS 93] FISHER P.F., “Algorithm and uncertainty in viewshed analysis”, Int. J.
Geographical Information Systems, vol. 7, no. 4, p. 331-347, 1993.
[FLO 99] DE FLORIANI L., MAGILLO P., “Intervisibility on terrains”, Geographic
information systems principles techniques management applications, Maguire Goodchild
Lonley, Rhind, p. 543-556, 1999.
[LAA 00] LAARIBI A., SIG et analyse spatiale, Hermes, p. 190, 2000.
 Evaluation of Covisibility of Planning and Housing Projects 221

[LEE 92] LEE J., SNYDER P.K., FISHER P.F., “Modeling the effect of data errors on feature
extraction from Digital Elevation Models”, Photogrammetric Engineering & Remote
Sensing, vol. 58, no. 10, p. 1 461-1 467, 1992.
[MAN 95] MANDELBROT B., Les objets fractals, Champ Flammarion, p. 212, 1995.
[MAT 94] MATHERON J.-P., Comprendre MERISE: outils conceptuels et organisationnels,
Paris, Eyrolles, p. 265, 1994.
[NAG 94] NAGY G., “Terrain visibility”, Computers and Graphics, vol. 18, no. 6, p. 763-
773, 1994 (http://www.ecse.rpi.edu/homepages/nagy/doclab.htm).
[RAN 94] RANDOLPH F.W., RAY C., “Higher isn’t necessarily better: visibility algorithms
and experiments”, in Thomas C. Waugh, Richard G. Healey (eds.), Advances in GIS
Research: Sixth International Symposium on Spatial Data Handling, p. 751-770, Taylor
& Francis, Edinburgh, 5-9 September 1994.
[RAN 99] RANDOLPH F. Wm., “Terrain elevation data structure operations”, 19e
Conférence Internationale de Cartographie, 1999, Tome 2, p. 1 011-1 020, 1994
(http://www.ecse.rpi.edu/homepages/wrf/research/index.htm).
[SER 97] SERRHINI K., “La métrique du paysage: deux indicateurs spécifiques au relief
pour l’aménagement de l’espace”, p. 259, Actes des troisièmes rencontres de Théo Quant,
Besançon 1997.
[SER 99] SERRHINI K., “Evaluation de l’impact et de l’insertion de projets d’aménagement
dans le paysage”, 19e Conférence Internationale de Cartographie, vol. 2, p. 1 639-1 646,
1999.
[SER 00a] SERRHINI K., “Le multiéchelles et la covisibilité pour une meilleure insertion des
projets d’aménagement dans le paysage”, Mari Europe 2000, Paris, 18-20 April 2000.
[SER 00b] SERRHINI K., Evaluation spatiale de la covisibilité d’un aménagement.
Sémiologie graphique expérimentale et modélisation quantitative, PhD Thesis, Tours,
2000.
[SER 01] SERRHINI K., “Intégration quantitative et multicritère du paysage lors de la
détermination de tracés d’un aménagement linéaire”, Mappemonde, no. 61, GIP-
RECLUS, p. 15-18, March 2001.
[THI 87] THIBAULT S., Modélisation morphofonctionnelle des réseaux d’assainissement
urbain à l’aide du concept de dimension fractale, PhD Thesis, INSA Lyon 1, p. 144 et
seq., 1987, 300 p., see also Part 3: Graph and Fractals.
 Chapter 11

Dynamics of Von Thünen’s Model:

Duality and Multiple Levels

The dynamic model1 presented below is developed on the basis of the

groundbreaking model of spatial analysis developed in 1826 by Johann Heinrich von
Thünen [THU 26]. Let us point out that the purpose of the von Thünen model is to
determine the location of cultivation in the space of the “isolated State”, that is, an
isolated plain containing at its center a market town. Von Thünen determines that
cultures will form succeeding rings around the town according to the income of the
farmer (Figure 11.1). This income is the difference between all the production costs
plus the cost of transport and the price on the urban market. The producer attempts
to maximize his revenue and the comparison of the potential income from each
product according to the price observed determines the production for the following
year.

The first issue, which calls into question a generally accepted hypothesis, is to
find out if spatial balance can result from a production that is strictly based on the
market prices of the previous year, by supposing that they were constant during the
season, and be maintained through time. Should this hypothesis not be verified, what
are the spatial consequences?

Secondly, generally speaking, what can be the spatial effects of the influence of
two or several market town centers whose production zones overlap?

Chapter written by Philippe MATHIS.

1 This chapter follows up on a paper delivered at the “European colloquium on theoretical and
quantitative geography”, 1997.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
224 Graphs and Networks

The stress is always placed on the spatial characteristics of the phenomenon, i.e.
the form of the supply zones of each town and of the production zones of each
product.

The considered space is discrete. In our simulations it comprises 600 properties

cultivating one of the three types of products (truck farming, silviculture, cereals)
according to the prices of the previous year, production cost and distance to the town
center which makes it possible to obtain the highest revenue.

Each year, the price in the town center(s) is a function of the ratio between the
quantity produced and the quantity required which we will assume to be constant
and dependent on the population.

The space is defined by a set of points divided in staged rows connected to each
other by a network describes by a graph with triangular faces. The properties are
located at the nodes of the graph and the delimitation of field surfaces is defined by
the dual of the graph which is formed by the transport network. This surface may or
may not be regular; here we chose the simplest solution of the equality of surfaces.
However, a different hypothesis may be taken into account.

The method used is simple and consists of a first description of space with the
definition of the characteristic points and the relations between these points: the
graph. Then the particular nature of the relation is taken into account by using the
von Thünen model. In the case of a diachronic comparison the process stops,
whereas it continues in the case of a dynamic by successive iterations, whose
resulting space in turn becomes the starting space and so on.

The method is perhaps less convincing than a mathematical demonstration, but it

is not limited to a few very simplified hypotheses and the systematic exploration of a
model, by freely modifying the characteristics of the spaces, makes it possible to
clearly delimit its field of validity.

Within this framework the von Thünen model is only a part of a larger group.

11.1. Hypotheses and ambitions at the origin of this dynamic von Thünen model

Two objectives have formed the foundations of this work. The first is to find a
dynamic use making it possible, among other things, to test the implicit hypothesis
of temporal stability. The second is the use of the von Thünen model in
conversational simulation while being able to modify any of its characteristics. We
call a simulation conversational when for each iteration it is possible to modify the
 Dynamics of Von Thünen’s Model 225

parameters on screen, to simulate decisions, to use the model practically as a game

where we are able to interfere in real-time with a certain number of parameters.

First of all, we wanted to test the hypothesis of temporal stability of the von
Thünen model in a dynamic version.

Von Thünen [HUR 94] does not specify the relation between the quantities
produced and the prices. However, it appears that this relation potentially contains at
least one instability related to the spatial nature of the phenomenon: if we adopt the
simplest hypothesis of proportionality, the variation of prices involves a variation of
the cost of transport and, thus, of the possible distance to cover between the place of
production and the marketplace.
Revenu
Net net
production
de production
income

1200 Truck farming cultures

Cultures maraîchères
1000

800

600 A
Intensive cultivation
Élevage intensif
400
Corn
200 B Maïs Extensive cultivation
C Elevage extensif
0
0 100 200 300 Km

Distancetoà the
Distance la ville
townde
ofBourg
Bourg

Lines expressing,
Droites exprimant, for pour
each chacun
of the four
des uses of usages
quatre land around
du solthe town
ofautour
Bourg,dethe
la net production
ville de bourg,income according
le revenu to the distance
net de production en from
the town.
fonction de la distance à la ville.
Courbe enveloppe,
Envelope exprimant
curve expressing en point
in each tout point le maximum
the maximum de profit
obtainable profit
réalisable
and et la spécialisation
the corresponding agricole
agriculture qui lui correspond.
specialization.
A,B,
A, B,C,C,points
pointsfor
pour lesquels
which le operating
the net revenu netprofit
d’exploitation
is the sameest
forletwo
même
uses.pour
These
deux usages.
points Cesradii
define the points définissent
of the les rayons
von Thünen rings. des cercles de Thünen.

Figure 11.1. Diagram of the producer’s income for each product depending on the price in
the market town minus all the production costs and the costs of transport

Effective production is a function of the cultivated surface since we suppose

constant returns, at least initially, and in the simplest von Thünen model this surface,
which can generally vary according to the case, is a ring around the town.
226 Graphs and Networks

Any variation in price according to the difference between the quantity required
and the quantity produced involves a modification of the producer’s income which is
defined as the difference between the price and all the production costs plus the cost
of transport. However, this is valid for each property. Since the production costs are
fixed, the income depends on the costs of transport. However, the total variation of
production is a function of the difference of the squares of the distances that these
prices make it possible to cover, when all other parameters are identical for each
producer.

Consequently, the stronger the variation in production will be, the greater the
distance from production and the smaller the cost of transport. It will also depend on
the product considered: the remotest production will be able to extend without
obstacles, whereas the most central production must shift the others and thus
generate an income that justifies the change of culture in the surface considered.

Thus, a priori, there is an amplification and instability process, all the more so
since, in particular for foodstuffs with a constant demand, the elasticity of the
quantity with respect to the demand is very low and, consequently, the elasticity of
the prices with respect to the quantities is normally much higher than 1. This further
amplifies the phenomenon and, thus, logically, its spatial consequences.

In addition to the development dynamics when we introduce cultivation rotation,

the development of yield remains to be determined. Similarly, it is necessary to be
able to introduce risks, differential developments in the productivity of the modes of
cultures, work, transport, soils, in the consumptions, in populations, in a global,
spatial, sector manner, etc.

The second ambition is to relax the hypotheses [HUR 94], to remove the most
constraining, which are mainly those related to the mathematical tool, in order to be
able to explore the capacities of this fundamental model in less simple
environments, less theoretical than those used by the majority of economists. This
implies, in particular, rejecting a uniform continuous space and the Euclidean
distance.

We wish to be able to:

– simulate in real-time and interactively, and to immediately have the figures of
production locations;
– freely change the characteristics of space: regular spaces of Christallerian type,
irregular spaces, random spaces generated by a model, etc.;
– change from the transport function to demand, whether proportional or not,
including load ruptures, but also the existence of several means of transport;
 Dynamics of Von Thünen’s Model 227

– vary the number of towns on request;

– analyze exactly the problems of production zones limits in the case of several
market towns;
– freely change the production costs, introduce qualities or specific “vocations”
of the soils;
– explore the influence of spatial scales.

Finally, it seems necessary to be able to explore the consequences of behavioral

inertia, instantaneous adaptation, training and anticipation. Therefore, this implies
having a basic dynamic structure, into which we can gradually integrate new
dimensions.

11.2. The current state of research

Due to the extent of the combination of the various hypotheses that have to be
explored, only initial results are presented here. The main objectives, such as the
simplification of the hypotheses and real-time simulation have been achieved. The
dynamization of the model made it possible to confirm the hypotheses of instability.
The introduction of risks is possible, but has not yet been analyzed because the
consequences of dynamization appeared significant. Culture rotations and behaviors
still need to be introduced: inertia, training, anticipation, instantaneous adaptation
being the selected hypotheses.

11.3. The structure of the program

The dynamization of the von Thünen model that we carried out a follow up on
the models of space description that constitute part of the D.LOCA.T.2 “software”
and uses a visual BASIC routine to implement the parameters, MAP3 software for
the “maps” and Excel for graphs.

2 D. LOCA.T is a patented software developed by members of the CESA laboratory.

3 MAP is a patented software developed by members of the CESA laboratory.
228 Graphs and Networks

Figure 11.2. Random space with 600 fields locations and 1 to N

possible towns. Seven towns were kept in this space

It is implemented in 32 bit FORTRAN, uses the Power Station Microsoft

compiler and runs on a desktop computer. The space presently used is either a
random space (600 – Figure 11.2 – or 1,225 zones or fields: both spaces look very
similar), or a space of the Christaller type (1,225 zones or fields – Figure 11.3)
which is generated by the RES4 program, which makes it possible to obtain
networks with planar graphs.

The current limit on the number of points is due mainly to the calculation time,
in particular, that of minimal distances because Floyd’s algorithm has a complexity
of O(n3). This doubling of the number of nodes involves a multiplication of the
calculation time by eight, which means a quadrupling of the size of the matrices and
thus of the necessary memory. However, the programs are conceived to be usable
regardless of the number of points used.

The program of the dynamized von Thünen model imports a file of properties (or
producers) locations, a file of markets towns and a file of distances from the
properties to the market towns. The necessary parameters are fixed by using an
interface, i.e. a window, in visual BASIC, which makes it possible to define on
screen the desired values for these various parameters, either at the beginning of the

4 RES is a program developed by Hervé Baptiste, which makes it possible to generate a space
where random locations are connected by a saturated planar network. Towns or central
squares can be assigned to these locations with possible population constraints. The model
can generate more than a million different spaces. See, by the same author, [BAP 99].
 Dynamics of Von Thünen’s Model 229

period, or during each iteration. The fields or the market towns are located at the
nodes and the arcs represent the circulation network (Figure 11.3).

Figure 11.3. Christallerian circulation network

The dynamic simulation space could be defined as: 1,225 locations, seven
kilometers between each and two central sites with 200,000 inhabitants each5. In this
simple and regular case the limits of the properties are defined by the dual of the
circulation network, which is a regular, planar and saturated graph. This network has
a graph with hexagonal faces as a dual.

On the basis of production and transport costs, distances and prices of products
in the various towns, the program determines the potential income from each
production for each property, and all the farms specialize in the production that
maximizes the private income.

The model respects the hypotheses 1 to 5 described by Huriot [HUR 94, p. 265 et
seq.]. Currently, the model exhibits myopic behavior, since it does not anticipate and
it maximizes the expected profit only for this period.

At the beginning the conditions are as follows:

– no soil use;
– no predetermined zones;
– by simplification all the farms are identical, but this is not necessary; specific
characteristics may be input for each farm (surface, quality of land for each product,
technical profitability, costs, etc.);
– three products are taken into account, but we can use more, if so desired;

5 See section 11.4.3.

230 Graphs and Networks

– no a priori capital constraints, but they could be easily introduced for each
farm;
– the market is not unique: there are as many of them as there are towns which
are taken into account. Thus, for the same product there is a specific price for each
point of sale, which is a function of the relation between local supply and demand;
– there is no exchange of products between various markets and demand is
confined to each one of them;
– the choice of the production for each farm is a function of the maximum
revenue expected income by considering the revenue, i.e. the prices of the previous
period minus production costs and the related transport costs. In the long term it is
envisaged that preserving inertia, learning and anticipation will be integrated.

Let us observe that the decision is made traditionally: each property chooses the
product that generates the highest income for the most interesting market in the case
of several market towns. Thus, there is a comparison of income for each product and
market. Naturally, there is no production at a loss. The choice is therefore more
significant for the producer when there are several markets, since by assuming that
there are three products and five towns, he has to take into account 15 potential
incomes! That renders the determination of the production zones of each town much
more interesting since these zones become all the more dynamic.

Determining the total quantities produced is simple when the product chosen for
each farm, the production surface, the output and the place where it will be marketed
are known. We can sum up for each market town the quantities of each product that
will be offered for sale there.

Thus, we have a cellular model where each farm is a “cellular agent” with a
behavior pattern that is the same for all in this case. Each farm individually decides
its production and the market where it will be sold according to its particular
interest: maximization of revenue.

The cellular model defining the nature and quantity offered for each market
makes it possible to move on to the global level of pricing for each market which is
obtained by comparison between local supply and demand. The difference in price
depends on the difference in quantities.

The prices in year T, which are determined overall for each market, will
influence the choices of production of the agents during year T+1 and thus the
production distribution space, but this will depend on all the prices in each market.
 Dynamics of Von Thünen’s Model 231

Consequently, the supplying zone of a town does not depend only on the prices
established there but also on prices in other markets that are a function of their
populations and distances.

The interrelationships are thus essentially of a spatial nature, since they make it
possible to pass from the level of the agent to the global level. The reasoning is not
the same: it is necessary to know all the production to define the price, whereas the
latter enables each agent to define his production in the next period, while remaining
completely unaware of the reaction of other agents: the agent’s rationality is limited.

11.4. Simulations carried out

The exploration of the model mainly related to the effects of the space scale
factor, the effects of transport costs, price trends, the effect of price/quantities
elasticities, the effect of the number of towns, the shifting of the borders of
production zones, which implied a significant number of simulations from 20 to 150
iterations.

Four simulations are presented here: the first two refer to a random space with
600 farms and with a single central town as in the original model, the difference
being in the value of the damper coefficient of the price/quantity relation. These two
simulations emphasize the extreme sensitivity of the model.

The third simulation refers to a regular Christaller space with two identical
towns. This “academic case” makes it possible to highlight the variations in the
spatial location of the border between the two provisioning surfaces and the strong
variations of production surfaces of the various products.

The last simulation again uses a random space of 600 fields where we find six
market towns of unequal importance, which, in a more complex case, makes it
possible to predict drive and spatial domination effects.

Each simulation comprises the various parameters used, one or more graphs of
the product’s price variations according to time and a simplified representation of
the production areas of the various products and the provisioning zones of the
various towns, if necessary.

Each simulation begins with specific characteristics: the space used, the various
files, the number of towns, products, iterations to be made, the useful agricultural
surface, a technical factor which is a coefficient of scale, the type of output, the type
of transport costs, the growth rate, the type of output of the land followed by
232 Graphs and Networks

transport costs per product, the unitary per capita consumption, the production costs
per product and the absorber or amplification coefficient of the price/quantity ratio.

The produced price/quantity ratio used in simulations has a very simple, linear
form:

pppactkj = pppantkj x (100 + epqkj x (-dpjk)/100) if dpjk < 0 [11.1]

pppactkj = pppantkj/(100 + epqkj x (dpjk)/100) if dpjk > 0 [11.2]

where pppactkj is the current price of the product k for the town J, pppantkj is the
former price of the product k for the town J, epqkj is the price/quantity absorber
coefficient and dpjk is the difference between the quantity produced and quantity
demanded expressed as a percentage.

Let us recall that the choice of production in year n is individual for each farm. It
is determined by the highest revenue that the prices of year n-1 generate. The
variation in production volume is the sum of all the individual choices.

11.4.1. The first simulation: a strong instability in the isolated state with only one
market town

The windows, through which the various parameters are implemented, are
preserved in the form of files. That makes it possible to define a tree structure of
simulations and to explore systematically.

The first simulations immediately showed the great sensitivity of the model and
its instability in time, contrary to the traditional assumptions of the existence of a
stationary balance [HUR 94]. The simulation presented below and illustrated by
Figure 11.4 is a typical example.
 Dynamics of Von Thünen’s Model 233

Prices
45

40

35

30

25

20

15

10

0 Years
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49

Figure 11.4. Price of the three products to the nearest constant in the market town,
with a price/quantity elasticity of 0.7

The first simulation uses a price/quantity elasticity of 0.7, which could have
been replaced by elasticity in its proper sense. The surprise comes from the fact that
traditionally the price/quantity elasticity is considered to be greater than 1 for
foodstuffs, whereas we systematically had to use a reverse multiplier between prices
and quantities.

Despite that, the fluctuations are very strong from one year to the next, and the
curve obtained exhibits a quasi-periodicity and a behavior that is unforeseeable in
the short term. This quasi-periodicity, which evokes a chaotic phenomenon, is the
result of the interaction of the three products chosen with individual rationality
according to the income of 600 agents that are independent, but localized spatially.

Here, the sum of the individual choices does not lead to the overall optimum.
The hypothesis that we can formulate on this subject is that any increase in price
involves the possibility of cultivating a product at an additional distance ∆ with
respect to the prior distance d. However, this additional distance ∆ from the market
center enables an additional production on a surface that increases proportionately to
d and is equal to:

δS = [π(d+∆)²-πd²] [11.3]

that is, δS = ∆π(∆+2d) [11.4]

234 Graphs and Networks

and production will thus be of type P = f(πΔ(Δ+2d)) [11.5]

Secondly, it is clear that the modification of one or several production surfaces

will consequently involve an adaptation of the other surfaces. Indeed, only the
growth of the third ring will have very little influence on the two preceding ones,
whereas a strong increase in the prices of this product will lead to an increase in
income and, thus, a reduction of the previous product in the space that has become
less profitable “to the space margin”.

year 1 year 2 year 3 year 4 year 5

year 6 year 7 year 8 year 9 year 10

year 11 year 12 year 13 year 14 year 15

year 16 year 17 year 18 year 19 year 20

Figure 11.5. Variation in the space of the cultivation zones of the three products

On the other hand, a variation in the price of the product cultivated in the second
and, all the more so, in the first crown will generate an adaptation of all the products
 Dynamics of Von Thünen’s Model 235

that can be reflected in time. Indeed, a priori, nothing makes it possible to consider
that the variation in the price of one or several products related to the inadequacy
between production and consumption will make it possible to solve the problem the
following year spontaneously and exactly. There are multiple reasons for this. The
main one stems from the spatialized character of production: an increase or
reduction in price involves an increase or reduction which is a function of the square
of the distance variation to the market.

We have here a mechanism of amplification where the disturbances will not

necessarily diminish, as observed over the first 20 periods of the simulation (Figure
11.4).

The effect on space is considerable, since we note a great instability not only of
the cultivation zones, but also of the cultures themselves, which will appear and
disappear according to market prices (Figure 11.5).

The maps indicate the zones of cultivation in the first years, but taking into
consideration the price graph clearly indicates that the phenomenon persists in time
with perhaps a quasi-cycle long superposition.

11.4.2. The second simulation: reducing instability

The second simulation differs from the first only by the value of elasticity which
is 0.3 here. We can immediately note the effect on the price curves (Figure 11.6),
curves whose variation amplitude is restricted, but always preserves this quasi-
periodic character with a duration of eight “years”, which evokes the Juglar cycle6,
during which the fluctuations seem to be limited in order to regain their amplitude of
the start.

In this simulation, space “competition” develops primarily between products 1

and 2, as we can observe from the graphs, which we have increased to this end.

The product whose fluctuation is the strongest is the one produced at the edge of
the town, i.e. truck farming, then comes silviculture and, finally, cereals on the
outside.

We note that only a considerable fluctuation in the price of product 1 would

enable an extension of the surfaces thus triggering a weaker fluctuation of product 2,
which will push back product 3 by generating a price hike the following year. This

6 A cycle highlighted by the French economist Clément Juglar in 1862 that has a duration of
seven to eight years and is an intermediary between the short and long Kondratiev cycles.
236 Graphs and Networks

one-year shift between the price maxima of each product is explained by the
increase in production surfaces following the increase in income (Figure 11.7),
which involves a reduction of the surface dedicated to the next product, whose
income has changed only a little.

We note a quasi-periodicity of change during the first 30 years which is followed

by a period of low amplitude fluctuation of only two years and then the phenomenon
restarts with strong amplitudes.

30

25

20
Prices

15

10

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49

Years

Figure 11.6. Price of the three products to the nearest constant in the market town,
with a price/quantity elasticity of 0.3

It is clear that such fluctuations could not happen so radically in reality, although
there are known cases where certain productions were partially supplanted by others,
such as, for example, the development of kiwis at the expense of apricots, or
completely as witnessed by the disappearance or quasi-disappearance of certain
types of fruit or domestic animal races. It would be necessary, however, to introduce
production cycles into the model, because certain cultures, such as silviculture, take
many years to develop and due to this fact the fluctuations are damped by a “delay
constraint”.
 Dynamics of Von Thünen’s Model 237

year 1 year 2 year 3 year 4

year 5 year 6 year 7 year 8

year 9 year 10 year 11 year 12

Figure 11.7. Variation in the space of the cultivation zones

of the three products with a damping effect

11.4.3. The third simulation: the competition of two towns

We used a Christaller space (Figure 11.3) conforming to the one described

above, with two equally important market towns, which are symmetricly located in
this homogenous space.

The first observation is that, even in this case where all the factors seem
identical, a very fast differentiation between the two towns appears, immediately for
one product and after two and four years for the two others (Figure 11.8).
238 Graphs and Networks

25

20
Prices

15

10

0
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49
Years

Figure 11.8. Price trends for the three products in both towns.
The prices are evaluated to the nearest constant. There are 1,225 sites

It should be noted that the quasi-periodicity remains a constant and that, during
the simulation period, the order of the fluctuations changes: the prices for products
in the same town are not the first to vary. That is particularly visible for product 1
(top curve) where the dotted lines represent the prices in the first town and
continuous lines represent those in the second. We observe a quasi-periodicity, but
with shifts between the towns.

We establish the origin of the fluctuation by carefully observing the production

zones: the number of fields between the two towns is odd and this is a phenomenon
to which we had not paid attention at the beginning also in programming, which is
necessarily discrete, the town on the left is selected first and therefore its production
zone is the largest, which triggers the fluctuation amplified by the appropriation of
the most distant quasi-median field for cereal production, due to the systematic
rounding-off. The town on the left benefits from three additional fields compared to
that on the right, that is 0.5%, from a total of 627 really productive fields for the two
towns out of a potential 1,225, when taking into account the selected parameters.
Space is intentionally oversized, so that the cultures are not restricted

This inequality of production surfaces triggers a differentiation in prices, which

in turn sustains the phenomenon, even more so since the layout of the fields in a
quincunx in turn introduces an inequality which we observe starting from the second
year: five extreme zones for town A and four for B. The inequality persists when the
zones of cultivation arrive at the limit of the expanse taken into account.

It seems that we are witnessing a “butterfly effect” at the start, which is later
amplified as it is clearly shown by the sometimes complex fluctuation of the zones
 Dynamics of Von Thünen’s Model 239

by product and of the border of the production zones of the two towns which can be
seen below.

The graph (Figure 11.9 in the color plates section), since it would be
inappropriate to call this a map, on the left represents the production zones of each
product for both towns: product one in blue, the second in green and the third in red.
The fields that do not provide any production to the urban market were left blank.
The graph on the right represents the production zones of each town: the fields
whose produce is sold in the town market on the left are in blue and those in the
town market on the right are in green.

The apparent symmetry of product zones disappears from the second year for
products 2 and 3 and the phenomenon is amplified throughout the later years until
the near disappearance of product 1 (truck farming) for the town on the right during
the 7th year, and the total disappearance during the 15th year for the same town. The
cereal zones also fluctuate alternatively between the 14th and the 17th year.

Moreover, we can note that the expansions and reductions of the surfaces
occupied by each culture do not fluctuate simultaneously, as it is clearly shown in
year 15 for the town on the right where we have the disappearance of truck farming
at the same time as a strong expansion of product 2 and a reduction of the cereal
zone. We could provide multiple examples of this type.

With respect to the border between the provisioning zones of the two towns, the
results here are very close to those of a Palander model between two companies, of
which we have here almost a partial simulation. Indeed, the shape of the border in is
the majority of cases a discrete hyperbole and the costs of transport are evaluated
from the fields to the markets, whereas for Palander they are estimated from the
production facilities to the market, which in both models symbolize the slope of the
costs of transport.

More generally, the graph of product’s prices for both towns alone does not
make it possible to measure the spatial complexity of the development of the
production zones, which in turn will modify local supply and therefore prices, thus
leading to new developments of the production zones.

11.4.4. The fourth simulation: the competition between five towns of different
sizes

This simulation has the same parameters and the same space as the first two, but
takes into account five towns of different sizes: 132,000, 53,608, 31,645, 21,771 and
16,289 inhabitants respectively, which are distributed according to Zipf’s law.
240 Graphs and Networks

This inequality of populations leads to the inequality of the quantities necessary

for subsistence and, consequently, of the prices in each market.

This more complex set approaches reality and, although it is more difficult to
interpret, it makes it possible to analyze the possible phenomena of drive,
domination, etc.

Considering the various graphs makes it possible to detect strong tendencies

(Figure 11.10).
30

25

20
Prices

15

10

0
10

13

16

19

22

25

28

31

34

37

40

43

46

49
1

Years

Figure 11.10. Price of each product for the five towns

At the beginning no surfaces are used and the prices per product is identical for
all towns are arbitrary for year 0.

The first observation is of a very different nature from the curves of product 1 for
the five towns, whose location is clearly visible in the graphs (see Figures 11.11 and
11.12 in the color plates section). While for the largest one the curve is relatively
regular in its quasi-periodicity, the others, on the other hand, seem to be subjected to
an influence that makes each of them specific, which means that the least populated
towns have the most irregular curves, i.e. the fifth town. We find the same type of
phenomenon attenuated for the second product, with a clearly marked irregularity
for the last three towns.

For the third product, although the amplitude of variations is rather low as a
whole, nonetheless, it appears stronger for the smallest, which may be explained by
the fact that its territory is the most constrained by the three most populated towns
that surround it. Its production zone is limited because the producers prefer to
market their produce in other markets where it can be sold at a higher price, which
 Dynamics of Von Thünen’s Model 241

repeats the development of the production zones. Everything occurs, as if the

smallest towns were, consequently, much more sensitive than the largest.

Indeed, a rapid analysis of these makes it possible to note that the first five, six
or seven years are dedicated to the growth of the production zones before an almost
regular fluctuation of the borders between the zones dominated by each town
appears.

We may also observe a fluctuation of the center-periphery type, which is

confirmed by the analysis of the price graph of product 3 for the five towns: to each
minimum of the most populated town center there corresponds a maximum for the
other four towns, which is not true for product 2, for which we have three maxima
for two minima and vice versa.

With regard to product 1 whose fluctuations are the most considerable and
whose role is essential, the periods are different and shifted in time.

This periodicity, as if the four peripheral towns did not reach their maximum
market price simultaneously, triggers a sort of a circular motion at the periphery,
which seems to be corroborated by taking into consideration the production surfaces
per product (Figures 11.11 and 11.12 in the color plates section).

On the one hand, product 1 does not reach its maximum expansion
simultaneously in the center and at the periphery, but, on the other hand, it does not
reach it simultaneously for all the towns at the periphery either. That seems to occur
as if we had a rotation of maxima for this production, which followed the
trigonometric direction.

11.5. Conclusion

The first conclusion is the corroboration of the initial hypotheses:

– on the one hand, the von Thünen model appears unstable in its present
configuration and with the simplest proportional price/quantity function. This seems
to stem from the fact that an increase in price and therefore in the distance at which
the goods can be produced involves an increase in cultivable surfaces as a function
of the square of this distance;
– on the other hand, not only is it perfectly possible to generalize the model, but
also the more finely a space is described, the more interesting it appears since the
complexity of data enriches the model and reveals other types of phenomena;
– the importance of the spatial dimension (location of production, analysis of the
limits of the production zones) is evident and it makes it possible to find and
242 Graphs and Networks

generalize the results and to join the Palander and Devletoglou models [DEV. 65,
p. 140-160].

The conclusion on instability must be refined because a certain number of

existing “dampers” have not yet been taken into account, such as, for example, the
fact that in silviculture once a piece of land has been planted its use is not modified
the following year and the necessary time is left for the trees to grow before cutting
them down and the fact that there are constraints in the succession of the cultures,
etc. Nonetheless, even by taking these significant nuances into account, it seems that
the invisible hand has limited efficiency. Due to production spatiality, the choice of
each producer according to the potential income is not enough to ensure a balance
between supply and demand in this case.

The second major conclusion that makes it possible to highlight dynamic

analysis is the chaotic behavior of the model: fluctuation, pseudo-periodicity,
butterfly effect.

A significant number of fields where production is determined individually and

rationally according to the price of the previous year and by maximizing their
income according to two or several markets in no way ensures either price stability
or production localizations stability.

The individual and immediate rationality of the behavior seems to increase the
total instability. However, the fluctuation mechanism generates its clear borders
itself: in general, the chaotic system is permanent, although certain parameter
constellations produced systems with “explosive” prices.

The total stability of the prices seems inversely proportional to the spatial
stability of the cultures and it also seems that the relative stability of cereal prices is
the consequence of the greatest capacity of this product to fluctuate in space. This
element appears fundamental: spatial disposition constitutes an absorber for
fluctuation, which is corroborated by “explosive” simulations that we have carried
out because in these specific cases, due to the inability of supply to adapt to demand,
prices constantly continue to increase. In this case, the resolution to the crisis lies in
other means than those described in the model.

The simulation makes it possible to show that if production is directly a function

of the income, that triggers fluctuations in prices and production zones. Indeed, the
variation of income implies a variation of the possible distances from production,
which involves a variation of the surface of production, which is a function of the
square of the radius, thus leading to a systematic amplification of production
margins and therefore of price and so forth. This triggers pseudo-periodic chaotic
development.
 Dynamics of Von Thünen’s Model 243

In the model, the spatial dimension comprising all the fields or cells makes it
possible to connect the agents who individually decide on their production according
to the revenue that enabled the prices of the previous period, thus exhibiting a short-
sighted behavior and a market mechanism which fixes the prices with respect to
supply and demand.

The simulation can be considerably enriched by taking into account various

types of constraints that make it possible to approximate real situations. We can thus
reproduce complex developments of production surfaces by type of good or market
town, or follow the developments of the borders of productions zones, etc.

Having said that, many developments are necessary to exploit all the possibilities
of the model, in particular the risks, the specific qualities and the introduction of
stabilizers, such as the constraints on production. It is necessary to explore other
price/quantity functions, but especially to introduce a price memory and a learning
capacity, to open the model to the outside, to really introduce multimodality of
transport, in particular mass transport between the various centers or market towns
with means of transport of great capacity, etc.

In conclusion, we have developed a traditional cellular model on the basis of the

traditional von Thünen model by using a description of space by graphs, which
clearly showed the concrete possibility of passage between a graph and a cellular
diagram, once the dual is used, as it is here, to take the surface into account. In this
case we only used the dual to delimit the borders of the fields by considering them
all as being equal in order to simplify matters. However, duality makes it possible to
vary these sizes7 considerably and, moreover, we can incorporate fields together to
simulate phenomena of land concentration or agreements.

In the present case, the use of multi-levels is limited to individual choices when
taking into account the effect of the overall balance or imbalance by evaluating the
local offer and during the spatial analysis of the structures generated by the cellular
representation.

Despite all the extremely strong limitations of this model, and beyond its
immediate objectives, we believe that its chief interest seems to lie in the beginning
of the corroboration of our fundamental hypothesis which was presented and
expressed throughout the third and the fourth simulation, for which passing from the
individual to the global level can only be achieved through the introduction of space.

7 See Chapter 10.

244 Graphs and Networks

11.6. Bibliography

[BAP 99] BAPTISTE H., Interactions entre le système de transport et les systèmes de villes:
perspective historique pour une modélisation dynamique spatialisée, Thesis, Tours, 1999.
[DEV 65] DEVLETOGLOU N.E., “A Dissenting View of Duopoly and Spatial
Competition”, Economica, vol. 32, p. 140-160, May 1965.
[HUR 94] HURIOT J.-M., Von Thünen, économie et espace, Economica, Paris, 1994.
[PUM 01] PUMAIN D., SAINT-JULIEN T., Les interactions spatiales, Collection Cursus,
Armand Colin, Paris, 2001
[THU 26] VON THÜNEN J.H., Der isolierte Staat in Beziehung auf Landwirtschaft und
nationalökonomie, vol I Hambourg, Perthes, 1826.
 Chapter 12

The Representation of Graphs:

A Specific Domain of Graph Theory

12.1. Introduction

A certain number of recent publications by the various teams of the GDR

“LiberGeo” have all simultaneously demonstrated the efficiency of graph theory in
describing networks and their various characteristics, such as speed [LHO 97], the
characteristics of transport supply and optimization [CHA 97], mass and value flows
[ROB 00], structural dynamics [BAP 99], but also its deficiencies in terms of
morphological indicators, which can be the effectively mitigated by using the
properties of fractals [GEN 00].

Upon reflection these various publications do not seem to be contradictory since,

although each attempts to develop its own tool in a relatively exclusive manner, they
are actually deeply complementary and close. It results from this observation that
fractals could be regarded as representations of graphs under certain conditions.

This hypothesis was implicitly reinforced by the fact that the representation or
drawing of a graph is an aspect that, curiously, has been completely left out in graph
theory because, apparently, it is not an important problem within the mathematical
research dynamics, whereas it appears essential for users and, in particular, for
network modeling.

Chapter written by Philippe MATHIS.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
246 Graphs and Networks

12.1.1. The freedom of drawing a graph or the absence of representation rules

The graphic representation of G is extremely simple: the location of the nodes in

the figure, i.e. drawing of the graphing, does not count, as opposed to the
representation of the relations between nodes. That obviously supposes that nodes
are simple points without declared attributes of the coordinate type. This is
interesting because it gives great freedom of graph representation. However, in order
to represent, for example, a transport network, and if we wish the result to
“resemble” the observation, i.e. to be similar to a map, it would be appropriate to
make this representation more precise by supplementing it with additional properties
or constraints so that the process of developing the representation can be repetitive
and the result reproducible.

However, the usual representations are done on paper or on a computer screen

and are therefore plane representations. This type of projection, whose nature is
implicit in the majority of publications on the subject, is actually neither obvious nor
traditional. Indeed, if we go back in time, the founding fathers of the theory showed
more diversity than what has been preserved through usage.

12.2. Graphs and fractals

It is not mentioned that the representation may be “projected” onto a plane or a

three-dimensional space, as done by Euler [EUL 1758], or be the projection of one
onto the other, as in the famous example of Augustin Louis Cauchy [CAU 76]
shown in Figure 12.1.

Figure 12.1. Cauchy projection (Figure 5.5 in [BIG 76])

 The Representation of Graphs: A Specific Domain of Graph Theory 247

Poinsot will develop work on polyhedra but, most of all, it is Lhuilier and
Headwood who will develop projections on a sphere and a torus (see Figure 12.2)
[HEA 86, LHU 86, POI 86], which is particularly interesting for the construction of
a dual.

Figure 12.2. Headwood projection (Figures 5.8 and 7.6 in [BIG 76])

It thus clearly appears that a graph can be drawing on a plane which can be
considered as a sphere of infinite radius. Therefore, on a sphere whose radius is not
infinite and on a transformed torus that is topological of the sphere, the draw must
specify it by taking into account the consequences that it may have. This is the
particular case for the definition and the drawing of the dual, which presupposes the
existence of an infinite face, which is obviously not always the case.

The implicit hypothesis of Berge therefore cannot be sufficient and it is not only
important to know how the nodes are connected, but also on which type of surface
the graphic representations of the graphs are drawing.

Let us consider the simplest case, i.e. that of Cauchy, where it is obvious that the
representation suggested is also a particular case: a projection onto a plane parallel
to one of the faces of the cube, whose projection center is located on the median of
the face considered.

Many other projections of a cube onto a plane are conceivable such as, for
example, the particular one where the projection center is located on the extension of
a diagonal of the cube and of course without counting all the intermediate situations
which generally produce regular forms. Thus, the number of possibilities grows
considerably without necessarily modifying the number of nodes and arcs, and all
may be the projection onto a plane of the same graph with three, four, or N
dimensions.

It is obvious that if the representation of the graph is automated and repetitive, it

is necessarily so by means of an algorithm, which must be accessible in its principle
248 Graphs and Networks

and its source code, lest the whole programming of the software used should be
questioned. However, the current algorithmic [BAN 01, COR 94] and discrete
geometry publications do not deal with the problem, so to speak.

The problem that consists of supplementing the traditional methods of producing

graphs so as to make them reproducible and therefore verifiable is, thus, relatively
obvious. This is essentially a problem of rendering objective a process whose
algorithms probably already exist in several software, the access to the source code
of which is unfortunately not yet free.

On the other hand, the initial problem of whether we can describe fractals in graph
form and create their graphic reproductions with the same properties of self-similarity
and internal homothety as the simple Mandelbrot fractals is much richer.

12.2.1. Mandelbrot’s graphs and fractals

Benoît Mandelbrot himself described in his work [MAN 77, p. 46 et seq.] the
generators of his fractals as linear graphs, without, however, naming or
individualizing the points that he very clearly makes appear at the ends of the
segments of straights as shown in Figure 12.3, which have been selected among
many examples.

N=3 N=8
1/r=√7 r=1/b=¼
D=log 3 / log (√7) ∼ D=3/2
1.1291

Figure 12.3. Mandelbrot’s fractal generators where N is the number of arcs,

R is the increase rate and D the fractal dimension [MAN 77]

We may consider without any difficulty that the generator is a simple graph,
whose drawing, however, has particular characteristics, such as flatness, the number
of elementary segments, their sizes and their successive geometrical positions (angle
to the previous, area of space, etc.).

1 2
1 1

2 1 1 2
.....................................................................................

Figure 12.4. A first stage graph (here a simple arc) and

the adjacency matrix associated with the graph
 The Representation of Graphs: A Specific Domain of Graph Theory 249

We can then individualize the nodes by naming them and this generator could
thus be implemented on multiple scales by studying the size of the elementary
segment.

Let us consider the first graph (Figure 12.4): we have the arc x, y: 1,2 which is
defined by the adjacency matrix. The graph drawing is member of a plane, the arc
has a length = l and it is horizontal.

Let us consider the adjacency matrix of the second graph (see Figure 12.5): we
proceeded to subdivide the graph which meant creating two intermediate nodes on
the arc of the previous graph, since the length of the new arcs is equal to 1/3 of the
previous arc, that is r = l/3.

Figure 12.5. Second and third stage: extension of the graph and replacement
of the central arc by two arcs of the same length in the positive part
of the plane and associated adjacency matrices

It has to be said that in the case of an extension of the graph by intrapolation, the
created nodes 11 and 12 must be inserted between the previous nodes. If this simple
constraint is followed, we note that the second matrix is self-similar to the previous
one and that, if we divide this matrix into five 2*2 submatrices, the three diagonal
sectors are identical to the first level matrix.

We will note that the graph drawing must follow the order of the adjacency
matrix.

Let us replace the central arc by two arcs of the same length, that is l/3, as
illustrated in the third diagram. We have created a new node 13, which is inserted
between nodes 11 and 12 of the previous matrix (Figure 12.6). The new matrix still
possesses the property of self-similarity.
250 Graphs and Networks

In order for the drawing to be reproducible, we specify the location of the new
node with respect to the previous segment which must be situated in the positive (or
negative) part of the plane.

Figure 12.6. First iteration of the graph of the von Koch curve

We have the first level of the von Koch curve and therefore we only have to
repeat the same sequence of operations for each arc to obtain the second level of the
von Koch curve, which is illustrated by the following diagram (Figure 12.7).

Figure 12.7. Second iteration of the graph of the von Koch curve
and self-similarity of the adjacency matrix

We can conclude from this first example that:

– the von Koch curve can be described in graph form;
– as long as the new nodes are inserted between the previous ones, the adjacency
matrix possesses the property of internal homothety and develops identically to
itself, since each indicated submatrix reproduces the adjacency matrices of the
previous level;
 The Representation of Graphs: A Specific Domain of Graph Theory 251

– the graph drawing must obey specific constraints in order to reproduce the
fractal in its traditional form:
- the drawing must be plane,
- the arcs are segments of a straight line,
- the length of the inserted arc has a ratio r with the original arc,
- the middle arc is replaced by two arcs of the same length,
- the decimal notation of the nodes created on an arc adds a figure to the
notation of the node on the left. In the case of replacing the initial arc by two arcs of
a length identical to that of the original arc and the creation of a node of the second
degree (subdivision), the notation is on the same level as that of the two nodes of the
replaced arc. This notation makes it possible to know the level of each element and
the length of each arc considered.

12.2.2. Graph and a tree-structured fractal: Mandelbrot’s H-fractal

The von Koch curve is a typical example of graph extension by intrapolation.

Rightly, the question arises about the possibility of graph extension by extrapolation.
We will tackle this question by using Mandelbrot’s H-fractal (Figure 12.8) [MAN
77, p.164].

Figure 12.8. Mandelbrot’s H-Fractal [MAN 77]

Let us consider Figure 12.9 starting from node O of the plane, where we
construct two co-linear arcs on both sides of this node. The adjacency matrix is
obvious: a symmetric (3,3) matrix.

We perform the extension of the initial graph by adding an orthogonal arc

(segment of a straight) at each of its ends, whose length is half of the previous one
252 Graphs and Networks

and whose number is that of the original node increased by a figure or a number
(classification of position).

The newly created nodes are listed in order but here, following the nodes of the
previous level, it is an extension by extrapolation.

As previously, it suffices to restart the procedure, i.e. the number of times needed
to develop the fractal graph and its adjacency matrix. Figure 12.9 shows us the result
of four successive iterations, which produce a fractal graph with 31 nodes on the
basis of the first graph with three nodes and two arcs. The adjacency matrix is self-
similar and the drawing under the stated conditions is identical to Mandelbrot’s H-
fractal.

0 1 2 11 12 21 22 111 112 121 122 211 212 221 222

0 1 1

1 1 1 1

2 1 1 1

11 1 1 1

12 1 1 1

21 1 1 1

22 1 1 1

111 1

112 1

121 1

122 1

211 1

212 1

221 1

222 1

Figure 12.9. The first four stages of the process of constructing the fractal

Let us note, however, that if we do not observe the condition of orthogonality of

the two new arcs and replace it by co-linearity for one and orthogonality for the
other by reversing the conditions according to the branch, we obtain a perfectly
 The Representation of Graphs: A Specific Domain of Graph Theory 253

regular tree structure which naturally is also a particular graph: a binary tree or a
complete m-ary tree with m = 2 in this case (Figure 12.10).

This example shows the obvious need for a precise and complete algorithmic
definition of the implementation in order for it to be reproducible.

Figure 12.10. The fractal is identical to the previous one; only the implementation
is different in one of its constraints: the orthogonality of the branches issued
from the same node instead of co-linearity

Figure 12.11. Mandelbrot’s fractal: graphs that are isomorphic to the two previous figures.
The adjacency matrix is the same and only changes the geometrical
constraints of implementation [MAN 77]

The graph drawing must follow specific constraints in order to reproduce the
fractal in its traditional form:
– the drawing must be flat;
– the arcs are segments of a straight line;
– the length of the extrapolated arcs has a ratio r with the original arc;
254 Graphs and Networks

– the arc is extended by two arcs of the same length, which form a given angle
with the previous arc;
– the decimal notation of the nodes created on an arc adds a figure or a number
to the notation of the original node.

This notation makes it possible to know the level of each element and the length
of each arc considered. If the arc is of level n and the initial arc has a length l, then
the rate of increase r is:

Ln = l1*rn-1

These conditions are valid only for a symmetric drawing, but it is possible to
have different ratios r for each branch, as well as unequal angles, as demonstrated by
the representation of a Pythagoras tree below.

12.2.3. The Pythagoras tree

The example of the Pythagoras tree [LAU 91] enables us to consider a more
complete use of the graphs by taking into account duality. The method of
construction of the tree is based on the first demonstration of the famous theorem.
We associate a square with each side of a right-angled triangle, which is placed with
the hypotenuse oriented downwards and we continue in the same way by assigning
to the opposite sides of the smallest squares a right-angled triangle, whose
hypotenuse corresponds to the length of the square considered and so on (Figure
12.12). In this figure overlaps are acceptable.

We then obtain the symmetric or asymmetric Pythagoras tree according to the

choice of the right-angled triangle; a choice that we may consider fixed at first.

Let us consider the quasi-dual1 of this figure, which we can describe like a
graph. The quasi-dual is defined by the centers of each figure or unit face, which
here are squares or isosceles right-angled triangles and the arcs joining them.

1 The quasi-dual is a dual graph whose arcs connecting the node located on the possible
infinite face are not taken into account (see section 12.3.1).
 The Representation of Graphs: A Specific Domain of Graph Theory 255

Figure 12.12. Pythagoras tree: construction of the first iterations [LAU 91]

We thus obtain Figure 12.13, which is a form of H-fractal obtained on the basis
of the same adjacency matrix but whose new arcs form an angle of ± π/4 with the
arc of the previous level.

The two manifestations of the Pythagoras tree are dual and the drawing is much
simpler and an esthetic by using the quasi-dual that corresponds to the “dual graph
of the map” defined by Rosen [ROS 91, p. 488].

The Pythagoras tree can just as easily be produced in an asymmetric fashion on

the basis of the same geometrical construction by removing the condition that the
right-angled triangle is isosceles and by fixing a value, or even by admitting a
random variable, either on the basis of a description by an adjacency matrix angle
and different or randomly variable angles.

Figure 12.13. The dual Pythagoras tree and a non-symmetric tree [LAU 91]
256 Graphs and Networks

12.2.4. An example of multiplane plotting

The previous adjacency matrix enabled us to show that the number of possible
graphs drawing is not limited, even if all are plane. However, this characteristic may
well be disputed and even cancelled. We can, for example, consider the H-fractal
whose two branches could be implemented on two orthogonal planes as shown in
Figure 12.14.

It suffices to state as a replacement for condition 1 that: the representation is

biplanar, branch 1… belongs to plane P1, branch 2… belongs to plane P2
orthogonal to P1, since the two planes have an intersection following the straight
supporting arc 12.

We see here a biplanar fractal. We can just as easily consider that each new arc is
orthogonal to the plane of the two arcs of the previous levels. We would then have a
multiplane fractal.

Figure 12.14. Biplanar fractal with orthogonal branches

12.2.5. The example of the Sierpinski carpet and its use in Christaller’s theory

Let us consider a ternary tree [LAU 91, p. 11-12]. On the basis of a node we
construct the graph extension by extrapolation, with each branch forming an angle
of 2π/3 with the following one with a ratio of r (Figure 12.15).

This tree enables us to define a quasi-dual, such that each of the four nodes of the
primal is the center of a dual face formed by an equilateral triangle, whose four
equilateral triangles also form an equilateral triangle. This figure itself gives us its
reduction ratio: each face of the dual is equal to ¼ of the face of the previous level
and each arc will be equal to the 1/x of the preceding arc.
 The Representation of Graphs: A Specific Domain of Graph Theory 257

Let us add a constraint to this drawing: we regard as full only a face with a
terminal node, i.e. a leaf.

By analogy, the face of the unique node will be regarded as also being an
equilateral triangle of identical size to that determined by four of the immediately
lower level. Since this graph has only one node, this will be ipso facto final and the
triangle full.

At the immediately lower level, only the faces associated with the three final
nodes will be considered as full and the central face will be empty.

During the following phase where the process is reproduced, each node of the
primal leads to a ternary extension and each full triangular face is divided into four
triangular faces, of which one is empty, one is the central face, etc. (Figure 12.16).

Figure 12.15. Ternary tree as a generator

Figure 12.16. The Sierpinski carpet: ternary tree third iteration and the quasi-dual: the
Sierpinski screen or carpet; full surfaces correspond to leaves of the tree [LAU 91]
258 Graphs and Networks

We thus obtain a Sierpinski carpet on the basis of a graph and his extension by
extrapolation (Figure 12.16).

Figure 12.17. Christaller’s hexagons obtained on the basis of a ternary generator

If the condition that only the final nodes correspond to full triangles is removed,
we obtain Christaller’s hexagons, which in the case (Figure 12.17) are of the market
principle K = 3. The hexagons are the dual of the dual of the ternary trees when the
space is tiled.

12.2.6. Development of networks and fractals in extension

Until now the fractals considered were convergent fractals due to a reduction
ratio less than 1. In this case these fractals have a higher limit and the internal
homothety applies only from an origin. If, on the other hand, we consider a
magnification ratio greater than 1 (Figure 12.18), we then have a tool to account for
the extension of networks, the growth of cities, etc. and we return to the percolation-
type phenomena.

Figures 12.18 and 12.19. Fractal in extension (Frankhauser [FRA 94, p. 257])
 The Representation of Graphs: A Specific Domain of Graph Theory 259

12.2.7. Grid of networks: borderline case between extension and reduction

The growth of networks is not achieved solely by extension and creation of new
external arcs, it is also carried out, in particular when the density increases, by
applying an increasingly fine grid. The problem of the fractal grid is in fact simple:
it is the borderline case between a tree-structured extension and a convergence
towards a limit. If we consider a reduction factor r ≤ 1 and a magnification factor
a ≥ 1, the borderline case, obviously, has the value 1, at which the fractal reproduces
identically to itself. The problem of the increasingly fine grid (possibly until a limit)
is solved by a reduction coefficient of the generator size with a possible delay. The
example of the H-fractal makes it possible to show it.

In this example we use Mandelbrot’s H-fractal whose reduction factor takes the
limit value 1. The first fractal has a generator size fixed at two. The second generator
with the same reduction factor has a size fixed at 1, that is, twice smaller. The
second process develops with a delay that is arbitrarily fixed here at 2 iterations.
When the first process arrives at the seventh iteration, the second process is only at
the fifth (Figure 12.20).

However, since this fractal is twice smaller in size, it only concerns the center of
the previous figure. We can thus interpret a first wave of urban allotment in a city
and then after two periods of time x a second denser grid develops giving, for
example, access to the center of the previous pieces of land in order to enable their
constructibility.

11111111

1111111 111111
1111112

11111112 11112
11111 1111 1121 2121

111 11 112 211 21 212

11121 1112 1122 2112 2122

1211 1 1221 2211 2 2221

121 12 122 221 22 222

12121 1212 12122 12211 1221 2212 22122 22221 2222

Figure 12.20. Example of a grid with delay by using a borderline reduction

factor =1 and a delay in the reduction of the generator size
260 Graphs and Networks

We could just as easily reverse the process: the first dense urbanization followed
by the second peripheral one with a less tight network, etc. However, this type of
fractal by construction has double points (mentioned in the figure above), which
should be eliminated from the adjacency matrix. Similarly, when two arcs are
superimposed over a single precedent, the latter two should be kept in the adjacency
matrix, since the intermediate node is identical to a subdivision of the previous
graph. This solution is preferable due to a finer definition of the network than
described by the graph, even though in terms of the incidence matrix this choice is
heavier, since it implies an additional row and a column. We could also use a
multigraph by making an expressway with fewer crossings coexist superimposed
with a normal road.

12.2.8. Application examples of fractals to transport networks

The example is very clear here: the networks of trunk and departmental roads
are, broadly speaking, saturated planar graphs, with a triangular mesh and without
possibility of adding an arc without it cutting through another. The internal
homothety is thus perfectly real, only the position of the nodes changes and overall
that implies a quasi-random variation. The succession on various levels of planar
graphs (department, region, France and Europe) thus constitutes a true fractal
structure, since not only are the graphs of comparable nature, but also the main cities
are the same on four levels. On the other hand, when we go down a level and the
scale increases, there appear new cities and graph nodes, as well as new arcs.

We have the case of an extension of the graph by intrapolation of the von Koch
type for a much less linear road diagram and of the double triangular grid type.

When we descend to sufficiently fine levels, we are obliged to take into account
the natural geophysical constraints, which are imposed on the network
infrastructure. The last two levels are disturbed, for example in the case of the Indre-
et-Loire department, by the presence of the Loire and the lack of bridges which on
this scale renders the planar network unsaturated.

In the agglomeration of Tours and especially in Tours center, the old rhomboidal
grid imposes itself and developed from the old town in urbanized areas starting from
the second half of the 19th century. Hence, this result in the unsaturated nature of the
graph and an illustration of the grid as a borderline situation in a dense space (Figure
12.21 in the color plates section). We may stress that we have a perfect illustration
of several types of grid in this zone, which corresponds exactly to the fractal grid
with the use of a borderline reduction factor and a delay for the grid with the largest
generator, as in the second possibility expressed above.
 The Representation of Graphs: A Specific Domain of Graph Theory 261

The second example drawn from the same hypothesis is also clear: we have an
irregular tree structure, which is apparently close to a tree structure of the stochastic
type linked to the definition of the shortest paths from Perpignan to the rest of
France (Figure 12.22 in the color plates section).

The irregularity of the tree structures results from geographical obstacles: Massif
Central, Rhone Valley, etc. In the case of this simulation, we only had matrix indices
at our disposal to compare the two solutions (number of nodes connected by the tree
structure, lengths of the paths, etc.), but no morphological indices.

The obvious fractal nature of this tree structure then makes it possible to use the
calculation of the fractal dimension and spread [GEN 00] of each branch.

12.3. Nodal graph

Graph theory in its current state does not enable a sufficiently precise description
of networks, in particular with respect to the nodes, vertices of the graph.

Indeed, the nodes are “elements” which are connected by relationships. In the
description of networks by graphs, the nodes represent places connected by
circulation network elements: roads, railroads, canals and, more generally, elements
of the natural or artificial infrastructure which enable the circulation of one or
several means of transport. It is what Chapelon refers to as the functional binomial
[CHA 97] by association of the infrastructure and means of transport.

Conversely, we may say that the node is what makes it possible to “connect”
several relations by describing the various interrelationships between the elements
and thus their independence, which is illustrated very precisely by the transfer or
fluence graphs [CHO 64, p. 39]. With respect to the description of networks, in the
same inversed manner, a node is what makes it possible to pass from an arc to
another, i.e. from one infrastructure element to another. Here the node is an
interconnection, an interface, or even a correspondence for the same or for different
means of transport.

However, in graph theory nothing defines this passage: we can implicitly pass
from an arc to any other incident arc to the node without any constraints apart from
those that are explicitly specified or those that can be easily deduced from the
definition of the characteristics of the nodes and/or the arcs, if they exist. These
cases are simple and the most important ones are determined by considerations, arc
capacities, optimal paths, etc.
262 Graphs and Networks

If the graph is directed, that imposes a constraint, that is, the existence of arcs
leaving so that the considered node, will not only be a dead end. However, one of
the conditions of existence of a circulation network is precisely strongly connected,
i.e. the possibility of going to any point and returning from there. However, certain
crossroads cannot be described by a node of an undirected graph: it is in particular
the case of the following semi-interchanges (Figure 12.23).

Indeed in this case, according to the direction of movement on the highway arc,
we cannot either leave or return on the highway arc from the crossroads considered,
which otherwise enables all the other possibilities.

It is necessary to descend to a large scale description in order to be able to

describe it. We must be able to pass automatically from one scale to another by
using an algorithm, so that the passage may be represented according to the needs.

Figure 12.23. Semi-interchanges

Moreover, the problem of maximum capacity in a graph describing a physical

network does not appear related to the maximum capacity arcs but more precisely to
that the nodes. Indeed, let us consider the simplest case: traveling along two axes
against the flow of traffic. If the maximum physical capacity of a car lane is 2,000
vehicles per hour, and we suppose that our arcs have only one lane and that the level
of saturation has almost been reached, it is clear that the problem will immediately
revert to a problem of traffic flow and that it is impossible to make the hypothesis of
a realistic flow capacity of 4,000 v/h, which would be already double the allowed
maximum for a lane on an arc with fluid flow. In the case considered, the capacity of
the crossroad – maximum theoretical capacity – would be, at the most, equal to half
the sum of two flows if the overlap was carried out without deceleration, which is
still an unrealistic hypothesis.

Such reasoning can be generalized to less simple crossroads than the chosen
example and, then, in order to determine a maximum theoretical capacity, it
becomes necessary to break up the flows into divergent and convergent flows.
Therefore, we will attempt to demonstrate that it is possible to mitigate this present
 The Representation of Graphs: A Specific Domain of Graph Theory 263

deficiency of graph theory, while remaining within the theoretical framework

defined beforehand, by using graph extension. This is referred as nodal zoom by
Chapelon, although here it would not be used to change the scale of representation,
i.e. to describe a city previously represented by just one node on an interurban
relation scale, but simply to study with precision how a specific node of the graph
functions or can function. This level of analysis is known as nodal.

Let us consider a simple node S of an undirected graph of the fourth degree

(Figure 12.24). We will define a subdivision of this graph that associates with it four
nodes of the second degree near to the considered node, i.e., S1, S2, S3 and S4
respectively. It is possible to go from S1 to S2, S3, S4, from S2 to S1, S3, S4, from
S3 to S1, S2, S4 and from S4 to S1, S2, S3 passing by S since these nodes are 2-
connected. These possibilities are described by the adjacency matrix of the second
order (Figure 12.24).

S S1 S2 S3 S4 S S1 S2 S3 S4

S 1 1 1 1 S

S2 S1 1 S1 1 1 1
S S2 1 S2 1 1 1
S3 S1
S4 S3 1 S3 1 1 1
S4 1 S4 1 1 1

Figure 12.24. Nodes and adjacency matrices of the first and second order

We reduce the graph by removing the intermediate node S and in order to ensure
the same connectivity we extend it by creating the arcs S1S2, S1S3, S1S4, S2S3,
S3S4 and S2S4. We thus transform the 2-connected relations into 1-related relations
by reduction of the previous graph. Thus, we obtain a nodal graph consisting of four
nodes of the fourth degree which enable the same passages from one arc to another
as in the original graph (Figure 12.25). The nodal adjacency matrix of the first order
has the same relations as the preceding adjacency matrix of the second order, but its
size has decreased by an empty row and a column.

It should be noted that this operation is merely the opposite of simplification of

flow graphs which were first used in France by Ponsard [PON 69] and then by
Lantner [LAN 74] in Dijon. It partially corresponds to the derivation of a
transformation by the method of nodes duplication [CHO 64, p. 39].
264 Graphs and Networks

S2 S1 S2 S3 S4
S1 S1 1 1 1
S3 S2 1 1 1
S4
S3 1 1 1
S4 1 1 1

Figure 12.25. Node reduced by the deletion of S and

the resulting adjacency matrix of the first order

In this case the nodal graph is a complete subgraph. If we apply this technique to
the previous case of the semi-interchange, we note that the S1 node has only one
connection to S2: S2S1. The same happens to S6, whose only connection to S5 is
S6S5. The nodal graph created in this way proves capable of describing with
precision the various possibilities and impossibilities of passing from one arc to
another. We note that the adjacency matrix cannot be made triangular: there exists a
circuit S5S4S3S2S5, as shown in the figure below.

S1 S2 S3 S4
S1 1

S2 1

S3 1

S4 1

Figure 12.26. Directed nodal graph and matrix

 The Representation of Graphs: A Specific Domain of Graph Theory 265

1 2 3 4 5 7 9 11 6 8 10 12
1 1
2 1
7 6
3 1

8 1 5 4 1
2
5 1 1
7 1 1
3 12 9 1 1

10 11 11 1 1
6
8
10
12

Figure 12.27. Extended G3 nodal graph and adjacency matrix with the nodes according to
the functional order: first the crossings then the divergences and convergences for each arc

Let us consider the same example but with a directed graph with four internal
semi-degrees and four external semi-degrees G1. The solution that consists of
breaking up this G1 graph into a subgraph with four nodes, like the G2 graph
(Figure 12.26) with two internal semi-degrees and two external semi-degrees, is not
satisfactory in terms of describing the flows and it certainly does not make it
possible to describe the semi-interchange considered above.

Nodal decomposition can be achieved in different ways according to at least

three solutions. The second solution is a variation of the first variation mentioned
above. The same method is thus reused: extending the graph by creating a
subdivision and a list of possibilities of passing from one to the other, i.e.
determination of the connectedness of the second order and a list of adjacency of the
second order (see Figure 12.28).
266 Graphs and Networks

1 2 3 4 5 7 9 11 6 8 10 12
1 1 1
2 1 1

8 3 1 1
4 1 1
10 3 7
5 1 1
2 7 1 1
11 1 6 9 1 1

5 11 1 1
12
6
8
10
12

Figure 12.28. G4 graph of a complete simple crossing and adjacency matrix: second solution

However, this second solution has a disadvantage that can limit its use: it
modifies the nature of the graph. Indeed, while remaining plane the graph is no
longer planar, since certain arcs intersect outside a node. There is an intersection in
the shape of an interchange, which transforms the planar graph into GRS due to the
absence of bijection between the graph itself and its projection onto a plane.

The third is a little bulkier but has the advantage of making the nodes
specialized. For a better analysis of the flows we introduce an additional constraint:
we consider only convergent or divergent nodes but not mixed nodes. We will call a
node divergent if from it a flow can split into two, and convergent if two separate
flows meet at this node. We will draw the graph as shown in Figure 12.29. However,
it should be mentioned that this graph is no longer planar. As before, its projection
onto the plane is only a projection of a GRS and there is not bijection between the
two. Using this presentation of the nodal graph, it is easy to describe any type of
crossroads.
 The Representation of Graphs: A Specific Domain of Graph Theory 267

D1 D2 D3 D4 D5 D6 D7 D8 C1 C2 C3 C4 C5 C6 C7 C8
D1 C4 D1 1 1
D5 C8 D4 D2 1 1
D3 1 1
C1
C5 D8 D4 1 1
D6 D5 1 1
D7 C7 D6 1 1
D7 1 1
D2 C6 C3
D8 1 1
C2 D3
C1
C2
Black nodes are divergent
C3
White nodes are convergent
C4
C5 1
C6 1
C7 1
C8 1

Figure 12.29. G5 graph with eight divergent nodes and eight convergent nodes

1 2 3 4 5 6 7 8 9 10 11 12 13 14

1 1 1

2 1 1
14 13
3 1

4 1 1
5 4 5 1 1
7 6 12
6 1

8 1 7
11
3
2 8 1

9
9 10
10 1

11

12 1

13

14 1

Figure 12.30. Nodal graph of the semi-interchange

268 Graphs and Networks

Generalization of the nodal decomposition

The nodal decomposition with divergent and convergent nodes is automatic and
can be carried out on a part of the graph or on its entirety at our discretion.

Figure 12.31. Any undirected graph G

Let us consider the graph G above (Figure 12.31) and the adjacency matrix
below.

1 2 3 4 5 6 7
1 1 1 1 1
2 1 1 1
3 1 1 1 1
4 1 1 1 1 1
5 1 1 1
6 1 1 1 1
7 1 1 1

For each node we can use the zoom to reveal the nodal graph by using the
technique described above, with specialized or unspecialized nodes (Figure 12.32).

The graph then appears in its extended version, but its properties are identical to
the previous one in terms of path layout. If the adjacency between two arcs was
achieved via a node in the first graph, the adjacency in the second graph is achieved
via a new arc and two nodes and the 1-adjacency has been transformed into 2-
adjacency. We must specify once again that the projection of the graph is not planar.
 The Representation of Graphs: A Specific Domain of Graph Theory 269

Figure 12.32. Graph with GDS nodal decomposition of the undirected graph G

Property: each node created to replace a primitive node in an undirected graph

has the same degree and so forth: this property is preserved by transformation.

The nodal matrix has the property of self-similarity with respect to the previous
matrix if we consider the sectors identified by the first figure of the node of the
nodal graph identical to that of the node which it replaces. On the other hand, the
diagonal of the previous matrix is replaced by the nodal matrix of each node.

The transformation of a node could be described by a complete m-ary tree,

where the number of wires is given by the degree of each root node.

The “generator” here is thus the transformation of the node into a nodal graph.
Mathematically, this generator could be used successively several times. Both
graphs may therefore be considered as fractals. The size of the diagonal submatrix,
which replaces the node of the simple graph, is the degree of the node in the case of
an undirected graph and the sum of external and internal semi-degrees in the case of
a digraph. The nodal transformation is therefore fractal.

The graph then appears in its extended version, but its properties are identical to
the previous one in terms of path layout. If only the adjacency between two arcs was
achieved via a node in the first graph, in the second graph the adjacency was
achieved via a new arc and two nodes. Moreover, an algorithm can easily be written
for this transformation.
270 Graphs and Networks

11 12 13 14 21 22 23 31 32 33 34 41 42 43 44 45 51 52 53 61 62 63 64 71 72 73
11 1 1 1 1
12 1 1 1 1
13 1 1 1 1
14 1 1 1 1
21 1 1 1
22 1 1 1
23 1 1 1
31 1 1 1 1
32 1 1 1 1
33 1 1 1 1
34 1 1 1 1
41 1 1 1 1 1
42 1 1 1 1 1
43 1 1 1 1 1
44 1 1 1 1 1
45 1 1 1 1 1
51 1 1 1
52 1 1 1
53 1 1 1
61 1 1 1 1
62 1 1 1 1
63 1 1 1 1
64 1 1 1 1
71 1 1 1
72 1 1 1
73 1 1 1

12.3.1. Planarity and duality

The property of duality of graphs is very interesting, but polysemous. Applied to

a particular draw, it allows us to make a network correspond to the faces that it
generates, which is fundamental and absolutely necessary for a space analyst.

Among various accepted variations we will specify the meaning of duality that
we employ, which is a sufficiently general definition for our purposes and
operational, so that the creation of a dual could be algorithmic. The objective of
 The Representation of Graphs: A Specific Domain of Graph Theory 271

taking duality into account is to become able to transform a graph into a cellular
network.

A graph is referred to as plane when it is drawn on a simple surface: a plane or a

sphere or a torus, as defined by Lhuillier [LHU 76], Heawood [HEA 76]. Let us
recall that a plane graph is known as planar, if none of its arcs are intersected outside
the nodes and that a planar graph is known as saturated or complete when it is
impossible to add an arc to it without it intersecting another arc outside the node.

All the faces of a saturated planar graph are triangular. Indeed, if one of its faces
was quadrangular, it would be possible to add to it a new diagonal arc and thus the
planar graph would not be planar saturated.

The problem of projection onto a plane of a three-dimensional graph. Let us

return to the historical example of Cauchy [CAU 76]. Each node is projected onto
one and only one point, and similarly each arc has one and only one projection.
However, it is not the same for the faces of a cube; indeed, the projection of the
lower face of a cube partially covers the projection of the upper face.

Let us consider point M belonging to ABCD, which is the upper face of the cube.
It is projected onto the plane in M’, just as point N located on the bottom face EFGH
is projected onto N’, points M, N, M’ are aligned, N’ merges with M’ and the point
is double. The same applies to any point of the cube projection where even the stops
of the projection squares contain double points, since they belong simultaneously to
two faces.

If we do not consider the cube formed only by nodes and arcs, the projection
does not contain a double point unless a particular choice has been made by the
projection center. If it is located on the extension of a diagonal there will be a double
node, if it is located on a diagonal plane passing through two sides, these will be
superimposed in the projection and, more generally, if it is located on a plane
passing through two sides, these may have a common projection. On the other hand,
if we take into account the faces of the cube, all the points of the projection are
doubled.

However, the case of the cube of Cauchy is not unique and many graph plottings
can be regarded as plane projections of three-dimensional graphs or more.
272 Graphs and Networks

B C
M
A
D
F G
N
E
H

Figure 12.33. Cauchy projection in [BIG 76]

The condition of existence of a 3D volume is four adjacent nodes of degree

d° ≥ 3. Let us consider Plato’s solids below (Figure 12.34) [ALD 00]: we note that
the simplest of them is the tetrahedron or pyramid with a triangular base whose
nodes are of the third degree, as those of the cube and the dodecahedron. On the
other hand, the octahedron and the icosahedron have nodes of the 4th and 5th degree
respectively.

tetrahedron octahedron cube icosahedron dodecahedron

Figure 12.34. Plato’s solids, three-dimensional graphs [ALD 00]

We can deduce from this that any graph with at least four adjacent nodes four by
four and thus of at least the third degree would contain a tetrahedron. We can also
deduce that irregular polyhedrons have similar properties.
 The Representation of Graphs: A Specific Domain of Graph Theory 273

tetrahedron octahedron cube icosahedron dodecahedron

Figure 12.35. Planar graphs of Plato’s solids [ALD 00]

Nothing forces us to only consider a three-dimensional space. Let us consider the

hypercube below (Figure 12.36).

1110 1111

1100 1101 1010 1011

0110 0111

1000 1001

0100 0101 0010 0011

0000 0001

Figure 12.36. Projection of a hypercube onto a plane

This graph shows us that all the nodes are of the 4th degree. Its projection onto a
plane is possible. In particular cases we may obtain a projection of the hypercube
which repeats Cauchy’s projection. The projection of a hypercube with a central
projection source yields two overlapping cubes connected by edges, which is
incidentally the plan of the Grande Arche in La Defense, Paris (Figure 12.37)
[DOL 01].
274 Graphs and Networks

Figure 12.37. Central projection of a hypercube in a

three-dimensional space – (Science Avenir)

The 3D graph thus obtained yields a plane graph with four overlapping squares
in a straight projection. If we have a volume, at least one face will be at least a
double face, even if the nodes and arcs taken separately the framework of the figure
are not. This is apparent for all of Plato’s solids and for their transforms; indeed, in
order for the projection not to have double points, it is necessary that the projection
center is internal to the volume considered.

12.3.1.1. Duality
The duality of the graphs is polysemous. For example, Beauquier, Berstel and
Chrétienne [BEA 92, p. 273 et seq.] use the concept of duality within the framework
of linear programs related to the determination of the minimum cost flow defining a
dual linear program… Band-Jansen and Gutin [BAN 01, p.150 and 667] use it in the
same sense and also in that of a dual of a matroid. Xuong uses it to define a
transposed binary relation and also within the framework “of what is referred to as
the principle of duality of directed graphs which is obtained by reversing the
direction of the arcs… [XUO 92, p. 69 and 329]”.

On the other hand, many other authors, for example, Cormen, Leiserson and
Rivest [COR 94], as well as Minoux and Bartnik [MIN 86] do not even mention it in
works on algorithms. Further on we will use and discuss a definition of duality in
line with that of Withney [WIT 32], Berge [BER 70, p. 20], Kaufmann [KAU 64,
volume 2 p. 358], Balakrishnan and Ranganathan [BAL 00].

Although Euler used a concept very close to duality, if not just a brilliant
anticipation thereof, for the bridges of Königsberg, Withney seems to be the first to
explicitly define the dual graph, which he calls the Geometric dual of G. However,
the idea goes back to Kemp in 1879 and Heffter in 1891, and, it seems, to Maxwell
even earlier…
 The Representation of Graphs: A Specific Domain of Graph Theory 275

The duality of a graph consists of associating the faces of the graph called primal
with the nodes of a graph called dual. Berge [BER 87] [BER 70] provides the
following definition for it. “Let us consider a planar graph G, connected and without
isolated vertex; we will make it correspond to a planar graph G* in the following
manner: inside every face s of G we places a vertex x* of G* and to every edge e of
G, we will make correspond an edge e* of G*. This will connect the z vertices x*
and y* corresponding to the faces s and t which are located on both sides of the edge
e. The graph G* thus defined is planar, connected and does not have isolated nodes:
it is called the dual graph of G”.

Figure 12.38. In Berge [BER 70]

The example of construction of the dual provided here by Berge (Figure 12.38)
is a particular case in the sense that the dual has the same nature as the primal: a
1-graph. The diagrams below (Figure 12.39) seems to make it possible to show a
property of the dual which has not yet been underlined.

*
F1

* * *
*
E1 E2 E3 E4
*
E5
V1 V2

*
F5 *
E6 F4
*

* V3
E9 *
F2
*
E8
*
F3
V6
*
E7

V4
V5

Withney Kaufmann Balakrishnan and Ranganathan

Figure 12.39. Property of the dual

276 Graphs and Networks

The Withney diagram [WIT 32, p. 148-149] is very simple and at the same time
paradoxical, as we will see below. The Kaufmann example [KAU 64, volume 2,
p. 358] is much more interesting because he shows that the dual in this case is a
multigraph, but does not mention it [KAU 64, volume 2, p. 357-358]. The
construction of the dual by Balakrishnan and Ranganathan [BAL 00, p. 165]
confirms Kaufmann’s.

The two latter examples enable us to deduce the following theorem:

– the dual of a 1-graph is a multigraph if and only if the primal has at least one
node of the second degree.

Indeed, by construction, the nodes of the dual belonging to faces which are
located on both sides of the two arcs connected by a node of the second degree will
be connected by at least two arcs. This is the case of the last two graphs. It is in this
sense that the example of construction provided by Berge is just a particular case,
like that of Withney, which is a particular case of a graph whose connected faces are
only separated by single arcs.

This enables us to specify that if the subdivision of the graph (by insertion of
second degree nodes) does not affect the planarity or non-planarity of the graph
[ALD 00, p. 260], on the other hand, the nature of the dual is changed depending on
whether this insertion is carried out between a peripheral face and the infinite face or
between any two adjacent faces.

This particularity poses a problem of algorithmic definition of the dual, since if it

is carried out by using an adjacency matrix, for example, the latter must be at least
three-dimensional and, thus, possibly very concave. In the case of a definition by
pointer, the case must be envisaged in such a way that the first value is not crushed
by the second, thus making a calculation for a line of optimization of the shortest
path or maximum flow type impossible. The algorithmic description of networks
and faces conceals another difficulty: the description of the infinite face of a plane
graph.

12.3.1.2. A return to the origins

This return to the origins is justified by an ambiguity in the definition of the
planar: the problem of the possible infinite face. Let us consider the famous problem
of Königsberg’s bridges (Figure 12.40). The population of this town in Eastern
Prussia found it amusing to try and find a path that would make it possible to cross
each bridge of the city by using every bridge just once. Some of them believed that
this was impossible but no proof of this impossibility had been given.
 The Representation of Graphs: A Specific Domain of Graph Theory 277

Figure 12.40. The bridges of Königsberg

In 1736, Euler presented the problem in its most general mathematical form
[EUL 1736], from where we draw the following chart (Figure 12.41).

Figure 12.41. Euler’s graphic representation

This presentation is most interesting partly because the map of the city is
replaced by a simple diagram showing its principal characteristics and it formulates
the problem in such a manner that the diagram becomes superfluous. However, the
diagram itself is very interesting since it poses, in fact and for the first time, the
problem of duality. Indeed, it associates each district of the city with a node and
connects these nodes by an arc which passes by one of the city bridges, thus defining
278 Graphs and Networks

a multigraph. By using this it passes from the primal, which is the map of the city
and its districts (finite), to a dual2, which is a simplification of the circulation
network. The fundamental problem of duality is thus posed: the comparison of a
network and the associated surfaces, which is our most important problem.

Both representations above are dual and planar. The graph describing the
communication network makes it possible to specify the dual of the sectors and, in
particular, its nodes. However, that immediately presents a fundamental problem:
that of existence or non-existence of the infinite face. If it exists, the dual of the
network defines three faces and an infinite, which may be chosen between C, B or
D. Nonetheless, that does not define four districts, as in a city. If we take four finite
delimited districts into account, the dual of the dual is not the primal and there is no
bijective correspondence between the two, since the dual of the dual according to
Berge should have an infinite face with an associated dual node and thus, it would
then have an additional node.

Let us stress here that the starting point seems to be the various districts, to
which we initially associate a node and we then connect by arcs passing through
each bridge. Moreover, the districts are finite, although the ambiguity is there, in the
problem.

The dual of the Cauchy cube

The problem presents itself differently when the graph is a three-dimensional
volume.

Definition of the dual of the cube

On the basis of the planar graph [ALD 00, p. 264], it is identical to the Withney
diagram mentioned below.

The determination of this dual corresponds perfectly to the definition of Berge.

We note, however, that this dual has a node which is external to the larges square, as
if face 6 was the outside face!

2 We find this procedure implicitly in Kaufmann [KAU 64, volume 2, p. 283] and in Rosen
[ROS 91, p. 488].
 The Representation of Graphs: A Specific Domain of Graph Theory 279

Figure 12.42. The dual of Cauchy projection

The definition of the dual on the basis of a cube in the three-dimensional space
and, more generally, of a convex polyhedron is as follows: “the dual of a convex
polyhedron can be constructed by placing a vertex at the center of each face of
original polyhedron, and joining a pair of vertices with a line-segment whenever the
corresponding faces of the original polyhedron are adjacent along an edge” [ALD
00, p. 273].

cube octahedron
8 vertices forming the dual 8 faces
6 faces 6 vertices
Figure 12.43. Cube and dual of the cube: the octahedron

Let us stress in this case that, if the dual nodes are located on the faces of the
primal, the arcs joining the nodes of the dual, on the other hand, do not intersect
with the arcs of the primal and instead only the projections intersect. That clearly
shows that the definition of Berge only applies to a plane and, moreover, planar
graph. It should also be added that the dual represented in this manner is a particular
case where the dual nodes are located in the middle of the faces of the primal. The
dual of a regular Plato’s solid is not necessarily a regular Plato’s solid.

Actually, if we specify that an arc connects the dual nodes of two adjacent primal
faces, the intersection of the arcs on the plane is simply a consequence of the
planarity of the graph in one case and of the projection of the graph, and thus of its
flatness, in the other case.
280 Graphs and Networks

A more difficult problem is that of the possibly infinite face, i.e. of the external
node (node 6 of Figure 12.42). Let us use again Cauchy’s projection but this time by
including the dual (Figure 12.44). We note the absence of an infinite face, but the
existence of a double node. The lower and upper nodes of the octahedron are
projected onto a single point of the plane in orthogonal projection, but if the
projection on the plane was oblique, these two nodes would be distinct.

Figure 12.44. Projection of a cube and of its dual onto a plane

Similarly, each external arc of the dual intersects with two arcs; this is due to the
fact that the projection center is too far away from the upper surface of the cube,
which is something that can be easily changed by increasing the upper square.
However, is this still a projection?

Figure 12.45. Straight projection of the cube and its dual by enlarging the upper square

In fact, if we project the cube and its dual orthogonally onto a plane which is
perpendicular to a diagonal plane of the cube, the central edge of the octahedron is
projected according to a point located in the middle of the projection of the face.
Therefore, in order for this edge to be projected beyond the basic node of the cube
face, the projection center has to be internal to the cube. Then, the projection cannot
 The Representation of Graphs: A Specific Domain of Graph Theory 281

be plane and it can be done onto a sphere whose center belongs to the space
delimited by the cube.

Figure 12.46. Projection of an octahedron and a square (left)

and any projection of the Cauchy cube (right)

Consequently, the projection of the dual of a cube onto a plane does not
correspond to the usual definition of the duality of planar graphs. The projection of
dual arcs intersects two projections of primal arcs and the node that is double for a
normal central projection is divided for any oblique projection, as above (the cubic
graph is represented in profile, the vertical of the projection center is materialized on
the square and the projected figure is seen from above to make it possible to observe
the deformations).

This projection with a double face satisfies the Euler relation between nodes,
arcs and faces without there being an infinite face:
– let G be a connected graph with e arcs and v nodes. Let r be the number of
areas in a representation of g;
– then, r = e-v+2.

In this definition we have removed the planar term because the projection of a
convex volume onto a plane is not planar, as shown in Figure 12.46. Taking into
account an external face would be against Euler’s relation in this case.

The non-orthogonal projections of the Cauchy cube and of its dual transform the
projection of the cube by giving it a trapezoidal aspect if the projection center is
located on a symmetric or quadrilateral plane, which is more irregular in other cases.
282 Graphs and Networks

However, the most visible result of this is mainly the separation of the node and
double arcs, which distinguishes the two upper and lower nodes of the octahedron,
as above. The projection is then planar and non-planar.

We thus obtain two different results on the basis of the same graph. The question
is then to find out whether there is possible isomorphism between the two dual
graphs (nodes and arcs) and whether there is isomorphism between the faces of the
two dual graphs.

Let us consider the dual obtained on the basis of the graph projected onto a plane
and that obtained by its projection onto a plane (Figure 12.47). We can make a node
of the dual D’ correspond to each node of the dual D. A pair of adjacent nodes of D’
corresponds to each pair of adjacent nodes of D and thus there is bijective
correspondence between the two graphs. Furthermore, if we move node 6 of D’
towards the exterior, will there still be bijection between the faces of D and D’
(Figure 12.48)? Is this operation possible?

Figure 12.47. D, dual of a plane projection (left);

D’, dual of the cube projected onto a plane (right)

Figure 12.48. Movement of node 6 of D’

 The Representation of Graphs: A Specific Domain of Graph Theory 283

Indeed, the surface of the cube is convex and the external part of the plane does
not belong to the projection. Due to this fact node 6’ of D’ no longer belongs to its
definition face, as we demonstrated above. Consequently, D can only be the dual of
the planar graph G, if and only if G is not a projection of a cube.

General definition of the dual

If we refer back to the two definitions, we note that the second is more general
than the first because it specifies “joining a pair of nodes with a segment of a
straight line each time the corresponding faces are adjacent along an arc”. This
applies to the three-dimensional graphs, whereas Berge’s definition specifies that “to
any edge e of G (the italics are our addition) will correspond an edge e* of G* that
will connect the z nodes x* and y* corresponding to the faces s and t which are
located on both sides of the edge e”.

Specifying “to any edge e of G” is not necessary since, on the one hand, it is a
simple projected consequence of the adjacency along an edge of the two faces
connected to a node of the dual and, on the other hand, it introduces a restriction,
which means that this definition of duality is valid only for planar graphs.

The second definition is more general because it applies to plane graphs and to
planar graphs, for which primal and dual arcs necessarily intersect and, of course, it
also applies to three-dimensional graphs, for which primal and dual arcs do not
intersect in space, but may do so once projected onto a plane.

12.3.1.3. Duality: a periodic fractal generator

Let us consider the following theorem.

If G is the drawing of a connected graph with n nodes, m arcs and f faces, then
G*, which is the dual of G, has f nodes, m arcs and n faces.

A tetrahedron has four nodes, four arcs and four faces. Its dual will also have
four nodes, four arcs and four faces and it will be a reversed tetrahedron written into
the original one and reciprocally.

We saw that dual of a cube, i.e. 8 nodes, 12 arcs and 6 faces, was an octahedron
with 6 nodes, 12 arcs and 8 faces, and vice versa.

The icosahedron has 12 nodes, 30 arcs and 20 faces. Its dual is the dodecahedron,
which has 20 nodes, 30 arcs and 12 faces.
284 Graphs and Networks

More generally, the dual of a convex graph with n nodes, m arcs, f faces in a
three-dimensional space has a dual, which has f nodes, m arcs and n faces and which
in turn has a dual with n nodes, m arcs, f faces.

Corollary: the graph and the dual of its dual have the same number of nodes, arcs
and faces.

Consequently:

THEOREM 12.1.
If the graph G is Plato’s solid and if the nodes of the subsequent duals are
defined by using faces, the dual G** of the dual G* will also be Plato’s solid of
a comparable nature to graph G.

The dual of a tetrahedron is a reversed tetrahedron written into the original,

whose dual is a reversed tetrahedron written into it, and so on…

The dual of a cube is an octahedron written into the original, whose dual is a
cube written into it, and so on…

The dual of an icosahedron is a dodecahedron written into the original,

whose dual is also an icosahedron written into the original, and so on…

tetrahedron octahedron cube icosahedron dodecahedron

Figure 12.49. Plato’s solids and their duals (according to [ALD 00], modified by us)

THEOREM 12.2.
Duality constitutes a periodic fractal generator for Plato’s solids with
internal homothety whose ratio r is constant, every two transformations (Figure
12.51).

The determination of the dual of Plato’s solid along with the definition
adopted until now leads to defining a dual is inscribed into the primal and, thus,
of a reduced volume. However, by taking into account the symmetry of the
 The Representation of Graphs: A Specific Domain of Graph Theory 285

transformation, we can reverse the process and look for the primal of Plato’s
solid which is itself considered as a dual.

Since the dual is inscribed, the primal will necessarily be ex-inscribed and
therefore have a larger volume.

THEOREM 12.3.
A primal Plato’s solid has Plato’s solid inscribed into as a dual, and is itself the
dual one of Plato’s solid inscribed around it with the same ratio.

The theorem above corresponds to the traditional definition of decreasing

fractality with a constant reduction ratio, i.e. 1/3 for the sides and 1/27 for the
volume, for a cube, on the basis of an original solid, with duality as generator. We
can reverse the process by adopting primality defined as above as generator on the
basis of this same solid. We will then have a fractal transformation with a constant
growth factor. By taking into account the perfect symmetry of the process, the
starting point of the fractal inclusion series can be found “in fine” at any level.

Figure 12.50. Periodic fractal by generation of duals

THEOREM 12.4.– for a volume whose faces constitute a closed convex surface, the
dual of a dual is never the primal. We will call ante primal the graph whose dual is
the primal considered. Any Plato’s solid can be regarded as the dual of an ante
primal. Thus, a primal would always have a dual inscribed into it and an ante
primal inscribed around it, and there is homothety between the two: the ante primal
and the dual, are therefore self-similar.

We can generalize the latter property to plane graphs (Figure 12.51):

– let us consider any planar graph G. We will call “quasi-dual” the dual G’
whose possible vertex located in the infinite exterior is not taken into account
(case of the projection of a graph with a non-convex surface);
286 Graphs and Networks

– the quasi-dual G’ of G, which thus has no external vertices, will be the dual
inscribed inside for the considered graph;
– G’ corresponds to the “dual graph of the map” defined by Rosen [ROS 91,
p. 488], or to the graph of the map of the 10 departments quoted by Kaufmann
[KAU 64, volume 2, p. 283];
– we will define by symmetry an ante primal, whose considered primal will be
the quasi-dual;
– the planar quasi-dual written inside is a subgraph of the ante primal.

Figure 12.51. Any quasi-dual graph inscribed and ex-inscribed

We will call graph periphery the chain of arcs delimiting the interior and the
exterior of the graph.

The quasi-dual has as many vertices as the primal has finite faces.

The quasi-dual has as many edges as the primal has internal edges.

The ante primal has the same number of internal arcs as the primal plus twice the
number of external arcs or the same number of arcs as the graph plus the number of
external arcs.
 The Representation of Graphs: A Specific Domain of Graph Theory 287

Figure 12.52. The division of Paris into 20 districts and the old city limits

The ante primal will have as many external faces as the dual written inside it.
The growth will be of comparable nature by addition of a constant number of faces
for each successive ante primal.

This process recalls the growth of cities as the plan (Figure 12.53) of the growth
of Paris under Haussmann shows.

Let G(1-12) be the quasi-dual exterior of G (1’, 2’, 3’, 4’, 5’), the ante primal.

The graph G(1’-5’) has G(1-12) as ante primal. The ante primal is the external
dual ex-inscribed of graph G. The graph G(1’, 2’, 3’, 4’, 5’) is a subdivision of the
graph G(1’, 5’).

13’ 14’ 15’ 16’ 17’

1 2 3 4 5 6

12’ 6’
1’ 2’ 3’ 4’ 5’

12 11 10 9 8 7
11’ 10’ 9’ 8’ 7’

Figure 12.53. Ante primal graph and the dual written inside it

Let us consider the annular graph and its projection onto a sphere: there is no
infinite surface but two possibly symmetric external points belonging to the two
opposite faces of the graph.

Plato’s solid is a figure written inside of or around a sphere and the graph of this
solid can be projected onto this sphere.
288 Graphs and Networks

Let us consider, for example, a regular graph projected onto a sphere and
covering it entirely, which is the case of a central projection of a cube onto a sphere,
as shown in Figure 12.54.

In this case the dual, which is a spherical plot of the central projection of the
cube, exists but cannot have an infinite face. In this case again Berge’s definition
does not apply. The dual of this figure will also be the central projection onto the
sphere of the dual tetrahedron of the cube.

The dual can be drawing either by using the spherical projection of the cube or
the projection of the dual tetrahedron of the cube. More generally, we can show that
the graph, which is the central projection onto the sphere of a convex volume, does
not necessarily consist of identical faces, as demonstrated by a soccer ball, or even
regular ones. This graph will have a dual projected onto the aforementioned sphere;
which is itself a projection of the dual of the convex volume obtained as indicated
by Labelle [LAB 81, p.98]: “let P be a convex polyhedron. Its dual is obtained by
taking as nodes the centers of the faces of P and by joining the centers of the
adjacent faces with segments”. This is still a particular definition in the sense that
the nodes of a dual are not necessarily in the middle of the faces but, on the contrary,
may be located anywhere on the face as long as they are mathematically distinct
from the nodes and arcs of the primal.

G’
F’

F G
A’ B’

B
A

E H

H’
E’ C
D

D’
C’

Figure 12.54. Central projection onto a sphere

This problem is not purely theoretical; it is an extremely concrete problem for

any geographer, for whom the ground is spherical in the first approximation. At a
certain scale a network will have a spherical projection and thus no infinite faces.
 The Representation of Graphs: A Specific Domain of Graph Theory 289

We arrive at the definition provided by Rosen [ROS 91, p. 488] of the dual graph
of the map: “any map can be represented by a graph. To establish this
correspondence each area of the map is represented by a node. Arcs will connect two
nodes, if the areas which they represent have a common border. Two areas which
only touch in a point are not regarded as adjacent areas. The resulting graph is the
dual graph of the map. From the manner in which we construct dual graphs, it is
clear that any map has a planar dual graph”, such as, for example, the projections
presented by Rosen (Figure 12.55).

However, this definition of the dual graph used by Headwood for the problem of
five colors and then of four colors of a map does not take into account the possibility
of an infinite face. Similarly, the number of arcs does not correspond between the
primal and the dual: external arcs do not have corresponding dual arcs. Let us add
that this dual is identical to the Eulerian tradition of the bridges of Königsberg. Due
to this fact this definition is different from that of the geometrical dual of Withney
consecrated later in other works.

Figure 12.55. Dual graph of the map, which is itself

a primal graph according to Rosen [ROS 91]

The same problem arises when we project the dual onto a torus as Headwood did
[HEA 76]. In this case an external node can be created, but it is not necessary.

Let us consider the projection onto a torus: in this case the dual will be very
specific. The external arc actually constitutes a loop, which is the same arc which
intersects the two primal arcs.

The problem of the torus grid is an old one, since it dates from the works on
perspective by Paolo Uccello in the 15th century (Figure 12.56).
290 Graphs and Networks

Figure 12.56. Torus grid by Paolo Ucello (Flammarion 1972)

The dual of this graph will be a possibly regular and identical graph obtained by
using either Withney’s or Rosen’s methods, which in this case both give the same
result.

We can conclude for the time being that duality remains a polysemous concept
whose definition remains to be specified in graph theory where two definitions
prevail, but where apparently only one of them makes it possible to obtain a cellular
graph.

12.4. The cellular graph

Cellular graph and its uses

The following discussion aims at enabling the passage from a traditional graph to
a cellular network thus making it possible to use a traditional cellular model, on the
one hand, and of a multiagent system, on the other hand, which may contain reactive
agents or artificial intelligence to simulate the behavior of network users.

Traditional transport system modeling by using a description in graph terms of

its physical network is based, as we saw, on precise assumptions, such as complete
rationality and total prior knowledge of the network, which is a necessary condition
for any search for any optimum in a network, be they the shortest paths in distance,
time, cost, maximum flow, the search for a Hamiltonian circuit, etc.

However, we have observed that the maximum capacity of a graph does not
depend, in general, on the capacity of the arcs, but on the capacity of the nodes. This
is a fundamental reason for the representation problems of the explanation of the
decomposition technique of nodes into nodal graphs.
 The Representation of Graphs: A Specific Domain of Graph Theory 291

We could observe, nevertheless, that the technique of nodal graphs led only to
their decomposition into convergent and divergent nodes, which is still insufficient
to determine the real capacity of a node because it does not take into account the
priority rules of various flows, their real overlap and interpenetration capacities, and
thus the variations in speed of flows, and, consequently, the capacity of the lanes,
which is itself a function of this speed, as well as the storage capacity of the arc
upstream of the node, which is traditionally defined as a queuing problem.

In order to be able to explore the range of possibilities with respect to the

circulation of traffic flows through a network, it seems necessary to go down to the
level of the agent, whose individual behavior will be determined, among other
things, by the rules of the highway code.

The chosen hypothesis is that the total state of the system results from the
behavior of all the agents, their capacities of training, adaptation and anticipation.

To simulate an MAS (multiagent system) [IRON 97] in a transport network, one

of the solutions is to have the definition of the network in the form of a set of cells
which make it possible to apply the rules of percolation-type behavior.

It is obvious that a method which makes it possible to pass from the network
modeled in the form of a graph to a cellular graph would constitute a powerful tool
employing few means and making it possible, among other things, to globally
simulate the transport system and simultaneously all its users through simple
behavior rules, and thus to dynamically study their interaction.

Our intention is to show that on the basis of the previously developed

considerations on duality and nodal decomposition, it is possible to define a cellular
graph supporting an MAS, and that this transformation can be expressed as an
algorithm.

Transformation of any arc into a cellular arc

S1 S2
Figure 12.57. Arc graph S1S2

1 2 3 4 5 6
Figure 12.58. Subdivision of the arc graph S1S2
292 Graphs and Networks

Let us consider any arc AB (see Figure 12.57). Arc S1S2 is an arc possibly
belonging to a plane graph but not necessarily planar. The coordinates of S1 and S2
are known: they are attributes of the nodes and, naturally, fixed. The length l of the
arc is therefore known. Let us perform a subdivision of the arc S1S2 with n nodes of
the second degree, so that they have between them an arc l/n+1=u. The length “u”
will be variable depending on the user, as we will see below.

Let us consider the ante primal of S1S2 with the preceding subdivision of n
intermediate second degree nodes (Figure 12.59).

1 2 3 4 5 6

S1 S2

Figure 12.59. Cellular arc

We have transformed the arc S1S2 into a cellular arc with n+2 cells of a length u,
which can correspond, for example, to the average statistical space occupied by a
vehicle or a passer-by.

The cells [LAN 97] which are created in this manner will have a statute that can
be specified at leisure and may or may not be occupied. However, let us
immediately remove an ambiguity in the case of a road arc, where a cell of a cellular
arc can only be occupied by one element at any time and in the majority of the cases
it would then be necessary to use a directed graph, digraph or directed multigraph
(Figure 12.60). The cellular arc could then be used as support for MAS micro-
modeling, for example, in the case of movement against the traffic flow. In the more
general case of a symmetric directed arc, the construction principle is identical.

11 12 13 14 15 16

S1 S2

26 25 24 23 22 21

Figure 12.60. Symmetric directed cellular arc or cellular digraph

 The Representation of Graphs: A Specific Domain of Graph Theory 293

The notation principle is simple: the first set of numbers indicates the node
towards which the arc is directed, the second indicates the node of origin of the arc,
the third possibly indicates the number of the arc in the case of a p-graph (or of the
lane, in the case of an arc whose capacity is defined by several lanes starting from
the lane farthest on the right) and the fourth indicates the number of the cell starting
from the destination node. This orientation of the cell number makes it possible to
automatically determine the distance from the destination node, which is necessary
for the behavior of user agents. We will associate various values with each cell, such
as occupation by a car, truck and public transport, which in this case can
simultaneously occupy several cells according to their sizes.

However, we are then obviously presented with the problem of the cellular
definition of the node because the previous pattern is not satisfactory at this level.
We will use the nodal graph developed into a cellular nodal graph: we transform the
node S (Figure 12.61) into four nodes S1, S2, S3, S4 (Figure 12.62), thus adding
four arcs of length u according to a rule opposite to that of simplification of flow
graphs, since here we do not simplify but rather break up [CHO 64, p. 39]. We take
as a starting point the derivation method of transformation by using the method of
duplication of nodes for the nodal graph.

Figure 12.61. Node of a digraph

S2 S1

S3 S4

Figure 12.62. Simplified decomposition of the digraph node

294 Graphs and Networks

S2 S1 S11 S12 S13

S33 S32 S31 S3 S4 n n- 1 n-2

Figure 12.63. Node of the cellular graph of a simple crossroads: one lane in each direction

Figure 12.64. Crossroads with two lanes in one direction

Figure 12.65. “T” or three-way junction

 The Representation of Graphs: A Specific Domain of Graph Theory 295

The size of the junction is automatically determined by the width of the cells and
the number of incidental arcs (entering + leaving) (Figure 12.65).

The same process is used for directed multigraph describing three lanes (Figure
12.64), four lanes, two in each direction (Figure 12.66), crossroads with three lanes
(Figure 12.65) and more, and the roundabouts, for which we may or may not use
convergent/divergent nodes.

However, it is preferable to use the simplest forms that are closest to the material
reality, with limited decomposition in order to introduce as little bias as possible into
the behavior of the simulated agents3, which is defined in the case of the ordinary
land vehicles by the rules of the Highway Code4. The objective here is different to
the nodal decomposition which we described previously. Indeed, the present
objective is to make it possible to simulate, under conditions close to reality, the
crossing of crossroads by reactive agents whose rules of behavior comply (or
possibly do not comply) with the rules of the Highway Code and the main driving
constraints. In the case of nodal decomposition with convergent and divergent
nodes, the objective is to determine not the capacity of each arc, but the deceleration
induced by flow divergence or convergence and, thus, the capacity of the node
according to speeds under the assumption of fluid flow.

With the cellular MAS modeling, the random simulation hypothesis of non-
compliance with rules or constraints and of the occurrence of an accident is taken
into account.

In this case, we no longer have to deal with flows and the determination of an
optimum under constraints, but with behaviors of possibly diverse, erratic or even
criminal agents.

Figure 12.66. Crossroads between two streets or roads with twice two lanes

3 Mathis, Philippe and Maupuis, Gregoire: Programming an automat VST-CESA, September

2000.
4 Decoupigny, Christophe, PhD “Modélisation fine des émissions de polluants issues du
traffic urbain” Tours, March 2006.
296 Graphs and Networks

One of the great advantages of this solution with respect to that of the nodal
graph is that the projection of the cellular graph remains planar and thus its
computerized description is largely facilitated: to each arc there corresponds a vector
of cells and each of them is clearly identified.

On the other hand, in the case of interchanges, i.e. of a very specific non-planar
graph, we also have to remain faithful to reality, which is more complex, and the use
of divergent and convergent cellular nodes makes it possible to describe such
interchanges with precision.

Nonetheless, we saw with nodal graphs that the decomposition follows a defined
process, an algorithm which is generating a fractal representation. This process can
be reversed on demand and according to the selected scale of the representation and
the user is free to consider the node traditionally, decomposed or in cellular form.

The application of this transformation of a network represented by a traditional

graph into a cellular graph makes it possible to take into account another level of
simulation: that of the user himself with his own constraints, his limited rationality,
his imperfect knowledge of the network, etc. The example of the graph of the center
of Tours clearly shows that the capacity and saturation problems are obviously due
to the morphology of the network itself: there are multiple intersections, the roads
are often one-way, with one or two lanes with problems of parking, obstruction of
lanes, etc.

12.5. The faces of the graph: from network to space

A graph is defined as a set of points and relations between these points. The
graph drawing, i.e. its representation on a surface, is in general a plane and therefore
it takes into account only such elements as the crossroads of a network, which is
represented by a node of the graph, and the relations between them, the roads or the
railroads, for example, which we will represent by an arc in the shape of a straight
line segment, a polyline or a curve.

Initially we only consider undirected graphs. The planar projection of the graph
G divides the plane into areas, of which one is open. These areas, which certain
authors call faces of the graph, have a status and rarely or poorly specified
properties. Mainly, these areas or faces appear, on the one hand, in the statement of
Euler’s formula, which is applied either to a planar graph with an infinite face or to a
projection of a convex volume without an infinite face:
 The Representation of Graphs: A Specific Domain of Graph Theory 297

– let G be a connected simple graph with e arcs and v nodes. Let r be the number
of areas in a representation of g;
– then, r = e–v+2;
– in addition, right from the start authors have been concerned with the problem
of colors. How can we apply colors, if the “faces” are not surfaces?

Finally, in the definition of the dual graph to each area or face we associate a
node of the dual graph, which can be localized not just in the middle of the faces but
anywhere in the face, since it is mathematically distinct from the nodes and the arcs
of the primal. This condition enables the dual node to cover, to the nearest ε, the
entire surface of the area considered.

This definition is valid regardless of the graph considered: planar or three-

dimensional polyhedron. However, this causes some difficulties due to the flatness
of a face, if it is in general supposed and real in the case of Plato’s solids, which is
not automatic once the number of arcs, when they consist of straight line segments,
is more than three. Indeed, a triangle is always flat, which is not the case of a
quadrangle or more, since the nodes do not necessarily belong to the same plane.

In the case of Plato’s solids, the dual is particular, since its nodes are implicitly
located at the center of the faces of the primal. The dual whose nodes are not at the
center of the faces of the primal would be irregular and would in turn have an
irregular dual, which would then not be isomorphous with the first, but rather its
transform, still with the same number of nodes, arcs and faces.

The face is possibly plane and can be traversed by a point. It is used as a support
for a possible dual node and has the role of separating the arcs and nodes from an
elementary cycle of the primal.

At first we make the assumption that the surface on which the graph drawing is
plane and that it is not the projection of a convex three- or more dimensional surface
and, thus, there are no double points.

When we look at the map of a city (Tours, Paris, Chicago, Barcelona), we note
that small islands are simple geometrical forms, a generally regular space paving:
very regular and formed by right-angled or square parallelepipeds for the town of
Millet where Hippodamos (5th century BC) had invented, according to legend, the
orthogonal plane, which we now find in Chicago, Manhattan or Tours; octagonal for
Barcelona with the plane of Cerda; square or triangular for Paris with the works of
Haussmann, etc. It is the same for the majority of pieces of land in the rural world
where Roman centuriations left generally clear traces.
298 Graphs and Networks

Thus, we have an self-similar form of sectioning where fractality has a limit in

urban planning: the size of the piece of land, which varies according to the times and
construction methods used, is perfectly visible in a large number of cities, such as
Tours, where the medieval fragmentation of Bourg Saint Martin is definitely smaller
than that of districts urbanized later on. Even when the pieces of land were large due
to their origin, i.e. gardens and parks of religious congregations, they are allotted
according to the contemporary socio-economic and technical considerations.
Frankhauser [FRA 94] and his team [GEN 00] showed how fractality could be used,
on the one hand, to describe and, on the other hand, to characterize the city.

On the basis of these faces and by using a fractal transformation it is possible to

define a cellular in order to approach the fragmentation as much as possible. The
cells or pieces of land thus defined will have attributes of various natures: built-up,
courtyard, park, private or public space, etc.

By considering cells and attributes we are in the domain of application of cellular

models. However, the two-dimensional graph then becomes insufficient to describe
a building or the city itself, which is not flat.

We then have the case of a GRS, which is the graph with spatial referencing
defined at the start, since in the case of a built-up cell or piece of land it is quite
obvious that the land use coefficient can be high, thus reflecting the existence of
buildings with many or fewer floors, and there again the simple graph with
geographical referencing is no longer sufficient.

This justifies in another way the taking into account of three-dimensional graphs
beyond the purely mathematical problem of the dual. Indeed, graph theory simply
brings together a set of elements with a collection of relations between these
elements. The dimension of the graph is neither specified, nor necessary for the
majority of algorithms. It is simply that the practice of representing graphs on a
plane used to prevail.

Three-dimensional analysis becomes necessary to simulate a city, since it has

three dimensions. A cellular graph lends itself admirably to this use as Serrhini had
suggested in his three-dimensional visibility model: the GRS is a must.

This method makes it possible to pass from a graph and the faces of a graph to a
set of cells which are clearly defined for arcs and nodes, and others for the faces of
the graph. It is then possible to use MAS and cellular modeling. The former enables
modeling and simulation of flows, whereas the latter enables the simulation of
stocks by taking into account the attributes of cells, simultaneously with a classical
four stages simulation on a more global scale.
 The Representation of Graphs: A Specific Domain of Graph Theory 299

12.6. Bibliography

[ALD 00] ALDOUS J.M., WILSON R.J., Graphs and Applications, an Introductory
Approach, Springer-Verlag, London, Berlin, Heidelberg, 2000.
[BAL 00] BALAKRISHNAN R., RANGANATHAN K., A Textbook of Graph Theory,
Springer, New York, 2000.
[BAN 01] BANG-JENSEN J., GUTIN G., Digraphs: Theory, Algorithms and Applications,
Springer-Verlag, London, Berlin, 2001.
[BAP 99] BAPTISTE H., Interactions entre le système de transport et les systèmes de villes :
perspective historique pour une modélisation dynamique spatialisée, Thesis, Tours, 1999.
[BEA 92] BEAUQUIER D., BERSTEL J., CHRETIENNE P., Eléments d’algorithmique,
Masson, Paris, 1992.
[BER 70] BERGE C., Graphes, Dunod, Paris, 1970.
[BER 87] BERGE C., Hyper graphes, Bordas, Paris, 1987.
[BIG 76] BIGGS N.L., LLOYD E.K., WILSON R.J., Graph Theory, Oxford University
Press, 1976 (1986).
[CAU 76] CAUCHY A.L., “Recherches sur les polyèdres”, Premier mémoire, J. Ecole
Polytechnique 9 (Cah.16) (1813), 68-86., in Graph Theory 1736-1936 Biggs N.L., Lloyd
E.K., Wilson R.J., Oxford University Press, 1976 (1986).
[CHA 97] CHAPELON L., Offre de transport et aménagement du territoire : évaluation
spatio-temporelle des projets de modification de l’offre par modélisation multiéchelles des
systèmes de transport, Thesis, Tours, 1997.
[CHO 64] CHOW Y., CASSIGNOL E., Théorie et application des graphes de transfert,
Dunod, Paris, 1964.
[COR 94] CORMEN T., LEISERSON C., RIVEST R., Introduction à l’algorithmique,
Dunod, Paris, 1994.
[DOL 01] DOLLARS A., “Le patron du Maître”, Pour la Science, no. 282, p. 92-94, 2001.
[EUL 1736] EULER L., Solutio problemats ad geometriam situs pertitnentis, commemtarii
Academia Scientiarum Imperialis Petropolitanae 8,128-140 = Opera Omnia, vol. 7, 1-10,
1736.
[EUL 1758] EULER L., Demonstreatio nonnullarum insignium proprietatum quibus solidra
hedris planis inclusa sunt praedita. Novi Comm. Acad. Sci. Imp. Petropl. 4 (1752-1753,
publié en 1758), 140-160 = Opera Omnia (1), vol. 26, 94-108, 1758.
[EUL 1758] EULER L., Elementa doctrina solidorum. Novi Comm Acad. SCI. Imp. Petropol.
4 (1752-3, published in 1758) 109-140 = Opera Omnia (1), vol. 26, 72-93, 1758.
[FER 97] FERBER J., Les systèmes multiagents. Vers une intelligence collective, Paris,
InterEdition, Paris, 1997.
[FRA 94] FRANKHAUSER P., La fractalité des structures urbaines, Anthropos, 1994.
300 Graphs and Networks

[GEN 00] GENRE GRANDPIERRE C., Forme et fonctionnement des réseaux de transport :
approche fractale et réflexions sur l’aménagement des villes, Besançon, 2000.
[HEA 76] HEADWOOD P.J., “Map Colour Theorem”, Quarterly Journal of Pure and
Applied Mathematics 24 (1890), p. 332-338, in Graph Theory 1736-1936 Biggs N.L.,
Lloyd E.K., Wilson R.J., Oxford University Press, 1976 (1986).
[KAU 64] KAUFMANN A., Méthode et modèles de la recherche opérationnelle, volumes 1
and 2, Dunod, Paris, 1964.
[LAB 81] LABELLE J., Théorie des graphes, Modulo éditeur, Mont-Royal Québec, 1981.
[LAN 74] LANTNER R., Théorie de la dominance économique, Dunod, Paris, 1974.
[LAN 97] LANGLOIS A., PHILIPPS M., Automates cellulaires, application à la simulation
urbaine, Hermès, Paris, 1997.
[LAU 91] LAUWERIER H., Fractals, endlessly repeated geometrical figures, Princeton
University Press, Princeton, 1991.
[LHO 97] L’HOSTIS A., Images de synthèse pour l’Aménagement du territoire : la
déformation de l’espace par les réseaux de transport rapide, PhD Thesis, Tours, 1997.
[LHU 76] LHUILLIER S.A.J., Mémoire sur la Polyédrométrie, Annales de Mathématiques 3
(1812-3), p. 169-189, in Graph theory 1736-1936 Biggs N.L., Lloyd E.K., Wilson R.J.,
Oxford University Press, 1976 (1986).
[MAN 77] MANDELBROT B., The fractal geometry of nature, New York, 1977.
[MIN 86] MINOUX M., BARTNIK G., Graphes, Algorithmes, Logiciels, Dunod, Paris, 1986.
[POI 76] POINSOT L., “Sur les polygones et les polyèdres”, J. Ecole Polytechnique 4
(Cah.10) (1810), p. 16-48, in Graph Theory 1736-1936 Biggs N.L., Lloyd E.K., Wilson
R.J. Oxford University Press, 1976 (1986).
[PON 69] PONSARD C., Un modèle topologique d’équilibre interrégional, Dunod, Paris,
1969.
[ROB 00] ROBERT D., Le réseau routier français dans la dynamique des échanges de
Marchandises de la France avec ses partenaires d’Europe Occidentale, PhD Thesis, Paris
1, 2000.
[ROS 91] ROSEN K.H., Mathématiques discrètes, Chénelière/McGraw-Hill, 1991 (1995,
1998).
[WIT 32] WITHNEY H., “Non separable and planagraphs”, Transaction of the American
Mathematical Society, 34 (1932), 339-362, 1932.
[XUO 92] XUONG N.H., Mathématiques discrètes et informatique, Masson, Paris, 1992.
 Chapter 13

Practical Examples

The elements presented above are not the result of chance or simple theoretical
research: they were necessary to solve a certain number of difficulties which the
modeling group had encountered and for which we had sometimes developed
empirical solutions or inferred the existence of a solution.

Below we initially present some of these intuitions or anticipations and then in

the second part we will show not only the applicability but the great interest that lies
in the results above within a large scale simulation of urban motor traffic, as well as
in the study of its characteristics at the level of agents and cells, whose results can be
used directly in urban planning.

13.1. Premises of multiscale analysis

13.1.1. Cellular percolation

The first example of partial use of cellulars and graphs carried out by the CESA
research group goes back to 1992. The very simple model simultaneously employs a
cellular frame to simulate urban expansion by percolation with or without the
existence of free space and a mass transit system, such as the suburban train
combining a graph to the cellular and therefore a bimodal approach with delay.

Chapter written by Philippe MATHIS.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
302 Graphs and Networks

In this example1, the fastest means of transport was public transport with fixed
stops of the train type (tunnel effect), from which there is normal diffusion, i.e. by 8-
connected This type of modeling shows diffusions in successive blots, the
development of the center and less quickly of the starting points of public transport
(which might well have been highway exchangers), all the more slowly as they were
further removed from the center, and a tendency for congestion as well as urban
diffusion in the Parisian suburbs (Figure 13.1).

This type of modeling is perfectly appropriate for the analysis of stocks, but the
process described remains very succinct. It is necessary to go beyond a simple,
purely mechanical description of diffusion.

Figure 13.1. Diffusion of housing by connectivity and according to public transport networks

This type of cellular modeling must be supplemented by MAS modeling to

simulate flows and we can then follow the definition of states at the local micro-
level, which is that of one or several pieces of land (stocks in dynamic system
modeling of the Forrester type [FOR 69]) as well as the traditional level of overall
optimization, and thus to detect variations, tensions, contradictions and the possible
mechanisms of resolution thereof.

The kind of spatial description enabled by this type of technique can be very
fine.

1 Multimodal percolation Address at the Sixth World Conference on Transport Research,

Lyon 1992.
 Practical Examples 303

13.1.2. Diffusion of agents reacting to the environment

The second preliminary application of these methods goes back to 1993, carried
out in a different spirit in order to model the behavior of tourists in a sensitive
natural environment: the crossing of the dune range on the Grande Côte at the mouth
of the Gironde. This is a dune range whose structure is often disrupted by trampling
and which required expensive dune restoration work, despite areas where entry was
prohibited to tourists, which are marked in red in the diagrams of the color plates
section.

After having observed and noted the behaviors of tourists during their journey to
the range from the parking lots and back, and perhaps their stay on the accessible
dunes, the objective of this simulation was to be able to simulate them in order to
avoid invasive trampling of the dunes and the creation of many new paths.

The size of a represented cell must be more than 1 pixel, however, screen
definition was too basic in 1993 to be able to display sufficient detail [MAT 92, 93],
as shown by the modeling diagram (Figure 13.2; see the color plates section)
describing the paths of tourists crossing the dune of the Grande Côte to reach the
range situated at the bottom from the parking spots on the sides of the road or from
the car park at the top of the diagram.

The diagrams in Figure 13.3 (see the color plates section) show two different
hardness hypotheses, the paths used and the results on the walkways according to
the behavior rules and the sensitivity of tourists: unconstrained movement of tourists
through the alleys between the dunes, with the principle of least effort (the most
direct and easiest path) between the parking on the edge of the main road and the
edge of the sea, and the location on the range, naturally, without superposition of
agents. The cells (used for the simulation) are of the size of a pixel due to the basic
screen resolution in 1993.

In these preliminary examples to true MAS the interaction between the tourist
and the physical state of the environment did not exist. The interaction between the
tourists was taken into account only on the range where the selected location was the
closest, but on a free cell.

In the next simulation we introduced stronger hardness on a path represented in

yellow above in order to simulate a visible obstacle on the path, but without
prohibition. We observe the detour taken by the tourists to reach the range. A current
simulation would take into account a probabilistic behavior according to the type of
user: age, accompanied by children of a certain age, outward or return journey, time
of day, etc. In 1993, the data of study could not have been exploited to this level of
details.
304 Graphs and Networks

In the last simulation the bold line at the top represents the creation of a car park
and a passage under the relatively busy access road, causing the concentration at this
level. Another difficulty was added to the trip by involving a concentration of flows
in a specific zone where dune erosion by trampling was considerable, and one of the
objectives was to reproduce this aspect of tourist behavior.

13.1.3. Taking relief into account in the difficulty of the trip

In the following simulations for the same area Serrhini [SER 00] used MNT, i.e.
a GRG to explore the influence of the slopes on the behavior of the same tourists by
simulating different sensitivities to the effort required for climbing a sand dune.

The diagram in Figure 13.4 (see the color plates section) is drawn for an
acceptable level of difficulty corresponding to a zero percent slope: tourists
considered in this example as carrying items or infants, or as old people avoid any
climbing, which perfectly describes the image in this figure.

In the second simulation the acceptable slope is changed to 14%, i.e. the double
of the maximum slope of a highway, which is a considerable slope on sandy ground:
14 meters for every hundred meters is the height of a five storey building (Figure
13.5; see the color plates section).

With this level of difficulty accepted the routes do not change with respect to the
previous assumption: only the same three paths cross the considered area.

Numerical exploration can then continue until a threshold is determined, from

which percolation through the represented dune massif starts, with only the slope for
an obstacle here.

With a threshold of 28% of slope acceptance, we note that practically all the
paths, indeed, cross this space (Figure 13.6; see the color plates section). This slope
is already very steep: it is a natural embankment. It is too much for old people and
parents carrying young children, especially since the only constraint taken into
account here is the slope and absolutely not the difficulty of the surface, such as
sand, for example.

These two old examples show how the traditional and cellular-MAS approaches
complement each other: the first is global and is an optimization procedure with
strong assumptions including a perfect rationality and a total knowledge of the area
considered. The second relates to an agent with limited rationality, but able to take
into account all the “roughness” of the terrain and the “specificity” of behaviors.
 Practical Examples 305

13.2. Practical application of the cellular graph: fine modeling of urban

transport and spatial spread of pollutant emissions

Taking into account the cellular and the attributes of cells makes it possible to
pass from the micro level of agents to the macro-level of the spatial system, by
making it possible, among other things, to evaluate “instantaneous stocks” that
modify the functional valuations of networks and spaces in terms of speed, capacity,
accessibility, but also constructability, etc. The software developed by Decoupigny
[DEC 06] makes it possible to better understand the often significant difference
between theoretical and practical performance of a system. We will see that its
contributions, even at the theoretical level, are numerous and significant.

From the results of the previous chapter Decoupigny worked out an algorithmic
method to transform a transport network represented by a graph into a cellular
graph.

We developed the following considerations: it is obvious that such a network

must be simulated on two levels, on the one hand, globally, as a problem of transport
network optimization, which is done at the level of a PUM (Plane of Urban
Movements), but also at the level of each arc and crossroads and that of the user.
From this stems the need for cellular simulation to establish the various states of the
network, its storage capacity (number of traveling cars as well as parked ones), its
possible saturation and its capacity to enable the level of usability expected by a
resident in his daily practice, and we add the characteristics of pollutant emissions of
vehicles on a very fine spatial scale.

13.2.1. The algorithmic transformation of a graph into a cellular graph at the

level of arcs

The first stage of the algorithm consists, as in the previous chapter, of

transforming the arc by subdivision and then determining the dual without taking
into account the external face, i.e. the ante primal, and to transform the faces into
cells whose length is the average length of a car, that is, five meters.
306 Graphs and Networks

Arc n
Cell Cell Cell Cell Cell
n,1 n,2 n,3 n,4 n,5

Node of the dual

Arc of the dual
Cellular arc
cellular node
Direction of movement

Figure 13.7. Cellular arc

The second stage as of the algorithms [DEC 06] makes it possible to supplement
the cellular graph obtained in this manner by adding one or more lanes in the same
direction and/or in the opposite direction, then possibly a right turn, as in the
example below, or a left turn, etc.

It is even possible to create exclusive rights of way and reserved lanes with
30km/h, with parking on alternating sides [DEC 06].

Figure 13.8. Cellular arcs, two lanes and a right turn, and a cellular bayonet arc

This example is a combination between contraction and widening, as all the

examples above are additive combinations on the basis of the first transformation
considering that two lanes in the same direction can be a directed 2-graph with
 Practical Examples 307

possible lateral communication of the cells following the formal characteristics of

the considered lane.

13.2.2. The algorithmic transformation of a graph into a cellular graph at the

level of the nodes

The algorithmic transformation of nodes into cellular subgraphs or cellular nodes

is more complex and generally includes two broad stages: the first determines the
number of nodal cells, whereas the second determines the characteristics of spatial
connection of the arcs at the considered nodes.

13.2.2.1. Determination of the number of nodal cells

This determination is based on the results of the previous chapter on the
decomposition of a node into a subgraph and on the search for its dual, which we
recall below.

(a) (b)

Figure 13.9. (a) Nodal digraph; (b) nodal digraph and its ante primal

As we may guess intuitively from the figures above, the number of cells is
determined by the number of arcs adjacent to the considered node and, more
precisely, by the number of lanes of each arc or, in other words, by the number of
adjacent arcs to the considered node of the local p-graph.

Algorithmically this number will be at least equal to the number of arcs (or
lanes) arriving at and leaving from the considered node, including the right and left
turns mentioned in the definition of arcs. For obvious traffic reasons cells cannot be
aligned except in the usual case of a crossroads with two one-way streets. In the
308 Graphs and Networks

general case the solution is a cellular subgraph with m rows and n columns of cells,
or as close to it as possible in the particular case of a roundabout.

13.2.2.2. Spatial connection of cellular arcs to the cellular node

Decoupigny [DEC 06] determines the connection of the arcs to any of the four
faces of a cellular sub-summit graph by analyzing the angle of the arc by taking the
trigonometric circle as a basis.

The number of arcs (lanes) connected to each face in this way makes it possible
to determine the m rows and n columns by considering each time the largest number
of arcs at the each opposed lateral face, as we see in the figure below.

(a) (b)

Cellular arc

Prohibited cell

Cellular nodes

Direction of movement

Figure 13.10. (a) Simple node; (b) complex node

For particular cases of interchanges and the roundabouts, as well as for a

presentation of the algorithms see the work mentioned above.

The set makes it possible to automatically define a cellular graph on the basis of
a transport network already transformed into a graph as shown in the figure and the
details below.
 Practical Examples 309

Figure 13.11. Cellular graph: area of Tours

Having shown the possibility of automatically representing the circulation

network by using a cellular graph, the following stage consists of defining the
behavior rules of the agents circulating in this network.

13.3. Behavior rules of the agents circulating in the network

“By assumption, traffic is composed of a discrete set of vehicles that share the
same space and the same conditional objective, which is that of arriving at the
destination point without accident. The journey time or the length of the path
followed are secondary.” The fundamental rule is thus to arrive at the destination
point without accident.

Moreover, “a motorist exhibits self-centered behavior, since he adapts his

behavior to his objective according to his immediate environment. He takes into
account the motorists driving before him, the constraints imposed by space (width of
the street, curve radius, etc.) and his capacity to evaluate these constraints.
Consequently, he does not have an overall rational view of the traffic, since his
rationality is limited by his field of vision and by his experience of the circulation
conditions, in a generally known space at a given time”.

On the basis of these fundamental hypotheses, Decoupigny defines, for a system

of agents, at least three sets of strict, elementary and behavioral rules.
310 Graphs and Networks

13.3.1. Strict rules

They combine spatial constraints with certain rules of the Highway Code, on the
basis of which we define a behavior which is referred to as borderline. They stem
from the fundamental rule.

The absolute constraint is of a physical nature. This is the speed limit of the
vehicle, which in a curved trajectory of a crossroads is defined by the centripetal
acceleration which is a function of the curve radius and the adherence of the vehicle.
This limit implies a maximum speed of entry in a crossroad, and thus, the taking into
account of the maximum negative acceleration defined by the vehicle’s
manufacturer, i.e. the stopping distance, which is maximum instantaneous speed at a
distance from the crossroads.

The speed on the arc will thus be limited, on the one hand, by the maximum
speed at the end of the arc and, on the other hand, by the speed of entry in the arc
and by the capacity of maximum positive acceleration of the vehicle.

Moreover, the maximum speed on the arc will be constrained by a traffic light or
a give way and/or stop sign at the crossroads and, thus, by zero speed of entry, lest
the fundamental rule is not complied with.

Adhering to this fundamental assumption imposes complying with certain rules

of the Highway Code, such as traffic lights, stop and give way signs in addition to
the speed constraint of entry at a crossroads, as well as speed limits, respecting one-
way streets and streets with assigned directions. However, these assumptions can be
removed or modulated.

13.3.2. Elementary rules

They correspond to standard actions on behalf of the agents such as: the security
distance depending on the braking capacity and relative speeds and therefore the free
space between agents, the rules of starting and accelerating, following the vehicle in
front and, thus, of the interaction between vehicles, of changing lanes, overtaking,
parking, etc.

These rules are the same for everyone and are mainly acquired automatisms.
They are defined to a limit. However, individual behaviors are variable according to
the capacities of each driver.
 Practical Examples 311

13.3.3. Behavioral rules

They depend on the evaluation of space, experience, different visual and hearing
perception capacities, as well as the nature of each agent. To simplify the model,
three classes have currently been defined, but that can be modified.

Competitive or aggressive drivers are typically at the extremes of the

acceleration and braking capacities and have shorter than average response times.

On the other hand, calm or careful drivers have slower reactions, higher braking
distance, weaker accelerations and a reaction speed definitely lower than the
previous category.

The intermediate category groups the largest number of agents and its
characteristics are in effect average.

Together this means that even with only three groups, a behavioral
differentiation among the agents is introduced.

On the basis of these basic elements and by integrating the data on pollutant
emissions2 Decoupigny has developed the TUREP software “URban Transport and
Emission of Pollutants” [DEC 06].

13.4. Contributions of an MAS and cellular simulation on the basis of a graph

representing the circulation network

The TUREP software, which not only uses the MAS, but also the attributes of
cellular agents, leads to a certain number of results expected a priori. However, it
also makes it possible to obtain much more general and theoretical results, if the
applicability of the laws considered as general is specified.

13.4.1. Expected simulation results

First of all, the TUREP software shows the possibility of automatically passing
from a graph representing a road network to a cellular graph. This makes it possible
to simulate motor traffic of the agents thanks to a small number of behavioral rules
and the description of phenomena which are permanently observed, but which are
badly or not at all described by the traditional approaches of the four stages model
type, like the stop and go, turnaround, the univocal influence of the vehicles upon

2 See Chapter 8.
312 Graphs and Networks

each other, which the MAS makes it possible to simulate perfectly well, as the
diagram below shows.

Figure 13.12. Universal influence of the vehicles upon each other

However, this also makes it possible to define the attributes of cells used as
circulation support, which is a phenomenon that until now could have only been
very coarsely approximated at the arc level after the phase of spatial allocation
[MAT 05].

The results provided by the use of a cellular graph, a cellular MAS and agents
has a totally different precision, it is micro simulation: space is defined by five
meter-long cells, each attribute of which is calculated and mapped. Among these
results appear the number of passes per cell, the speed and acceleration of vehicles,
the total time of occupation, gas or particulate emissions, all this for a variable
number of vehicles, from one to a hundred or even much more depending on the
computer used.

13.4.2. Limits of application of laws considered as general

The most important results, confirming the intuition of the authors, are the
descriptions of the limits at the level of microsimulations of the application of laws
which are considered as general and, more precisely, the determination of the
maximum flow in the graph, which no longer corresponds to the strict application of
the Ford-Fulkerson theorem, the uselessness of the queuing theory, the non-
applicability of the flow speed curve in its reciprocal form.
 Practical Examples 313

13.4.2.1. Asymmetric application of the flow speed curve

The simulations of Decoupigny show very clearly that a low speed is not
necessarily the result of dense vehicle flow: there is a very clear network effect
which does not enable high speed, especially, in an urban area. The consequence
could be heavy pollutants emissions in zones with high density and low speed
circulation. On the other hand, dense flow implies a speed that decreases the more
flow is disturbed by turnaround, which are at the origin of the stop and go
phenomena. However, these phenomena may just as well be the result of network
effects, which appear dominant in an urban zone.

13.4.2.2. The non-applicability of the Ford-Fulkerson maximum flow in a graph theorem

In a fine representation of the circulation network, such as that made possible by
the cellular graph, capacity is not limited by that of an arc, except for vehicle
parking or an accident, but by the capacity of crossroads, network nodes, graph
nodes. However, we saw that this aspect is not at all dealt with by traditional graph
theory, for which the node is neutral. The simulations made with the TUREP
software obviously reproduce what any driver knows and experiences
systematically: congestion is initially and primarily caused by crossroads.
Decoupigny shows that the capacity of crossroads does not merely depend on their
physical shape, but that the rules of priority play a fundamental part in it by
modifying the various degrees of freedom imposed on flows stemming from each
incidental arc. It can also be seen in the example of an interchange that congestion
can extend by percolation from node to node.

That is a very operational aspect for the layout of urban transport systems as well
as for the forecast of pollutant emissions, without these two aspects necessarily
being correlated.

It appears very clearly that the circulation rules are a primary influence on
network flows and constitute an essential means of traffic control.

A perfectly clear illustration is given by the simulation of a speed bump in a

street and the consecutive modification of pollutant emissions, whereas average
speed remains almost identical, but the acceleration and braking phases change.

13.4.2.3. The uselessness of the queuing theory for the simulation of urban circulation
The precision with which crossroads and the use of simple agent behavior rules
are reproduced renders the use of the queuing theory useless. We see them form
naturally and develop in the same way, according to the priorities of each crossroads
and the characteristics of the agents.
314 Graphs and Networks

13.5. Effectiveness of cellular graphs for a truly door-to-door modeling

This type of approach is also perfectly adapted to the modeling of phenomena

which had been ignored until now, such as the simulation of pedestrian movements
in a city, the study of parking phenomena and initial and final paths.

Models which are currently qualified as door to door are only very partially so
because the final paths are only estimated on average. Currently, 20% of urban car
journeys are less than 1 km. However, the distances covered for parking and leaving
the parking spot may be between 200 and 400 meters: the walking part is sometimes
longer than the driving stretch! Another example is that it takes at least two hours to
go from the University of Tours to the Ministry of Infrastructure, including 55
minutes by TGV for 230 km and more than an hour of initial and final trips on foot
and in the subway over 10 km.

All multimeans journeys have a pedestrian component, which until now was
neither modeled nor really simulated. Multilevel simulation using graphs and
cellular graphs would make it possible to really do this.

Similarly, the simulation of semi-pedestrian paths and mixtures of means such as

public transport, pedestrians, cyclists and rollerskaters is not presently carried out, so
far for lack of adapted tools. The cellular graph must make it possible to mitigate
these difficulties.

The problem of parking vehicles on the public road is not dealt with, neither with
respect to the process of searching for a spot, nor with respect to the effects on the
speed of circulation and the capacity of the roads, which is what the cellular MAS
model makes it possible to simulate.

Finally, more traditionally, movements for multiple reasons are currently

increasing and the traditional methods of optimization are not very adapted to solve
them, whereas the use of the MAS and a search for a satisfactory individual sub-
optimum make it possible to approach it more efficiently

13.6. Conclusion

After developing certain aspects of graph theory to obtain a cellular

representation of arcs and nodes in order to be able to use cellular MAS and agents
with a view to carry out very precise spatially spread micro-simulations, we have
shown using some examples that these theoretical developments had to some extent
been anticipated by experience.
 Practical Examples 315

Multiscale existed already, thus anticipating the theoretical developments and

requiring them in order to really have a foundation. By using two examples of
modeling we have shown that each of these characteristics and theoretical properties
had their counterpart and their need in the theory which based on them the already
used concrete possibility.

The work conducted since the publication of the French version of this book by
Decoupigny within the laboratory, as we have rapidly mentioned above, made it
possible to show the practical possibility of such a method and its efficiency in
micro-modeling.

Indeed, the expected results have been obtained: algorithmic passage from the
graph to the cellular graph, development of an MAS in this space and simultaneous
use of cellular agents in order to arrive at flows, on the one hand, and, on the other
hand, at stocks at the moment t characterized by the various attributes of cells. We
have thus returned to the spirit of dynamic system models of Forrester [FOR 69], but
finely spatially spread here.

However, most interesting and also, perhaps, least expected are the most general
results limiting the fields of validity in micro-simulation of the law, which is a
theory and theorem hitherto used systematically.

The use of TUREP enables a true current and perspective simulation of transport
networks.

It makes it possible to test structural modifications of the network itself in terms

of the priority rules, one-way streets and speeds, which the traditional technique of
four stage models did not enable, since it is based on econometric relations and
therefore on the fundamental condition “all other things being equal”. This set is
thus also a model of spatial perspective, which enables a more justified, more
“objective”, reproducible and more verifiable representation in the sense given by
Waldo Tobler.

13.7. Bibliography

[DEC 06] DECOUPIGNY C. “Modélisation fine des émissions de polluants issues du trafic
en milieu urbain” PhD Thesis, March 24, 2006, Tours.
[FOR 69] FORRESTER J.W., Urban dynamics, 1969 MIT, French translation Dynamiques
urbaines, 1979, Economica, Paris.
[MAT 92] MATHIS P., LIARD V., “Percolation et diffusion dans une espace sensible”,
Address at the International Conference: Tourisme en Façade Atlantique, Royan 14, 15,
and 16 October 1992.
316 Graphs and Networks

[MAT 93] MATHIS P., LEROI B. and LIARD V., “Diffusion des touristes dans un espace
dunaire sensible”, ASRDLF-CESA Conference, Tours 1993.
[MAT 05] MATHIS P. (ed.), ESPON Project 1.2.1, Final Report, 2005, www.espon.lu.
[SER 00] SERRHINI K., “Intégration quantitative et multicritère du paysage lors de la
détermination d’un aménagement linéaire”, Mappemonde, 2000.
 PART 4

Graph Theory and MAS

 Chapter 14

Cellular Graphs, MAS and

Congestion Modeling

In his thesis, Christophe Decoupigny shows that the cellular transformation of

road graphs enables the use of agent-centered modeling which was, in fact, the
objective of this extension of theory.

In more general terms, cellular graphs enable the development of a totally

bottom up approach. But this change of point of view in relation to the four steps of
the top down, traditional, and generally disaggregation models make it possible to
address new problems not directly involved in most traditional models:
− How are agent movements determined?
− What are the consequences of the hypothesis of the driver agent and his
learning ability?
− How are local and global congestions created and what theoretical
consequences result from them?
− How can we validate a model based on individual behavior rules which are
necessarily chosen and simplified?

The doctoral research work done by Jean-Baptiste Buguellou contributes many

elements of response to these different questions1.

Chapter written by Jean-Baptiste BUGUELLOU and Philippe MATHIS.

1 Buguellou Jean-Baptiste Analyse multi-niveaux des flux de transport: modélisation multi-
agents de l'interaction entre l’offre et la demande de transport under the direction of Philippe

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
320 Graphs and Networks

14.1. Daily movement modeling: the agent-network relation

Local movement modeling must take individual choices into account. The point
of view adopted here is that of a bottom up modeling or individual-centric approach.
The level of the individual has long been ignored in traditional four-step transport
models, including those inspired by the Wardrop equilibrium. It is therefore
necessary to focus on agent timetables and a diagram of their activities. The flows
are the result of specific localizations, as well as individual characteristics and time
values. They are characterized by an activity diagram and a schedule for each
simulation agent.

14.1.1. The modeled space: Indre-et-Loire department

The modeled space, although a department (département, a French

administrative region), is based on functional space: the community of Tours and its
“hinterland”.

14.1.1.1. Daily mobility spaces

The daily functional mobility space of the Tours community corresponds to that
of the Indre and Loire department as shown in the household-work relations map of
Tours. This is the space where we model agent movements.

The graph is transformed into a cellular graph using the same method as
Christophe Decoupigny but the cell size was reduced to 1 meter. The goal of this
reduction is to ensure accelerations and stops are realistic. In fact, with a 5 m cell, an
acceleration of 1 cell/second equals a Formula 1 racing car, and an acceleration of
half a cell complicates algorithms. In this case, the acceleration of these vehicles is 2
cells per second for those exceeding the speed limit who presumably have powerful
cars. For normal drivers, it is randomly between 1 and 2 cells per second, and for
slow drivers, the acceleration is 1 cell per second, for example, heavy trucks. This is
relatively consistent with the capabilities of modern vehicles.

The second consequence is that the vehicles occupy several cells; this does not
represent a major drawback, unlike heavy trucks, because we can then account for
their length. A vehicle is located by the head and end cell.

Mathis, Land Management Degree, Ecole Polytechnique de l’Université de Tours –

Aménagement Tours Department, 2008.
 Cellular Graphs, MAS and Congestion Modeling 321

Figure 14.1. Characteristic data used by the model (department of Indre-et-Loire)

14.1.2. Diagram of activities: a step toward the development of a schedule

The activity diagram for the agent defines his theoretical path for the
development of all occupations imposed on him or his household, such as work, or
chosen, such as sport. They constitute a list of localized “constraints” in the network
and determine the daily path of each agent.
322 Graphs and Networks

By activity, we mean the main occupation in a given place and time where the
duration includes waiting times before and after the activity. Tbis is localized in one
place and excludes the movement that is attached to it. Since the activity locations
are not identical, accomplishing these daily tasks requires travel.

14.1.3. Typology of possible agent activities

We retained a certain number of types of activities for which the location

depends on the characteristics of the individual. These activities are categorized by
order of priority going from an activity called “main” or “mandatory” to
“secondary” or “optional”2,3. Congestion simulation starts with two elements: the
generation of movements from the sum of daily schedules and their effect on the
transport network.

The activities used in these simulations were deliberately limited to the

workplace, shopping and household activities:
− Work: daily activity for active people, localized in a vertice in the diagram, has
a duration and schedules related to the social and occupational category of the
individual.
− Shopping: localized activity in one of the vertices of the diagram locally
closest to the previous activity enabling the “shopping” activity. Its duration is
randomly determined between a minimum and a maximum. Its priority is secondary.
− Home: the home is the starting and ending point of the daily cycle with the
possibility of a return in the middle of the cycle if time constraints allow it.

Several activities are attributed to a single individual who will, depending on his
schedules, build a diagram of “possibilities” in a day. There is no spatial or temporal
sequence information, but only a series of tasks to do during the day. The individual
must be “characterized” in the sense that each occupation must have a time and
place in order to become an activity in the individual’s schedule.

Creating a schedule requires the integration of travel time between each activity.
Based on the creation of an “empty” Dijkstra (algorithm of the shortest route
between different vertices in a diagram taking no account of possible congestion),
we have a theoretical knowledge of travel time. This is an optimal schedule. All
activity locations of agents are spatialized, but their accomplishment is also subject

2 The problem of the nature of tasks involved, their allocation to a member of the household
was described in the notable report: “Cohérence entre les politiques de transport et les
politiques d’aménagement – le cas de Mulhouse” research report, 2008 PREDIT-ADEME.
3 Laurent Guimas, ongoing thesis.
 Cellular Graphs, MAS and Congestion Modeling 323

to time constraints linked to the schedules of each activity, including starting and
ending times.

Knowing the schedule of individuals provides us with information on daily

movements, represented by their origin/destination pairs and times of
departure/times of arrival. We then have a time matrix of daily movements made up
of cycles in the transport diagram: departure from a starting point, travel through
several locations and return to the starting point. All these matrices make it possible
to determine global flows for all destination origins and by units of time.

From this matrix, the agents can carry out their travels in the network. We go
from the theoretical estimation of travel times to their evaluation by “experience”
according to the density of consecutive traffic in the routes used by all agents in a
section of the network at a certain period of time. This traffic density can trigger
knowledge for an agent leading to modifications of the path, modes of
transportation, sequence of tasks and, in the end, localization.

Agents with a limited rationality have the capacity for learning, enabling them to
widen the network used and the evaluation of each facet The implementation of the
decision process for each agent is conditioned by their non-satisfaction with the
route that they just took, as well as by the uncertainty linked to the unknown when
change occurs. The level of tolerance is based on being early or late4, and on the
route being longer or shorter. This all depends on the types and categories of agents
as well as their constraints.

Some uncertainly that cannot be overcome remains, because of the blind

behavior of agents who make these decisions individually, resulting in congestion in
the network. This uncertainty decreases with experience of the agent but can never
be eliminated.

14.1.4. Individual behavior mechanism: the daily scale

14.1.4.1. The choice of sequences

The sub-group of possible sequences is independent of any considerations linked
to the transport network. It can be stored by each agent.

Only the schedules enabling us to sequence all activities are retained. Then, for
each iteration, by associating each sequence with a time for the activity and travel,
each agent determines his own schedule.

4 [DEP 01]
324 Graphs and Networks

14.1.4.2. Sequences timed for the “best” course

Based on his belief, each agent will have to determine the route to take to reach
his objective: the one that will minimize total time in the transport network. This is
an “expected” value based on inaccurate knowledge by agents.

The different calculations of necessary minimum routes use two algorithms

developed for this purpose.

The first one involves the mode of road transport. It is based on the Dijkstra
algorithm, but in a modified version said to be “timed road Dijkstra” for integrating
the fact that travel time for a given edge depends on the time of day (discretized by
10 minute time steps) at which it is taken (traffic speed variation according to traffic
density).

The precision allowed makes it possible to overcome the traditional dichotomy

between rush hours and lowest hourly traffic, as well as to simplify: during a period
of time, the modification of traffic speed was similar throughout the network. On the
other hand, because of this method, the spatio-temporal heterogeneity of slowdown
intensity is taken into consideration.

The departure time from the household toward the first activity corresponds to a
shortest path calculation in a timed diagram with arrival time constraint.

14.2. Satisfaction and learning

What is the difference between intention and accomplishment leading to the

search for a new route that is better adapted?

14.2.1. The choice of an acceptable solution

Following the example of J.G. March and H.A. Simon [MAR 64], we will also
distinguish between “routine responses” and “responses obtained from the resolution
of a problem” assuming that the distinction depends on the shift perceived by
individuals between expected course and the course actually taken.

During this shift, two directions are available to the individual for the following
“day”: he either deems the perceived shift “unacceptable” and starts his decision
process all over again relying on new beliefs, or he considers it “acceptable” and
settles for making the same choices. In this last case, it is, in the end, the risk in
terms of uncertainty about a different path and the “cost of innovation” [MAR 64] in
 Cellular Graphs, MAS and Congestion Modeling 325

cognitive terms that dominates the choices of the individual who prefers to stay in a
situation that he feels is “known”.

As the consideration of limited rationality of agents authorizes a certain tolerance

from the agents in terms of their choice of path, we will assume that this difference
can be characterized as insignificant without being nil because of the gap between
the presumed traffic conditions and the real ones.

The learning matrix is translated into real behavior by the formulation of a

perceived time of travel integrating with the history of experiences. The last
experience modifies this perceived time. The weight of history and
instantaneousness in terms of perceived time is integrated by two weighing factors.
For an edge and time interval experienced by an agent, we obtain:
– Texp(t): perceived travel time of the edge according to the time interval for
agent X at the end of the current iteration.
– Teff(t): actual travel time (allocation module) of edge A the last time according
to the time information for agent X at the end of the allocation module.
– Texp(t-1): the perceived travel time at the end of the previous iteration.
– α and β of weighing factors.

β T exp ( t −1) + α Teff (t )

T exp ( t ) =
β +α

When the agent first explores an edge in a given time interval, the (t-1) perceived
travel time takes the value of the edge’s idle theoretical travel time.

In accordance with the literature on relative costs of an early or a late arrival

[DEP 01], this choice clearly puts the early arrivals into perspective by making a 10
minute difference or a 20% portion of the travel value acceptable compared to only
5 minutes or 10% respectively for the delay. If the delay or time saved are included
in this interval, then the agent is satisfied; otherwise there is a modification of the
schedule by which the agent arrives earlier – earlier departure or change in itinerary.
This solution is suggested by many authors who only consider a specific proportion
of dissatisfied people.5

The two forms of early dissatisfaction do not have the same consequences: the
delay implies a questioning of the itinerary, whereas being early reinforces this
itinerary but requires questioning of schedules associated with the movement in the

5 For example [NAG 92 ; NAG 98a; NAG 98b; MAT 08]

326 Graphs and Networks

schedule (modification which can have a series of consequences on the other

activities in the schedule, making the modification significant).

The solution that seems the most “well balanced between the actual travel time
and what is perceived with the previous iteration” is α=0.5 and β=0.5, i.e. learning
characterized by a memory weight equal to that of the actual experience.
Convergence is then faster. In fact, if we can assume that at the end of the first
iteration, the least satisfied agents are those who have used the main roads during
the busiest times, then these roads will be automatically off-loaded of some of their
users and the drivers still using them will see an improvement in their travel speed.
During the different consecutive daily iterations, the slowdown phenomena will lean
toward a “user” state of balance in which some agents are satisfied.

In the model, each agent has his own very limited knowledge of the network and
does not share it with any other agent. The agents implement their personal network
with the actual valuation of explored edges and only these edges. The table is made
up of 96 steps in time, for a total of approximately 166,356 agents × 5,435 edges ×
96 time intervals.

It is the actual principle of MAS, autonomy and emergence of a global structure

from choices and local actions, that is the main difference from traditional
approaches.

14.2.2. Collective learning and convergence of the model toward a balanced

solution

The model converges in such a way that the number of unsatisfied agents
asymptotically leans toward a limit as indicated in Figure 14.2.

Learning is then the result of the non-satisfaction of the conditions described

above. However, different agents may be unhappy from one iteration to the other,
because their discontent is the result of local conditions encountered, and more
precisely, of the importance of the flow using a particular edge at a particular time.

The agents who exit the network first are those who contribute to its total
congestion. They are reintroduced during the iterations.

Clearly, this implies that the network is unable to support the flows determined
by the allocation of agents according to the optimal solution in terms of time-
distance, for each of them. As we will see below, the global optimum is not equal to
the sum of individual optima determined by algorithms which do not take account of
the different types of congestion.
 Cellular Graphs, MAS and Congestion Modeling 327

The greater the number of unsatisfied agents, the more the agents tend to explore
the graph, and consequently, the agents have more difficulty in finding an acceptable
solution.

Figure 14.2. Evolution of agent’s satisfaction

14.2.3. Examination of the transport network

Figure 14.3. Modification of routes due to learning

328 Graphs and Networks

Figure 14.4. Modification of routes in tour center due to learning

14.2.3.1. Agent knowledge: the different household-work roads used

The three maps in Figure 14.4 show an example of how an agent gains
knowledge of the area between his household in the west of the department and his
work south of central Tours in the north and Cher in the south. The maps display the
difference between the route during the first iteration, the theoretical “shortest” in
time and consecutive routes attempted by the agent to decrease lost time because of
congestion. Memorizing these routes and the time they take makes it possible to
constitute the network experienced, i.e. spatial and temporal memory of the agent,
his understanding of the network in the Latin sense of the term “cum scire”.

The different sources of congestion with which the agents are confronted during
their daily travels have still to be explained in detail.

14.3. Local congestion

The agent’s learning process is triggered by the observation of an unsatisfying

difference between the theoretical time calculated without taking account of the
other users, and “real” time observed during the “movement”.

Here, we are encountering a well-known problem: that of the absence of an

invisible hand in the sense of Adam Smith, where the global optimum is not equal to
the sum of individual optima. This is the result of the blind behavior of agents, who
all begin by using the optimum solution in a short space of time, thus saturating an
infrastructure, which has a limited capacity.
 Cellular Graphs, MAS and Congestion Modeling 329

But, in our opinion, this problem with the capacity of a graph is not well defined,
in the sense that this capacity is measured in edges by methods from the Ford-
Fulkerson algorithm, because in the traditional theory of graphs, the peaks are
neutral. Even in the dynamic allocation models, the problem of intersections is
underestimated. At best, they are managed by queue algorithms and average
throughputs. At worst, they are not taken into consideration at all.

The use of cellular graphs and reactive agents makes is possible to understand
the formation of part of a congestion front on an edge that the traditional theory
already confirmed6 as shown in Figure 14.32, but especially to model the capacity of
different types of peaks and their saturation mechanisms. These different sources of
congestion explain global congestion and the great difficulty of traditional
approaches in accounting for it.

14.3.1. The peaks represent different types of intersections

14.3.1.1. Street crossings and traffic circles

The network is represented by a graph that we can call quasi-plane in a misuse of
language, in the sense that all street, avenue and boulevard crossings are plane.

These intersections currently fall into two categories. The first is the traditional
intersections where traffic axes cross based on different methods:simple right-of-
way, give way and traffic lights. These, unfortunately, tend to be auto-blocking. The
second is traffic circles for which the priority (in France) is to the left, which makes
them auto-unloading since, in contrast with the first category, the priority goes to the
one who wants to exit the traffic circle.

Unfortunately, in practice, it is difficult to simulate the use of intersections

because the data on temporal calibration of traffic lights and priorities (stop signs)
does not exist or is unpublished, whereas we have abundant information on edges.

The problem is primarily linked to the absence of homogenous databases on a

space involving intersection priorities (number of lines, light traffic times, lane angle
of incidence, etc). It is therefore impossible to correctly integrate weight of
intersections into travel times and choices of itineraries. This is even more
problematic as the capacities and mode of clearance of intersections are not constant
and depend on their inherent properties. In fact, the partial integration of the
properties of intersections could introduce a bias by triggering a reinforcement of
performances relating to certain itineraries to the detriment of others. That is why it

6 [MAE 06]
330 Graphs and Networks

is sometimes simpler to model intersections with the help of a traffic circle7 instead
of a crossing. But we will demonstrate that this simplification is not neutral, since it
introduces two systematically opposed errors, mainly in a dense, almost saturated,
urban space without being able to predict the direction of the result, or its constancy
or intensity.

14.3.2. The emergence of congestion fronts on edges

The dynamic formation of a congestion front appears with the evolution of the
concentration of vehicles in a road section.

The diagram in Figure 14.5 is a speed “time-space diagram” on an edge. It is

made up of N cells (5 m cells for visualization or 5000 m by y seconds). The speed
of vehicles over occupied cells is indicated by the shading: the darker the shading,
the slower the speed. The gradient of the first series of occupied cells indicates the
speed of the first vehicles in the simulation.

As Jean Baptiste Buguellou writes, the previous illustration retranscribing “the

emergence of a congestion front in the edges” is caused by the simple interactions
between vehicles. Driver actions on an edge are mainly based on rules of pursuit.
Consequently in heavy traffic, a slight fluctuation of the pace, of a vehicle at time T
and a given location, triggers a chain reaction on the vehicles behind. Because of the
small space between vehicles, this chain reaction is realized by a generalized
adjustment of the instantaneous speeds of each vehicle and in fine by the formation
of a slowdown front. Depending on the density of vehicles in front, this front can
become durable.

The driver who causes this phenomenon is in essence microscopic8. The

microscopic action of a driver is what forms the initial conditions of the mobile
congestion process. The scope and persistence of a congestion front then depend on
the environment in which the driver evolves and notably the configuration of the
environment beforehand.

7 [CHO 03].
8 The macroscopic model cannot retranscribe such phenomena because it does not integrate
the behavioral dimension of the driver and the variations in time of the vehicle aspects. The
consideration of this phenomenon could only be anticipated by these models (making it more
complex) and not the result of the initial model hypotheses.
 Cellular Graphs, MAS and Congestion Modeling 331

Figure 14.5. Evolution of traffic density and traveling speeds in a

road section: dynamic formation of a front

In this simulation, the drivers are separated into three classifications: the careful,
the normal and sportsmen, categorized by acceleration, stopping and reaction speed.
This illustrates perfectly the impact of the various behaviors, leading to
heterogeneity of vehicle movement speeds, on the traffic dynamic.

The heterogeneity of behaviors increases the impact of interactions by

modulating the stopping speed force and evaluation of security distances for each
driver.

In this way, the simple presence of three driver categories triggers the
implementation of a congestion front on the edge. This front, once established, is
particularly marked in space (400 m). Over time, it tends to widen9, grouping an
increasing number of vehicles together. The phenomenon caused by this simulation
corresponds explicitly to the mobile congestion: the front is relatively stable in space
but the vehicles making up this front keep moving in the network. The front’s
stability is then the result of the connection between the number of exiting vehicles
and the number of entering vehicles.

9 If the traffic conditions and, particularly, the inter-vehicle time do not change, the
congestion front will tend to keep widening until it fills up the road section.
332 Graphs and Networks

This phenomenon corresponds perfectly to stop-and-go; the vehicles entering the

front slow down and move at slow speed. Over time, the vehicle continues to move
forward in the congestion front and can exit and accelerate to reach a speed that is
almost free. We can also observe the formation of small waves of congestion after
the main front.

This congestion front develops with the introduction of heavy trucks. In fact, the
simultaneous presence of different types of vehicles with non-identical acceleration,
stopping and speed characteristics thus increases the heterogeneity of the flow
regularity. The consideration of heavy trucks in the same simulation conditions only
reinforces the congestion front and increases its size.

The stronger the heterogeneity of traffic, the greater the possibility that the
phenomenon will occur. It is generally seen at the level of road edges. Changes in
lanes increase the probability of the stop and go phenomenon. The irregularity of
flow caused by the heterogeneity of behaviors and vehicle categories, multiplied by
lane changes (and arrivals at interchanges) creates the initial conditions for the
formation of a congestion front. Logically, the probability increases when the
concentration of vehicles is greater, when the space between vehicles increases this
probability is zero, and when it is shorter, then the probability is stronger.

Figure 14.6. Evolution of vehicle concentration and speed over a road section
integrating the presence of heavy trucks
 Cellular Graphs, MAS and Congestion Modeling 333

The driver’s behavior on experiencing this wave can then amplify it (a sort of
resonance) or decrease it. The inter-vehicle proximity makes amplification
increasingly probable.

The wave stops when the distance between two vehicles is long enough to
decrease its amplitude, or when consecutive drivers modulate their behavior
according to the context without applying an automatic speed adjustment
corresponding to density; thus increasing the possibility of temporarily reducing
inter-vehicle space.

With these space-time diagrams, we can observe results that were already known
(Sven Maerivoet 2006) constituting an additional element of reinforcement of the
model used. These congestion fronts partially contribute to the dispersion of
observations for throughput-speed type curves.

14.3.3. Intersection modeling

The most significant contribution from Jean-Baptiste Buguellou’s thesis involves

the capacity of intersections: different types of crossings and traffic circles.

In these different cases, the increase of the complexity involves the programming
of vehicle movements: acceleration and deceleration rules, respect of priorities,
stopping distances, straight trajectories, right or left turns and their relative
proportions, as well as the complexity of the intersection itself.

In fact, in our models, there is no average speed, the driving agents accelerate
and brake according to their environment as well as their own characters. Thus they
constitute several different groups, even with distractions (as we will see) and
varying degrees of respect for driving rules such as the use of indicators.

The only absolute rule observed is to arrive at the destination without an

accident. This implies the absolute rule that two vehicles are not in the same square,
thus obeying rules of speed and limit stopping, i.e. stopping capacities indicated by
the manufacturer, inter-vehicle distances and maximum speed in an axis such that
the vehicle can obey end of edge constraints – such as a stop sign or a stop light. In
summary, the drivers behave like model citizens! There is still an overestimation of
throughputs compared to reality.

14.3.3.1. Modeling of the road intersection

The transformation of a peak in a traditional intersection is located in the line of
Christophe Decoupigny’s applications: passage from a four degree peak to a summit
digraph followed by definition of the ante primal representing intersection cells.
334 Graphs and Networks

We have simplified the graph to focus only on the flows by representing the
flows and their priorities (see Figure 14.7): priority 1: the flow has priority over all
the other cases of a trajectory turning right, priority 2: the flow must give way, the
case of a straight trajectory on a non-priority axis, and priority 3: the case of a
trajectory turning left still in the simple case of a cross-threaded crossing without
stop lights or general right-of-way, etc.

Figure 14.7. Example of crossing with vehicle flows, straight trajectories and no left or right
turns with priority because he must give way to vehicles in the
other axis coming from the right

14.3.3.2. Traffic circles

Even though the traffic circle in [CHO 03] is created by direct cell assembly:

Figure 14.8. Structure of traffic circles [CHO 03]

 Cellular Graphs, MAS and Congestion Modeling 335

The technique used by Jean-Baptiste Buguellou is slightly different from that of

Christophe Decoupigny in the sense that it only uses right connectivity.

This technique doubles the four peaks of the digraph introducing a space in the
center and limiting movements in the diagonal edges, connecting the peaks of two
orthogonal peaks in the reverse trigonometric direction to respect the left priority.
To create the number of desired cells, it is necessary to add the number of peaks
must be of the order of two, according to the desired size of the traffic circle because
of duality.

Figure 14.9. Traffic circle construction model

We can observe that if we add the central edges (inside the traffic circle), this
form corresponds exactly to an interchange (planar problem). The number of degree
2 vertices (small circles) defines the size of the intersection.
16

4 12

Figure 14.10. Movements close to and in a traffic circle

336 Graphs and Networks

14.3.4. Limited peak capacity: crossings and traffic circles

14.3.4.1. The capacity of traffic circles

First, consider a specific case where all the drivers go straight ahead and there
are no right or left turns. The sum of free throughputs is measured from a constant
inter-vehicle space. If all vehicles are separated by two seconds, we have a
theoretical throughput by edge of 30 vehicles, or 120 for all incident edges.

Figure 14.11. Average traffic circle flow according to the intervals of vehicle generation
(the time between the appearance of two vehicles)

Figure 14.11 summarizes the series of simulations in traffic circles. The

simulations are, relatively, simpler than those of crossings, since in a traffic circle,
by programmed rule, a driver cannot enter if he does not have enough space.
Consequently, he does not block the traffic circle. The flows are therefore in the
same direction, and the exit is on the right and it has priority. We can clearly see
however, that the average nodal throughput goes through a maximum for a
generation of vehicles every two seconds, i.e. 1,800 vehicles per hour per axis, or
120 vehicles per minute in total, but that the capacity of the traffic circle is less than
80 v/mn maximum ,i.e. only two thirds of the capacity of entering edges.
 Cellular Graphs, MAS and Congestion Modeling 337

This is summarized in Figure 14.11 with free flow up to the maximum of one
vehicle every three seconds in each axis, after which congestion occurs and there is a
drop in throughput.

The nodal throughput in Figure 14.12 is calculated for the most simple flows,
with no right or left turns. When different trajectories are introduced, this diagram is
no longer completely valid; these are Qcg and Tcg movements. This simple
introduction of trajectories with different degrees of priority modifies the capacity of
the traffic circle through the placement of the critical throughput, it is one of the
conclusions that make static allocation models ineffective.

Figure 14.12. The simulated nodal throughput of a traffic circle in relation to the sum of
theoretical edge throughputs10 Qcg: critical throughput of the traffic circle11 Tcg: Inter
vehicle time of critical generation for a traffic circle

10 We consider here the theoretical throughput of an edge through the constant preservation
of the time interval separating each generation of vehicles in the network. For a three second
interval generation, the throughput per minute is necessarily 20 vehicles per edge, the space
between vehicles representing a distance-time of 3 seconds from the speed of the vehicle.
11 This critical throughput is a function of the number of exits used by drivers. Where all
trajectories focus on a single exit, the maximum throughput is of approximately 20 vehicles
per minute (in the context of simulation hypotheses). When the four exits are used, the critical
throughput is then of 80 vehicles per minute.
338 Graphs and Networks

In the more favorable case, that of the traffic circle, as we can observe, the
maximum capacity of the intersection is lower than the sum of the capacities of
entering edges. When the frequency of emergence decreases, this difference
decreases until it almost disappears, but then throughputs are clearly lower than the
maximum throughput.

This aspect is never taken into consideration by traditional models. It is true that
this is relatively insignificant for inter-urban traffic where traffic circles are large,
relatively widely spaced and where the main axes have priority over crossings.

On the other hand, in the case of urban space where the length of edges is short
and peaks are numerous, this limitation reaches its full force because summit
congestion is gradually distributed. MAS models in cell graphs integrate this aspect
by construction, and therefore are not biased like more traditional models.

14.3.4.2. Crossing capacities, flow complexity and consequent complexity of their

systems
The traffic circles are relatively simple: with left priority at entrance they are
self-unloading as long as the exiting edge is not saturated. On the other hand,
crossings with right-of-way are auto-blocking regardless of the direction of vehicles
and the existence of the trajectory turning left implies the blocking of one of the two
lanes of the axis involved.

We have chosen to discuss only the example of a traditional intersection with

right-of-way with flows that cross in a perpendicular way for the four straight
destinations with no right or left turns. The flows have priority at intersection
entrance but must give way to the flow coming from their right side.

We must add that the drivers are polite, meaning that they will not enter the
intersection if there is a possibility of blocking. This is a strong hypothesis because
we know by experience that it is not always, and perhaps only rarely, the case. The
model presented has optimistic capacities, the driver behavior is very favorable,
efficient and polite.

The traffic dynamic seems extremely chaotic in all simulations. In the

configuration of perpendicular flow, each origin-destination pair dominates for short
periods. Theoretically, the incident edge flows should be unstable but the
fluctuations should return weaker amplitudes.
 Cellular Graphs, MAS and Congestion Modeling 339

The average throughput of incident edges is slightly higher than 12 vehicles, but
is very irregular.

The immediate domination of a flow is mainly due to the first rule of not
blocking the intersection; the oscillations are then a local function at time t of spaces
between vehicles. At no time, however, do we witness the formation of a saturated
stationary system – that is, a system where the flow of at least one edge is almost
zero over a long period – mirroring the traffic dynamic measured in a traffic circle.

Figure 14.13. Throughput measure of incident edges per minute according to traffic density
340 Graphs and Networks

Figure 14.14. Throughput measure of incident edges per minute according to traffic density

The nodal throughput also seems chaotic, illustrating the interferences between
the different trajectories, and finally the absence of explicit domination of a flow
(non-stationary situation) when traffic density is strong. From the time when the
generation of the vehicle interval is less than six seconds, the vertice throughput
fluctuates between 28 and 58 vehicles per minute. When this generation interval is
higher than or equal to six seconds, the vertice system becomes stationary and non-
saturated, and returns a constant throughput of approximately 40 vehicles per
minute. Beyond this generation interval, the vertice throughput decreases with the
reduction of traffic density.
 Cellular Graphs, MAS and Congestion Modeling 341

Figure 14.15. Vertice throughput per minute according to traffic density

We can observe obvious differences in relation to the measures taken at the

traffic circle.

The traffic circle throughput was stable, regardless of traffic density. The stages
represented the type of stationary system of the traffic dynamic internal to the
intersection. In the context of an intersection based on four rights-of-way, the
chaotic phase seen with incident edges can be directly reported in the nodal
throughput without necessarily having compensation effects In the context of a
right-of-way, it is obvious that interferences between the trajectories of vehicles
influence the nodal throughput despite the rule of not blocking.

The maximum throughput logged is clearly lower for an intersection based on

four rights-of-way. In fact, the latter is of approximately 58 vehicles per minute
whereas for a traffic circle, we observed constant nodal throughputs of
approximately 80 vehicles per minute when traffic density nears TCG.

It appears that when traffic density is relatively high, all intersection exits cannot
have a maximum flow.

Generally speaking, and from the series of simulations, the constraints inherent
in priority rules, multiplied by the form of the intersection (trajectory crossing in the
center of the intersection and no longer just at the extremities of an edge) implies a
limited functional crossing capacity, and in any case, clearly lower than that of a
traffic circle.
342 Graphs and Networks

The traffic dynamic in an intersection based on four rights-of-way favors the

formation of queues in incident edges. The simulations register a throughput peak of
approximately 60 vehicles per minute for a generation interval of three seconds.
Based on this interval, free edge throughputs are of 20 vehicles per minute for a sum
of 80 vehicles for all incident edges. We can then observe a difference of 20 vehicles
per minute appearing as queues in the intersection.

Similarly, for inter-vehicle space of two seconds between each vehicle, the sum
of free throughputs in incident edges is theoretically 120 vehicles per minute. The
maximum throughput registered is then 55 vehicles. In this configuration, only one
in two vehicles can go through the intersection.

The crossing constitutes a bottleneck in edge flows.

Figure 14.16. Average throughput of a vertice for four rights-of-way over one hour for a
three second interval and trajectory variation

We should note that the complication of trajectories and notably the proportion
of left turns will influence the maximum flow of the vertice by decreasing it in
comparison with the example given, because in fact, the flows now only have one
level three priority, i.e. they must give way twice in a row.

A total right turn in the four axes is the only hypothesis that provides good
throughput although not at 100% of edge capacity because we must then take into
account the necessary slowdown for a tight 90° turn implying lower speed than the
50 kmh normally authorized in the city.
 Cellular Graphs, MAS and Congestion Modeling 343

Figure 14.17. Comparing the three types of crossings

For constant or dense traffic, i.e. with time intervals lower than seven seconds
between vehicle arrivals, the traffic circles have a higher throughput, and are clearly
superior for intervals of four and three seconds, where the capacity is almost
doubled.

For an identical circulation context (vehicle trajectories, traffic density), the

modes of operation of the three types of intersections are specific, and if we
integrate the throughput of each edge, the distinction between each intersection
increases even more. The properties of intersections, and notably the status of the
priority, define a specific traffic flow and consequently directly influence the flow
capacity of incident edges at the intersection.

It is, in fact, very difficult for traditional models to create this distinction
between the intersections.

14.3.4.3. The human factor

The hardware structure and regulatory attributes of the peak are not the only
factors. Although they are essential, the human factor is just as important. This is
shown by the following results simulated from a traffic circle, taking into account
the use of signaling by indicator, allowing a driver to “anticipate the behavior” of
the vehicle coming from the left. This, and the use of the hypothesis of politeness
are important: the driver who previously did not enter the roundabout when there
was a possibility of his blocking it will do so, if the one coming from the left signals
that he is turning right.
344 Graphs and Networks

Average throughput of the vertice according to starting intervals12

In fact, the nodal throughput of this form of intersection follows four distinct
phases.

For all right trajectories13

The first phase is illustrated in Figure 14.18.

Figure 14.18. Flow of incident sections of a traffic circle for only right trajectories

12 The road environment corresponds to the generation hypotheses formulated for the
simulation of strictly perpendicular flows, or toward the four possible exits of the intersection.
13 Please refer to the annexes for intermediate evolutions of the throughput of edge
throughputs (33% and 66% the proportion of drivers using their indicators).
 Cellular Graphs, MAS and Congestion Modeling 345

In another case, signaling enables the driving agents to enter the intersection.
The system is then chaotic and maintained.

For all straight trajectories

The second phase is illustrated in Figure 14.19.

Figure 14.19. Flow of incident sections of a traffic circle for only straight trajectories

Without signaling, the crossing system is completely chaotic. With signaling,

two axes become dominant.
346 Graphs and Networks

For all left trajectories

The third phase is illustrated in Figure 14.20.

Figure 14.20. Flow of incident sections of a traffic circle for only left trajectories
 Cellular Graphs, MAS and Congestion Modeling 347

The worst case is 100% of left turns without indicators; the flows only have a
level three priority. Without priority, one of them dominates, taking advantage of the
polite nature of drivers, whereas with signaling, nobody dominates. Chaos is strong,
however, benefiting a relatively dominant flow.

For all mixed trajectories

The fourth phase is illustrated in Figure 14.21 (equiprobability between each
trajectory).

Figure 14.21. Flow of incident sections of a traffic circle for mixed trajectories
348 Graphs and Networks

In the case of equiprobable mixed trajectories, signaling limits some movements

without eliminating the chaotic character of flows.

It is perfectly clear from these few examples that human behavior adds to the
physical structure of the peak to limit, or alternatively, facilitate throughput, an
aspect that macro simulation models obviously cannot address.

Microscopic models alone (in particular MAS), based on a finite description of

the transport network (cellular graph) can show the flow disparities in intersections.
These disparities are based on environmental contexts specific to each moment in
real time. A simple modification of this context can result in a flow mode that is
radically different at the vertice and each incident edge in particular (chaotic
dimension when traffic is very dense).

Consequently, it appears that to model daily mobilities, particularly urban

mobilities, these forms of modeling are the only ones that can be used. In fact, this
type of mobility implies a finite modeling of movements and their traveling times
(see [MAT 08]). This subtlety can only be achieved by modeling crossings. The
capacity will then be the result of this modeling, which will be mechanically taken
into account. This can only be done from this type of modeling, because the
capacities of edges vary according to their properties and environmental contexts.

14.3.4.4. Crossing chains

The preceding elements in edges and crossings simulate congestion at a micro
spatial scale. What about a more meso-spatial scale in which we consider a series of
edges and crossings, and notably when they are coordinated to create a “green
wave” that can only work in one direction?

The problem is as follows: a road uses a series of intersections where the stop
lights are either not synchronized and may be random, or they are synchronized to
allow a driver to go through them consecutively without stopping. In this case, we
speak of synchronized stop lights with a gap designed to allow vehicles to cross the
distance between two stop lights at the legal speed and be able to cross the second
and subsequent without having to stop. But since it is easier, and even vital in
certain cases, that the cross line has green lights on both sides at the same time, a
green wave can only be effective in a single direction, favoring vehicle distribution
only from a point of origin and in a determined direction.

This raises two problems: that of the gap between stop lights in the direction of
the green wave and that of cross flows, notably when they increase or decrease the
flow of the axis involved.
 Cellular Graphs, MAS and Congestion Modeling 349

14.3.4.4.1 The gap between stop lights

Consider the diagram in Figure 14.22 and the following hypotheses.

Figure 14.22. Structure of modeled axis

All the vertices have a length of 200 for a maximum capacity of 40 vehicles.

The maximum speed is 14 m/s or 14 seconds to cross without slowing down.

The drivers are separated into three classifications to vary their capacity of
acceleration, stopping and the maximum adapted speed.

The vehicles appear at outside vertices (vertices 1 and 7).

The vehicles enter the network when the available space on the vertice is
sufficient (a vehicle every two seconds if the flow is not restricted by traffic lights).

The drivers all go straight, to arrive at vertice 1 the destination of the vehicle is at
vertice 7 and conversely for arrivals at vertice 7.

In this modeling, all green lights last for 30 seconds (on the axis and
perpendicular to the axis) from vertice 7 to vertice 1, and the temporal gap between
lights is 15 seconds, the minimum traveling time of an edge.

Figure 14.23. Vehicle flow in edges between vertice 1 and vertice 7

The far left arrow (see Figures 14.23 and 14.24) shows that the vehicles that
initially start first can cross all the intersections without stopping. The last ones are
the only ones who cannot cross the first stop light. The temporal gap of 15 seconds
allows most vehicles who crossed vertice 6 to cross all the intersections. Some
350 Graphs and Networks

drivers must stop at the last two stop lights, but this represents very few vehicles
(three in total).

Figure 14.24. Vehicle flow in edges between vertice 7 and vertice 1

Figure 14.25. Close up of the flow of vehicles on the arcs between node 7 and node 1
 Cellular Graphs, MAS and Congestion Modeling 351

In total, 29 vehicles crossed the first intersection for a green light lasting 60
seconds, one vehicle every two seconds, the normal speed of a vehicle in an edge.
This throughput is necessarily not constant, otherwise queues would not form in the
intersections and these queues would not increase. The difference in throughput on
vertices in the same period is what leads to the formation of groups at stop lights.

The zoom makes it possible to see the movement of vehicles. The time gap
theoretically enables the passage of several intersections without stopping, but it has
a negative effect: the reduction of the queue in the following intersection is not
necessarily effective when the first vehicle is at the end of the intersection (in the
previous vertice). It consequently involves the slowdown of the vehicle (5 m/s speed
in the speed log file or 18 km/h).

It appears that this form of synchronization is relevant for the axes not presenting
dense traffic, or conversely, for axes with only few perpendicular entrances (limiting
the number of arrivals on the axis and consequently the queues).

We can observe that the successions of crossings in an axis, with a

synchronization of stop lights with gap, facilitate the flow in a direction. This creates
a situation partially resembling that of an axis, with the appearance of congestion
fronts which are created here by the stop lights or slowdowns before or after a red or
yellow light. The drivers able to adapt their speed to the synchronization have a
chance of crossing all of them, as long as they are not slowed down by the
acceleration of other vehicles when the stop light turns green. On the reverse side of
the green wave, the flow is not favored because synchronizing with the rhythm of
stop lights is very difficult.

If vehicles leave the flow by turning left or right without slowing down or
blocking a lane, this will improve the flow. In contrast, the arrival of vehicles from
cross streets can only slow down the flow by increasing the heterogeneity, and
decrease inter-vehicle distances, triggering an amplification of stop and go reactions
and tendencies.

14.3.5. In conclusion on crossings

The use of cellular graphs and reactive agents enables us to understand the
formation of a congestion front in an edge on one side that confirms the traditional
theory [MAE 06]. It especially enables us to model the capacity of the different
types of peaks and the mechanisms of saturation (the absence of intersections in the
traffic dynamic is also the result of problems in formulating an intrinsic physical
capacity of intersections: each intersection has its own (and especially not constant)
capacity according to vehicle trajectories, traffic density, driver behavior, etc.). The
352 Graphs and Networks

advantage of cellular graphs multiplied by a (human) microscopic simulation of the

movement makes it possible to directly integrate these factors (they form the
movement environment) without first considering a theoretical maximum capacity,
which is difficult to evaluate since it is not constant. Since the nodal capacity
illustrated by nodal throughputs is not constant, the constraint determination
algorithms are ineffective, unless we make them dynamic, for calculating the best
roads with non-constant nodal constraints. This aspect is one of the main
contributions of the thesis by Jean-Baptiste Buguellou [BUG 09]. These different
sources of local congestions explain global congestion and the major problem of
traditional approaches to investigating it.

14.4. From microscopic actions to macroscopic variables14 a global validation

test

14.4.1. The appropriateness of the model with traditional throughput-speed,

density-speed and throughput-density curves

The allocation model uses microscopic actions and decisions by agents, that is,
drivers in their road environment. However, this type of model must be validated
because of the simple fact of the multiplication of possible actions and situations.
We then choose to validate the model parameters from the capacity of the model to
transcribe the macroscopic efficiency variables representing a vehicle flow, all the
more so because they initially come from empirical considerations. In fact, the
empirical measures illustrate explicit relations between the three macroscopic
variables of a flow of vehicles including speed, throughput and density of vehicles
(or concentration).

But beyond, these different elements of data, throughput-speed, density-speed

and throughput-density relations, are basically traditional top-down modelings and,
in particular, four-step models.

Finding the fundamental relations in fine of traditional models from simple

individual behavior rules shows the validity of the emergence process and at a
minimum is a reinforcement of Popper’s validity hypothesis.

14 The following development on the throughput/speed curve, congestion fronts and the
capacity of intersections as well as their different systems are taken from the thesis from Jean-
Baptiste Buguellou [BUG 09].
 Cellular Graphs, MAS and Congestion Modeling 353

Figure 14.26. Fundamental diagram pattern [ELB 05]

The first form of Greenshields (1935) returns a perfect parabola. Consecutive

empirical observations make this perfect image more complex and correct it to
formulate three phases in the throughput-speed relation: an uncongested phase, a
transitory state and finally a congested phase characterizing the simultaneous
decrease of throughput and speed.
80

70
104
60 18 13
7
39 12
50
Km / H

40

30

20
51
10

0
0 400 800 1200 1600 2000 2400

Veh / H

Figure 14.27. The Greenshields (left) and Hall et al. (right) curve
354 Graphs and Networks

Generally, the literature gradually moves away from the perfect Greenshields
image, and especially emphasizes the problem of the transitory phase that is difficult
to model, and chaotic to say the least, with a branch point between a fluid and
saturated system. It perfectly illustrates the human and unpredictable character of the
traffic dynamic during critical situations, whether it be in a congestion front or at an
intersection.

The calibration of the allocation module and the different parameters linked to
acceleration, deceleration and the inter-vehicle space management necessitate a
return to the relevance of results obtained by the model according to macroscopic
operation traffic rules. The validity of the allocation module itself is then measured
in relation to the aggregation of microscopic behaviors with proven macroscopic
relations, such as with validation of behavioral driver parameters.

Figure 14.28. Throughput – speed curve obtained from the allocation module

The resulting curve corresponds relatively well to the one proposed by Hall et al.
(1992)15 and goes back to a compliant image in terms of the relation of throughput
and speed, clearly highlighting the three zones of the curve with a transition
spreading over a large speed range.

15 Note that the Hall curve now refers to HCMs (Highway Capacity Manual) of the
Committee on Highway Capacity and Quality of Service of the U.S. Transportation Research
Board.
 Cellular Graphs, MAS and Congestion Modeling 355

Figure 14.29. Density-speed curve obtained from the allocation module by

releasing space constraints

The density-speed curve corresponds to that of the fundamental Elbers diagram

(Figure 14.26).

Figure 14.30. Greenshields proposition (Bourrel, p.43) and current (Maerivoet, p.49)

The different later works complete the perfect parabola of Greenshields, leading
to a slight deterioration. The critical throughput is more quickly reached than
Greenshields proposed. Similarly, the transition phase between fluidity and
congestion returns a chaotic image (in the image of the throughput-speed curve).

The results from the allocation module correspond relatively well to the general
form of the throughput-density relation. However, the gradient in the first phase is
more important than the macroscopic relation prefigures. We should note that this
curve deteriorates when considering heavy trucks because of differences in speed
and possible passing problems.
356 Graphs and Networks

Figure 14.31. Throughput-density curve obtained from the In Buguellou allocation module

14.4.2. The distribution of traffic density over time

Figure 14.32 shows the traffic density curve obtained in Tours from 6:30 am to
10:00 pm.

Figure 14.32. Number of agents in the network according to time

 Cellular Graphs, MAS and Congestion Modeling 357

The curve has a perfectly traditional form with a morning rush hour peak that is
much higher than the afternoon one. We should note that this hourly distribution is
naturally influenced by the hours of the start of work activity and that these
beginning hours are not well known. The estimates16 were considered as non-
modifiable in the short term.

Figure 14.33. Difference between the number of agents per time interval of iteration 1 and
the last iteration

This distribution curve is also the product of knowledge of agents’ who seeing
the congestion leave earlier and/or use other roads.

14.4.3. The measure of lost transport time by agents because of congestion

During the first iteration before gaining knowledge of the conditions, agents,
considering the network from a theoretical and narrow point of view, i.e.
independently from actions of other agents, each use the fastest road. This result is
congestive, hence the search for other satisfactory itineraries. At the first iteration,
the sum of distances traveled results in 3,560,571 km traveled with total expected
time of 3,241,555 minutes, whereas actual time will be 4,469,242 minutes or
1,227,687 minutes more, representing only 7.38 minutes per agent. This is relatively

16 INSEE Survey on Work and Time Schedule.

358 Graphs and Networks

low and normal in a space where congestion is very localized to large axes,
especially in the middle of the urban community.

During the 18th iteration, for example, the global distance traveled is at 1/3000th
equal, whereas the actual time decreased by 224,407 minutes and the difference
between expected time and actual average time goes down to 3.84 minutes,which is
within the acceptance zone. Knowledge made it possible to globally gain
approximately 5% of travel time, and especially made it possible to plan all the
movements. If we consider an hourly cost of only 7€/h, the global gain resulting
from knowledge is 40,180€ per day, or more than 8 million€ per year!

We have here an interesting result going back, with other methods, to the second
Wardrop principle.

A balance appears, but that is not a function of an optimal solution for each
agent. It only represents an acceptable and satisfying short-term solution in a precise
context for most agents. Long term, the problem is different, relocalization becomes
possible.

Because of global variables, the MAS model seems consistent with observations
which are upstream of traditional models and enable their calibration. In the case of
MAS-cellular models, these characteristics are the product of the model function,
which is satisfying because if global results match observed reality, we can consider
that the model acquires a higher status, and that is it reinforced in the Karl Popper
sense.

14.4.4. Spatial validation

One last test can be done: the spatial test to analyze the coherence of the model’s
results with the observations done on the field.

The most specific characteristic of the map shown in Figure 14.34 is the north-
south axis between Loches in the south, Amboise on the Loire and Château Renaud
to the north. This axis only connects small towns but enables trucks going to Paris to
avoid Tours and a few tolls by using a shorter route than the highway. The
consequence is automatic, since the speed of trucks is legally lower than that of cars.
This causes slowdowns and congestion fronts because of on the one hand the
relative narrowness of this axis and, on the other hand, because of the absence of
routes avoiding this axis. Specific vehicle drivers are then slowed down by trucks
without much opportunity to pass.
 Cellular Graphs, MAS and Congestion Modeling 359

This is a locally known phenomenon confirming what theoretical simulation

predicted and what the model simulates. This corroboration is interesting because it
is located at the meso local level between local micro and global.

Figure 14.34. Example of accessibility in Indre-et-Loire from Loche

14.5. Conclusion

The goal of the use of an MAS on cellular graph in a “bottom up” approach was
to exceed a certain number of limits which clashed with the traditional top down
four-step model approaches in transport modeling.

We can observe that this method makes it possible to start with agents or even
households, and to simulate their daily movement cycles according to agendas with
spatial precision and especially time precision, which the more traditional methods
cannot match.
360 Graphs and Networks

In addition, this method highlights the non-applicability of the Ford-Fulkerson

theorem in an urban space which assumes the peaks are neutral and calculates the
maximum capacity by a cross section in the edges, implying that congestions caused
by intersections are not taken into account although they are the most significant.

Jean-Baptiste Buguellou demonstrates well the existence of micro local

congestions in the edges, for example, intersections and chains of intersections with
maximum capacities clearly lower than edge capacities, and in addition, these
capacities are very variable for the different axes according to relative flows and
driver trajectories.

The model seems reinforced by the production of global density-speed-

throughput data of the same nature and characteristics as those used upstream of
traditional models as inputs for their calibration. There is validation, or at least a
corroboration of the Buguellou model. Corroboration is confirmed by a first spatial
analysis of results.

As with non-applicability in urban traffic of the Ford-Fulkerson theorem, the

model also shows that the sum of individual optima is not equal to the global
optimum if we consider the capacities of the network in its different components.
The model does this with a certain optimism in the hypotheses. A state of
equilibrium is reached through learning which enables the realization of all agendas
and a global economy in relation to the theoretical reference situation. There is,
therefore, emergence of a structure and an organization, but even more important, of
an efficient organization reaching an acceptable short-term dynamic equilibrium by
almost all agents.

Beyond the modeling objective reached, the model also leads to significant
theoretical results.

14.6. Bibliography

[BOU 03] BOURREL, E., Modélisation dynamique de l'écoulement du trafic routier: du

macroscopique au microscopique, PhD Thesis, INSA, 2003.
[BUG 09] BUGUELLOU J.-B., Analyse multi-niveaux des flux de transport: modélisation
multiagents de l'interaction entre l’offre et la demande de transport, PhD Thesis, Tours
University, France, 2009.
[CHO 03] CHOPARD B., DUPUIS A. “Cellular automata simulation of traffic: a model of
the city of Geneva”, Networks and Spatial Economics, no. 3, pp.9-21, 2003.
[DEC 06] DECOUPIGNY C., Modélisation fine des émissions de polluants issues du trafic en
milieu urbain, PhD Thesis, Tours University, France, 2006.
 Cellular Graphs, MAS and Congestion Modeling 361

[DEP 01] DE PALMA A., FONTAN C., “Elément d'analyse de la composante horaire des
déplacements: le cas de la région Ile-De-France”, Les cahiers scientifiques du transport,
vol. 39, p.55-86, 2001.
[ELB 05] ELBERS J., Development of an indicator for traffic flow stability with application
to ramp metering, PhD Thesis, University of Twente, The Netherlands, 2005.
[GRE 35] GREENSHIELDS, B.D., BIBBINS J. R., CHANNING W. S., MILLER H. H., “A
study of highway capacity”, Proceedings Highway Research Record, Volume 14, Part 1,
Transportation Research Board, Washington DC, pp. 448-477, 1935.
[HAL 92] HALL F. L, HURDLE V. F., BANKS J. H., Synthesis of Recent Work on the
Nature of Speed-Flow and Flow Occupancy (Or Density) Relationships on Freeways.
Transportation Research Record 1365, Transportation Research Board, National
Research Council, Washington DC, pp. 12-18, 1992.
[MAE 06] MAERIOVOET S., Modelling Traffic on Motorways: State-of-the-art, Numerical
Data Analysis and Dynamic Traffic Assignment, Catholic University of Louvain,
Belgium, June 2006.
[MAR 64] MARCH J.G, SIMON H.A, Les Organisations: Problemes Psycho-sociologiques.
French translation of the book Organizations, Translation by J.C. Rouchy, Dunod, Paris,
p. 244, 1964.
[MAT 08] MATHIS P. (ed.), Cohérence entre les politiques de transport et les politiques
d’aménagement – le cas de Mulhouse, Research Report, PREDIT-ADEME, France, 2008.
[NAG 92] NAGEL K., SCHRECKENBERG M., “A cellular automaton model for freeway
traffic”, Journal de physique, vol. 2, no. 12, pp. 2221-2229, 1992.
[NAG 98a] NAGEL K., SIMON P. M., “Simplified cellular automaton model for city traffic”,
Phys. Rev. E, vol. 58, p.1286-1295, 1998.
[NAG 98b] NAGEL, K. et al., “Two-lane traffic rules for cellular automata: a systematic
approach”, Phys. Rev. E, vol. 58, pp.1425-1437, 1998.
 Chapter 15

Disruptions in Public Transport and

Role of Information

In the two previous chapters, information was only partially taken into
consideration. Information appears implicitly in the rules of behavior of agents in
the cellular graph with no real explanation of the process defining the calculation of
distances between vehicles. In reality, the agents have access to information on the
position of the vehicle in front through the simulator’s memory files. Other
mechanisms much closer to reality are being developed. On the other hand, before
engaging into a traffic circle, the vehicle is considered as having perceived the use of
indicators by those already occupying it, without the mechanism really being
individual perception.

Moreover, in the case of learning, the agent measures time spent in each edge
and implements his own knowledge of the network, from current and past
experience remembered, where he uses global knowledge: the theoretical values of
the network independently of their use, and thus of any congestion.

This last type of “shared knowledge” is often called the blackboard technique,
which can be generally developed and more economical in terms of memory space,
but risky, leading to congestion amplifications. The two processes were used in the
lab with the consequences still being explored.

Chapter written by Julien COQUIO and Philippe MATHIS.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
364 Graphs and Networks

One of the vital advantages of the research by Julien Coquio1 is to integrate

explicitly the information perceived by travelers in public transport and its
consequences in management and the effect of disruption.

The second major advantage is to show explicitly that this subject can and must
be handled from two points of view, one in a “traditional” way, that of the carrier
and global regulation of the transport system, implying a very systemic top down
vision, and the other from the point of view of the traveler who can find his way in
the transport system according to inherent criteria which can be completely different
from those of the system management, based on a bottom up logic.

These points of view are like both sides of a coin, different, sometimes opposite
but inseparable. They are different because the criteria reflect the rationalities used:
total theoretically for the manager who optimizes according to his own criteria, and
partial, limited for the travelers. However, any decision of the former affects the
latter and the decisions of the latter will modify the choices of the former to a greater
or lesser extent.

This dynamic co-determination process based on broadcast information is what

Julien Coquio is exploring. This analysis is part of a more global vulnerability
measures context of a public transport system.

15.1. The model and its objectives

15.1.1. Public transport

This notion of public service, an important element in the organization and

operation of public transport, refers in particular to the guarantee of the individual’s
right to access vital goods and services. Public service is the expression of Greater
Good and one of the main reasons that public authorities grant themselves the power
to create public policies.

This leads to several obligations: continuity of the service, equality of access and
handling, adaptability and security.

In public transport systems, performance can be addressed in its territorial

dimension, with the notion of accessibility. The levels of accessibility are mainly
based on five inseparable parameters: the density of the territory delivery, frequency

1 Julien Coquio, La performance adaptative d’un système de transports en commun:

modélisation et mesures de vulnérabilité. Vers une prise en compte de la vulnérabilité
fonctionnelle, PhD thesis, University of Tours, 2008 [COQ 08]. This research was the result
of a partnership between Université de Tours and SNCF.
 Disruptions in Public Transport 365

of circulation, scheduling of services, the number and duration of possible transfers

and the performance of node-node services2.

Here, we study the adaptive performance of public transport systems, by

focusing on the internal vulnerability, i.e. on transport system dysfunctions
following internal or external hazards, leading to a non-respect of theoretical
circulation schedules.

Figure 15.1. Basic links in the chain of risks3

The travelers moving in the network are taken into consideration, transport
supply and demand mutually structuring at different spatio-temporal scales.

The vulnerability is addressed in its structural (deterioration of the connection

between network vertices) and functional4 (deterioration of the actual service given
to travelers, with consideration of resources limiting disruptions) aspects.

On the functional level, we particularly focus on the study of “on demand”

regulations, with the help of traveler information, as a possibility for optimizing the
use of the supply5 and thus having a role in the mechanisms of adaptation. Taking

2 Chapelon (ed.) “Conception de services régionaux de transport public et optimisation de

l'offre”, National research and innovation program in ground transportation “New services for
users”, Operational group no2 “Mobility services”, 2005, p.213.
3 Source: Theirry Prost, “La ville, les risques et le génie urbain”, DEA memoir in “Méthodes
de conception en bâtiment, aménagement et techniques urbaines”, INSA de Lyon, 1994.
4 During his thesis, J.F. Gleyze determined three types of vulnerabilities: material, structural
and functional (J.F. Gleyze, La vulnérabilité structurelle des réseaux de transport dans un
contexte de risques, PhD thesis in theoretical analysis and epistemology in Geography,
University of Paris 7, IGN, 2005).
5 This is one of the objectives for developing Intelligent Transport Systems.
366 Graphs and Networks

the functional dimension into consideration is vital because “the finality is not that
of the network but of the system able to ensure communication”6.

Increasing information needs, to which operators must respond, are variable

depending on the context. In a normal situation, the information is made up of a
group (schedules, prices, itineraries, etc.) enabling travelers to prepare for traveling
or moving. In an expected disrupted situation, travelers must be notified of the
extent and consequences of the disruptions to their travel beforehand. In an
unexpected disrupted situation, when an incident occurs, travelers want to be
notified quickly and regularly concerning the nature of the disruption, the measures
taken, possible substituting itineraries as well as consequences on their routes.

This raises a question: are the objectives of operators (regulating traffic)

compatible with those of travelers (arriving as quickly as possible)? The problem
can be viewed in two ways: from the point of view of the operator or the traveler. In
the operator approach, traveler information is “an activity aimed at controlling the
crowd, above all”7 [BAY 95]. The information is a tool that can help with good
flow management. But, the information is also a way to increase the rationality of
traveler choices; it makes it possible for travelers to redirect their route or to
optimize their available time.

15.1.2. Hypotheses to verify

The objective is to test hypotheses relating to a general problem of the

vulnerability of public transport, and more specifically, to the impacts of
broadcasting information to travelers. The disruptions taken into consideration were
traffic interruptions in parts of the network and strikes.

Questioning relating to the H5, H6 and H7 hypotheses make it possible to

address the question of compatibility of traveler and operator outcomes. In
particular, hypothesis H6 is geared toward travelers, whereas hypothesis H7
corresponds more to an operator’s vision.

The methodological choice for testing these hypotheses is modeling. Since no

existing model could respond exactly to our needs, we developed a new model
(PERTURB) and a simulation platform on the basis of existing projects in the lab to
answer our questions. The approach is exploratory by nature and makes it possible
to study the operation of a public transport system during disrupted situations.

6 M Chesnais., Réseaux en évolution, Caen, Paradigme, 1991, 167p.

7 D Bayart, “La machinerie de l'information voyageurs en Gare du Nord” in Joseph et al.,
Gare du Nord, mode d’emploi, Programme de recherches concertées Plan Urbain-RATP-
SNCF, Paris, RATP-Editions Recherches, 1995, pp.253-324.
 Disruptions in Public Transport 367

Field of structural vulnerability

H1: The structural impacts linked to disruptions in a section of the network are
strongly connected to the network links and whether it is possible to bypass the
disruption.

H2: The network vertices experience disruptions in different ways because of

the differences in their situation in the network (presence of alternative supply).

H3: Because of the time slot of the disruption, the structural impacts are
different because of the scheduling structure linked to the supply of public
transport.

H4: The structural consequences of a strike vary greatly depending on the

strike program and priorities of circulation.

Field of functional vulnerability

H5: The consequences of broadcasting information to travelers are strongly
linked to the network connection and to the scheduling structure of axes in the
network.

H6: Traveler information enables them to optimize their routes with

knowledge of network disruptions and by adapting their behavior.

H7: Capacity problems can greatly decrease the impact of broadcasting

information, or even cancel the positive effects.

Box 15.1. Hypotheses to test

15.2. The PERTURB model

15.2.1. Theoretical fields mobilized

Three main theoretical fields were mobilized for the creation and development of
the model: the theory of systems, the theory of graphs and the multi-agent paradigm.

The systemic approach makes it possible to highlight the major elements that
make up public transport: network, vertices, edges, schedules, vehicles, lines and
travelers8. Since the system is open, there are regulation and adaptation processes,

8 We can find these elements in the constituting classifications of the PERTURB model.
368 Graphs and Networks

which are greatly dependent on the variety of behaviors and configurations that the
system can adopt. The system is regulated by an information system and a decision
support system.

The theory of graphs is used to model the transport supply and to measure
certain characteristics. Indicators describe the graphs and nodal indicators and at a
more finite scale, the network: monopolar, multipolar accessibility, link between
two vertices9 [CHA 97]. The theory of graphs enables us to structure the network
and analyze it simultaneously.

These approaches are completed by the use of multi-agent systems in order to

consider the transport supply and demand simultaneously (necessary in the
functional approach discussed earlier). The travelers are the active unsubordinated
system10. The decision support system considers the travelers globally and its aim
consists of managing flows, whereas individually, travelers follow their own aims.
These two can be antagonistic when capacity problems emerge. In this way, we find
the two points of view mentioned earlier theoretically (operator and traveler).

We reason with origin-destination (OD) movement sequences. The itineraries

correspond to schedules in the sense of the theory of graphs. Our approach specifies
that we consider the reorientations of agents during their travel when they
experience a disruption.

In our case, the rationality of agents is conditioned by the information they have,
notably during disrupted situations. Their actions affect the characteristics of the
system (bottom up approach).

15.2.2. Working hypotheses

PERTURB was created on the basis of Working Hypotheses (WH), which we

must explain (for model transparency) because of their influence on the results
obtained. They are justified by the necessary simplification of reality inherent to any
modeling work.

HT1: when travelers have correct information on the transport supply, they
make a totally rational choice and choose the optimal path from a strictly travel time

9 L Chapelon, Offre de transport et aménagement du territoire, évaluation spatio-temporelle

des projets de modification de l’offre par modélisation multi-échelles des systèmes de
transport, p.558. PhD thesis in space and land management, Université François Rabelais de
Tours (CESA), under the dir. of Ph.Mathis, 1997.
10 They could however be considered as external to the system.
 Disruptions in Public Transport 369

point of view11. The decision processes are made according to the law of All or
Nothing12.

HT2: when a traveler is notified of a situation, we assume that he takes this

information into account in his decision process.

HT3: we assume that travelers do not change destination during the trip, even in
the case of disruption(s).

HT4: a traveler does not choose a route with over x transfers, but because of a
disruption, this can happen.

HT5: we assume that transfer times of a route chosen cannot be lower than y
minutes (configurable value) for any traveler.

HT6: in the context of the SDI module, we assume that the vehicles have an
unlimited capacity. This hypothesis is cancelled in the SDIC module, integrating the
capacity constraints of vehicles13.

HT7: the network vertices and transfer corridors; transfer corridors have an
unlimited capacity14.

HT8: traffic interruptions on sections of one line have no consequence on

undisrupted section schedules. When the section of a line is cut, the rest of the line
keeps its theoretical schedules when the traffic resumes. There is no progressive
return. These choices were necessary in order to simplify the modeling problem.

15.2.3. Functionalities

The model developed in C++ is structured in five application modules described

below. In the general interface, a map representing the RER and train network in Île-
de-France is shown.

11 In reality, the criteria can vary and be multiple: shortest route, speed, comfort, cost,
security, agreement, pollution, etc.
12 Nevertheless, the route taken can be different from the one found initially (reorientations,
capacity problems, etc.).
13 See later the different PERTURB modules. The difference between SDI and SCDIC
modules is the inclusion of capacity constraints (effective only for SDIC).
14 With road traffic, capacity problems are more linked to network vertices than to sections.
For public transport systems, the problems are more linked to vehicles: it is at the level of
movement from peak to the vehicle that problems for the traveler occur most often if all
places are taken.
370 Graphs and Networks

The Prelim module makes calculations based only on the transport supply in the
absence of disruption: node-node, monopolar and multipolar.

The S module makes calculations based only on the transport supply with
calculations based only on the transport supply with consideration of disruptions:
node-node, monopolar and multipolar.

The SD module simulates traveler movements in expected disrupted situation

(e.g. maintenance on rail infrastructure). The agents make their itinerary calculation
with knowledge of the disruptions.

The SDI module, more adapted to unexpected disrupted situations, adds

parameters linked to the distribution of information (delay, areas of distribution,
etc.).

Figure 15.2. General PERTURB interface

From this interface we can access the different functions of the model,
integrating some parameters (Figure 15.3).
 Disruptions in Public Transport 371

Figure 15.3. The 5 PERTURB application modules

The SDIC module adds constraints linked to the capacity of vehicles, compared
with the SDI module.

We decided to simulate disruptions affecting specific sections of a network line

with interruption of traffic, starting and ending at a precise time. The parameters of a
disruption are as follows: starting and ending times, disrupted line, first vertice of
the disrupted section, last vertice of the disrupted section, and indication if the
disruption affects both directions.

Figure 15.4. Disrupted section in a public transport line

In the case of a strike, we simply have to replace the theoretical transport supply
by the one corresponding to the new transport plan.

For the calculation of the shortest roads used in all modules, the main algorithm
used is the Dijkstra algorithm15. The three SD, SDI and SDIC modules simulate the
circulation of vehicles and traveler routing in the network. They have a few specific

15 We will not detail how this algorithm functions, it is traditional in the theory of graphs (see
for example P Lacomme, C Prins and M Sevaux, Graph Algorithms, 2003, Paris, Eyrolles,
p.411.
372 Graphs and Networks

functions (vehicle circulation, entrance and exit of travelers, walk during a transfer,
etc.). The simulations use steps of 1 minute.

15.3. The simulation platform

The PERTURB model is part of a simulation platform specifically implemented

to answer our questions. The objectives were the following: to easily enable the
variation of the parameters relating to transport supply and demand and to be able to
quickly and dynamically visualize (video animations) the results obtained.

The main interface is by MAPNOD software, developed by Alain L’Hostis and

Laurent Chapelon, ex-PhD students of Philippe Mathis. It was adapted to our
needs16. Four applications enable us to generate transport supply and demand.
Finally, the PERTURB model makes calculations and simulations succinctly as
discussed earlier.

The architecture of the simulation platform is summarized in Figure 15.5.

Figure 15.5. Architecture of the simulation platform

Some adaptations were made to the MAPNOD software. We created a specific

module for the implementation of parameters to launch the different applications
and a dynamic visualization module was developed for observing the phenomena
over time. This dynamic module visualizes the prioritization process of a network by
suppressing edges, the line creation process, the variability of itineraries between
two network vertices and the flows of travelers in the network. This seemed
necessary because of the significant importance of imaging in the field of networks,
as visualization of a complex structure, as a diagnostics and research tool, and
finally as a means of communication.

16 The original version of MAPNOD is available only at the following address:

mapnod.free.fr.
 Disruptions in Public Transport 373

The RES model, developed by Hervé Baptiste17 enables random generation

under constraints of vertices with allocation of population and a road transport
network connecting them within a theoretical space (see Chapter 4).

The Meshing application makes it possible to choose the degree of extensivity of

operation (meshing) by the suppression of edges in the network obtained with RES.

The degree of intensity (frequency of traffic) and node-node travel times are then
implemented with the help of the Linecreation application. Because of Meshing and
Linecreation applications, it is possible to generate the “two complementary aspects
of structuring a network”18: extensivity and intensivity.

The temporal demand of traveling in the network is achieved through the

Demandgeneration application, from a traditional gravity model. It is the last step
in our ex-nihilo design of a transport system. It is possible to carry out manual
operations for the different transport system design steps.

The constraints used in the different applications enable us to vary relevant

choice factors in terms of transport supply and demand creation in a territory, even
though it is impossible to completely and faithfully simulate such decision
processes. This system generation, which can be completed and improved later, may
be useful in several research projects beyond our specific study.

The platform enables us to easily vary the input parameters of the PERTURB
model and to visualize the results.

15.4. Simulations in real space: Île-de-France

Throughout this research, the Île-de-France region seemed a relevant field of

study. The public transport network is dense. These numbers provide an overview of
the significance of the supply: five Transilian train networks19, five RER lines, 16
subway lines, three tram lines, one tram-train line, OrlyVal and over 1,000 bus lines.

Nevertheless, we must be aware of the great diversity of situations in Île-de-

France, notably between the Paris Intra-Muros offer, in the first suburb and beyond.

17 H Baptiste., Emphasis on urban hierarchies through a graph generated randomly, p.62.

DEA memoir, under the direction of Ph. Mathis, University of Tours, 1994.
18 CERTU, Les modes de transports collectifs urbains. Eléments de choix pour une approche
globale des systèmes, 2004 Collection Références, Lyon, 192p.
19 Some inter-city Corail and TER trains also make it possible to carry out trajectories
internal to Île-de-France.
374 Graphs and Networks

This diversity in the supply density, and in meshing is interesting for the context of
our approach.

Data relating to the theoretical transport offer was obtained in the context of our
communications with the AMIVIF20, an organization whose main mission is to
aggregate data from the SNCF, the RATP and OPTILE21, in order to offer multi-
modal information services22. The AMIVIF database is the most up to date for the
transit system supply in Île-de-France: it integrates all carrier schedules, transfers,
and position determination technology of stops. From this standpoint, we extracted
schedules for a Regular Business Day (RBD), and only worked on the day supply
for busy modes23. The reference day chosen was May 17, 2006. The number of
network vertices was 912.

15.4.1 Disruptions simulated in the Île-de-France public transport

To define the disrupted supply, we chose to work on the following disruptions:

traffic interruption on RER-E between Magenta and Haussmann Saint-Lazare24,
traffic interruption between Plaisir-Grignon and Mantes-la-Jolie on line N, traffic
interruption on RER-C between Austerlitz and Invalides (the case of Castor
maintenance on rail infrastructure in the summer).

As mentioned earlier, we also worked on strike programs on RER-C25

corresponding to important and unimportant social movements: program U68, U35
and U16.

These scenarios enabled us to work on axes, in generally meshed zones. We also

studied the vulnerability variations of the system over time. The approach is
simultaneously spatial and temporal.

20 Association Multimodale d’Information des Voyageurs en Île-de-France (Worldwide

association of Traveler information in Île-de-France), grouping carriers. On January 1, 2008,
the AMIVIF became part of the Syndicat des Transports of Île-de-France.
21 Association of private road carriers in Île-de-France.
22 Notably internet sites www.transport-idf.com and www.transilien.com.
23 The night network (Noctilian) was not studied.
24 In this type of scenario, some RER were directed toward the East station. Since we did not
integrate the supply in our approach, we did not take this parameter into consideration.
25 It is quite common that only one or just a few network lines are affected by a strike.
 Disruptions in Public Transport 375

Figure 15.6. Locating disruptions

The objective is not a complete study on each scenario. The simulations provided
elements of response in relation to the hypotheses formulated.

15.4.2. Node-node calculations: measure of the deterioration of relational

potentials between two network vertices

15.4.2.1. Traffic interruption between Magenta and Haussmann St Lazare on RER-E

Figures 15.8 and 15.9 show the relational deterioration (increase of transfer time)
of RER-E stations during a traffic interruption on RER-E between Magenta and
Haussmann St Lazare. Every minute, we calculated the time necessary to go from
one vertice to another. The time is the sum of necessary waiting time for the first
vehicle and travel time. The greater the frequency of traffic, the lower the variation
in duration. We can then see differences between lowest hourly traffic and rush
hour. A possible explanation for this type of diagram is given in Figure 15.7.
376 Graphs and Networks

Figure 15.7. Evolution of transfer times between two network vertices over time

This is a zone where the transport supply is dense. Subway lines enable us to
bypass the disruption. In addition, line A of RER, serving Val de Fontenay, enables
us to bypass the interrupted section before we reach it. Loss of time is not usually
significant even though it can be at certain times. These diagrams show the diversity
of impacts in the disruption depending on time of day and the vertices involved. At
certain times, there is no impact because an alternative itinerary is as fast (and
sometimes more so) than the itinerary using the disrupted section. The impact can
change between nil and significant at a moment’s notice.

Figures 15.8. Time to reach St Lazare from Bondy

 Disruptions in Public Transport 377

Figure 15.9. Time to reach Val de Fontenay from Haussman

V15.1 and V15.2 videos help us visualize the itineraries with disruptions
according to the time of day.

15.4.2.2. Social movement in the RER-C

The simulations are performed once a minute during the day. First, the
calculations only consider the RER-C offer. Then, the other lines are integrated into
the calculation to evaluate how they can partially compensate the RER-C
dysfunctions (by partial redundancy).

15.4.2.2.1. Calculations not taking into account the other lines in the network
The diagrams in Figures 15.10 and 15.11 highlight the variation of the impact of
the disruption based on the strike program. Program U16 is very restrictive and only
ensures a supply during rush hour. This is particularly clear in Figure 5.11. The
Invalides to St Martin d’Etampes route only operates during the evening rush hour.

In general, outside of rush hour periods, the differences are the most clear. This
corresponds to a prioritization on the times that traffic is the most dense.
Consequently, average time increases according to the severity of the strike. In the
case of the Ermont-Eaubonne to Invalides route, it is 45 minutes during normal
times, 55 minutes in U68, 83 minutes in U35 and 100 minutes in U16. This is not
connected to the travel time itself but to the differences in traffic frequency.26

26 Travel time can also have an impact in cases where a direct train is replaced by a bus.
378 Graphs and Networks

The last time at which the movement is possible (connectedness is no longer

effective) also varies according to the program. It is instantiated in the diagrams with
a vertical line to the axis of abscissa.

Beside these temporal aspects, prioritization also occurs in stations involved27. In

program U16, the number of stations not involved is 33 (in 86 stations).

Figures 15.10. Time to reach Invalides from Ermont-Eaubonne

Figure 15.11. Time to reach St-Martin d’Etampes from Invalides

27 We excluded certain vertices from calculations in order to compare the different programs.
 Disruptions in Public Transport 379

15.4.2.2.2. Calculations taking account of other lines in the network

The diagrams in Figures 15.12 and 15.13 show the influence of other network
lines in minimizing the drop in quality in the connection between network vertices:
respectively Pontoise to Invalides and Juvisy to Invalides, with a U35 level strike.

We can observe the significance of having different lines available for the
network vertices when one of them experiences disruption.

Figure 15.12. Time to reach Invalides from Pontoise

Figure 15.13. Time to reach Invalides from Juvisy

380 Graphs and Networks

Specifically, the Pontoise station has two lines going to Paris besides RER-C:
line H on Paris Nord and line J on the Paris Saint Lazare network. For Juvisy, RER-
D partially compensates for the disruptions in RER-C.

Figure 15.14. Pontoise and Juvisy situation in the network

Video V15.3 and V15.4 animations help us to visualize itineraries (with

disruption) according to the time of day.
 Disruptions in Public Transport 381

15.4.3. Unipolar calculations: measures of the deterioration of traveling

opportunities from a network vertice

15.4.3.1. Traffic interruption between Mantes-la-Jolie and Plaisir-Grignon

The diagram in Figure 15.15 shows a superimposition of times necessary to
reach the vertices of the network starting at Mantes-la-Jolie (with a zoom in the most
accessible Mantes-la-Jolie vertices, in a non-disrupted situation) at 7:47AM, with
and without disruption. This time corresponds to a train departure from Mantes-la-
Jolie en route to Paris-Montparnasse via Plaisir-Grignon.

The vertices are classified by increasing travel time in the absence of disruption.
Most connections are not disrupted although some are (the peaks in the diagram)
where the disruption is significant (up to one and a half hours for Plaisir les Clayes).

At 7:48AM, there is no impact. In fact, since the traffic frequencies are not
significant in the Mantes-la-Jolie–Plaisir Grignon axis, it is always faster to go
through line J of the Paris Saint Lazare network.

Figure 15.15. Time necessary to reach the network vertices for a

departure from Mantes-la-Jolie at 7:47AM

In this example we measure all the characteristics of the vulnerability of a public

transport system compared to individual road transport.

Maps 15.1 and 15.2 enable us to spatialize the information, to visualize:

– connections impacted by the disruption;
– minimal transferring of paths in the other axes.
382 Graphs and Networks

We can observe that the impact of the disruption goes beyond the part of the
network experiencing dysfunctions (indirect disruption effects). Near Paris
Montparnasse, we are at a crossroads between disrupted and non-disrupted parts of
the network. These are vertices where travel time is almost the same whether we use
line N or line J.

Map 15.1. Lost itme to reach network vertices for departure at 7:47AM from Mantes-la-Jolie
(traffic interruption between Mantes and Plaisir-Grignon)
 Disruptions in Public Transport 383

Map 15.2. Lost time to reach network vertices for departure at 7:47AM from Mantes-la-Jolie
(traffic interruption between Mantes and Plaisir-Grignon)

15.4.3.2. Traffic interruption between Austerlitz and Invalides

The maps and diagram in this section show some of the consequences of a traffic
interruption between Austerlitz and Invalides on the RER-C.
384 Graphs and Networks

Map 15.3 shows lost time to reach network vertices for a departure from
Austerlitz at 8:00AM. Some connections to the north of Paris are greatly disrupted.
We again measured to what degree the impact has an effect beyond the portion of
the network directly involved. But here we address a point that is very specific to
public transport linked to train, buses or trams, etc. timetables. It is probable that
certain connections are greatly affected because part of the itinerary has low traffic
frequencies. In fact, loss of time at the start because of a traffic interruption between
Austerlitz and Invalides can then make us miss an important transfer by just a few
minutes.

Map 15.3. Lost time to reach network vertices for departure at 8:00AM from Austerlitz
(traffic interruption between Invalides and Austerlitz)
 Disruptions in Public Transport 385

Map 15.4. Lost time to reach network vertices for departure at 8:00AM from Austerlitz
(traffic interruption between Invalids and Austerlitz)

The second map is a zoom of Paris, because the previous map does not enable us
to highlight the differences very clearly We can see that the vertices which are most
affected are those on or close to RER-C, which is what we would expect.
386 Graphs and Networks

The diagram in Figure 15.16 shows the transfer of minimal routes. We can
observe that they mostly involve lines 1, 5 6, 10 and 14 of the metropolitan network.

Figure 15.16. Number of minutes per minimum route line to reach the network vertices for
departure from Austerlitze at 8:00AM

15.4.4. Multipolar calculations: global measures of structural impacts

Multipolar calculations make it possible to provide information at a higher level

than the one given up to now. Accessibility calculations for each network vertice
(with and without disruption) are possible with PERTURB. We can also give the
difference (with and without disruption for each network edge) and evaluate the
number of routes affected by a disruption.

In the present case, for the three traffic interruption scenarios we calculated: the
number of connections between interrupted origins-destinations, the number of
connections including the extended time and the average of this extension for
departure times set at 7:00AM. The values are presented in Table 15.1.

The traffic interruption between Austerlitz and Invalides seems to be the one that
disrupts the network the most: the number of ODs affected is high. Nevertheless,
traffic interruption between Mantes-la-Jolie and Plaisir-Grignon leads to more
extreme consequences. The number of disconnected ODs is more significant and the
average additional time for the disrupted ODs is higher (38 minutes compared to
8 minutes). This is linked to the differences in network meshing for the zones
involved. The interruption between Magenta and Haussmann Saint-Lazare is a
 Disruptions in Public Transport 387

typical situation: relatively high number of ODs affected but additional average time
low because of the possibilities of bypassing the disruption.

Average
Global
Number of Number of additional
additional time
disrupted disconnected time for
for disrupted
ODs ODs disrupted
ODs
ODs

Haussmann − Magenta 9367 76994 0 8

Mantes-la-Jolie −
689 26308 7240 38
Plaisir-Grignon

Austerlitz − Invalides 16545 129949 1813 8

Table 15.1. Global evaluation of structural impacts of the three scenarios studied at 8:00AM

Figures 15.17 and 15.18 show the additional times generated by the disruption
for each OD. We can see that additional travel time is clearly less in the case of the
interruption between Austerlitz and Invalides than in the one between Mantes-la-
Jolie and Plaisir-Grignon.

For example, the number of ODs where additional time is 15510 (94%) in the
first case and 272 (39%) in the second case.

Figures 15.17. Lost time between origins-destinations of the network for departure at
7:00AM (traffic interruptions between Austerlitz and Invalides)
388 Graphs and Networks

Figure 15.18. Lost time between origins-destinations of the network for departure at 7:00AM
(traffic interruptions between Mantes-la-Jolie and Plaisir-Grignon)

15.5. Simulations in theoretical transport systems

Theoretical transport systems were designed using RES and Linecreation

applications and manual operations. In this case, we did not use Meshing and
Demandgeneration applications.

The simulation platform provides us with an almost unlimited combination of

simulations. This is the reason for directing the simulations according to the
hypotheses defined, particularly those for which few responses could have been
given with our study on Île-de-France.

We carried out simulations integrating the capacity parameter of vehicles, in

contrast to the study done on Île-de-France. Our goal is to show the role of this
parameter and thus the advantage of taking account of it in real cases.

15.5.1. The initial network and line creation

First, a network of 100 vertices is defined with the RES model. The network is a
saturated planar graph (Delaunay triangulation). Each vertice has identical weight
(2,000 people). The network is thus homogenous.

With the Linecreation application, we create network lines. They must cover all
the edges by avoiding superimpositions. The circulation time slot is 6:00AM to
12:00PM. The interval between two routes is 10 minutes (except for line 1 which is
9 minutes) and travel time between two points of 5 minutes (4 minutes for line 1).
Thirty-eight lines are thus created.
 Disruptions in Public Transport 389

Figure 15.19. The initial network

The process can be visualized dynamically in video animation (V15.5). Figure

15.20 presents their traffic potential, by the sum of value-edges of each edge in the
line. As we move through the process, the lines get shorter and have a lower traffic
potential.

Figure 15.20. Traffic potential of lines created

390 Graphs and Networks

In order to work on different networks while being able to compare them easily,
we keep all intermediate states of the line creation process. Figure 15.21 shows 10
and 20 line networks.

Figure 15.21. 10 and 20 line networks

15.5.2. Studied disruption

Consider a traffic interruption on line 1 between vertices ACA and ABL. For
multipolar cases, we do not consider vertices located between these two vertices (in
order to obtain comparable values because they cannot be reached in certain network
configurations).
 Disruptions in Public Transport 391

Figure 15.22. Location of the disruption

15.5.3. Multipolar calculations

15.5.3.1. Systematic exploration

We carry out multipolar calculations at 8:00AM for networks obtained (from
four to 38 lines)28. The diagram in Figure 15.23 shows the percentage of additional
time linked to the disruption and meshing according to the number of lines taken
into consideration in calculations. Meshing is measured by the number of edges
divided by the number of vertices.

28 Four lines was the minimum number of lines. In fact, the disruption in the network in three
lines makes the diagram unrelated (the vertices located on the disrupted section are excluded
from calculations).
392 Graphs and Networks

Figure 15.23. Structural vulnerability and network meshing (multipolar calculations for
departure at 8:00AM with exclusion of vertices in disrupted section)

We can observe that an increase in meshing does not necessarily lead to a

decrease in additional travel time (in percentage terms) because of the disruption29.
In particular, the addition of lines 10 and 11 paradoxically leads to an increase in the
percentage of the additional time linked to disruption. The diagram in Figure 15.24
provides some explanation. It shows that the addition of these two lines reinforces
the importance of line 1 and thus network dependence in relation to it.

Figure 15.24. Relative importance of line 1 in a normal situation

29 Nevertheless, we must remember that more significant meshing can help in connecting the
vertices of the disrupted section, not considered here.
 Disruptions in Public Transport 393

In this way, the vulnerability to a disruption could, in certain conditions, grow

with an increase in the network. We will discuss this point in more detail.

15.5.3.2. Study in the 4 line network

We will now discuss the question of meshing influence by studying the effects of
adding a line to the four-line network.

15.5.3.2.1 Addition of a line reinforcing line 1

First we add a line between vertices ABL and ABN.

The addition of this line results in reinforcing the importance of line 1. We can
observe Figure 15.26: the number of minutes on this line for multipolar calculations
(at 8:00AM here) is more significant with five lines than with four.

Figure 15.25. Addition of line ABL-ABN

394 Graphs and Networks

Figure 15.26. Number of minutes per line for departure at 8:00AM (multipolar calculations
excluding the vertices of the disrupted section)

The result is a paradoxical increase, counter intuitive, of the vulnerability. The

percentage additional time linked to the disruption is more important when there are
five lines for each hour. This is emphasized in the diagram in Figure 15.27.

Figure 15.27. Strucural impact of the disruption (multipolar calculations

excluding the vertices of the disrupted section)

Consequently, the gap between the normal situation and the disrupted situation
(and thus vulnerability) increases with the addition of this line. We should still note
 Disruptions in Public Transport 395

that any measure of vulnerability is partial and the addition of this line may decrease
vulnerability to other disruptions. In addition, this does not mean that the situation
with five lines is not as good, but it increases the gap between normal situation and
disrupted situation in the case studied.

15.5.3.2.2 Addition of a line decreasing the importance of line 1

We add a line between ACA and ACG vertices (see Figure 15.28).

Figure 15.28. Addition of line ACA-ACG

Figure 15.29. Number of minutes per line for departure at 8:00AM (multipolar calculations
excluding the vertices of the disrupted section)
396 Graphs and Networks

In contrast with the previous case, the addition of line ACA-ACG results in a
decrease in the importance of line 1, evidenced in the diagram in Figure 15.30.
Unlike the previous case, the number of minutes on line 1 decreases with the fifth
line.

In contrast with the previous case, the percentage of additional time linked to the
disruption is lower when there are five lines at all times. Line 5 partially
compensates for the disruption to line 1.

Figure 15.30. Structural impact of the disruption (multipolar calculations

excluding the vertices of the disrupted section)

The influence of network meshing on vulnerability is linked to its form (kind of

meshing, i.e. shape of the network) and cannot be limited to a simple number of
edges/number of vertices relation. Reinforcing the importance of an axis by the
creation of feeder lines can result in an increase in vulnerability in relation to it.

15.5.4. Simulations integrating capacity constraints

By using SDI and SDIC modules, we can simulate agent movements and
measure the role of information. We only use these results on theoretical systems
because as the data available for Île-de-France does not allow us to properly
integrate capacity constraints.
 Disruptions in Public Transport 397

15.5.4.1. Studied network and disruption

Remaining in the same initial network, we add a line representing a shuttle
between ACT and ACU vertices. The capacity of vehicles is 20 seats. We vary the
frequency of traffic in this line from 5 to 30 minutes.

Figure 15.31. Addition of the ACT-ACU shuttle

Fifty travelers depart from vertice ADM and 50 from vertice ADC (top left) and
go toward the vertices beyond the section disrupted. They must use the section that
is disrupted (the ACA-ABL section).

The disruption occurs when they are between vertices ADC and ACT. In this
way, if they are notified (if the information is broadcast in vehicles and they receive
the information in the vehicles), they can go to line ACT-ACU (we will call this line
6) by getting out of the vehicle.

15.5.4.2. First case: all travelers receive information in the vehicles

The diagram in Figure 15.32 shows travel time of agents according to route
frequency on line 6. For line frequencies of 5 to 20 minutes, all the agents get out of
the vehicle to use the shuttle (line 6). Nevertheless, because of capacity problems,
many agents cannot enter the first or second vehicle30.

30 We can observe that the higher the agent’s number is, the greater the capacity constraints
he will experience. This is linked to the model’s rules of operation. When several agents
arrive in the same vertice, we start processing those with the smallest identifier. Nevertheless,
the definition of vertice queues enables agents who have not been able to get on a vehicle to
have priority when the second one arrives.
398 Graphs and Networks

Figure 15.32. Travel time for agents according to the circulation frequency of line 6
(all travelers get information in vehicles)

After 25 minutes, the scheduling structure is such that the travelers, even if they
are notified, are not redirected because travel time on line 6 is not quicker. This is
identical to a situation where the information is not broadcast in vehicles.

The speed flow from vertice ACT is based on the frequency of line 6. The
diagram in Figure 15.33 illustrates the number of agents present in vertice ACT
where travelers exited, over time.

Figure 15.33. Number of travelers transferring at vertice ACT over time

 Disruptions in Public Transport 399

We can see this in Figure 15.34, which illustrates global travel time in the
network according to the frequency of line 6. We can also see that for frequencies of
10 to 20 minutes, global travel time in the network is lower if the information is not
broadcast in vehicles. Nevertheless, some travelers are lucky (we can see this in
Figure 15.32 where certain travelers have a travel time that is shorter to a situation
similar to absence of an information broadcast) hence the advantage in combining
aggregated and disaggregated values in the results analysis.

Figure 15.34. Travel time in the network according to circulation frequency of line 6
(all travelers get information in vehicles)

The video animations show the route of travelers:

– Video 15.6: capacity constraints and information broadcast in vehicles not
taken into account;
– Video 15.7: information broadcast in vehicles, line 6 frequency of 5 minutes,
and takes account of capacity constraints;
– Video 15.8: information broadcast in vehicles, line 6 frequency of 20 minutes,
and takes account of capacity constraints;
– Video 15.9: information not broadcast in vehicles.

15.5.4.3. Second case: half the travelers do not receive information in vehicles
Travelers not receiving information do not change course. Consequently,
capacity problems in line 6 are less important.
400 Graphs and Networks

The diagram in Figure 15.35 shows differences in shorter travel time than in the
previous case.

Figure 15.35. Travel time of agents according to circulation frequency of line 6

(1 traveler in 2 does not get information in vehicles)

The video animations show the path of travelers:

– Video 15.10: capacity and information broadcast in vehicles not taken into
consideration;
– Video 15.11: information broadcast in vehicles, line 6 frequency of 5 minutes,
and consideration of capacity constraints;
– Video 15.12: information broadcast in vehicles, line 6 frequency of 20
minutes, and consideration of capacity constraints.

15.5.4.4. Comparison of the two cases

The diagram in Figure 15.36 compares the two cases studied. We can see that
lower global travel time is obtained when the frequency of line 6 is 5 minutes and
that certain travelers do not get the information (slightly better than when the
information is not broadcast in vehicles). With one third of travelers receiving
information, global travel time would be even lower. In addition, the lower the
frequency on line 6 (from 5 to 20 minutes), the greater the difference linked to the
fact that all travelers do not get information.
 Disruptions in Public Transport 401

Figure 15.36. Global travel time: influence of broadcast parameters and possibility
of receiving information by agents

15.6. Discussion on hypotheses

The calculations and simulations carried out on Île-de-France and theoretical

transport systems enabled the production of quantitative results using the PERTURB
model. It is now possible to discuss the hypotheses proposed.

15.6.1. Field of structural vulnerability

H1: The structural impacts linked to disruptions in a section of the network are
strongly connected to network meshing and to whether it is possible to bypass the
disruption.

The multipolar calculations carried out in the three disruption scenarios in Île-de-
France made it possible to demonstrate the importance of meshing. We are able to
show that the impacts in a slightly meshed zone were very important (number of
disconnected ODs and high additional average time). Nevertheless, we must also
take into account the number of disrupted ODs, which can be significant in a central
meshed zone. Multipolar calculations carried out in theoretical networks also show
that an increase in meshing could lead to an increase in impacts. This problem needs
to be addressed.

H2: Network vertices experience disruptions in different ways because of the

differences in their situation in the network (presence of alternative offer).
402 Graphs and Networks

This hypothesis, linked to hypothesis H1, is verified. We showed to what degree

the connections between the network vertices are disrupted (notably the unipolar
calculations in Île-de-France). We also showed the specific role of bypass feeders, in
the case of connections with the important vertices in the network (e.g. Pontoise
with three train lines: line J of the Paris Saint-Lazare network, line H of the Paris
Nord and RER-C network). Exchange poles have the advantage because there are
different possibilities for getting there or for leaving.

H3: Depending on the time slot of the disruption, structural impacts are different
because of the scheduling structure linked to the public transport supply.

This hypothesis is verified. We emphasized the differences in impacts (increase

in travel time) of a disruption over time. Certain disruptions have a zero impact at
certain times for a given connection and significant consequences at other times in
the day. This phenomenon is almost chaotic: the impact can suddenly change
significantly. This is an aspect specific to public transport.

H4: Structural consequences of a strike vary according to the strike program and
traffic priorities.

We verified this hypothesis by calculating three strike programs on the RER-C.

We measured the differences in impact according to connections between nodes
taking, or not taking account of other network lines. The more significant the strike,
the more choices of service then tend to favor rush hour periods and the busiest axes.
It would be interesting to continue these analyses by looking at a strike affecting the
whole network.

15.6.2. Field of functional vulnerability

H5: The consequences of information broadcast to travelers are tightly linked to

network meshing and to the scheduling structure of axes making up the network.

The simulations of traveler flows in the theoretical transport systems enable us to

validate this hypothesis31. In fact, because of the presence of a disruption bypass
feeder and its schedules, the information has a generally important and decisive role.
In certain cases, the information enables travelers to find a different route, but in
other situations, it does not, because the alternative offer will not make the traveler
save time. The role of information is then directly linked to the context of the
traveler and the transport supply.

31 Also see in [BAP 94] other simulations validating this hypothesis.

 Disruptions in Public Transport 403

H6: Traveler information enables the traveler to optimize his route with
knowledge of network disruptions and by adapting his behavior.

Elements of response given in relation to hypothesis H5 are also relevant here.

The importance of traveler information is linked to criteria which depend on the
transport supply and demand. The advantage for each traveler varies greatly.
Nevertheless, we can say that the information potentially has a very important role.
We observed how the travelers’ route can be modified because of the broadcast of
relevant information. But beyond the criteria mentioned, the fact that the travelers’
route can be modified because of the broadcast of relevant information also depends
on the capacity of bypass itineraries.

H7: The capacity problems of substitution itineraries can greatly decrease the
impact of information broadcast, or even cancel the positive effects.

Capacity constraints lead to competition between agents for access to vehicles.

Consequently, Greater Good is not equal to the sum of individual interests. The
information enables some travelers to optimize their route, but for others, it can
make it worse (travel time may be longer than in the absence of information).
Traveler information can therefore either be beneficial or harmful to the traveler.
Consequently, at a global level, the increase in individual agent rationalities can lead
to a global non-optimal solution (global travel time in the network).

This assumes that the operator has a responsibility for his choices. Must he try to
reach a global optimum to the detriment of a certain number of travelers who will be
penalized by limiting the information given to them? Must he ensure total
transparency in information, knowing that this can also lead to penalizing a certain
number of travelers?

15.7. Conclusion

The simulations carried out enable us to reach only partial conclusions because
only specific cases were studied and after the model of K. Popper. These are only
elements that may or may not be reinforced by other studies and simulations with
other models. Nevertheless, some temporary conclusions can be drawn.

The first conclusion that we can logically draw is that nodes are well connected
or badly connected, and travelers have a good or bad situation on the network:
meshing (and thus possible redundancy) is different depending on the locations,
particularly between outlying areas and the city center, even without consideration
of capacity constraints. Consequently, the disruptions affect individuals differently
depending on their situation in the network and the time. Equality of access and
404 Graphs and Networks

processing, which is an objective that a public service must try to achieve, is not
achieved due to reasons of cost linked to many factors including population density
and distribution, the form of the region, etc.

The second temporary conclusion is that the obligation of adaptability is more

often achieved by the necessary redundancy (even partial) of elements, leading to
over-capacities, but it is costly for the operator and thus for all users. Nevertheless, a
lack of over-capacity can be restrictive for the network operator and for travelers
when certain parts of the network experience disruptions because there is no modal
transfer possible in bypass itineraries because of lack of redundancy.

The third conclusion, which clearly appears from the simulations, is that the
interests of the operator and travelers are sometimes different. The objective of one
is to ensure the best operation possible for the transport network, including in the
case of disruption. This implies reasoning in terms of a system favoring the
objective, i.e. transporting the majority of travelers to the possible detriment of some
of them without always explaining the criterion of choice. The consequence is an
increase in the inequalities and possible inability for some in obtaining a satisfactory
solution.

The fourth temporary conclusion highlights the sometimes ambiguous nature of

information at the level of the operator or the traveler. Because of the capacities
available to him, the operator can tend towards partial broadcasting of information
so as not to trigger a sudden transfer effect.

If the traveler receives complete and transparent information, he will search the
optimal route by hypothesis. But in this case, all travelers with the same destinations
are in the same boat, and there would be a complete transfer to another line that will
not necessarily be able to absorb the additional flow. Consequently, a certain
number of travelers will not be satisfied and the solution will no longer be optimal
for travelers.

The presence of several operators makes the problem even more complex: the
information on possible bypass itineraries can lead to a rerouting of travelers to
saturated lines from another operator. In this case, the traveler situation is disrupted
because of lack of global information on the different networks and on the capacity
of lines that could handle the surplus of travelers resulting in the disruption.

The public transport service obligations mandate that the operator makes the
treatment of users equal, but how do you satisfy this obligation when it is impossible
to ensure an optimal solution for all?
 Disruptions in Public Transport 405

A first possible response is the decrease of capacity constraints. Consequently,

the adaptive performance seems to be on the same path as the territorial performance
with increase in redundancy and capacity available when needed. But this solution
conflicts with the search for economic and in fine performance, since its cost is
supported by all travelers and even citizens.

Another possible response to study would consist of including the possible

capacity problems in the information broadcast to travelers, as is increasingly the
case in road transport (an individual can make itinerary calculations on a site as well
as get information on traffic conditions). In this way, travelers can make their choice
with more complete and equally distributed information than has been considered
here.

We would then have to evaluate how traveler choices would lead towards a
global optimum. This leads to an improvement in choice mechanisms in the model,
possibly because of the multi-agent approach developed, but would raise two
problems: first, that of the possible amplifying effect of information for agents who
have not got all the information about the network and disruptions; and second, that
of the possible anticipation capacity of agents. This also requires that the questions
linked to possible optimums for operators be more deeply explored in order to
compare them with agent-oriented results.

15.8. Bibliography

[BAP 94] BAPTISTE H., Emphasis on urban hierarchies through a graph generated randomly,
DEA memoir (under the direction of Ph. Mathis), University of Tours, 1994.
[BAY 95] BAYART D., “La machinerie de l'information voyageurs en Gare du Nord” in J.
ISAAC (ed.), Gare du Nord, mode d’emploi, Programme de recherches concertées Plan
Urbain-RATPSNCF, Paris, RATP-Editions Recherches, pp. 253-324, 1995.
[CER 04] CERTU, Les modes de transports collectifs urbains. Eléments de choix pour une
approche globale des systems, La Documentation Française, Lyon, 2004.
[CHA 97] CHAPELON L., Offre de transport et aménagement du territoire, évaluation spatio-
temporelle des projets de modification de l’offre par modélisation multi-échelles des
systèmes de transport, p.558. PhD Thesis (under the dir. of Ph. Mathis), François Rabelais
Tours University (CESA), 1997.
[CHA 05] CHAPELON L. (ed.), “Conception de services régionaux de transport public et
optimisation de l’offre”, National research and innovation program in ground
transportation, New Services for Users, Operational group no. 2 “Mobility services”,
2005.
[CHE 91] CHESNAIS M., Réseaux en évolution, Caen, Paradigme, 1991.
406 Graphs and Networks

[COQ 08] COQUIO J., La performance adaptative d’un système de transports en commun:
modélisation et mesures de vulnérabilité. Vers une prise en compte de la vulnérabilité
fonctionnelle, PhD Thesis, University of Tours, 2008.
[DEM 04] DEMORAES F., Mobilité, enjeux et risques dans le District Métropolitain de
Quito, PhD Thesis, University of Savoy, 2004.
[FER 95] FERBER J., Les systèmes multi-agents: vers une intelligence collective,
InterEditions, Paris, 1995.
[GLE 05] GLEYZE J.F., La vulnérabilité structurelle des réseaux de transport dans un
contexte de risques, PhD Thesis, Paris 7 University, IGN, 2005.
[LAC 03] LACOMME P., PRINS C., EVAUX M.S., Graph Algorithms, Eyrolles, Paris,
2003.
[PRO 94] PROST T., “La ville, les risques et le génie urbain”, DEA memoir in Méthodes de
conception en bâtiment, aménagement et techniques urbaines, INSA de Lyon, 1994.
[SAN 07] SANDERS L. (ED.), Models in Spatial Analysis, ISTE, London, 2007.
[STA 97] STATHOPOULOS N., La performance territoriale des réseaux de transport,
Presses de l'Ecole Nationale des Ponts et chaussées, Paris, 1997.
 Conclusion

A Method of Multi-level
Simulation and Visualization

Since its origins, graph theory had the aim, among other things, to model
networks, to allow simple representation of them, and to use the theory for problem
setting and resolution of fundamental issues.

Since the beginning of the last century mathematicians of great talent, such as
Euler, Poinsot, L’Huillier, Cauchy, Hamilton, Cayley, Maxwell, Brunel, Heawood,
have started to concentrate on essential problems which are still here today: network
representation and the determination of its characteristics with an operational aim.

Our point of view is very close to that of operational research, which, according
to Brunet [in KAU 72], was introduced in 1937: “war risks and the conduct of
military operations have provided favorable conditions for the development of
operations research”, just as two thousand years before the war between Rome and
Syracuse had encouraged Archimedes to use his discoveries in physics to defend his
city. The use of graph theory in urban development is also fully accepted1: it is a
powerful modeling tool using and developing traditional results of operational
research [FIN 02].

However, for us graph theory is also, and therein lies, perhaps, the originality of
this book, a tool of automated cartography which enables the creation of verifiable

Written by Philippe MATHIS.

1 Though we cannot ignore the possibility of its use in numerous different hypotheses, such
as, for example, the search for sets of graphs which are vulnerable to natural or human-caused
events with an aim of making them safe, but unfortunately the hypotheses can be reversed.

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
408 Graphs and Networks

and reproducible computer-generated images. It is a research tool which enables

multi-level simulation in its most developed sense including the simultaneous use of
multi-rationality.

In Part 1, we presented the traditional applications of graph theory, which are

adapted to the modeling of transport networks with particular developments: use for
multi-means, temporal networks with multiple spatial and temporal scales. The
examples presented by various authors have shown that the existing applications are
capable of much more than their traditional uses.

Chapelon, who is the creator of the nodal zoom method, applies his techniques to
the problem of road congestion on a meso-space level, after having developed a
specific multi-means design by taking into account network specificity and the
nature of the vehicles that use it.

Baptiste uses a temporal graph applied to public transport with exclusive right of
way and scheduling and develops Floyd’s algorithm, so that it takes into account the
constraints of temporized graphs whose arcs exist only at certain moments in time.
These are considered here for short time spans of about a minute. In a study carried
out by the laboratory for the MELT2 the simulation of the SNCF network applies
graphs to temporal constraints. Many transport problems can be dealt with by these
temporized graphs, such as, for example, the transport of goods by truck with travel
prohibitions on weekends, but also of traffic in estuaries subject to the tide, for
example.

F. Decoupigny introduces multi-means simulation on two spatial levels: on the

one hand, car access to natural tourist sites and, on the other hand, pedestrian
displacement at the said sites, i.e. the simulation of two types of behavior with two
extremely different means for the same journey motivation.

For model pedestrian movement Serrhini uses a Cellular Graph with

Geographical References, since it simulates in Part 3 the crossing of a dune range by
tourists, by taking into account the slope of the path.

The RES and RESDYNAM models by Baptiste explicitly use multi-level to

generate networks with determination of scale automatically, with pseudo-random
auto-similarity which is provided by the “random” function and therefore
reproducible. Similarly, the simulation of the growth of cities according to the
improvements of the network at their level of projection and the increases in speed

2 Ministère de l’équipement de l’infrastructure du tourisme (French Ministry for the

equipment of the tourism infrastructure).
 Conclusion 409

and duration of circulation lead to an allometric growth, which is similar to the

works of Berry.

In the historical simulation of the change of axis of the city of Tours in the 19th
century, taking into account the dual surfaces of the nodes constitutes the most
interesting phenomenon and marks the change of logic of the model. To some extent
the attribute of each dual surface of a network node becomes essential whether it is
occupied or not, i.e. built-up or not, using a maximum of density as the limiting
value of occupation.

In all of these models and simulations the various authors pragmatically

anticipated a certain number of later developments of graph theory, to some extent
posing an implicit conjecture. Since we worked a lot with them it was up to us to
elaborate the first synthetic approach: to go beyond and attempt the first theoretical
justification of this implicit but functional multi-level conjecture, where spatial and
temporal aspects are not expressed in same manner.

If we find ourselves in a position where the only intervention is that which

determines the initial conditions, after which the process is entirely automatic, i.e.
determined without human intervention, we obtain the framework of model closure.
Here there will be convergence towards a result as for RES-DYNAM or URBAN
DYNAMIC by Forrester or, possibly, a phenomenon of continuous oscillations, as
in the dynamic von Thünen model. This assumption is rejected by Martouzet [MAR
93] for whom urban development must be ethically and practically open.

Consequently, we must consider that we are permanently in a state of transition,

in which intervention is, and must always be, possible, following the transposition to
urban development of the thesis on policy ethics of Jonas. The intervention in the
model must therefore always be possible, so that it stays a model of urban
development simulation with a didactic, exploratory or predicting role.

This hypothesis of possible and/or necessary intervention in the area, which

modeling would make it possible to highlight, is largely corroborated by the
simulation of the development of Tours in the 19th century where we note that the
simultaneous modification of the intensity of traffic on the two main axes, i.e. the
river and the road, initiates a modification of urban development by shifting the
main axis of the city.

It is clear that in this case we are in one of the many transitional stages of the
system, which has since continued its development under the influence of other
internal and external factors. In this case, it should be stressed that the modification
taken into account occurs on several levels: a transformation of the conditions of
410 Graphs and Networks

regional and national circulation causes a local transformation of urban dynamic,

which, in turn, contributes to reinforce this dynamic.

The RES-DYNAM model consolidates this point of view by showing that human
interventions in the form of network improvement play the part of positive or
negative amplification in the development of cities depending on their population
and location.

Another conclusion is indicated to us by these various models: spatial and

temporal multi-levels are not expressed symmetrically. Although, the passage from
one level to another is, indeed, expressed a priori by a change of the spatial scale, on
the other hand, in time it is expressed by a change in the nature of the observed
phenomena: flows that are considerable in the short term, in the long term are
replaced by the more significant variations in structural changes. Therefore, design
develops towards a cellular-type model.

What becomes essential is the attribute, occupied or not, and which level of the
dual face is associated with the primal node of the graph of the transport network.

In this case the development on the mid-term or long-term cellular structure of

the frame necessarily described by a GSR in 3D is modified by the short-term
development of the movement flows. Temporally, there is interaction between
various levels. Taking into account the saturations demonstrated by Chapelon
generates behavior modifications, but in the very short term, when that is possible,
as well as network modifications for various reasons, but of a more long-term
nature.

Throughout all these works, we gradually went from empirically using very
traditional algorithms to taking into account the spatial and temporal multi-level.

The second original object of graph theory was, as we stressed, to enable the
simple representation of networks of any nature, specifically including the problems
of cartography where the conjecture has become the theorem of four colors: coloring
of the nodes, the arcs and the faces of a planar graph.

We dedicated Part 2 to the problem of network representation, which is largely

overlooked by the majority of modern authors3.

The lack of interest in the problems involved in the representation of graphs is

undoubtedly explained by the research dynamics of the mathematicians specializing
in graphs, for whom this problem was of no real scientific interest, and by the large

3 One exception should be noted: the recent work by Michael Kaufmann and Dorothea
Wegener [KAU 01].
 Conclusion 411

fields of application of operational research where the spatial aspect is often non-
existent or at best merely secondary [FIN 02].

The field of computer-generated images has thus developed in parallel because

of the demand for representation by using various contributions: discrete geometry,
fractals, lattices, etc.

We could note that since graph representation was extremely unregulated, its
constraints were often, at best, implied and this problem did not appear as such in
the eyes of many users of the theory, whose concerns were not at this level, or who
did not consider it as a problem, and in any case not a problem of graph theory.

The MARIO model by Larribe completely exploits the possibility of drawing

graphs without constraints other than to represent the elements of the graph by nodes
and the relations between these elements by edges or arcs, and to plot it on a plane
for greater ease.

Indeed, the specific problem is the representation of a network of actors. The

software uses the property of isomorphism of graph plots to be able to modify the
representation, until the moment when the obtained figure suggests a possible
structure, another way of seeing the set of data, or a lack of information, etc. This
result is made more likely by the possibility of representing only partial graphs
and/or subgraphs depending on the known attributes and relationships of the actors.
Thus, attributes of multiple natures and therefore of the social multil-evel are used.

The work of Larribe builds on the numerous works developed by sociologists or

socionomists and is located between the organizational diagram and the sociogram.
We also note that the graph used is actually a multi-graph: there may be several
types of relationships between pairs of actors and in addition actors have attributes
which can be classified and quantified.

We saw the usefulness of such a self-centered procedure, whose body of

hypotheses is particularly restricted in cases of limited rationality; it is only the
information possessed by the executor or even the drawing of the graph since it is
possible to gather information coming from several agents. The hypothesis of
limited rationality is used on a different level to that of the MAS. It is not an issue of
limited rationality of the agent in his space because of his own ignorance of a
network or space which is external to him, but rather of the designer of the graph
who draws one or more successive projections of his own graph resulting from his
personal and necessarily incomplete knowledge of the network of actors.

The progression to a representation including space, which we have called a

Graph with Geographical References (GGR), has been achieved by Khaddour in
412 Graphs and Networks

order to represent the phatic networks of the town of Latakia. The projection of
these graphs introduces a constraint with respect to the situation of Larribe: the
graph must have a bijection with the geographical space represented, in particular,
with houses and phatic spaces, which is inherent to GGR.

Here, the introduction of space and its temporal development makes it possible
to highlight different levels of movement of the agents towards the visited phatic
spaces and the author shows that these spatial scales cover the differences between
agents. Moreover, these relationships have a comparable nature, but on different
scales: a new spatial fractality is highlighted by means of property.

The projection constraints are even stronger when it becomes a question of

representing “eurocorridors”, evaluating traffic and representing pollutant emissions,
as done by Decoupigny by taking into account a non-plane GGR. In the latter case
multi-scale is expressed by taking into account the third dimension, i.e. the altitude
of the various network nodes and, thus, of the slopes and their effects on truck
pollution. With two dimensions we can only use average pollutant emissions,
whereas with three dimensions we can approach real emissions according to the
nature of the truck.

In his thesis Chapelon used an index of network sinuosity, whereas for

Decoupigny the precision of network description increases with the integration of
the third dimension: the network is no longer plane, even if it remains flat and not
convex.

In fact, this solution is adopted by default, since it would have been necessary to
be able to actually have a longitudinal profile and use the subdivision of arcs to
achieve the necessary precision. Moreover, between the graph drawing and the real
network there must be more than isomorphism, that is homothety, since through
altitudes the form of the surface must be preserved at the level of nodes and arcs.
Here we are very close to a digital terrain model.

However, these situations are still relatively simple compared to the development
of computer-generated images illustrating anamorphoses or simply valuation
changes.

Let us consider the works of L’Hostis and, in particular, his chronomaps or, to be
precise, his map valuated in time distance. This type of “map” is subjected to several
constraints: the bijection of sites/nodes and the representation of an arc between two
consecutive nodes with a length proportional to the time to cover the distance
according to the various means operating in the network. He used a straight line for
the fastest, which, therefore, determines the unitary length of journey represented by
 Conclusion 413

the segment of the straight line “the shortest path from one point to another”
following the Euclidean paradigm.

Here, it is no longer a map in the traditional sense, since the faces are only useful
to give an impression of the “relief”, but a complex surface in a three-dimensional
space, at least for “chronomaps” in minimum mode. On the other hand, for a
chronomap in multi-mode, L’Hostis uses a three-dimensional GSR and no longer a
surface managed by a GGR. The importance should be stressed not only of the
parameters but also of their signs: the chronomap in minimum mode, with a
“traditional” height, and that of the opposed sign with a depth, show completely
different aspects of the same projection.

And yet, the creation of this type of computer-generated image is perfectly

transparent since the code which is open, available on the net, is reproducible and
therefore verifiable.

We considered that the zoom method was fractal and therefore we must consider
for the same reasons that the method of chronomaps is also fractal when scale or
level are changed. L’Hostis clearly demonstrates this.

When we move from the map with 150 nodes to one with 528 nodes there is
auto-similarity between the two situations: the same structure of saturated planar
graph, the nodes of the first graph are preserved in the second, the distances along
the main roads are also preserved, etc., in the same way as there was auto-similarity
with Chapelon’s use of nodal zoom, which can also be used for chronomaps.

Serrhini uses graphs with several levels within the framework of determining
covisibility. At first, he determines a digital terrain model on the basis of a 1/25,000
map which is which is defined by a regular graph with a square grid. In this digital
model he uses multi-resolution, when that is required, in order to obtain meshes of a
finer size, where it is necessary. The process is of the same nature as the zoom, but
applies to surfaces and not to arcs. At this level it is a Graph with Geographical
References because it defines a surface in a three-dimensional space. Between this
surface and its straight projection onto a plane there is a bijection.

On the other hand, in certain cases it is necessary, in order to better define

visibility when there are wooded areas creating a screen, a large building or a
viaduct under which the line of sight can pass, to use a 3D surface, which along with
its projection onto the plane has a surjection, i.e. several points of the surface can be
projected onto only one point on the plane. We then obtains a Graph with Spatial
References (GSR). From the GGR or GSR we can calculate more precisely the
minimal paths between two points, or avoid certain fragile zones with conservational
constraints [SER 00], but also the indices of visual spatial pollution, etc.
414 Graphs and Networks

Transforming the traditional von Thünen model into a dynamic model, makes it
possible to show that the chaotic aspect can be obtained and maintained in the long
term between certain limits when certain hypotheses are fulfilled and, in particular,
the existence of large numbers of actors deciding according to data from the
previous period.

Most of all, we also have a cellular model using the dual structure of a network,
where the surface is explicitly taken into account and generates dissymmetry. This
cellular space model supplements and explains the erratic variations of the overall
price model.

Developments of the techniques presented above are currently in progress4. We

are dealing with taking into account, beyond the network in a strict sense, the faces
of the primal or dual graph in order to represent surface values following a different
method from the traditional “cartogram”, which represents them proportionally to
the plane surface of the mesh, in order to be able to compare the same values,
densities, incomes on different scales.

Moreover, these models corroborate the hypothesis according to which it would

be possible to pass from the level of micro-simulation to the level of macro-
simulation as that is, in fact, carried out in the dynamic model of von Thünen.

Part 2 shows the considerable importance of imagery in the field of networks, as

a visualization of a complex structure, an efficient mode of presenting the results, a
diagnostic and research tool and, finally, as means of communication. “What I have
not drawn, I have not understood” said Johann Wolgang Goethe. However, even in
these various works we gradually see the graph drawing become more specific,
written as an algorithm, many issues still remain to be cleared.

Part 3 had the objective of consolidating certain empiricisms by using

demonstration tests in order to more rigorously assess the various developments
presented earlier.

A certain number of questions arose and, in particular, the problem of spatial

multi-scale: what is its justification?

Was the fractality hypothesis justified? Could we connect graph and fractal
theories?

4 Cartogram Program of the DATAR; see the contributions of L’Hostis, Decoupigny and
Mathis.
 Conclusion 415

The second important question was that of the duality of graphs, which are
powerful means, but, perhaps, insufficiently specific, in particular, for an easy use of
surfaces on the basis of networks.

The third question which we tackled, was that of a better description of nodes,
for which the theory offered no description for certain types of crossroads or
interchanges observed in the networks.

Lastly, we were presented with the question of the possibility of moving from a
graph to a cellular space by using algorithms, so as to be able to simulate flows on
various scales including the largest scale, i.e. that of the vehicle itself. Descending to
this scale was a necessary condition in order to consider certain questions, such as
saturation, thanks to a better description of the nodes, such as that of waits at
crossroads and traffic lights and also that of the effects of parking on capacity, etc.

The first point made it possible to show that on the basis of a graph we could
easily develop a fractal with the property of auto-similarity, not only in terms of
form of the graph plotting, but also at the level of the adjacency matrix. This
connection to fractals, furthermore, makes it possible to place a global
morphological indicator at the disposal of the graph user, whose effectiveness has
been demonstrated by Cyrille Genre Grandpierre in his thesis, after Frankhauser and
Thibault [THI 87].

Chapelon’s zoom is simultaneously an illustration and an anticipation of the use

of the property of auto-similarity between levels: use of planar graphs, which are
saturated insofar as the nature of the terrain enables it, centering of the n-1 level on a
level n node. The chronomaps of L’Hostis are similar.

The fractal graph not only underpins the zoom method, but also made it possible
to define methods of graph extension, i.e. a grid method with a neutral reduction
factor, with limits which appear related to temporality and are expressed spatially. In
fact, the modes of land use change and thus involve different urban or rural parcel,
which modifies the size of the generator. For example, fixing a minimal size for the
constructability of a piece of land or its size for allotment, has enabled, under
Hausmann’s supervision, an increase in the size of new Parisian buildings. In rural
areas the regrouping of lands is linked to the techniques of exploitation and involves
a reorganization of the pieces of land. This modification of the size of the pieces of
land is not unregulated, but rather reflects the social evolution at the time, which, in
turn, should be able to be modeled and simulated by a multi-agent system of
management. This assumption makes it possible to account for the existing networks
by showing that the fractal development is not only iterative, but that these iterations
are also temporal and they are rendered in modeling in a simplified manner by a
periodic phenomenon, which is a possible function of a random dummy variable of
416 Graphs and Networks

the Boolean type. Moreover, it is not the process of fractalization that changes, but
the socio-economic conditions, which determine its parameters.

This hypothesis has the advantage of enabling a perspective. The systematic

exploration of socio-economic conditions whose effects are thus summarized would
make it possible to discover their spatial consequences, which would in turn
feedback at the global level.

The problem of parceling involved in network fractalization necessarily led us to

take into account the surface dimension of the faces of the grid graph, of cubes, from
which, inevitably we started taking into account the duality of the graph.

The problem of duality was examined from the point of view of graph drawing
with an aim of being able to carry out repetitive and verifiable cartographies and
algorithms in the line of the work of Tobler. We could highlight that the definition
of a dual is often implicitly valid only for the planar dual, whereas this definition
also relates to dimensional volumes or three and more.

The definition of the planar dual uses a node on the external face which is
presumed infinite on the plane and finite on the sphere, since the definitions of a
planar dual are valid for projections onto a sphere. We saw some difficulties that
arose from this. Moreover, if the primal is a convex volume, its projection onto a
plane implies double surfaces and the projection of the dual of this convex primal
does not have an infinite face. An interesting illustration is provided by the dual of a
cube: the octahedron written into it, which in turn has a cube for a dual and thus ad
infinitum, just as the cube is the dual of an octahedral written around it and thus ad
infinitum, from which we have a periodic fractal volume. On the one hand, this
converges to zero and, on the other hand, it tends towards infinity, but without a
starting point as opposed to the simple fractals of the von Koch type and Sierpinski’s
carpet or even Menger’s sponge, whose fractal volume is defined on the basis of a
right-angled parallelepiped.

This example of periodic fractality with neither a starting point other than an
arbitrary one nor a limit between zero and infinity, perfectly introduces and justifies
the concept of increasing fractality or, by extension, as well as the definition of the
dual of Rosen’s map and thus the subsequent concepts of ante primal and quasi-dual.
Indeed, strictly speaking, the dual of Rosen’s map that does not respect Euler’s ratio
does not strictly constitute a dual graph, which is why we prefer the name of a quasi-
dual for the figure written in and ante primal for the figure written out.

The use of the quasi-dual and of the ante primal is fully justified with respect to
the simulation by the simplification that it introduces into the algorithmic passage
from one to the other by removing the infinite face, as well as what it can involve as
 Conclusion 417

problems. Let us note in this respect the highlighted difficulty, amongst other things,
for the modeling of the possible change in the nature of the graph between the
primal and the dual, since the latter is not necessarily of the same nature as the
former: passing from a 1-graph to a p-graph and vice versa.

The nodal graph that we obtain by subdivision, i.e. by creation of the arcs
defined by the adjacency matrix of the second order then by the suppression of the
original node, makes it possible to study the flows inside the node, for which graph
theory offered no solution. That makes it possible, among other things, to describe
certain specific nodes which do not necessarily correspond to the traditional
assumptions of the theory and to follow flows and their orientations even inside the
node while remaining within the strict framework of graph theory. We thus
transform a node into a potential “interchange”. The nodal graph has its principal
use with respect to traffic analysis of track flows, and of breaking them up into
divergent and convergent flows for the study of nodal capacity, decelerations and
crossing saturations. This enables a micro-spatial approach which is different from
the graph of crossroads conflicts [FIN 02, pp. 64-65].

Let us note, however, that the graph obtained in this manner is no longer a planar
graph, but merely a plane graph, since nodes cross without intersecting as for a
highway interchange.

This decomposition which is similar to nodal zoom also makes it possible to

solve the problem of the nodes of the cellular graph.

The cellular graph aims to make it possible to carry out MAS and cellular
modeling on the basis of a traditional graph.

The problem was to algorithmically transform a directed graph into a succession

of series cells and vice versa, thus making it possible to simulate the behavior of a
set of agents circulating in this network following rules, for example, those of the
Highway Code.

As Decoupigny shows, [DEC 06] the usefulness of this transformation is double.

On the one hand, it enables MAS modeling, which is a simulation at the micro level,
by taking into account the agents individually and no longer the whole of the
transport system for an optimization of the Operational Research type. On the other
hand, this makes it possible and requires a complete change in the kind of
assumptions by substituting the necessary total rationality of optimization. This is a
very heavy and not always realistic assumption, for the limited rationality at the
level of the agent, who in his behavior (that of a driver, for example) does not need
to know the entire network, but only the part which is useful to him. In extreme
cases this could be only his destination and his immediate environment, from which
418 Graphs and Networks

he can develop a strategy of trial and error movement that we all employ in cities or
districts that we do not know at all or not sufficiently.

In the case of individuals seeking a path to reach a place in a generally known

network, the mechanism is different, in the sense that it remains essentially
individual5.

The partial knowledge that the individual has of his environment can be very
diverse. It can be limited to his position and an approximate location of his
destination, as in the case of the TUREP software or until learning a route which
could be optimal or not, but which is considered satisfactory by the agent because it
allows him to carry out his agenda, as in the case of the J.B. Buguellou model. It
may be developed, updated, specified according to the time of use with regard to the
capacities and speeds and therefore time adjusted through a learning process.

The adaptation can involve two different modeling mechanisms, one being a
strictly individual learning process without knowledge sharing, and the other based
on the “blackboard” technique implemented by all experiences and source of
knowledge shared by all the agents. We used both, and shared knowledge seems to
have an amplifying effect in group movements as used by all GPS system agents
without any consideration of congestion.

The range of the possible strategies increases considerably with the mechanisms
of learning, adaptation, use of external information and, in particular, in the case of
congestion or an accidental or non-accidental event affecting the normal operation of
the network with all the adverse effects that this can produce, as demonstrated by the
Perturb model of Julien Coquio. From these adaptive mechanisms the agents can
develop mechanisms of anticipation constituting one of the next modeling steps.

We immediately perceive the importance of the phenomenon of learning, which

enables the agent to approach optimization for a criterion or a combination of
criteria that he chooses. However, he can approach optimization in a short-sighted
manner, without taking into account the reactions of others or, on the contrary, by
integrating them with a blackboard, in a prediction or a point of view, i.e. arriving at
a prediction with all the strategic consequences that this can involve at the individual
and global levels.

The problem raised by Julien Coquio then appears, which point of view do we
use: that of the manager, and possibly the transport system controller, in the case of
“intelligent” systems but then we are placed in the context of a top down type model
even though we are using agents. But these agents are also guided, “controlled in the

5 We eliminate here the case of a system using GPS, radio or other navigation aids.
 Conclusion 419

sense of the theory of systems”; or on the contrary, the point of view of the agents
themselves, independent and free from their bottom up criteria of choice influencing
the system? Here is the difference between both of these possibilities: sharing of
information or not enhanced by the possibility for the controller to use the
blackboard for his own purposes. This additionally raises the problem of the
criterion of choice, whether it be by Pareto or others, i.e. the problem of choosing
between balance and efficiency, which is primarily political.

The MAS finds a field of extraordinary usefulness because it makes it possible to

simultaneously use several criteria of rationality according to the various groups of
agents identified, and not only the traditional minimization of distance and/or time
between an origin and a destination, but also, for example, an esthetic maximization,
chaining of several movement motivations for a quasi-optimization over the entirety
of movements of a period. This does not necessarily imply the optimization of each
movement considered separately, which is a strategy increasingly employed by
agents, i.e. a better use of discretionary time per period.

In this context, the models reveal another important aspect: even though in the
different models, including those currently in application [MAT 08], the agents use
the optimal solution in either distance or time without considering congestion, the
system is saturated, and the situation is not only globally not optimal, but it can also
be impossible, at least for some agents.

In fact, the congestion problem is complex and micro-local originally, as is

shown in a congestion front cellular graph on an edge, with congestion that is not as
heavy at intersections regardless of whether they are traffic circles, traffic lights or
priority crossings. We notice that in an urban vehicle traffic system, the Ford-
Fulkerson theorem on flows in graphs does not apply because that is not the capacity
of the edge, but mainly the capacity of nodes which is vital, or they are much lower
than the capacity of input edges. The cellular graph technique automatically takes
this aspect into account because it allows the modeling of all intersections.

The cellular graph can also be used in a totally different manner: no longer
simply as support and indicator of the location of agents with their own rules of
behavior and their strategies, but as a set of spatial cells with attributes, whose
variations are studied according to various temporal scales.

The simulations carried out by Decoupigny showed the effectiveness of the

method of cellular graphs with regard to the use of cells by agents and the
characteristics of flows, in particular, according to the network and the priority rules
as well as for the fine spatial analysis of pollutant emissions with the obvious
applications of all these results to space management and urban planning. For
example, we may thus carry out a study of saturation at the micro-local level to
420 Graphs and Networks

supplement the approach described by Chapelon and “to test” the effects of the
modifications suggested, locally of course, and to consider their overall
consequences, in particular, as regards deferment, temporal or modal spatial transfer.

We immediately notice the considerable developments that this brings to

multiple strategic, spatial, temporal as well as social levels6.

For example, following the ESPON 1.2.1 project on transport networks in

Europe [MAT 05] (bearing upon the 29 countries of Europe), the problem of the
prospective transport model arises with all its consequences. Taking into account the
existing infrastructures, the construction delays (15 years for highways or railroads)
and the inevitable congestion phenomena, this model must necessarily take into
account a structural change, new infrastructure, but also the desired or unintentional
radical modifications of the hitherto constant trends with respect to relative growth
of the means, means transfer, energy costs and particularly the assumption of the
high or increasingly high price of an oil barrel, space congestion and scarcity of
space, reaction of populations to infrastructural projects, sensitivity to the
environment, etc. The traditional four stage models can no longer be applied for two
main reasons: the structure is not constant anymore and thus the legitimacy of these
models based on econometric relations no longer exists and these models take into
account neither the space, in particular, in terms of network capacity, nor the
populations.

It is thus necessary to pass by fine descriptions of intermediate space between

graph and cellular, using logistical MAS, which determines the means used taking
account of the various economic and other constraints. Then, we consider the road
haulier determining the routes and therefore the areas crossed, where there are other
types of traffic and limited capacities, populations exposed to risks, management
MAS defining the standards of transport and traffic regulations, environmental
protection, social conditions, etc.

The multi-scale must therefore develop definitions of space at various

intermediate or meso-spatial levels, develop MAS for decision makers (logisticians
and transporters, as well as more general management systems) and, finally, take
into account the various timescales and their specific interactions due to their
fundamental orientation.

Thus, by means of crossing cellular spaces and agents we are able to find a
solution for the problem, not strictly speaking of aggregation, but that of the
relations between levels, which requires defining not only each of these levels, but
also the relations between levels.

6 We refer here, among other things, to Schelling’s model of urban segregation.

 Conclusion 421

The dynamic model of von Thünen, i.e. that of Tours, but especially Turep,
Bugellov and Perturb have all provided an outline, each in its field, for the model’s
level and temporality.

A certain number of research projects are currently in progress in our laboratory,

and in others, working on descriptions of intermediate space between micro-cellular
and the graph, on taking into account time, as the increased use of networks
according to the daily time, on the use of information as a means of increasing the
reliability of networks and reducing their vulnerability, for example, on the
coherence between the Plan of Urban Movement and the usage of space by
sustainable urban growth.

Beyond the networks, the surface aspect of graph theory, which has been present
since the beginning, is taken into account to a very limited extent in the problem of
coloring the faces and in that of duality. However, the most complete description of
space requires considerable progress in this direction. This complete description of
space is necessary, in particular, to take into account the locations of residential and
business areas that are the origins and destinations, medium-term variables, etc.

The perspectives for simulation and multi-level representation by using graph

theory to describe space are rich because the problems change in nature and
rationality according to the levels and space can be used as a point of passage
between them.

At the end of the book we note that not only does graph theory make it possible
to solve many problems of Operations Research, but it also makes it possible to
tackle entirely new problems bottom up and no longer top down; these are temporal
and spatial problems of the emergence of order, which are similar to Schelling’s
model of urban segregation.

In addition, these tools make it possible to create many types of reproducible and
objectified computer-generated images and cartographies, in the sense used by
Tobler.

Bibliography

[DEC 06] DECOUPIGNY C., Modélisation fine des émissions de polluants issues du trafic en
milieu urbain, Thesis, Tours University, Tours 2006.
[FIN 08] FINKE G., (ed.), Operations Research and Networks, ISTE, London and John
Wiley &Sons, New York, 2008.
[KAU 72] KAUFMANN A., Méthodes et modèles de la recherche opérationnelle, Dunod,
Paris, 1972.
422 Graphs and Networks

[KAU 01] KAUFMANN M., WEGENER D. (eds.), Drawing Graphs, Methods and Models,
Springer-Verlag, Berlin, Heildelberg, New York, 2001.
[MAR 93] MARTOUZET D., Recherche du fondement de l’éthique de l’aménagement,
Thesis, Tours, 1993.
[MAT 05] MATHIS P., (ed.), Transport Services and Networks: Territorial Trends and Basic
Supply of Infrastructure for Territorial Cohesion, ESPON Project 1.2.1. www.espon.lu,
Luxembourg, 2005.
[MAT 08] MATHIS P. (ed.), Cohérence entre les politiques de transport et les politiques
d’aménagement – le cas de Mulhouse, Research Report, PREDIT-ADEME, France, 2008.
[SER 00] SERRHINI K., “Intégration quantitative et multicritère du paysage lors de la
détermination d’un aménagement linéaire”, Mappemonde, 2000.
[THI 87] THIBAULT S., Modélisation morpho-fonctionnelle des réseaux d’assainissement
urbain à l’aide du concept de dimension fractale, Thesis, Lyon, 1987.
 List of Authors

Manuel APPERT
UMR Espace
CNRS
University of Montpellier
France

Hervé BAPTISTE
CESA
Centre de recherche VST-ART
University of Tours
France

Jean-Baptiste BUGUELLOU
Cabinet Nestear
Gentilly
France

Laurent CHAPELON
UMR Espace
CNRS
University of Montpellier
France

Julien COQUIO
SCNF
Tours
France

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
424 Graphs and Networks

Christophe DECOUPIGNY
CESA
Centre de recherche VST-ART
University of Tours
France

Fabrice DECOUPIGNY
UMR Espace
University of Nice
France

Ossama KHADDOUR
CESA
Centre de recherche VST-ART
University of Tours
France

Sébastien LARRIBE
Centre de recherche VST-ART
University of Tours
France

Alain L’HOSTIS
INRETS-TRACE
Lille
France

Philippe MATHIS
CESA
Centre de recherche VST-ART
University of Tours
France

Kamal SERRHINI
Urban Systems Engineering Department
Compiègne University of Technology
France
 Index

0-9 automobile equivalent 12

1-adjacency 268
B
2-adjacency 268
8-connected 302 butterfly effect 242

A C
access movement 37 capacity 13, 14
activity diagram 320, 321 cells 292
actors 111 cellular
adaptive graph 48 arc 292
adjacency matrix 249 graph and its uses 290
agent 319, 320, 321, 323, 324, 325, model 243
326, 327, 328, 357, 358 modeling 298
activities 322 centroids 204
knowledge 328 chain of risks 365
movements 319, 320 choice mechanisms 405
agent-centered modeling 319 chronomaps 186
agent-network relation 320 collective learning 324
alternating migration 82 communication flow 131
amplification and instability process congestion
226 fronts 330, 333, 351, 352, 358
ante primal 285 simulation 322
antecedents 51 convex
anticipate the behavior of an actor polyhedron 288
116 surface 283
attributes 117 cooperation-conflict 112
automaton 39 covisibility 193, 202

Graphs and Networks: Multilevel Modeling, Second Edition Edited by Philippe Mathis
© 2010 ISTE Ltd. Published 2010 by ISTE Ltd.
426 Graphs and Networks

D functional binomials 47
D.Loca.T 99, 227
G
dampers 242
Department of Indre-et-Loire 321 game 112
deterioration generator 248, 269
of traveling opportunities 381 global
diffusions 302 congestions 319
digitalization 164 measures of structural impacts 386
divergent and convergent cellular optimum 403, 405
nodes 296 validation test 352
door to door movement 49 gradient of the road 147
driver agent 319 graphic distance 181
dual 103, 278, 288 gravitation model 36
graph of the map 289 GRS 266, 298
duality 270, 285
dynamic H
allocation 100
generalized visibility 211 H-fractal 255
model 224 heterogenity of emissions 168
module 69 hypotheses to verify 366
process 73
simulation 95 I
impacts 41
E impoverished metrics 180
individual
emissions 147
behavior rules 319, 352
extension
choices 320
by extrapolation 251
infinite face 280
of the graph 249
initial network 388, 389, 397
evolution
interconnection constraints 49
of vehicle concentration 332
intrapolation 249
intersection modeling 333
F
faces of the graph 296 L
field of functional vulnerability 367,
401, 402 leaf 257
flows 12 learning 319, 322, 325, 326, 327,
Floyd’s algorithm 48 328, 360
fluctuation 242 ability 319
fractal transformation 298 level of tolerance 323
fractals in extension 258 limited rationality 323, 325
FRED model 35 line creation 372, 388, 390
 Index 427

local congestion 328 performance 26

lost transport time 357 PERTURB model 367, 372, 373, 401
phatic spaces 132
M Planar 271
plotting of graphs 178
MAP 100, 227 primal 278
MAS 298 primality 285
macro-level of the spatial system 307 principle of least effort 303
macroscopic variables 352 profile along each arc 165
maximum capacity 262 profiles 170
meta-game 112 public transport 364
micro level of agents 307 Pythagoras tree 255
microscopic actions 352
model of cellular type 230
Q
movement
at sight 39 quasi-dual 254
modeling 320 quasi-periodicity 236
multigraph 278
multi-levels 243 R
multimeans 100
multipolar calculations 386, 391, 401 recreational visits 31
multiscale 100 reducing instability 235
relief 189
representation 178
N
RES 228
Network RESCOM 131
analysis 116 rights-of-way 340
structure 37 rules of behavior of agents 361
nodal 263
matrix 263 S
non-symmetric metric 180
satisfaction and learning 324
number of the cell 293
saturation thresholds 75
scale 33
O
self-centered 120
opening 101 self-similar matrix 249
organizational diagram 114 Sierpinski carpet 256
simulation platform 366, 372, 388
P simulations in real space 373
simulations integrating capacity
peak capacity 336 constraints 396
pedestrian diffusion 31 slope 148
percolation 301 social networks 111, 115
percolation-type behavior 291 sociogram 114
428 Graphs and Networks

sociograph 119 traffic density 323, 324, 331, 339,

sociography 119, 124 340, 341, 343, 352, 356
space-time relief maps 186 transitivity 114
spatial validation 356 traveler choices 366, 405
street crossings 329 tree-structured fractal 251
structural analysis 116
structure of traffic circles 334 U
studied disruption 390
subdivide 249 unconstrained movement 303
surface modeling (MNS) 199 unipolar calculations, 379
symmetric directed arc 292
V
T visible landscape 193
terminal node 257 visual length 181
theoretical transport systems 388, von Koch 250
401, 402
three-dimensional anamorphoses 184 W
time-delay graph 47, 48 Wardrop equilibrium, 320
traffic circles 329 weight of the empty vehicle 150