Algorithms and Complexity 9th

International Conference CIAC 2015

Paris France May 20 22 2015
Proceedings 1st Edition Vangelis Th.
Paschos
Visit to download the full and correct content document:
https://textbookfull.com/product/algorithms-and-complexity-9th-international-conferen
ce-ciac-2015-paris-france-may-20-22-2015-proceedings-1st-edition-vangelis-th-pasch
os/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Experimental Algorithms 14th International Symposium

SEA 2015 Paris France June 29 July 1 2015 Proceedings
1st Edition Evripidis Bampis (Eds.)

https://textbookfull.com/product/experimental-algorithms-14th-
international-symposium-sea-2015-paris-france-
june-29-july-1-2015-proceedings-1st-edition-evripidis-bampis-eds/

Algorithms and Complexity 10th International Conference

CIAC 2017 Athens Greece May 24 26 2017 Proceedings 1st
Edition Dimitris Fotakis

https://textbookfull.com/product/algorithms-and-complexity-10th-
international-conference-ciac-2017-athens-greece-
may-24-26-2017-proceedings-1st-edition-dimitris-fotakis/

Software Technologies 10th International Joint

Conference ICSOFT 2015 Colmar France July 20 22 2015
Revised Selected Papers 1st Edition Pascal Lorenz

https://textbookfull.com/product/software-technologies-10th-
international-joint-conference-icsoft-2015-colmar-france-
july-20-22-2015-revised-selected-papers-1st-edition-pascal-
lorenz/

Web and Wireless Geographical Information Systems 14th

International Symposium W2GIS 2015 Grenoble France May
21 22 2015 Proceedings 1st Edition Jérôme Gensel

https://textbookfull.com/product/web-and-wireless-geographical-
information-systems-14th-international-
symposium-w2gis-2015-grenoble-france-
Data Management Technologies and Applications 4th
International Conference DATA 2015 Colmar France July
20 22 2015 Revised Selected Papers 1st Edition Markus
Helfert
https://textbookfull.com/product/data-management-technologies-
and-applications-4th-international-conference-data-2015-colmar-
france-july-20-22-2015-revised-selected-papers-1st-edition-
markus-helfert/

Smart Cities Green Technologies and Intelligent

Transport Systems 4th International Conference
SMARTGREENS 2015 and 1st International Conference
VEHITS 2015 Lisbon Portugal May 20 22 2015 Revised
Selected Papers 1st Edition Markus Helfert
https://textbookfull.com/product/smart-cities-green-technologies-
and-intelligent-transport-systems-4th-international-conference-
smartgreens-2015-and-1st-international-conference-
vehits-2015-lisbon-portugal-may-20-22-2015-revised-sele/

Cloud Computing and Services Science 5th International

Conference CLOSER 2015 Lisbon Portugal May 20 22 2015
Revised Selected Papers 1st Edition Markus Helfert

https://textbookfull.com/product/cloud-computing-and-services-
science-5th-international-conference-closer-2015-lisbon-portugal-
may-20-22-2015-revised-selected-papers-1st-edition-markus-
helfert/

Web Information Systems and Technologies 11th

International Conference WEBIST 2015 Lisbon Portugal
May 20 22 2015 Revised Selected Papers 1st Edition
Valérie Monfort
https://textbookfull.com/product/web-information-systems-and-
technologies-11th-international-conference-webist-2015-lisbon-
portugal-may-20-22-2015-revised-selected-papers-1st-edition-
valerie-monfort/

Language and Automata Theory and Applications 9th

International Conference LATA 2015 Nice France March 2
6 2015 Proceedings 1st Edition Adrian-Horia Dediu

https://textbookfull.com/product/language-and-automata-theory-
and-applications-9th-international-conference-lata-2015-nice-
france-march-2-6-2015-proceedings-1st-edition-adrian-horia-dediu/
 Vangelis Th. Paschos
Peter Widmayer (Eds.)
LNCS 9079

Algorithms
and Complexity
9th International Conference, CIAC 2015
Paris, France, May 20–22, 2015
Proceedings

123
Lecture Notes in Computer Science 9079
Commenced Publication in 1973
Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen

Editorial Board
David Hutchison
Lancaster University, Lancaster, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Friedemann Mattern
ETH Zürich, Zürich, Switzerland
John C. Mitchell
Stanford University, Stanford, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
C. Pandu Rangan
Indian Institute of Technology, Madras, India
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max Planck Institute for Informatics, Saarbrücken, Germany
More information about this series at http://www.springer.com/series/7407
Vangelis Th. Paschos · Peter Widmayer (Eds.)

Algorithms
and Complexity
9th International Conference, CIAC 2015
Paris, France, May 20–22, 2015
Proceedings

ABC
Editors
Vangelis Th. Paschos Peter Widmayer
LAMSADE Inst. of Theoretical Computer Science
Université Paris-Dauphine ETH Zürich
Paris Cedex 16 Zurich
France Switzerland

ISSN 0302-9743 ISSN 1611-3349 (electronic)

Lecture Notes in Computer Science
ISBN 978-3-319-18172-1 ISBN 978-3-319-18173-8 (eBook)
DOI 10.1007/978-3-319-18173-8

Library of Congress Control Number: 2015936678

LNCS Sublibrary: SL1 – Theoretical Computer Science and General Issues

Springer Cham Heidelberg New York Dordrecht London

c Springer International Publishing Switzerland 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broad-
casting, reproduction on microfilms or in any other physical way, and transmission or information storage
and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known
or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the
editors give a warranty, express or implied, with respect to the material contained herein or for any errors or
omissions that may have been made.

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media
(www.springer.com)
 Preface

This volume contains the papers presented at CIAC 2015: 9th International Conference
on Algorithms and Complexity held during May 20–22, 2015 at the Université Paris-
Dauphine, Paris. This series of conferences presents original research contributions in
the theory and applications of algorithms and computational complexity.
The volume begins with invited papers and continues with contributed papers al-
phabetically arranged by the last names of their authors. There were 93 submissions for
this CIAC edition. Each submission was reviewed by at least three Program Committee
members. The committee decided to accept 30 papers. The program also includes four
invited talks.
We thank all the authors who submitted papers, the members of the Program Com-
mittee, and the external reviewers. We are also grateful to the four invited speakers,
Bernard Chazelle (Princeton University), Haim Kaplan (Tel Aviv University), Elias
Koutsoupias (Oxford University), and Mikkel Thorup (University of Copenhagen), who
kindly accepted our invitation to give plenary lectures at the conference.
We gratefully acknowledge the support from Université Paris-Dauphine, its Labo-
ratory LAMSADE, the AGaPe Research Group of the LAMSADE, and the EATCS.
We would also like to thank the Local Organizing Committee and in particular
Katerina Kinta, Cécile Murat, Olivier Rouyer, and Juliette de Roquefeuil for their ac-
tive participation in several organization tasks. Many thanks also to Mireille Le Barbier
for her valuable help in allocating all the necessary rooms in the University, for all the
conference main and side activities.
We finally would like to thank EasyChair for providing us a very friendly environ-
ment for handling the contributions and editing the proceedings of CIAC 2015.

March 2015 Vangelis Th. Paschos

Peter Widmayer
 Organization

Program Committee
Hannah Bast University of Freiburg, Germany
Vincenzo Bonifaci IASI-CNR, Rome, Italy
Jérémie Chalopin CNRS and Université d’Aix-Marseille, France
Victor Chepoi Université d’Aix-Marseille, France
Marek Chrobak University of California, Riverside, USA
Pierluigi Crescenzi University of Florence, Italy
Jurek Czyzowicz Université du Québec en Outaouais, Canada
Yann Disser Technische Universität Berlin, Germany
Thomas Erlebach University of Leicester, UK
Bruno Escoffier Université Pierre-et-Marie-Curie, France
Irene Finocchi Sapienza University of Rome, Italy
Fedor Fomin University of Bergen, Norway
Pierre Fraigniaud CNRS and University Paris Diderot, France
Herman Haverkort Technische Universiteit Eindhoven,
The Netherlands
Matúš Mihalák ETH Zürich, Switzerland
Luca Moscardelli University of Chieti-Pescara, Italy
Yoshio Okamoto University of Electro-Communications, Japan
Vangelis Th. Paschos Université Paris-Dauphine, France
David Peleg Weizmann Institute of Science, Israel
Marie-France Sagot INRIA and Université Lyon 1, France
Piotr Sankowski University of Warsaw, Poland
Maria Serna Polytechnic University of Catalonia, Spain
Paul Spirakis University of Liverpool, UK and CTI, Greece
Dimitrios Thilikos CNRS, LIRMM, France and UoA, Greece
Roger Wattenhofer ETH Zürich, Switzerland
Peter Widmayer ETH Zürich, Switzerland
Gerhard J. Woeginger Technische Universiteit Eindhoven,
The Netherlands
Organizing Committee

Additional Reviewers

Adamczyk, Marek Barbay, Jérémy

Albers, Susanne Bein, Wolfgang
Angelini, Patrizio Belmonte, Rémy
Bampas, Evangelos Bilò, Vittorio
VIII Organization
Blum, Christian
Bonnet, Edouard
Bonsma, Paul
Borassi, Michele
Boria, Nicolas
Brandes, Philipp
Brandstadt, Andreas
Brandt, Sebastian
Brunetti, Sara
Bulteau, Laurent
Calamoneri, Tiziana
Campelo, Manoel
Chikhi, Rayan
Chuzhoy, Julia
Couëtoux, Basile
Crochemore, Maxime
D’Angelo, Gianlorenzo
Dabrowski, Konrad Kazimierz
Das, Shantanu
Demange, Marc
Dürr, Christoph
Elbassioni, Khaled
Emek, Yuval
Estellon, Bertrand
Fanelli, Angelo
Feige, Uri
Fernau, Henning
Foerster, Klaus-Tycho
Garg, Naveen
Garnero, Valentin
Gaspers, Serge
Gavoille, Cyril
Gawrychowski, Pawel
Giroudeau, Rodolphe
Godard, Emmanuel
Goldwasser, Michael
Golovach, Petr
Gu, Qianping
Habib, Michel
Hackfeld, Jan
Jansen, Klaus
Keller, Barbara
Kellerer, Hans
Kim, Eun-Jung
Klimm, Max
Knauer, Kolja
Kowalik, Lukasz
König, Michael
Labourel, Arnaud
Langner, Tobias
Lauria, Massimo
Lazarus, Francis
Liotta, Giuseppe
Lv, Yuezhou
M.S., Ramanujan
MacQuarrie, Fraser
Mamageishvili, Akaki
Markakis, Evangelos
Marques-Silva, Joao
Mary, Arnaud
Meissner, Julie
Melideo, Giovanna
Messeguer, Arnau
Mnich, Matthias
Monaco, Gianpiero
Montanari, Sandro
Mäkinen, Veli
Naves, Guyslain
Nicolas, Bourgeois
Nisse, Nicolas
Nutov, Zeev
Otachi, Yota
Pacheco, Eduardo
Pajak, Dominik
Pascual, Fanny
Pasquale, Francesco
Préa, Pascal
Radoszewski, Jakub
Raffinot, Mathieu
Ravelomanana, Vlady
Rigaill, Guillem
Rossi, Gianluca
Russo, Luis M.S.
Sacomoto, Gustavo
Sau, Ignasi
Saurabh, Saket
Schaefer, Marcus
Schewior, Kevin
Seidel, Jochen
Sgouritsa, Alkmini
Shioura, Akiyoshi
Sikdar, Somnath
 Organization IX

Sikora, Florian Uno, Yushi

Sinaimeri, Blerina Uznański, Przemysław
Sirén, Jouni van Stee, Rob
Spieksma, Frits Vaxès, Yann
Stachowiak, Grzegorz Viglietta, Giovanni
Stamoulis, Georgios von Niederhäusern, Leonard
Stolz, David Wenk, Carola
Suchý, Ondřej Wiese, Andreas
Telelis, Orestis Winkler, Peter
Uitto, Jara Young, Neal
 Fast and Powerful Hashing Using Tabulation
(Extended Abstract)

Mikkel Thorup∗
University of Copenhagen, Department of Computer Science, Universitetsparken 5,
2100 Copenhagen East, Denmark

Abstract. Randomized algorithms are often enjoyed for their simplicity, but the
hash functions employed to yield the desired probabilistic guarantees are often
too complicated to be practical. Here we survey recent results on how simple
hashing schemes based on tabulation provide unexpectedly strong guarantees.
Simple tabulation hashing dates back to Zobrist [1970]. Keys are viewed as
consisting of c characters and we have precomputed character tables h1 , ..., hq
mapping characters to random hash values. A key x = (x1 , ..., xc ) is hashed to
h1 [x1 ] ⊕ h2 [x2 ]..... ⊕ hc [xc ]. This schemes is very fast with character tables in
cache. While simple tabulation is not even 4-independent, it does provide many
of the guarantees that are normally obtained via higher independence, e.g., linear
probing and Cuckoo hashing.
Next we consider twisted tabulation where one character is "twisted" with
some simple operations. The resulting hash function has powerful distributional
properties: Chernoff-Hoeffding type tail bounds and a very small bias for min-
wise hashing.
Finally, we consider double tabulation where we compose two simple tabula-
tion functions, applying one to the output of the other, and show that this yields
very high independence in the classic framework of Carter and Wegman [1977].
In fact, w.h.p., for a given set of size proportional to that of the space consumed,
double tabulation gives fully-random hashing.
While these tabulation schemes are all easy to implement and use, their anal-
ysis is not.

This invited talk surveys result from the references below.

References
1. Dahlgaard, S., Knudsen, M.B.T., Rotenberg, E., Thorup, M.: Hashing for statistics over k-
partitions. CoRR abs/1411.7191 (2014), http://arxiv.org/abs/1411.7191
2. Dahlgaard, S., Thorup, M.: Approximately minwise independence with twisted tabulation. In:
Proc. 14th Scandinavian Workshop on Algorithm Theory (SWAT). pp. 134–145 (2014)
3. Pǎtraşcu, M., Thorup, M.: The power of simple tabulation-based hashing. Journal of the ACM
59(3), Article 14 (2012), announced at STOC’11
4. Pǎtraşcu, M., Thorup, M.: Twisted tabulation hashing. In: Proc. 24th ACM/SIAM Symposium
on Discrete Algorithms (SODA). pp. 209–228 (2013)
5. Thorup, M.: Simple tabulation, fast expanders, double tabulation, and high independence. In:
Proc. 54th IEEE Symposium on Foundations of Computer Science (FOCS). pp. 90–99 (2013)

Research partly supported by an Advanced Grant from the Danish Council for Independent
Research under the Sapere Aude research carrier programme.
 Contents

Communication, Dynamics, and Renormalization . . . . . . . . . . . . . . . . . . . . 1

Bernard Chazelle

‘‘Green’’ Barrier Coverage with Mobile Sensors . . . . . . . . . . . . . . . . . . . . . 33

Amotz Bar-Noy, Dror Rawitz, and Peter Terlecky

A Refined Complexity Analysis of Finding the Most Vital Edges

for Undirected Shortest Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Cristina Bazgan, André Nichterlein, and Rolf Niedermeier

Orthogonal Graph Drawing with Inflexible Edges. . . . . . . . . . . . . . . . . . . . 61

Thomas Bläsius, Sebastian Lehmann, and Ignaz Rutter

Linear time Constructions of some d-Restriction Problems . . . . . . . . . . . . . 74

Nader H. Bshouty

Efficiently Testing T-Interval Connectivity in Dynamic Graphs . . . . . . . . . . 89

Arnaud Casteigts, Ralf Klasing, Yessin M. Neggaz,
and Joseph G. Peters

Competitive Strategies for Online Clique Clustering . . . . . . . . . . . . . . . . . . 101

Marek Chrobak, Christoph Dürr, and Bengt J. Nilsson

Scheduling with Gaps: New Models and Algorithms. . . . . . . . . . . . . . . . . . 114

Marek Chrobak, Mordecai Golin, Tak-Wah Lam, and Dorian Nogneng

MinMax-Distance Gathering on Given Meeting Points . . . . . . . . . . . . . . . . 127

Serafino Cicerone, Gabriele Di Stefano, and Alfredo Navarra

Evacuating Robots from a Disk Using Face-to-Face Communication

(Extended Abstract) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
J. Czyzowicz, K. Georgiou, E. Kranakis, L. Narayanan,
J. Opatrny, and B. Vogtenhuber

Planarity of Streamed Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

Giordano Da Lozzo and Ignaz Rutter

Clique-width of Graph Classes Defined by Two Forbidden Induced

Subgraphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
Konrad K. Dabrowski and Daniël Paulusma

Randomized Adaptive Test Cover. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

Peter Damaschke
XIV Contents

Contraction Blockers for Graphs with Forbidden Induced Paths . . . . . . . . . . 194

Öznur Yasßar Diner, Daniël Paulusma, Christophe Picouleau,
and Bernard Ries

On the Complexity of Wafer-to-Wafer Integration . . . . . . . . . . . . . . . . . . . 208

Guillerme Duvillié, Marin Bougeret, Vincent Boudet,
Trivikram Dokka, and Rodolphe Giroudeau

Label Placement in Road Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

Andreas Gemsa, Benjamin Niedermann, and Martin Nöllenburg

Discrete Stochastic Submodular Maximization:

Adaptive vs. Non-Adaptive vs. Offline . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
Lisa Hellerstein, Devorah Kletenik, and Patrick Lin

Parameterized Algorithms and Kernels for 3-Hitting Set

with Parity Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
Vikram Kamat and Neeldhara Misra

Simple Strategies Versus Optimal Schedules in Multi-agent Patrolling . . . . . 261

Akitoshi Kawamura and Makoto Soejima

Sharing Non-Anonymous Costs of Multiple Resources Optimally. . . . . . . . . 274

Max Klimm and Daniel Schmand

Algorithms Solving the Matching Cut Problem . . . . . . . . . . . . . . . . . . . . . 288

Dieter Kratsch and Van Bang Le

End-Vertices of Graph Search Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . 300

Dieter Kratsch, Mathieu Liedloff, and Daniel Meister

Deciding the On-line Chromatic Number of a Graph with Pre-Coloring

Is PSPACE-Complete . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
Christian Kudahl

A Lex-BFS-Based Recognition Algorithm for Robinsonian Matrices . . . . . . 325

Monique Laurent and Matteo Seminaroti

Mixed Map Labeling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339

Maarten Löffler, Martin Nöllenburg, and Frank Staals

Optimal Online Edge Coloring of Planar Graphs with Advice . . . . . . . . . . . 352

Jesper W. Mikkelsen

Approximability of Two Variants of Multiple Knapsack Problems . . . . . . . . 365

Shuichi Miyazaki, Naoyuki Morimoto, and Yasuo Okabe

Block Sorting is APX-Hard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377

N.S. Narayanaswamy and Swapnoneel Roy
 Contents XV

An Opportunistic Text Indexing Structure Based on Run Length

Encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
Yuya Tamakoshi, Keisuke Goto, Shunsuke Inenaga, Hideo Bannai,
and Masayuki Takeda

PSPACE-Completeness of Bloxorz and of Games with 2-Buttons. . . . . . . . . 403

Tom C. van der Zanden and Hans L. Bodlaender

Advice Complexity of Fine-Grained Job Shop Scheduling . . . . . . . . . . . . . . 416

David Wehner

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429

Communication, Dynamics, and Renormalization

Bernard Chazelle(B)

Department of Computer Science,

Princeton University Princeton, Princeton, NJ 08540, USA
chazelle@cs.princeton.edu

Abstract. This paper explores a general strategy for analyzing

network-based dynamical systems. The starting point is a method for
parsing an arbitrary sequence of graphs over a given set of nodes. The
technique is then harnessed to carry out a dynamic form of renormaliza-
tion for averaging-based systems. This analytical framework allows us to
formulate new criteria for ensuring the asymptotic periodicity of diffusive
influence systems.

Keywords: Dynamical systems · Networks · Renormalization ·

Influence systems

1 Introduction
There is by now a wide, well-established body of techniques for decomposing
networks into smaller pieces [3,8]. These include methods based on spectral par-
titioning, SDP relaxation, diffusion, coarsening, flows, metric embeddings, local
search, etc. By comparison, the cupboard of decomposition tools for dynamic
networks looks bare. Allowing edges to come and go puts basic connectivity
questions in a new light and calls for a novel set of tools [11]. This need, of
course, hinges on the relevance of dynamic graphs in the first place. It is easy
to argue that, in practice, networks rarely come with a fixed set of edges. On
the Web, for example, hyperlinks are added and deleted all the time. The same
is true of social networks and virtually any large graph subject to failure. The
present motivation for investigating dynamic networks emanates from a specific
concern, however: the dynamics of multiagent systems and, more ambitiously,
the emergence of collective behavior in living systems.
Think of how fireflies synchronize their flashes, birds form V-shaped flocks,
ants find shortest paths, and bacteria perform quorum sensing. The standard
approach to modeling such systems is to look at the individual organisms as
agents endowed with two kinds of rules: communication rules to specify which
agents “listen” to which ones under what circumstances; and action rules to
This work was supported in part by NSF grants CCF-0963825 and CCF-1420112.
Part of it was done when the author was an Addie and Harold Broitman Member
at the Institute for Advanced Study, Princeton, NJ.


c Springer International Publishing Switzerland 2015
V.Th. Paschos and P. Widmayer (Eds.): CIAC 2015, LNCS 9079, pp. 1–32, 2015.
DOI: 10.1007/978-3-319-18173-8 1
2 B. Chazelle

instruct the agents on what to do with the information they acquire. Communi-
cation between the agents is channeled through a dynamic network whose topol-
ogy changes endogenously as the system evolves over time. Before we describe
our approach to analyzing such systems, we provide a few words of intuition.

1.1 Parsing Dynamic Networks

The analysis relies on specific methods for tracking the propagation of informa-
tion at all scales. This means monitoring how often any two groups of agents of
any size communicate with each other. This, in turn, opens the door to divide-
and-conquer. To take an extreme example, suppose that the agents can be par-
titioned into two subsets that never exchange information. If so, each subgroup
is decoupled from the other and can be analyzed separately. Somewhat trickier
is the case of two groups A and B with no edges pointing from B to A. Since
information runs in the reverse direction of the edges,1 the group B is decoupled
and can be analyzed on its own. The same is not true of A, however, since the
group relies on information from B for its dynamics. What to do then? The
key observation is that, should the system B eventually settle around a fixed
point or a limit cycle, its predictable behavior will allow us to treat the group
B, after a suitable period of time, as a single block-agent: think of it as some sort
of super-agent with static or periodic behavior. In this way, the original system
can now be analyzed by first working on B and then turning our attention to a
block-directional system of |A| + 1 agents (+1 because of the block-agent). Think
of an Ancient Régime monarchy where the royal court B has its own internal
dynamics, one that is vital to the dynamics of A yet entirely oblivious to it.
Breaking down a system into its decoupled and block-directional parts is a
process that can be iterated recursively. One difficulty is that the decomposition
scheme (ie, the choice of A and B) cannot be fixed once and for all but must
adapt to the changes in the communication network. This is unusual. In statisti-
cal physics, for example, this kind of decomposition hierarchy is known as real-
space renormalization (more on which below). Unlike here, the decomposition
scheme is fixed and known a priori. In the planar Ising model, for example, the
communication graph is a grid, so block-spin renormalization naturally results
in a hierarchy of subgrids. The process can be specified without knowing any-
thing about the dynamical system itself. By contrast, we have here a feedback
loop between the dynamics of an influence system and the topology of its com-
munication channels, so that it is impossible to build any hierarchy before we
know which graphs will crop up when. This is the reason we call this brand of
renormalization algorithmic, to distinguish it from its closed-form version from
statistical physics and quantum mechanics.
How do we choose the partition of the agent set into A and B? There are many
considerations at play but one of them stands out: updating A and B should
be as infrequent as possible. To take a fanciful example, consider a basketball
1
An edge pointing from me to you indicates that I am listening to you and therefore
that the information flows from you to me.
 Communication, Dynamics, and Renormalization 3

game. It might be sensible to choose one team to be A and the other one B
on the grounds that the ball stays within a given team more often than not.
On the other hand, this might not be true when the action is near the basket
and possession switches rapidly between shooters, blockers, and rebounders. One
could then imagine changing the choice of the groups A and B every now and
then in order to keep the interactions between the two groups to a minimum.
Yet to find the absolute minimum is not an option. The choice must not only
mirror the flow of information across the networks but also proceed on-line: in
particular, one should not have to look ahead into the future to decide how to
split the agents into groups A and B.2
The dynamic assignment of A and B partitions the timeline t = 0, 1, 2, . . .
into a sequence of consecutive intervals within which the assignment is time-
invariant. Within each such interval, the intra-communication among the agents
of A (or B) could be itself quite complex, thus prompting us to partition A and B.
Proceeding in this way recursively  produces a hierarchical decomposition of the
timeline, ie, a parse tree such as (01)(234) (5)(67)(8 · · · ) . In this example, the
timeline forms the root of the tree. The root has two children associated with the
time intervals 01234 and 5678 · · · , which themselves have respectively two and
three children, with intervals 01 and 234 for one and 5, 67, and 8 · · · for the other.
As we show below, this allows us to view the transmission of information across
the agents at all relevant timescales and “renormalize” the system accordingly.
The parsing procedure is a message passing algorithm—as are, we should point
out, most spectral methods and belief propagation algorithms used for fixed
networks (a word we use interchangeably with “graphs”).

1.2 Temporal Renormalization

The systems are deterministic, so a given starting configuration of the agents
yields a single orbit. The question is under which conditions is the orbit attracted
to a limit cycle. Whereas attraction to a fixed point can often be analyzed by
looking at the orbit of interest and setting up a suitable Lyapunov function for
it, asymptotic periodicity does not lend itself to this kind of investigation. It
is often necessary to reason about the space of all orbits, which is, of course,
a source of complication: as though trying to understand the behavior of an
infinite sequence of networks was not hard enough, we must consider the space
of all possible such sequences at once. How do we even encode such a structure
so we can reason about it? Each orbit gives rise to a parse tree, so the challenge
is how to organize the set of all possible parse trees into a single structure:
this is the role of the coding tree. By way of analogy, consider the set of all
English sentences. Via lexicographic ordering, we can organize this set as an
infinite coding tree whose paths are in bijection with the sentences. Such data
structures are known in computer science as prefix, digital, or radix trees. Being
2
At the risk of belaboring the obvious, we mention that this concerns only the analysis
of the dynamical system: the choice of A and B has no incidence on the dynamics
itself.
4 B. Chazelle

associated with a sentence, each path can be parsed in accordance with the rules
of English grammar. The key insight is that sentences whose paths share long
prefixes will have parse trees with big overlaps: this in turn allows us to infer a
hierarchical decomposition of the coding tree itself.
In the case of influence systems, the coding tree is infinite and each path in it
corresponds to a (chronological) sequence of communication networks. The trick
is to infer from the parse trees associated with the paths a recursive decomposi-
tion of the entire coding tree itself. This can be thought of as a form of temporal
renormalization not carried out in closed form but algorithmically (a remark the
discussion below should help clarify). What makes the coding tree useful for the
analysis is that it is not merely a combinatorial structure but a geometric one:
indeed, a path of the coding tree corresponds to a whole set of nearby orbits
that share the same sequence of networks. These look like “tubes” in spacetime
Rn × N. For example, an influence system with 2 agents would produce a cod-
ing tree in three dimensions that might look a bit like a real tree of the sort
we encounter in nature. Paths are branches whose cross-sections have an area
(or volume in higher dimensions). For reasons we discuss below, bounding the
rate at which new branches are formed (entropic growth) and how thin they
get as we move further from the root (dissipation) is the key to the analysis.
Roughly speaking, bushy trees with thick branches correspond to chaotic sys-
tems. We have proven in a suitable model that a tiny random perturbation of the
input sends the agents into a limit cycle almost surely [6]. We follow a different
tack here and establish a more general result under plausible heuristic assump-
tions. To replace these assumptions by established facts appears to be a major
mathematical challenge which is left as an open problem. As a side-benefit, our
approach gives us a platform for working out the renormalization scheme in full,
something that was not needed in [6].
We define the model of diffusive influence formally in the next section (§2)
and discuss specific constructions as a warmup. In §3, we show how to parse
an arbitrary sequence of networks. The section is of independent interest. By
generalizing the idea of breadth-first search, this gives us a principled way of
tracking the propagation of information (eg, rumors or contagion) in dynamic
graphs. We sketched the idea earlier [6] but we give it the full treatment here.
In §4, we lay down the foundations of algorithmic renormalization and explain
in §5 how to use the framework to mediate between the two “forces” driving
the system: entropy and dissipation. Finally, we show how to analyze a diffusive
influence system in §6 by setting up the relevant recurrence relations. It bears
mentioning that the first three sections are highly general. It is only in §4 that
the piecewise-linearity of the system is used and in §5,6 that the “averaging”
nature of diffusive influence systems is exploited. In particular, the algorithmic
renormalization scheme applies to any discrete-time network-based dynamics.

2 The Model
Our model draws from the classic Hegselmann-Krause model of opinion dynam-
ics [9]. Part of its appeal is the simplicity of its definition: Fix a real parameter
 Communication, Dynamics, and Renormalization 5

r > 0 and initialize n agents on the real line R. At each time step, each agent
 r or
moves to the mass center of its neighbors, in this case, any agent at distance
less. In other words, the position xi ∈ R of agent i becomes xi ← |Ni |−1 j∈Ni xj
at the next step, where Ni = { j : |xi − xj | ≤ r }. The updates are carried out
synchronously and repeated ad infinitum. Numerical simulations suggest fast
convergence. Although the typical relaxation time has yet to be pinned down, it
is known that, within polynomial time, the agents end up frozen in single-point
clusters at least r away from each other [1,5,10,12–15].

2.1 Generalized HK-Systems

There are three natural ways to extend the original Hegselmann-Krause (HK)
model. One of them is to lift the agents into higher dimension instead of confining
them to the real line. Regardless of the dimension, the agents will still converge to
a fixed configuration in polynomial time [1]. Another modification is to replace
the update rule with a weighted mass center. Assuming nonzero self-weights,
convergence is still guaranteed but it might become asymptotic. To see why,
consider two agents at distance less than r moving toward each other one third
of the way at each step: x1 ← 13 (2x1 + x2 ) and x2 ← 13 (x1 + 2x2 ). In the phase
space R2 , any orbit is attracted exponentially fast to the line x1 = x2 . The
third type of extension is to redefine what it means to be a “neighbor.” Despite
massive empirical evidence that the system should converge to a fixed point, a
change as simple as allowing a different threshold ri for each agent i produces
a dynamics that is still unresolved. On the other hand, certain minor variants
are known to produce periodicity and even chaos [6]. To grasp the subtleties at
play, we need a more expressive palette to work with. We begin with a slight
generalization of HK systems (lin-DNF) and then push the generalization to its
natural limit (diffusive influence systems). Though looking vastly different to the
naked eye, this is an illusion: the two formulations are in fact equivalent (modulo
an adjustment in the number of agents).
A lin-DNF is a set of linear constraints expressed as a disjunction of conjunc-
tions, ie, P1 ∨P2 ∨· · · , where each Pl is of the form Q1 ∧Q2 ∧· · · and each Qk is a
halfspace uT x ≤ v (or uT x < v), where u, x ∈ Rn . We define the communication
graph G(x) by associating a node to each agent. The edges of the n-node graph
G(x) depend on x = (x1 , . . . , xn ): for each pair i = j, we choose a lin-DNF φij
and we declare (i, j) to be an edge of G(x) if φij (x) is true. A natural extension
of HK systems is provided by the update rule: for i = 1, . . . , n,
1 
xi ← xj and Ni = { j | (i, j) ∈ G(x) }. (1)
|Ni |
j∈Ni

Note that the original HK system is put in lin-DNF form very simply by setting
φij (x) as
   
xj − xi ≤ 0 ∧ xi − xj ≤ r ∨ xi − xj ≤ 0 ∧ xj − xi ≤ r . (2)
6 B. Chazelle

2.2 Diffusive Influence Systems

The definition of a diffusive influence system is identical to that of an HK system,

with the only difference coming from the communication network, specifically the
criterion used to include a pair (i, j) as an edge of G(x). We equip the pair with its
own first-order predicate φij (x) and make (i, j) an edge of G(x) if and only if φij (x)
is true. Recall that a first-order predicate over the reals is a logical sentence con-
sisting of universal and existential quantifiers bound to real variables y1 , . . . , ym ,
along with polynomial inequalities from Q[x1 , . . . , xn , y1 , . . . , ym ] tied together by
Boolean connectives. Formulation (2) is a particularly simple instance of a first-
order predicate, as it lacks quantifiers and bound variables, and uses only linear
polynomials.
Do we really need the full first-order theory of the reals? The answer is yes.
Here is a simple example taken from the field of robotics [2]. The agents are repre-
sented by points in a room full of obstacles and G(x) is defined as the “constrained
Delaunay graph”: this means that (i, j) is an edge of G(x) if there exists a sphere
passing through the agents i and j with no agent or obstacle protruding inside.
Instead of single real numbers, a variable xi is now a point in R3 and an edge (i, j)
is characterized by the truth-condition of a first-order formula with both existen-
tial and universal quantifiers. In plain English, the formula reads as follows: “There
exist a center and a radius such that, for all points p on an obstacle or at an agent,
p does not lie in the corresponding ball.” This is formally expressed as a first-order
predicate over the reals: the formula contains the symbols ∃, ∀, ∨, ∧, a number of
bound variables, and a set of polynomial inequalities with rational coefficients.
We see that, even for a simple communication network from robotics, alternating
quantifiers are already necessary.
Whereas the update rule itself (1) is kept deliberately simple, it is the high
expressiveness of the language in which the communication network is specified
that gives diffusive influence systems their distinctive feature. Because virtually
any edge selection rule is acceptable, the definition meets the primary objective of
the model, which is to allow the agents to have their own, distinct communication
rules.

2.3 Equivalence of the Models

Phrased in the language of first-order logic, diffusive influence systems seem

far removed from the lin-DNF formulation of generalized HK systems. The two
definitions are in fact equivalent. Indeed, any diffusive influence system can be
put in lin-DNF form after suitable transformation. This involves a number of
steps including quantifier elimination, linearization, tensoring, addition of new
agents, etc. (See [6] for details.) It might seem surprising that polynomials of
arbitrary degree can thus be replaced by linear ones, but this is made possible
by the piecewise linearity of the underlying dynamics.
We can go further and rid the lin-DNF formulation of all its Boolean con-
nectives. To do that, we consider the set D of hyperplanes formed by the linear
 Communication, Dynamics, and Renormalization 7

constraints in the formulas φij . By adding a variable if necessary,3 we can always

assume that these hyperplanes, called the discontinuities, are of the form uT x = 1.
They form an arrangement whose full-dimensional cells c, the continuity pieces of
f , are each assigned a directed n-node graph Gc . We extend the labeling to all of
Rn by defining G(x) as the graph with no edges if x lies on a discontinuity and
G(x) = Gc if x lies in cell c.
For convenience (specifically, to avoid cluttering the notation), we assume
that the number of hyperplanes in D is at most polynomial in n, so that the
number of graphs is bounded by nO(n) . After a translation along the all-one vec-
tor 1 and rescaling if necessary, we can always choose the unit cube X  [0, 1]n
as the phase space, so the continuity pieces are bounded open n-dimensional
polyhedra. To summarize the discussion,

Definition 1. A diffusive influence system (f, X) is a piecewise-linear system

specified by a map f from X to itself, x → P (x)x, where P (x) is the incidence
matrix of the communication graph G(x) augmented with self-loops, with each
row rescaled so as to sum up to 1. The matrix P (x) is constant over the cells of
a hyperplane arrangement in Rn .

We assume a positive diagonal to avoid spurious periodicities of no mathe-

matical interest.4 We define P (x) as the identity matrix for any x on a disconti-
nuity or on the boundary of the unit cube. Our discussion generalizes easily to
update rules based on weighted mass centers, and we use uniform weights across
each matrix row only for notational convenience.
A more substantive simplification in this paper is our assumption that the
systems are locally coupled, meaning that each φi,j depends only on xi , xj and
not on the other agents. Local coupling means that the presence of the edge (i, j)
in G(x) depends only on the sign-condition of x with respect to discontinuities
uT x = 1 for which only the coefficients ui and uj may be nonzero. To summarize,
a diffusive influence system (f, X) is specified by an arrangement of hyperplanes,
where each continuity piece c (an open n-cell) is labeled by a directed graph
whose edges (i, j) depend only the projection of c onto the (xi , xj ) plane. Figure 1
illustrates the case of two agents.

2.4 A New Brand of Renormalization

Most dynamic regimes can be observed in low dimensions (eg, fixed-point attrac-
tion, periodicity, quasi-periodicity, chaos, strange attractors), which is why research
in dynamics has had a tendency to focus on systems with few degrees of freedom.
At the opposite extreme, statistical mechanics prefers infinite-dimensional systems
for the mathematical benefits it can draw from the thermodynamic limit. Influence
3
For example, x1 = 0 might becomes x1 + x0 = 1, with x0 set to 1. Perturbation now
involves both x0 and x1 .  
4 01
For example, consider the two-node cycle G(x) with P (x) = .
10
8 B. Chazelle

Fig. 1. A map f for n = 2. Note that the continuity piece c is mapped continuously
but f (c) is not. It is this sort of fragmentation that makes the dynamics difficult to
analyze.

systems sit in the “mesoscopic” middle: many agents, but still too few, too diverse,
and too autonomous for any classical treatment. These network-based systems seek
to model the emergence of collective behavior through the multiplicity of individ-
ual interactions. But how do we go about analyzing the endogenous interaction of
many diverse agents?
The physical analogue would be to allow each particle in a gas to follow
its own laws of physics. While this feature alone may push influence systems
beyond the purview of statistical mechanics, the concept of renormalization will
play a key role (just as it does in the study of one-dimensional systems like the
logistic map). The basic idea is to rescale a system while retaining its dynamics.
In our case, scaling is with respect to both time and the number of agents.
Renormalizing an influence system is to produce another one with fewer agents
and a behavior resembling a coarse-grained version of the original one. If the
communication graph was fixed then graph clustering techniques might suggest
a way to do that: agents interacting with their √ neighbors
√ in a grid, for example,
can be clustered into
 subgrids,
 so that a n × n grid of n agents can be
renormalized as a n/k × n/k grid of “block-agents” consisting of k agents
each. Naturally, this decomposition can be repeated recursively.
This form of block-spin renormalization, famously introduced by Kadanoff,
works only if the interaction among the block-agents can be understood rea-
sonably well without having to track the precise behavior of their constituent
agents. In other words, a block-agent needs to be able to “hide” the internal role
of its agents from the outside. The other requirement is that the coarse-grained
system should look like a “blurred” version of the larger one. To achieve this,
it is customary to view the coarse-graining process itself as a dynamical system
mapping an influence system to another, simpler one. The goal is then to adjust
the coarse-graining parameters to home in on a fixed-point attractor and thus
allow the basic behavior to be retained throughout the renormalization steps.
What is the role of time in all of this? Presumably, a block-agent has a char-
acteristic timescale: the time it takes its own agents to settle in their long-term
 Communication, Dynamics, and Renormalization 9

mode. If this time is the same for all the block-agents, then we can rescale the
time axis by redefining its basic unit at every step in the coarse-graining hierar-
chy. To carry out this plan, of course, one needs to cope with the ever-changing
topology of the communication graph. Worse, the changes being endogenous,
one cannot postulate a prior on the graph distribution. Determinism is not nec-
essarily an impediment to using tools from statistical mechanics [4] because the
bifurcation analysis needs to focus on the “edge of chaos,” a region replete with
pseudorandomness.
In [6] we outlined an approach to network-based renormalization and derived
sufficient conditions for the asymptotic periodicity of diffusive influence systems.
We revisit the method here and show how to replace a certain timing assumption
by a general-position heuristic under which the critical region can be shown to
be of measure zero—the details are given below in §6. This is an opportunity
to develop the renormalization framework to its fullest, something our earlier
timing assumption [6] allowed us to bypass. The algorithmic renormalization of
dynamic networks is a general, powerful idea and one, we believe, likely to be
useful elsewhere.

2.5 Mixed Timescales

Diffusive influence systems have been shown to span the entire range of dynamic
modes, from fixed-point attraction to chaos. Predicting their behavior can be
undecidable [6]. It is no surprise therefore that they should exhibit periodic orbits
of any length. Remarkably, long limit cycles with large basins of attraction can
be manufactured as well: think of them as extremely long periodic orbits robust
under perturbation. We give such a construction below: the period is a tower-of-
twos of height proportional to the number of agents. This type of behavior may be
pathological but it touches upon a phenomenon at the heart of influence systems:
the mixing of timescales. If one regards attraction as a process of amnesia, then to
be caught in a limit cycle means to forget one’s origins. Indeed, an agent begins
with an unbounded amount of information (encoded in its initial position) but,
once trapped in a limit cycle, carries only a few positional bits from then on.
This near-total loss of memory is what we term “amnesia.”
In a diffusive influence system, some agents can “collude” to form memory
devices that can hold information over periods of time much higher than their
own characteristic times. These gadgets allow agents to recover their memory
just as they are about to reach amnesia, which in turn allows them to extend
their internal characteristic time and delay the eventual attraction to a limit
cycle. This archival mechanism can be nested hierarchically to create bigger
and bigger timescales. Remarkably, such constructions are robust to noise. The
existence of hierarchical schemes that combine to create new dynamical regimes
points to the need for time-based renormalization, which is the main theme of
this work. The ability to create higher scales both in length and time seems
a necessary component of any living system, and the relevance of the present
investigation should be appreciated in this light.
10 B. Chazelle

The Very Slow Clock. We describe an n-agent diffusive influence system with
a limit cycle of length roughly equal to a tower-of-twos of height n/2. This is
“two to the two to the two...” repeated n/2 times. The design relies on a small
gadget which converts an (n − 2)-agent system with period pn−2 = k into an
n-agent system with period pn ≈ 2k . All gadgets share two static agents a, b
positioned at −1 and 1, respectively; for convenience, we do not count them in
n. Each gadget also has its own pair of mobile agents. Applying the construction
(roughly) n/2 times leads to the desired result. As an inessential but convenient
exception, we allow zero entries in the matrix diagonals. The n (mobile) agents
are at positions x1 , . . . , xn . The update rules will automatically ensure that,
for all i: xi ≤ 0 precisely when the time t is −1 (mod pn ); and xi = (−1)i
precisely when t = 0 (mod pn ). For n = 2, we use a, b to (easily) engineer the
system so that (x1 , x2 ) follows the periodic orbit (−1, 1), (1, 1), (1, −1), (−1, −1).
In general, to form G(x), we extend the communication graph of the (n − 2)-
agent system by adding to it the two nodes n and n − 1, together with the edges
specified below:
(i) If xi ≥ 0 for some i < n − 1, then add the self-loop (xn−1 , xn−1 ) and the
edge (xn−1 , xn ). Note: agent n − 1 moves half way toward agent n; this case
is the most frequent, as it occurs whenever t = −1 (mod k).
(ii) Otherwise, add (xn−1 , a); if xn ≤ 0 then add (xn , b) else add (xn , xn−1 ).
Note: in the first (resp. second) case, agent n moves to the location of agent
b (resp. agent n − 1) and agent n − 1 moves to −1; this occurs only when
t = −1 (mod k), which is typically very rare.
The mechanism of the clock is quite simple. The n-agent system consists of a sub-
system of n−2 agents (the “subclock”) and a gadget that interacts only with the
static agents a, b. There is never any edge between the gadget and the subclock:
their interaction is mediated entirely through the positional constraints.
At time 0, the odd-labeled agents are at −1 and the others at 1. Rule (i)
kicks in for the next k − 1 steps, pulling agent n − 1 toward agent n, so they
end up separated by ≈ 2−k . At time k − 1, the subclock is entirely in negative
territory, so rule (ii) kicks in. If agent n (which stays always to the right of agent
n − 1) is still at a positive location, then it slides left to the position of agent
n − 1 while the latter is repositioned at −1. This is the key moment: in k steps,
the system has managed to move agent n to the left by as little as ≈ 2−k . As
long as xn > 0, we can repeat these k steps, which means the subclock can run
through roughly 2k cycles. (The leftward slide of agent n varies a little in length
at each iteration but always remains on the order of 2−k .) Note that we do not
rush to reset the subclock as soon as xn ≤ 0 but, rather, allow it to complete its
natural cycle. The initial position of the agents does not require fine calibration
and the astronomical period is robust under perturbation.
The recursive construction of the Very Slow Clock (VSC) suggests how renor-
malization should proceed. The n-agent system VSCn results from nonlocal-
coupling between a clock VSCn−2 and a gadget consisting of two private agents
and two shared ones. The characteristic timescale of the gadget is exponentially
higher than that of the VSC with which it is coupled. Renormalizing the system
 Communication, Dynamics, and Renormalization 11

is easy because it was essentially put in by hand. In general, to tease out the
subparts of the dynamics that can be “factored out” is no easy task. In the end,
renormalization is about dimension reduction. When the graph never changes,
the dynamics can be expressed by the powers of a fixed matrix and the dimension
reduction is done by breaking down the system into its eigenmodes. When the
communication topology keeps changing, however, linear algebra is of little use.
The dimension reduction cannot be carried out in “closed form.” It would seem
that it can only be performed as a step-by-step process that evolves alongside
the dynamics of the system. This is what algorithmic renormalization tries to
achieve.
Theorem 1. For any n > 2, there exists an n-agent diffusive influence system
with a periodic orbit of length equal to a tower-of-twos of height proportional to
n. The dynamics is robust under perturbation.

3 Parsing Graph Sequences

The algorithmic renormalization of influence systems relies on a general proce-

dure of independent interest: a method for parsing arbitrary graph sequences.
Recall that the analysis of influence systems rests crucially on our ability to
decompose an infinite sequence of communication graphs hierarchically. By build-
ing a tree on top of the network sequence, we are thus able to break down the
original system into subsystems. This is similar to forming the parse tree of an
ordinary English sentence. While this operation is driven by the rules of syn-
tax, what sort of rules will we use then to parse a sequence of communication
graphs? In a nutshell, the idea is to track the propagation of information across
the networks (think of a virus spreading in a population) and use the pauses (for
example, due to poor connectivity) as breakpoints to guide the construction. The
parsing algorithm can be seen as a grand generalization of breadth-first search
for dynamics graphs.
Given a (finite or infinite) sequence G = G0 G1 G2 · · · of directed graphs over
the same set of n labeled nodes, make the lowest-labeled node “wet” and declare
wet any node pointing to a wet node in G0 . Next, switch to G1 and call wet any
node that is either already wet or points to a wet node via an edge of G1 . Water
flows only to immediate neighbors in a single step and wet nodes remain wet.
We use water as an anology but we could substitute rumor propagation or viral
contagion. Transmission runs in the reverse direction of the edges, following the
principle that information is acquired only by pointing to it. As soon as all the
nodes become wet (if ever), the subsequence of graphs considered up to this point
is said to form a wave. When this occurs, we dry up all the nodes and repeat
the entire process, restarting from the current point in the graph sequence. It
might happen that some nodes never get wet or that wetness propagation is
interrupted for long periods of time before resuming again. What to do in such
cases is at the heart of the parsing procedure, which is called the flow tracker.
12 B. Chazelle

3.1 Making Waves

2
We treat the sequence G as a word in a language over an alphabet of 2n letters,
where each letter represents an n-node graph (or, equivalently, an n-by-n 0-
1 matrix). Parsing the sequence means producing a parse tree T (G): this is a
rooted tree whose leaves are associated with the graphs G0 , G1 , G2 , . . . from
left to right and whose internal nodes reflect important transitions during the
flooding. As we shall see, even when the sequence G is infinite, the depth of the
tree remains finite. For example, if all the graphs are strongly connected (or if
the graphs are connected but undirected), then5
 seq
T (G) = (G0 · · · Gi1 )wave (Gi1 +1 · · · Gi2 )wave · · · , (3)
where ik is the time at which the graph becomes entirely wet for the k-time. The
tree T (G) has depth only two because it is a simple case. The superscripts indicate
the nature of the nodes: the node (· · · )seq is the root of the parse tree of the origi-
nal sequence; its children (· · · )wave correspond each to a wave and their number is
infinite; the leaves of the tree are given by G0 , G1 , etc. In this particular case where
all the graphs are assumed to be strongly connected, the waves are less than n-long
but in general there is no a priori bound on their lengths. The first two levels of any
parse tree look like (3); most trees will have higher depth, however, and recursive
rewriting rules will be needed to express them.
It is time to put this informal description on a more rigorous footing. Recall
that G is an arbitrary (bounded or not) sequence of directed graphs over n
labeled nodes. We denote by B the set of nodes initialized as wet. So far, we
have considered only the case where B is the singleton consisting of the lowest-
labeled node. A variant of the procedure will soon require wetting more than
one node at the outset, hence the use of the set B. We put together a partial
parse tree in the form of (wavemaker (G, B))seq , which we then proceed to refine.
This is how wavemaker works:
wavemaker (G, B) “ G is a graph sequence and B ⊂ [n] ”
W0 ← B and print ‘ ( ’
for t = 0, 1, . . .
print ‘Gt ’
Wt+1 ← Wt ∪ { i | ∃j ∈ Wt such that (i, j) ∈ Gt }
if |Wt+1 | = n then Wt+1 = B and print ‘ )wave ( ’
print ‘ )wave ’

We bend the usual conventions of programming language theory by allowing

the last print statement to be executed even if the sequence G, hence the for-
loop, is infinite.6 The output of wavemaker looks like the right-hand side of (3).
5
Throughout this paper, to save space, we represent trees not as stick figures but
as nested parenthesis systems. Here is a treewhose
 root
   has
 two children, with the
leftmost one having two children of its own: ( )( ) .
6
In the odd chance that wavemaker prints a single spurious ( )wave at the end, we
simply erase it.
 Communication, Dynamics, and Renormalization 13

A wave can be either complete or incomplete, depending on whether all the

nodes get wet: a wave is complete if it is terminated by the successful outcome
of the conditional “if |Wt+1 | = n.” For that reason, an infinite wave is always
incomplete but the converse is not true: a finite sequence G might simply run out
of graphs to flood all the nodes, hence leaving an incomplete final wave. If G is
infinite, then either all the waves are finite but their number is not (as is the case
with strongly connected graphs) or there are a finite number of waves but the last
one is infinitely long. In this case, the partial parse tree (wavemaker (G, B))seq is
of the form:  seq
(G0 · · · Gi1 )wave · · · (Gil +1 · · · )wave . (4)
Why isn’t this the end of the story? The reason is that waves themselves need
to be parsed, especially when they are very long. For example, suppose that B is
a singleton and its node is always disconnected from the other ones. In that case,
water never flows anywhere and wavemaker parses G as ((G0 G1 G2 · · · )wave )seq ,
which is as simple as it is useless. Clearly, parsing must be carried out even
during flooding delays. We explain now how to do this.

3.2 Parsing a Wave

A flooding delay implies that the current “wet” set Wt stops growing: it could
be momentary or permanent (there is no way to tell ahead of time). During any
such delay, the graphs Gk have no edges pointing from the dry nodes to the wet
ones, ie, from [n] \ Wt to Wt . This motivates the definition of a block-sequence as
any subsequence of G0 G1 G2 · · · such that, for some partition [n] = A ∪ B of the
nodes, no edge from the subsequence points from B to A: the block-sequence
is said to be of type A → B. Let G = G0 G1 G2 · · · be a wave. Recall that, in
the case of a complete wave, this means that all the nodes are wet at the end
but not earlier. As usual, B is set to the singleton consisting of the lowest-
labeled node. By definition of a wave, wavemaker (G, B) terminates at or before
the first time |Wt+1 | = n. Let t1 , t2 , . . . be the times t at which Wt ⊂ Wt+1 (ie,
strict inclusion). These coupling times indicate when the water propagates. They
break down the wave G into block-sequences Gk via (blockseqmaker (G, B))wave .
Using superscripts to distinguish the wave nodes from the block-sequence kind,
the output is of the form:
 wave
(GA0 →B0 )bseq Gt1 (GA1 →B1 )bseq Gt2 (GA2 →B2 )bseq Gt3 · · · . (5)

blockseqmaker (G, B) “ G is a wave and B ⊂ [n] ”

W0 ← B and print ‘ ( ’
for t = 0, 1, . . .
Wt+1 ← Wt ∪ { i | ∃j ∈ Wt such that (i, j) ∈ Gt }
if Wt = Wt+1 then print ‘Gt ’
else print ‘ )bseq Gt ( ’
bseq
print ‘ ) ’
14 B. Chazelle

As was the case with wavemaker, we remove empty parenthesis pairs ( )bseq ;
so, say, in the case of strongly connected graphs, the output of blockseqmaker
for a wave will simply be Gt1 Gt2 · · · (ti = i − 1). The block-sequence GAk →Bk
is a maximal subsequence of the wave G that witnesses no water propagation:
it is of type Ak → Bk , where Ak = Wtk +1 , Bk = [n] \ Ak , and B ⊆ Ak ⊂ Ak+1 .
The initialization requires setting t0 = −1 so that A0 = W0 = B. Note how
the letters A and B switch roles (more on this below). Until now, B has been a
singleton: parsing block-sequences, our next topic, will change all that.

3.3 Parsing a Block-Sequence

We have built the first three levels of the parse tree so far. The root is associated with
the full sequence and its children with its constituent waves. The root’s grandchil-
dren denote either single-graphs (leaves of the tree) or block-sequences. Parsing the
latter repeats the treatment of general sequences described above with two small
but crucial differences. Let GA→B denote a block-sequence of type A → B. First,
we break it down into block-waves by applying wavemaker(GA→B , B). The output
is of the form
 bseq
(G0 · · · Gi1 )bwave (Gi1 +1 · · · Gi2 )bwave · · · . (6)
We note that B is now wet at the outset. This contrasts with the circumstances
behind the creation of a block-sequence of type A → B, which feature wet A,
dry B, and delay in water propagation. This radical change in wetness status
explains the need for a recursive method that encapsulates wetness status by
local variables hidden from the outside. It is in that sense that the renormaliza-
tion is truly algorithmic.
The block-waves constitute the children of the node associated with the
block-sequence GA→B . Let H denote any one of the subsequences of the form
Gik +1 · · · Gik+1 (i0 = −1). To extend the parse tree further, we replace in (6)
each occurrence of the block-wave (H)bwave by (blockseqmaker (H, B))bwave .
Each block-wave is thus parsed as
 bwave
(GAB )dec Gt1 (GA1 →B1 )bseq Gt2 (GA2 →B2 )bseq Gt3 · · · . (7)
The parsing of a block-wave differs from that of a wave (5) in two ways:
• First, we notice the difference in the first term GAB . As we scan through the
sequence H, we may have to wait for a while before some edge actually joins
A to B. Since there is no edge from B to A, the two sets are decoupled until
then (hence the superscript dec). In other words, A and B form a cut in all
the graphs in the sequence GAB : such “decoupled” sequences are parsed as
a single leaf in the tree.
• The second difference is more subtle. When parsing an ordinary wave, recall
that the block-sequence GAk →Bk satisfies Ak = Wtk +1 . In the case of a block-
wave, instead we have Ak = Wtk +1 \ B (and, as usual, Bk = [n] \ Ak ). Note
that it is still the case that Ak ⊂ Ak+1 . This assignment of Ak satisfies the
main requirement of a block-sequence of type Ak → Bk , namely the absence
of any edge from Bk to Ak .
 Communication, Dynamics, and Renormalization 15

We had to amend the old invariant Ak = Wtk +1 because of inductive soundness.

Here is an example to illustrate why: Let G0 = a → b c ; G1 = a → b → c ;
G2 = a ← b → c ; G3 = a b → c; and G4 = G1 . In this notation, G0 has a single
edge (a, b). The block-wave G0 · · · G4 of type {a, b} → {c} is parsed as
 bwave
(G{a,b}{c} )dec G1 (G{b}→{a,c} )bseq G4 , (8)

where G{b}→{a,c} = G2 G3 . The reason why setting Ak = Wtk +1 is a bad idea

is that the block-sequence G2 G3 would then be interpreted as G{b,c}→{a} , which
would have the effect of making a block-sequence of type {b, c} → {a} a child of
another one of type {a, b} → {c}. To ensure the soundness of the recursion, we
need the cardinality of the A-set to drop by at least by 1 as we go from from
parent to child: this property is ensured by setting Ak = Wtk +1 \ B since this
implies that Ak ⊂ A.
There is a subtlety in the recursion. The arrival of G4 brings in the edge (a, b),
which is incompatible with a block-sequence of type {b} → {a, c}. It is therefore
no surprise to see Gb→ac in (8) terminate and give way to the single-graph G4 .
Actually this is not the true reason for the termination of the block-sequence.
Indeed, if G4 were of the form b → a → c, the block-sequence would still terminate
regardless of its continuing compatibility with the type {b} → {a, c}. The true
cause of termination is the growth of Wt . This shows that a node can only be
parsed if the relevant parameters of its ancestors are already known.

3.4 Rewriting Rules

It is convenient to think of the flow tracker (the name for the entire parsing
algorithm) as the application of certain productions of the sort found in context-
free grammars: seq, wave, bseq, bwave are used as nonterminal symbols and sg,
dec as terminals; here, sg is shorthand for “single-graph.” Via the formulas (3,
5, 6, 7), the flow tracker yields the following productions:
⎧
⎪
⎪ seq =⇒ (wave, wave, . . . )
⎪
⎨ wave =⇒ (bseq, sg, bseq, sg, . . . )
(9)
⎪
⎪ bseq =⇒ (bwave, bwave, . . . )
⎪
⎩
bwave =⇒ (dec, sg, bseq, sg, bseq, sg, . . . )
How sensible is it to make GAB a leaf of the parse tree? On the face of it, not
very much. Indeed, suppose that G is a long, complicated sequence with a rich
parse tree and V is its node set. Form a new sequence G  by adding an isolated
node a (with no edge to or from V ). Further, suppose that a is given the lowest
label so that it defines the starting singleton B. The flow tracker will parse G  as
the 6-node single-path tree: T (G  ) = ( ( (G{a}→V )bseq )wave )seq , with G{a}→V =
( ( GaV )dec )bwave . Adding a single node hides the richness of T (G) by producing
a completely trivial parse tree. Of course, it is quite possible that picking another
node for B besides a would produce a more interesting parse tree. But optimiz-
ing the selection of the starting singleton is out of the scope of our discussion.
16 B. Chazelle

As for the parallel treatment of A and B in GAB , this is something for the renor-
malization of the dynamical system itself to handle. The reason we delegate this
task is that A and B may not operate along the same time scale. Indeed, in a
block-sequence of type A → B, the set B, unlike A, can be handled separately
over the entire length of the sequence. It follows from our previous discussion
that the parse tree has depth at most 2n + O(1), as a typical downward path
reads
seq, wave, bseq, bwave, bseq, bwave, . . . , bseq, bwave, dec/sg.
The number of waves or block-waves that are the children of a given parent
can be unbounded, however. In the case of diffusive influence systems, large
degrees express structural properties of the dynamics and play a key role in the
analysis. We conclude our discussion of graph-sequence parsing with a simple
example of the flow tracker in action. Let G0 = G1 = G4 = G6 = a ← b ← c,
G3 = G5 = a → b → c, and G2 = a b → c. Setting G = G0 · · · G6 and choosing a
to form the initial singleton B, we have
 wave  bseq  bseq wave seq
T (G) = G0 G1 ((G2 )dec G3 )bwave G4 (G5 )bwave G6 .

4 Algorithmic Renormalization
Let (f, X) be a system as in Definition 1. The coding tree Tf (X) of (f, X) cap-
tures both its geometry and its symbolic dynamics [6]. We defined it informally
as a combinatorial object tracing the graph sequences of the orbits and we made
a passing remark about the geometric information it encodes. We formalize these
ideas now. The levels of the coding tree correspond to the times t = 0, 1, 2, etc.
Combinatorially, its paths encode the entire symbolic dynamics of the system
by listing in chronological order all the possible communication graph sequences
formed by the orbits. In addition, nodes carry geometric information about the
corresponding orbits. Each node v of the coding tree is associated with two cells
Uv , Vv , where tv is the depth of node v: the cell Uv is a continuity piece of f tv ,
ie, a maximal connected subset of X = [0, 1]n over which f tv is continuous; and
Vv = f tv (Uv ) (which is not necessarily n-dimensional). The coding tree is defined
inductively by starting with the root, Uroot = Vroot = X, and attaching a child
w to a node v for each of the cells c into which the discontinuities of f break Vv .
We then set Vw = f (c) and Uw = Uv ∩ f −tv (c).7 We define Vt = tv =t Vv and
note that Vt includes all the points reachable in t steps. Any point reachable in
t+1 steps can also be reached in t steps (just start at the second step); therefore,

Vt+1 ⊆ Vt . (10)

The nesting time ν = ν(Tf ) is the smallest t such that Vt intersects no discon-
tinuity. This number cannot be bounded if the system is chaotic; on the other
7
Recall that f is the identity along the discontinuities so there is no need to define
children for them. This causes a node to become a leaf if Vv falls entirely within one
of the discontinuities, which can be handled separately and hence assumed away.
Another random document with
no related content on Scribd:
hearted fellow, and does not scorn old friends. Many’s the nod and
word he has given me from the billiard-room window at Sidney, when
I have been passing, let who would be there. Everybody is glad of
the prosperity of John Lawe, Esq.”
Somebody having made inquiry about the voyage, Green went on,
“O, that’s the worst part of it. It’s horrid enough, to be sure, to be
cooped up for months on board, and all so solemn and dull, and no
getting out of the way of the clergyman. But it’s not so bad as it used
to be, when they treated such folks as we like so many wild beasts.
They paid the captain so much a head for the people embarked, and
never asked how many he landed; so he starved as many as he
pleased, and stowed them so close that scores were stifled by the
way. It was mighty dull work then for those that got safe; the labour
was so hard, and no liberty. There was little encouragement to go to
the colonies then. But now that they don’t kill one by inches by the
way, it is worth putting up with the passage, for the chance of making
one’s fortune at the end of it.”
“Particularly for them that have friends in power to get fine
situations for them,” said Jerry pertly. “Bob and I are going to have
good care taken of us, I hear. But it’s a great plague that the old
ones are going to be spies over us. It will spoil our sport terribly,
unless we can manage to cut them.”
“That’s better than having them whining and praying after us all
the way from here to the ship, as the old folks mostly do,” said
Green. “When I went before, my father behaved as if he was
following me to the gallows. He knows better now. He gave me the
wink yesterday for a sharp chap that knew how to take care of
myself. He said,—true enough,—that the worst blunder I ever made
was coming back when once I was well off.”
“Aye, aye, Green; a certain person knew how to take care of
herself as well as you. She knew better than to keep herself single
five years for you. ’Tis a fine feather in her cap to have brought you
so far on a fool’s errand.”
Green tried to conceal his visible passion under an appearance of
indifference, while he muttered that a better one than he came for
would follow him out very soon, if the judge did not baulk them of the
sentence they meant to get pronounced upon her.
 “Here they come, lads!” he cried, interrupting himself. “All is ready:
our carriage at the door! Put a bold face upon it, boys! Now for it!
Don’t have anything to say to the whiners at the gate. Curse all spoil-
sports! Give them three cheers, boys! Hurra! hurra! hurra!”
And gibing, jeering, laughing, shouting, went the batch of convicts
through a throng of relatives and former companions, and gazing
strangers; some of whom were pale and weeping, others signing and
winking, and more gaping in wonder and pleasure at the scene;
speculating upon whether the largest share of punishment did not
rest with those who were left behind. Bob, and one or two other
scowlers, were almost overlooked in the company of adventurers,
who seemed to be going forth merrily to cheat the law, and seek their
fortunes in a land of plenty.
 Chapter IV.

NEW HOMES.

Ellen was the first of the family that arrived at Hobart Town in Van
Diemen’s Land. Next came the convict-ship, which was sent round to
Launceston to disembark its passengers; that port being nearer the
district where the convict labour was to be employed. When the
batch of parish emigrants arrived, a fortnight afterwards, Frank
found, on application to the proper government officer, that his sister
had landed in good health, and had received a high character from
the clergyman and his lady who had come over as superintendents
of the expedition; that Ellen had been forwarded, with a few of her
fellow-passengers, to the district where a service had been procured
for her as dairy-maid on a settler’s farm; and that care had been
taken that her parents and brother should be indentured to farmers
in the same neighbourhood. So far, all was well. Frank could learn
nothing about his brothers, except that they were to be landed at
Launceston, and that Launceston was within thirty miles of the spot
where he was to be located. The officer he was speaking to had
nothing to do with the arrangements respecting convicts: his
business was to take care of emigrant labourers on their arrival.
Castle himself could not help being pleased at the appearance of
things at Hobart Town, when he and Frank took a walk, the evening
after their arrival. The only objections he could think of were, that
there were few shops; that it was not at all likely that the country
inland should be half so civilized as what he saw; and that it was a
thing he had not been used to, to have Christmas fall at the hottest
time of the year, and the trees green all the winter through. It was
now May; and they told him that winter was coming on, and yet that
the woods would look as green as now all the time; and that the
snow, if there was any, would not lie more than a day on any ground
but the mountain tops, and a bleak common here and there. They
told him that for more than three hundred days in the year the sun
would shine all day, and the air be dry and pure, and seldom too hot
or too cold. All this was what he had not been used to, and did not
know how to believe. His son supposed that if it came true, he would
not object; as one of the consequences of such a climate is that
English people have much better health, and live, on the average, a
good deal longer at Van Diemen’s Land than at home. Castle
peevishly laughed at all talk about life and health, when it was, in his
opinion, doubtful whether they might not be starved to death within
three months. His son left this point to be demonstrated by time
rather than by argument; and meanwhile observed that there were
few signs of starvation about Hobart Town, in which, besides the
government residence, there are nearly eight hundred houses, most
of which are surrounded with gardens; the dwellings having been
originally built on separate allotments of land, of a quarter of an acre
each. The streets cross at right-angles, and command fine views of
the neighbouring country, and afford cheering evidences of the
success of the industry which has sought employment there. A dock-
yard is seen on the river’s brink; and corn-mills, tanneries, breweries,
a hat-manufactory, &c., are conspicuous in the midst of the town. An
amphitheatre of green hills rises to the westward, the crowning
summit of which is 4000 feet high; and from these hills descends a
fine stream of water, flowing through the town into the Derwent,
which, with its varying expanse and beautifully wooded bays and
sloping shores, forms the eastern boundary. This view was little
enough like what Castle had fancied in opposition to all that he had
been told. He was for ever picturing to himself a region of wild
woods, or bleak plains covered with snow; and he was now as much
surprised at the sight of meadows, hills, dales, and a thriving town,
with a blue sky overhead, as if he might not have known as much
before. He had complained of his hard lot in being indentured as a
shepherd; and no wonder, while he thought his flocks were to inhabit
a dreary wilderness; but now that he found he had nothing to fear
from storms and snow-drifts, that the pastures were excellent, the
springs plentiful, and the sheep as fine as the world can produce, he
began to think he might be worse off in point of occupation; though
he would give nobody the satisfaction of hearing him say so. His wife
was to be a domestic servant in the same farm where he was
shepherd; and even little Susan was carefully stipulated for; the
labour of children being valuable at almost any age, in a place where
much more assistance is wanted than can be had.
The first part of their journey to the Dairy Plains, (the district where
they were to settle,) was through the very choicest portion of the
island, both as to beauty and fertility. It is not surprising that those
who first surveyed this tract, and took it as a fair sample of the island
at large, should have represented Van Diemen’s Land as a terrestrial
paradise, and been suspected of exaggeration through the
favourableness of their report. The fact is, the island is supposed to
contain about 15,000,000 acres,—one-third of which is considered
arable, another third fit for sheep-pasture, and the rest unprofitable
at present. The country between Hobart Town and Launceston
consists of green hills and fertile plains, among which towns and
villages and solitary dwellings are interspersed. Rivers wind between
their wooded banks, and streams flow down from the high grounds.
Excellent macadamized roads run through the whole district, and
branch off to the growing settlements on either hand of the main
track. It was a great amusement to Frank to compare whatever he
met with that partook of the civilization of his own country with
whatever looked new and strange. Before leaving Hobart Town, he
had been all the more struck with its printing establishments, its
Mechanics’ Institute, its Book Society, and schools, from observing
the strangeness of the natural productions that he met at every step.
In the gardens he beheld tea trees where he had been accustomed
to see lilacs and laburnums; and cotton plants, myrtles, and
geraniums growing as tall as himself, and spreading out into bushes.
The very grass grows differently;—not stringy in the roots and
carpetlike in the surface, as in England; but in tufts. Parrots, instead
of canaries, were the pets of young ladies; and the bandicoot was
offered for sale instead of the rabbit. Cockatoos instead of crows
were to be frightened away from the fields and gardens; and flocks
of pigeons among the stubble looked as much like partridges as
pigeons; only more beautiful,—with their gold-dropped wings,—than
either species in England. On the road, in like manner, the freestone
bridge over the Jordan, the postman on horseback, the tillage and
inclosures, looked British; but the evergreen woods, in the midst of
which arose the peppermint tree to a lofty height;—the herds of
kangaroos coming out of their covert into the dewy plains at sunrise;
—the spotted opossums climbing and descending the forest trees;—
the black swans sailing on the lakes, and uttering cries like the
creaking of an old sign-board;—all appeared foreign, and scarcely
belonging to the people who had settled among them.
A sight of a yet different character met the eyes of the travellers
near the close of the second day, when they were drawing near their
future abode in the province called Norfolk Plains, in the centre of
which lay the Dairy Plains, where Ellen was expecting them. They
had for some time quitted the broad road, and were following a track
along which their waggon proceeded with tolerable convenience. At
last they came to a point beyond which it had not been carried, and
where a gang of labourers was at work roadmaking;—not as in
England, each man intent upon his own heap of stones, free in limb
and thoughtful in countenance;—not as in Ireland, where some are
lounging and all are joking;—but charged with the fetters of felons,
and superintended by an armed taskmaster. As Frank looked upon
these wretches, with their hardened or woful countenances, he felt
indeed that he was not in England, but in one of her penal
settlements,—breathing the air of one of the places where her vice
and misery are deposited. His very soul became sick when, as the
labourers turned to stare at the somewhat uncommon sight of a
waggon full of travellers, he met the eyes of his convict brothers. He
hoped that his companions would not perceive them; but he soon
found that his father did, by his testy complaints of the jolting of the
cart, of cold and heat, and what not. The unhappy mother looked on
her outcast children with as much curiosity as compassion. Bob
turned away, and stooped to his work, never looking up till they were
out of sight; but Jerry waved his cap and shouted, and dared Frank
to a wager which of them would first be free to work for themselves:
whether it would take longest to work out his sentence, or to pay for
Frank’s passage and settlement. This supplied a new theme of
complaint to Castle, who wrought himself up into a passion about his
being a slave, with all his family. Frank, who hated bondage as much
as any man, thought it could hardly be called slavery to contract to
work for one person for a certain time, in return for advantages which
could not otherwise be obtained. If disappointed of these
advantages,—of sufficient food, clothing, shelter, and money wages,
—the contract was void, and no harm done; if not disappointed, the
object was gained. The evil lay, not in their case as labourers; but as
honest men. Felons ought not to be let off so easily, (because their
labour happened to be more valuable than at home,) as to make
disgrace, for which many of them did not care, their only punishment;
their worldly circumstances being actually bettered by their removal
to a new colony. It was not that labourers need be better off than
their family would probably be, four or five years hence; but that
felons ought not to be placed in as good circumstances as the
honest emigrant at the end of the same period.
Frank was not yet aware (as he afterwards became) that, for want
of knowing the rate of wages at the colonies, emigrants often bind
themselves for a much lower rate than they might obtain if they went
free, or if they were properly informed of the existing state of things;
and thus think themselves deceived, and are tempted to break their
contract when they find how matters stand. This evil is to be referred
to the ignorance of emigrant labourers, quite as much as to the close
economy of the settlers, and should induce all who have heard of it
to obtain such information before concluding their bargain as will
save them from repentance afterwards, and guard them against
quarrels on this score with their new masters;—quarrels, which,
always a great evil, are most so in newly settled countries, where all
hands and hearts are wanted to work together for the common good.
As it is, a British artisan jumps at the offer of a plentiful subsistence
and 2s. a-day besides for five years, out of which the expenses of
his removal are to be paid; and for this rate he binds himself. When
he gets to his destination, he finds that this plentiful subsistence,
including meat, bread, beer or spirits, tea, sugar, comfortable
clothing, and a convenient dwelling, costs no more than 2s. a-day,
and that, if free, he might earn, being a good workman, from 7s. to
12s. a-day, or even 15s., if he be a superior mechanic of a scarce
class. It is mortifying to find that he has sold himself, however much
higher than formerly, for less than he is worth in his new position;
and hence arise discontents which embitter the first years of his new
life, if they do not occasion a breach of contract. The friends of a
rational plan of emigration should do their utmost to make known to
as many as it may concern, to what extent labour is wanted in the
colonies,—what is the rate of money wages in each, and what those
money wages will procure. The official information on these points
transmitted from Van Diemen’s Land in 1827, was, that common
labourers earn 3s. per day; common mechanics 7s.; better
mechanics, from 8s. to 12s.; best ditto, from 12s. to 15s.; and
persons of peculiar qualifications, fitted to superintend farms or other
undertakings, 1l. a-day. Since that time, wages are understood to
have risen. The price of wheat is 7s. a bushel; meat, 2d. or 3d. per
lb.; sugar, from 3d. to 6d.; and tea, from 1s. 6d. to 4s. per lb.—No
wonder that, amidst all their gratitude at being well provided for,
many such workmen as Frank are vexed and mortified to find how
much more they might have made of their labour.
Far other feelings, however, than those of discontent were
awakened in Frank by the aspect of his new abode. It was almost in
a state of nature, his employer, Mr. Stapleton, having preceded him
to take possession only a few days before: but it was far from being
a desolate spot in the midst of a waste, as settlers’ farms are wont to
be in colonies where the unwise object is to disperse the inhabitants,
instead of bringing them near to enjoy the advantages of a division of
labour and reciprocity of consumption. The Dutch government at the
Cape of Good Hope formerly forbade settlers to approach within
three miles of each other; and thus effectually prevented the full
improvement of the land, the construction of roads, and the opening
of a market for exchanges. Hence the Dutch settlers at the Cape are
to this day deprived of many advantages of civilized life. They have
too much of whatever they grow, and too little of what they would fain
buy; and are debarred the comforts of society and mutual help.
These evils are likely to be avoided by the method of disposing of
land now adopted by our government in Australia; the land being
sold on terms which make it the interest of the settler to improve his
tract, and to take advantage of a neighbourhood which may relieve
him of some of his produce. Mr. Stapleton, having been obliged to
choose his land carefully, and to pay 9s. an acre for it, (instead of
6d., or nothing at all, like some of the earlier inhabitants,) was not
tempted to wander away into the wilderness, and sit down where he
might happen to like the prospect, or to be smitten with some new
discovery of fish-ponds, woods, and meadows. He made his choice
instead among the lands of a certain district; and selected such, as
to extent and quality, as would on the whole best suit his purposes,
in conjunction with the privileges of a neighbourhood. His land,
though not of the very first quality, was good enough to have fetched
15s. per acre, if it had lain somewhat more to the north or east,
where the country was rapidly becoming better peopled; but it
stretched towards the unoccupied districts at the foot of the western
mountains, and was less valuable than if it had been surrounded by
civilization, instead of only bordering upon it.—It consisted,—not of
jungle and forest ground, where room must be made by the axe
before seed could be sown and sunshine visit it; but of a lightly
timbered and undulating surface of grass land, wanting only a single
burning to render it fit for the plough, or for a new growth of pasture.
The trees were not of the nature of copse and thicket; but growing in
clumps a hundred feet apart, and with clear stems, measuring ninety
feet or more to the lowest branch; thus affording spots for shade and
shelter without interfering with tillage. The boundaries, where not
formed by natural streams, were fixed by marking the trees; and
there was no immediate need of fences where neither man nor beast
was likely to trespass, and where there was at present no live stock
that could be in danger of straying. No one was near who could be
tempted to steal; for none were poor. The wild cattle, which in former
days did great mischief on the grounds of the settler, had long ago
been driven among the mountains, where it was supposed the race
had died out, as none now appeared. The few oxen and horses that
Stapleton brought with him were kept near the dwelling; and the rest
of the stock was not to follow till all was in readiness for its reception.
A rude shed had been hastily constructed for shelter, under a clump
of trees; and a few sawn planks were lying about: by which Frank
saw that the materials of his business were ready for him to begin
upon without delay. Tools and utensils were stowed away in corners,
or heaped under the trees, till their proper places were provided for
them; and a goodly row of casks and packages of provision stood in
the background. Frank had believed that his spirit of enterprise had
died within him under the hardships of his own country; but he now
felt it revive in a moment; and was anything but dismayed at the
prospect of what he had to do in his capacity of carpenter, before the
scene before him could put on the appearance of a snug and well-
managed farmstead. He saw in fancy the day when a little hamlet of
weather-boarded cottages would be sending up their blue smoke
among those trees; when cattle-sheds and sheep-pens would stretch
out behind the dwellings, and the busy forge and creaking timber-
wain would drown the cry of the quail, and scare away the
kangaroos that were now leaping over the plains. He did not forget to
add a very superior workshop and timber-yard to his picture of his
own dwelling; or to imagine his father looking down from among his
flock on the hills, or Ellen within sight, going forth in the bright early
morning with her milk-pail.
As if to answer to his thought, Ellen now appeared. She had stolen
half an hour to run in search of Mr. Stapleton, to ask once more how
soon he thought Frank might possibly arrive. Mr. Stapleton was
almost as eager for the event as herself; but he knew no more, and
was just dismissing her, disappointed, when the waggon was heard
approaching. While she waited a moment, straining her sight to
make out whether it was the right party, before she ran to meet them,
her brother jumped out, and even Castle started up with more
alacrity than he had shown since they landed. Before they could well
greet one another, Stapleton came up to ask where Frank’s tools
were, and to tell him that he was wanted very much indeed. He could
not refuse him permission to go forward one mile, in order to deposit
Castle and his wife at their new abode; but he lent a hand towards
emptying the waggon of his workman’s packages, and gave him
notice that he should be glad to see him back the first possible
moment.
“You will soon find what great people such as we are here,” said
Ellen, laughing. “This is the place to grow proud in. No more
lounging about the fields, Frank; no more leaning over gates
chewing straws, while nobody inquires for one. You will never need
to touch your hat and ask for work here; people will come begging
you to be so very kind as to put up a door for any pay you please.
This is the place to grow proud in.”
Frank observed, with a grave smile, that pride was dangerous to
one in Ellen’s place.
 “Well, then, I will be proud of you, and you shall be proud of me;
and no harm can come of that.”
The first time that the brother and sister could obtain a few
minutes’ conversation without being overheard, Frank inquired,
“Now, Ellen, tell me straight forward. How do you like your
change?”
“Why, I more than half like it; but there are some things I do not
like.—It is a fine thing to be so well off, and to know that I shall be so:
and I do not mind the work, though it is rather hard, to be sure; and
my cows are nothing but a credit to me, and I have seen no animals
to be afraid of when I go out milking, though some of them leap
about very strangely indeed; and my mistress makes much of me, as
I told you; and her little worries are not much to be wondered at
when one thinks of the confusion we live in just now; and I dare say
there will be an end of them when we get our soap and candles
made out of the house, and another hand or two to help in the
brewing and washing. And then to think that father and you are so
well off——”
“But tell me what there is that you do not like.”
Ellen almost shuddered when she whispered that her fellow
servant, who ate at the same table, and slept in the same room, and
was her companion almost all day, was a convict, and had been sent
to this country for robbing an aged mistress who had confided in her,
and deserved gratitude instead of treachery from her. To be
compelled to hold daily and hourly intercourse with such a person
was a very great evil, and one which doubled Frank’s anxiety about
his sister. He was glad to hear that there was a probability of the
woman marrying as soon as she could obtain a remission of her
servitude by steady conduct.
A half smile which he perceived on Ellen’s lips when this part of
the story was being told, made him question her further respecting
the evils of her situation, or the trials which she might not be
disposed to consider exactly as evils. The idea in her mind was that
which he suspected,—that she might quit her service before her
convict companion.—Frank looked graver than ever. Who—what—
where was he,—the person that seemed to have made advances in
Ellen’s good graces already? She was eager to explain that there
was no one in particular yet. It was too early for her to have looked
about and settled her mind yet;—but there was this one, and that
one, and the other one, that carried her pail for her, morning and
evening, however busy he might be; or was ready to teach her how
to clean and card wool; or showed her what a pretty little homestead
he was about to have in the neighbourhood, and intimated how
happy she might be as the mistress of it.
“They hinder my work sadly, and their own too,” continued she,
blushing, “for all I tell them that I have nothing to say to any body yet.
I am so afraid any of them should have been convicts, (though I am
sure Harry Moore never was;) and I dare not ask mistress any thing
about them.”
“Ask her, by all means,” said Frank. “Or I will ask your master, if
you wish it. They only can tell us, and it is a point we must find out.
Meantime, keep to your business as quietly as you can. What makes
you so sure that Moore (is not that his name?) was never a convict?”
Ellen could give no better reason than that she could wager her
life upon it. She thought her brother grown very pertinacious on a
sudden, because this was not perfectly satisfactory to him; but Frank
was not pertinacious—only wary and affectionate.
 Chapter V.

THE CASTLES AT HOME.

It was very well for a man of Castle’s irritable temper to be made a

shepherd, instead of a labourer at home, within sight and hearing of
all the bustle and difficulty occasioned by much pressure of work and
few hands to do it. He could not have borne to be, as his wife said,
driven from pillar to post,—called off from one thing before he had
done it, to do something else to which he was altogether
unaccustomed. It suited him much better to be out upon the downs
after the sheep; though even in that quiet place he had his troubles.
The sheep-walk was too extensive to be under the management of
one person; and Castle’s brother-shepherd was not a very congenial
companion. He was a gentleman convict;—a young man who had
gamed away his little fortune, and then taken to swindling, for which
he had been transported. Being unequal to hard work, and having no
mechanical skill, he was sent out to tend sheep; an employment as
little suitable as might be to his social dispositions and active habits.
The two reluctant companions agreed only in their inclination to
grumble.
“They call this a fine scene,” observed the young man, “but it does
not suit my taste. I had rather see our sheep in the Smithfield pens
than on these downs. Then one misses the London cries, however
much the magpies chatter here. As for the cooing of the doves, it
really depresses the spirits. People talk of the stars being so brilliant
here,—like golden lamps; but I like real lamps better. A row of them
in Pall Mall is worth a hemisphere of stars.”
“I don’t know much about lamplight,” replied Castle, “having been
too poor to burn candles at home, and so going to bed in the twilight;
but this place is so lonesome, I sometimes wonder whether it is in
the world or out of it. All this view is like an old deserted park, to be
sure; but where is the squire’s house, or the church steeple, and the
children coming out of school? There is no public-house far or near;
and no parson or his lady to speak a word to one: only a young man
that comes to read prayers on Sundays in a shed or on the green,
and away again to do the same thing somewhere else. Not such a
thing have I seen since I came as a carriage with ladies in it; and
they say there are no hunts. With all the game there is here, no
scarlet jackets ever come in sight from the woods.”
“That is the worst of it,” responded the other grumbler. “We have
all the dulness of a country life without its solace of amusement. It
was really too tantalizing lately, to see a kangaroo hunt which I could
not join. If they would let me take my turn, I might be of some use to
them as an experienced huntsman. I should like to hunt opossums
till I could get skins enough to make your pretty daughter a cloak
worthy to be worn; and——”
Castle here moved off impatiently, having too much paternal pride
to listen to convict wooing on Ellen’s behalf. The young man followed
him, continuing,
“The snipe-shooting is very choice, I’m told, in the marshes
yonder. I must have leave when winter comes on, to go and try my
luck. But the hunts are the best things,—more spirited perhaps than
you are aware of.”
“Hunts! hunts!” cried Castle. “I see neither deer nor fox. An odd
sort of hunting, if you mean killing any of these leaping things, with
their queer ways. Why, the little ones don’t run beside their dams, as
is natural, but she pops them into her bag, and off she hops, as if
she had only two legs. The first I saw, I thought she had happened
an accident, and had her fore legs cut short; and I thought she got
on wonderfully well considering; and then in a minute appeared a
whole herd of them, with their young in their bags.”
“It is a pretty sight to see them come down from the woods at
sunrise to feed in the plains. Then is the time to hide behind a
thicket, and make sure of one’s game. Which do you prefer, as a
bottom dish, kangaroo or bandicoot? In a pie, properly seasoned, it
is difficult to say which is the best. I have given many a hint down
below that either is much more palatable to me than rations of salt
meat.”
 Castle, who thought no man need desire more in the way of diet
than to eat meat every day, looked with contempt on the grimaces of
his companion over his ample supply of beef, wheaten bread, and
cider.
“If you want to hunt,” said Castle, “I wish you would kill off the vile
beasts that have been making havoc among my lambs. I might have
got at one, but I was downright scared with its ugliness.”
“Was it the hyena or the devil?”
“O, the devil, to judge by its looks. It is as big as a middle-sized
dog, with the head of an otter, crowded with teeth. It moved very
slow, but I could do no better than stare at it.”
“They call it the devil here,” replied the gentleman. “You should dig
little pits, and set your dogs upon it when it has fallen in. It will go on
worrying your lambs, unless you keep on the watch.”
“Another thing that puts me out,” observed Castle, “is that the
beasts are one below another here, as if they were bewitched. In
England, we have a horse of one size, and a dog of another, and a
rat of another; and none of them is like the rest; but here we have a
big kangaroo, and a kangaroo the size of a dog, and another no
bigger than a rat; and these last are not real kangaroos. I declare it
makes my hair stand up to see a rat leaping like a real kangaroo: just
as it would to see a mouse shaking its mane and trotting and
cantering like a horse. I have not been used to such freaks, and this
is a country I can’t understand.”
“I hope to understand it better,” replied the convict. “I was always
fond of roving, and in time I may have explored farther than we can
see from these green hills that we both find so dull. What do you
mean to do when you get free?”
“They may settle that that got me bound,” replied Castle, testily.
Then, struck with a sense of his own ingratitude, he added, “To be
sure, if there is no squire’s house, there is no workhouse either; and
if I see no acquaintance, there is nobody to taunt me with misfortune;
but, on the contrary, they make much of me at home. And there’s
——”
“Your daughter.”
“What; my little Susan! Yes, they make a handy little thing of her
already, and——”
 “I mean the other handy one, Ellen.”
“She will do well enough, Sir, I assure you. She has a fine spirit
and a steady mind of her own, and a proud brother to take care of
her; and that is better than a broken-down father; though it should go
hard with me but I would protect her, Sir, if there was no one to do it
better.”
So saying, Castle walked off, showing by his manner that he was
not sociably disposed.
His wife was much more altered within a short time by her change
of circumstances than he. The first thing that seemed to affect her
favourably was the use that was made of her little daughter in the
household arrangements. When the farmer’s wife found that her new
domestic was indolent and indifferent, she endeavoured to make the
best of a poor bargain by squeezing as much work as she could out
of Susan. The child was willing enough, and proud to find herself of
so much use; but her mother was jealous on her account, and began
for the first time to show symptoms of tenderness for her. She not
only argued in her defence, but helped her when she was more
disposed to proceed with her work than to “go and play;” words
which had little charm for a child whose associations with play were
those of hunger, scolding, mockery, and all the miseries of pauper
life. When the farm servants rose at daybreak to go forth for the day,
Susan was always ready to jump up at the first word, to replenish the
wallets and fill the cans, though her mother turned round in bed, and
muttered that it was too soon to get up. She needed no reminding
about tending the house-lamb, and feeding the poultry, and dusting
as much of the coarse furniture as she could reach. After breakfast,
if any one would lift her upon the dresser, or lay the utensils and the
bowl of water on the floor, she would wash up without breaking
anything: and she was always at hand to carry messages into field or
farm-yard, or to help with dinner and supper, or to carry letters to the
spot where they were to be deposited in readiness for the postman’s
weekly call; and when not able to do anything better, she could scare
away the crows and cockatoos from the fields and garden. Her
mother thought this a hard life for a little girl: but Susan was stout,
rosy, and merry; and the farmer himself found a few minutes now
and then to take her on his knee and teach her the alphabet, in
preparation for the time when a schoolmaster could be brought
within reach. The first thing Mrs. Castle did heartily was washing up,
one day when Susan had nearly scalded her fingers. She took more
and more of the child’s work from her, and still Susan turned to
something else; so that ere long, both were pretty fully employed;
and in proportion as the once reckless and lazy pauper became
interested in the occupations going on around her,—in proportion as
she bestirred herself to get the baking done while the house was
clear of the men, and the washing over in time to have a chat in the
evening,—she grew like the active and tidy housewife she would
never have become in her own land.
A circumstance which hastened this change was the opportunity
she now had of gratifying one taste,—almost the only taste she ever
had, and which seemed to have died out under the hardships of her
condition: a taste for gardening. When a girl, she had had a garden;
and as long as her husband had owned an acre of ground, she took
possession of a corner of it for her pinks and roses, under pretence
of growing vegetables for the family. From that time to this, nobody
had heard her mention fruit or flowers; but Ellen bore in mind her
love for them now when the remembrance might be turned to some
purpose. She mentioned, in her step-mother’s presence, that her
master was trying what he could do in the management of vines, for
the growth of which the climate was peculiarly favourable; and that
whether he got any wine or not, his trouble would be more than
repaid by his profits from his other fruits. The peaches, to be sure,
were not of the best sort, though so plentiful as to lie rotting on the
ground, after bushels had been thrown to the pigs; but the apricots,
and yet more the raspberries, which grew to such a size and in such
quantity as no English person would believe without seeing them,
were likely to prove a good speculation, being sent to a distance in
the form of jam. Sugar being remarkably cheap in this country, there
was little risk in trying a batch of sweetmeats, which were to be sent
to India for sale by a vessel from Launceston. The idea was caught
up as Ellen expected it would be; and as the farmer and his wife did
not take to the scheme of fruit-growing as heartily as was desired,
the emigrant family tried whether they could not get a garden of their
own. The small part of their wages which they were yet at liberty to
use was applied to the purchase of a plot of ground, and Frank found
time to work in it, and Ellen procured wherewith to stock it, and their
step-mother haunted it early and late, before and after work,—and
Castle himself relaxed his brow, and spoke in a tone that was not
querulous, as he looked round upon that which, however small, was
so much more than he had ever expected to possess again,—a
family property.
“Look at father,” whispered Ellen to her brother. “I have not seen
him and her arm-in-arm since I was no bigger than Susan.”
“He is like a prisoner that has been quite shut up coming for the
first time into the gaol court,” said Frank. “The feel of the air makes
him push his hat up from over his eyes. Only set him quite free, and
he will uncover his brows, and lift up his head like a man.”
“And so he ought,” replied Ellen, “since it is for no fault of his own
that he has been bound down to poverty.”
“Ah! poverty is a cold and dreary prison, Ellen. That puts me in
mind,—have you seen, I wonder, any thing that has surprised you
very much lately,—any thing that you would like to tell me if you were
sure of not being overheard, or of not being thought fanciful?”
“The word ‘prison’ put me in mind of something that I have been
wanting to talk to you about almost this month past.—I don’t know
how to believe my own eyes about it, but I am sure I have seen Jerry
in our farm-yard at night, and lurking about among the sheds before
most of our folks are stirring in the morning.”
“Aye; but was he all alone, Ellen?”
“Bob was never with him, that I could see; but he seemed to me to
be waiting for somebody off our premises, and I thought it must
naturally be me. So twice I ran out to catch him; but once I was
crossed by two of our people that I don’t choose to come in the way
of; and the other time he was whispering with the same two, so that I
dared not go near. How could he get liberty? and what could he be
about?”
“Something very deep, I am afraid,” replied her brother. “As to the
liberty,—it is no difficult matter for convicts who behave pretty
regularly to get hours of liberty at the beginning and end of the day;
and the lads being employed on roadmaking so near, accounts for
our getting a glimpse of them sometimes. But what I am uneasy
about is Jerry’s having so much to say to the convicts at your place
and mine;—for I have seen him at Stapleton’s oftener than you have
among your people. I am afraid of some plot——”
“O, mercy!” cried Ellen; “what sort of a plot?”
“That is more than I can say. Sometimes they plot, I hear, for
nothing worse than to escape; but some have had to do with the
natives, (who are little better than wild beasts), and have brought
them down upon the farms, setting them to steal and even to
murder; for which they pay the poor savage creatures by helping
themselves with their wives.”
Ellen trembled while she asked whether any of the natives could
be in the neighbourhood.—Her brother hoped not, as the
government had declared that they were driven back among the
mountains, where they must soon die out as their wild cattle had
done: but as long as any convicts were disposed to bush-ranging,—
and some did actually escape every year,—he could not, for his part,
feel quite secure. He thought he should speak to Stapleton about it.
Meanwhile, he desired Ellen to drop not a syllable that should alarm
her father, or anybody else.
“I hope, sister,” he continued with some hesitation, “I hope Harry
Moore has no acquaintance, more or less, with Jerry, or any other
such people.”
Ellen’s eyes flashed as they used to do when she was a
passionate little girl at school.
“Harry!” she cried. “Harry Moore have any sneaking doings! Harry
Moore keep bad company! You don’t know Harry a bit better than the
very first day,—the day when you thought he might be a convict
himself.”
“No need to be angry, Ellen. He might just know him enough, you
see, to say ‘How d’ye do?’ when they meet, and to judge how often
Jerry might fairly be here.”
“After all,” said Ellen, sighing, “it is my father’s own son that I flew
off about his being acquainted with; so there is no need for me to be
so proud. No; Harry does not know either of the lads, even by sight;
but I shall tell him what you have been saying, though nobody else,
Frank.”
 “Certainly. Conceal nothing that weighs upon your mind from
Harry, any more than if he was your husband already. I look to him to
help me to keep an eye upon Bob, who may be made something of,
they say, little hope as there is for Jerry. Bob works within bounds at
spare hours, instead of roving into the bush, or prowling about the
settlers’ farms, where he has no business. Bob must be saving
money fast, unless he has unseen ways of spending it. He works
hard, and is well paid for his extra labour. He may have the
advantage of me, after all; and settle on a place of his own before
me.”
“Because he got a free passage as a punishment. That is really
very hard, Frank.”
“Harry Moore will be the first at liberty, however, Ellen; and that I
am glad of on your account. I am soon to begin building you a
house, at over hours; and you may depend on my doing my best to
have it all complete by the time the six months are gone.”
“Six months!” cried Ellen.
“Why, I do not mean that you need wait till then. You may fairly
marry as soon as you like;—and many in our own country would be
glad to have that said to them. I only mentioned six months as the
time when Harry would be all his own master. Then I shall hope to
see you milking a cow of your own.—Meantime, till I have found out
more about Jerry, be cautious how you get out of reach of those that
will take care of you.”
Ellen sighed, and smiled, and wondered which was the strangest
world,—the one she had left behind, or the new one which seemed,
after several months, nearly as foreign as when she had entered it.
She had no doubt which was the pleasantest. How could she, when
a vague fear and thorough dislike of some of the people in the
neighbourhood were the only set-off against the prosperity of all
whom she loved, and her own bright prospects with such a husband
as Harry Moore had promised to be?