Ebook Handbook of Automata Theory Volume Ii Automata in Mathematics and Selected Applications First Edition Jean Eric Pin Online PDF All Chapter

Handbook of Automata Theory Volume
II Automata in Mathematics and

Selected Applications First Edition
Jean-Éric Pin
Visit to download the full and correct content document:
https://ebookmeta.com/product/handbook-of-automata-theory-volume-ii-automata-in-
mathematics-and-selected-applications-first-edition-jean-eric-pin/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
Handbook of Automata Theory Volume I Theoretical

Foundations First Edition Jean Éric Pin
https://ebookmeta.com/product/handbook-of-automata-theory-volume-
i-theoretical-foundations-first-edition-jean-eric-pin/
Mathematical Theory of Switching Circuits and Automata

Sze-Tsen Hu
https://ebookmeta.com/product/mathematical-theory-of-switching-
circuits-and-automata-sze-tsen-hu/
Automata Theory and Formal Languages Fundamental

Notions Theorems and Techniques First Edition Alberto
Pettorossi
https://ebookmeta.com/product/automata-theory-and-formal-
languages-fundamental-notions-theorems-and-techniques-first-
edition-alberto-pettorossi/
FORMAL LANGUAGE AND AUTOMATA THEORY 2nd Edition Singh

Ajit
https://ebookmeta.com/product/formal-language-and-automata-
theory-2nd-edition-singh-ajit/
Quantum Dot Cellular Automata Circuits for
Nanocomputing Applications 1st Edition Trailokya Nath
Sasamal
https://ebookmeta.com/product/quantum-dot-cellular-automata-
circuits-for-nanocomputing-applications-1st-edition-trailokya-
nath-sasamal/
Cellular Automata and Groups 2nd Edition Tullio

Ceccherini-Silberstein
https://ebookmeta.com/product/cellular-automata-and-groups-2nd-
edition-tullio-ceccherini-silberstein/
Introduction to Computation Haskell Logic and Automata

Donald Sannella
https://ebookmeta.com/product/introduction-to-computation-
haskell-logic-and-automata-donald-sannella/
Linear Algebra and Optimization with Applications to

Machine Learning Volume II Fundamentals of Optimization
Theory with Applications to Machine Learning 1st
Edition Jean Gallier
https://ebookmeta.com/product/linear-algebra-and-optimization-
with-applications-to-machine-learning-volume-ii-fundamentals-of-
optimization-theory-with-applications-to-machine-learning-1st-
edition-jean-gallier/
An Introduction to Formal Languages and Automata 6th ed

Peter Linz
https://ebookmeta.com/product/an-introduction-to-formal-
languages-and-automata-6th-ed-peter-linz/
Handbook of
Automata Theory
Volume II
Automata in Mathematics and Selected Applications
Edited by Jean-Éric Pin

Handbook of
Automata Theory
Volume II
Automata in Mathematics and Selected Applications
Edited by Jean-Éric Pin

Editor:
Jean-Éric Pin
Institut de Recherche en Informatique Fondamentale (IRIF)
Université de Paris and CNRS
Bâtiment Sophie Germain, Case courier 7014
8 Place Aurélie Nemours
75205 Paris Cedex 13
E-mail: Jean-Eric.Pin@irif.fr
Volume II:
2020 Mathematics Subject Classification: 68Q45; 03B25, 03B70, 03C13, 03D05, 11A67, 11B85,
11J81, 11U05, 20E05, 20E08, 20F10, 20F65, 20F67, 37B10, 37B20, 68Q10, 68Q12, 68Q15,
68Q17, 68Q19, 68Q45, 68Q60, 68Q70, 68Q85, 68R15, 68T50, 68U10
Keywords: free groups, Stallings automata, automatic groups, self-similar groups, automatic
sequences, numeration systems, Cobham’s theorem, symbolic dynamics, synchronous automata,
finite model theory, fractal image generation, communicating automata, model-checking, Church’s
problem, distributed synthesis, timed automata, recursion schemes, Markov decision processes,
infinite-state systems, natural language processing, temporal logic, branching time logic, quantum
finite automata
ISBN Vol. I 978-3-98547-002-0

ISBN Vol. II 978-3-98547-003-7
ISBN Set 978-3-98547-006-8 (set of both volumes)
Bibliographic information published by the Deutsche Nationalbibliothek

The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie;
detailed bibliographic data are available on the Internet at http://dnb.dnb.de.
Published by EMS Press, an imprint of the

European Mathematical Society – EMS – Publishing House GmbH
Institut für Mathematik
Technische Universität Berlin
Straße des 17. Juni 136
10623 Berlin, Germany
https://ems.press
© 2021 EMS Press
Cover drawing of Jacques de Vaucanson’s digesting duck (canard digérant) published in

Scientific American Vol. 80 (3), 1899. Fractal tree on the first page by Nicolas Janey.
Typeset using the authors’ LaTeX sources: Marco Zunino, Savona, Italy
Printing and binding: Beltz Bad Langensalza GmbH, Bad Langensalza, Germany
♾ Printed on acid free paper
987654321
Dedicated to the memory of
Professor Zoltán Ésik
(1951–2016 )
Preface
The Handbook of Automata Theory has its origins in the research programme Au-
toMathA (Automata: from Mathematics to Applications, 2005–2010), a multidisci-
plinary programme of the European Science Foundation at the crossroads of mathe-
matics, theoretical computer science, and applications. It is designed to provide a broad
audience of researchers and students in mathematics and computer science with a com-
prehensive overview of research in automata theory.
Automata theory is one of the longest established areas in computer science. It
was born over sixty years ago, with the seminal work of Kleene, who first formalised
the early attempts of McCulloch and Pitts, and was originally motivated by the study
of neural networks. For many years, its main applications have been computer design,
compilation of programming languages, and pattern matching. But over the last twenty
years, applications of automata theory have considerably diversified, and now include
verification methods to cope with such emerging technical needs as network security,
mobile intelligent devices, and high performance computing.
At the same time, the mathematical foundations of automata theory rely on more
and more advanced parts of mathematics. While only elementary graph theory and
combinatorics were required in the early sixties, new tools from non-commutative
algebra (semigroups, semirings and formal power series), logic, probability theory, and
symbolic dynamics have been successively introduced, and the latest developments
borrow ideas from topology and geometry.
It was time to gather these mathematical advances and their numerous applications
in a reference book. The Handbook of Automata Theory is intended to serve this
purpose. It comprises thirty-nine chapters, presented in two volumes:
Volume I: Theoretical foundations

Volume II: Automata in mathematics and selected applications
Together, the two volumes cover most of the topics related to automata. Volume I
presents, in the first part, the basic models of the theory: finite automata working
on finite words, infinite words, finite trees and infinite trees, transducers, weighted
automata and max-plus automata, and two-dimensional models. In the second part,
complexity and algorithmic issues are discussed extensively, including connections
with circuit complexity and finite model theory. In the third part, the algebraic and
topological aspects of automata theory are treated.
Volume II first offers a wide range of connections between automata and mathe-
matics, including group theory, number theory, symbolic dynamics, finite model the-
ory, and fractal-type images. Secondly, selected applications are covered, including
viii Preface
message-passing systems, symbolic methods, synthesis, timed automaton model, veri-

fication of higher-order programs, analysis of probabilistic processes, natural language
processing, formal verification of programs, and quantum computing.
Much of this material had never been published in a book before, making the
Handbook a unique reference in automata theory.
Due to the length of the Handbook, the chapters are divided into two volumes. For
the convenience of the reader, the front matter and the index appear in both volumes
(paginated with roman numerals).
As this project started over ten years ago, some recent developments may not have
been addressed. Nevertheless, the reader will be able to find updates and possible
corrections on
https://ems.press/isbn/978-3-98547-006-8
Acknowledgements. I would like to thank the European Science Foundation, and in

particular the Standing Committee for Physical & Engineering Sciences (PESC), for
funding the research programme AutoMathA within their Research Networking Pro-
gramme (2005–2010). The Handbook would not have been possible without their moral
and financial support. The programme AutoMathA brought together a research commu-
nity of wide scope; its joint work and efforts have been vital for composing the present
handbook. The AutoMathA project was initially launched by Jorge Almeida (Lis-
boa), Stefano Crespi Reghizzi (Milano) and myself. Let me also thank the other mem-
bers of the AutoMathA Steering Committee: Jacques Duparc, Jozef Gruska, Juhani
Karhumäki, Mikołaj Bojańczyk, Søren Eilers, Stuart W. Margolis, Tatiana Jajcayova,
Véronique Bruyère, Werner Kuich, Wolfgang Thomas, and Zoltán Ésik. Sadly and un-
expectedly, Zoltán Ésik passed away during the final stages of the Handbook project.
We dedicate the Handbook to the memory of this great scientist and friend. The constant
support of the AutoMathA Steering Committee during the preparation of this handbook
was an invaluable help. Narad Rampersad’s assistance during the early stage of the
Handbook was also particularly appreciated.
All the authors are particularly indebted to Jeffrey Shallit. As one of the few
native English-speaking authors of the book, Jeffrey has accepted the daunting task of
reviewing all the chapters in their entirety. He not only detected a considerable number
of English mistakes, but he also greatly improved the style and mathematical content
of the chapters. I would therefore like to express my deepest thanks to him.
The advisory board consisting of Søren Eilers (Copenhagen) and Wolfgang Thomas
(Aachen) was instrumental in defining the early version of the Handbook project. I am
particularly indebted to Wolfgang Thomas for his advice and constant encouragement
and help during the long gestation period of this handbook.
Of course, this handbook would not have been possible without the authors of the
thirty-nine chapters. I would like to thank them all for their high quality scientific
contribution and their perseverance during the chapter review process.
Preface ix
For their patience and extreme care in the production of the Handbook, I would
like to thank the typesetter Marco Zunino and all the people of EMS Press I have been
working with: Apostolos Damialis, Sylvia Fellmann, Thomas Hintermann, Manfred
Karbe, Vera Spillner, and Simon Winter. Special thanks to Nicolas Janey who kindly
designed the fractal tree on the first page.
Jean-Éric Pin
Managing editor
Paris, 2021
Contents
VOLUME ONE
Preface vii
List of contributors xvii
Part I
Foundations
Chapter 1
Finite automata 3
Jean-Éric Pin
Chapter 2
Automata and rational expressions 39
Jacques Sakarovitch
Chapter 3
Finite transducers and rational transductions 79
Tero Harju and Juhani Karhumäki
Chapter 4
Weighted automata 113
Manfred Droste and Dietrich Kuske
Chapter 5
Max-plus automata 151
Sylvain Lombardy and Jean Mairesse
Chapter 6
!-Automata 189
Thomas Wilke (revised by Sven Schewe)
Chapter 7
Automata on finite trees 235
Christof Löding and Wolfgang Thomas
xii Contents
Chapter 8
Automata on infinite trees 265
Christof Löding
Chapter 9
Two-dimensional models 303
Stefano Crespi Reghizzi, Dora Giammarresi, and Violetta Lonati
Part II
Complexity issues
Chapter 10
Minimisation of automata 337
Jean Berstel, Luc Boasson, Olivier Carton, and Isabelle Fagnot
Chapter 11
Learning algorithms 375
Henrik Björklund, Johanna Björklund, and Wim Martens
Chapter 12
Descriptional complexity of regular languages 411
Hermann Gruber, Markus Holzer, and Martin Kutrib
Chapter 13
Enumerating regular expressions and their languages 459
Hermann Gruber, Jonathan Lee, and Jeffrey Shallit
Chapter 14
Circuit complexity of regular languages 493
Michal Koucký
Chapter 15
Černý’s conjecture and the road colouring problem 525
Jarkko Kari and Mikhail Volkov
Contents xiii
Part III
Algebraic and topological theory of automata
Chapter 16
Varieties 569
Howard Straubing and Pascal Weil
Chapter 17
Profinite topologies 615
Jorge Almeida and Alfredo Costa
Chapter 18
The factorisation forest theorem 653
Thomas Colcombet
Chapter 19
Wadge–Wagner hierarchies 695
Jacques Duparc
Chapter 20
Equational theories for automata 729
Zoltán Ésik
Chapter 21
Language equations 765
Michal Kunc and Alexander Okhotin
Chapter 22
Algebra for trees 801
Mikołaj Bojańczyk
Index xxiii
xiv Contents
VOLUME TWO
Preface vii
List of contributors xvii
Part IV
Automata in mathematics
Chapter 23
Rational subsets of groups 841
Laurent Bartholdi and Pedro V. Silva
Chapter 24
Groups defined by automata 871
Chapter 25
Automata in number theory 913
Boris Adamczewski and Jason Bell
Chapter 26
On Cobham’s theorem 947
Fabien Durand and Michel Rigo
Chapter 27
Symbolic dynamics 987
Marie-Pierre Béal, Jean Berstel, Søren Eilers, and Dominique Perrin
Chapter 28
Automatic structures 1031
Sasha Rubin
Chapter 29
Automata and finite model theory 1071
Wouter Gelade and Thomas Schwentick
Chapter 30
Finite automata, image manipulation, and automatic real functions 1105
Juhani Karhumäki and Jarkko Kari
Contents xv
Part V
Selected applications
Chapter 31
Communicating automata 1147
Dietrich Kuske and Anca Muscholl
Chapter 32
Symbolic methods and automata 1189
Bernard Boigelot
Chapter 33
Synthesis with finite automata 1217
Igor Walukiewicz
Chapter 34
Timed automata 1261
Patricia Bouyer
Chapter 35
Higher-order recursion schemes and their automata models 1295
Arnaud Carayol and Olivier Serre
Chapter 36
Analysis of probabilistic processes and automata theory 1343
Kousha Etessami
Chapter 37
Natural language parsing 1383
Mark-Jan Nederhof and Giorgio Satta
Chapter 38
Verification 1415
Javier Esparza, Orna Kupferman, and Moshe Y. Vardi
Chapter 39
Automata and quantum computing 1457
Andris Ambainis and Abuzer Yakaryılmaz
Index xxiii
List of contributors
Boris Adamczewski (Chapter 25) Jason Bell (Chapter 25)

CNRS, Université de Lyon Department of Pure Mathematics
Institut Camille Jordan University of Waterloo
43 boulevard du 11 novembre 1918 Waterloo, ON N2L 3G1
69622 Villeurbanne Cedex Canada
France jpbell@uwaterloo.ca
Boris.Adamczewski@math.cnrs.fr
Jean Berstel (Chapters 10, 27)
Jorge Almeida (Chapter 17) Laboratoire d’Informatique Gaspard-Monge
CMUP, Departamento de Matemática Université Paris-Est Marne-la-Vallée
Faculdade de Ciências 5, boulevard Descartes
Universidade do Porto Champs-sur-Marne
Rua do Campo Alegre 687 77454 Marne-la-Vallée Cedex 2
4169-007 Porto France
Portugal Jean.Berstel@gmail.com
jalmeida@fc.up.pt
Henrik Björklund (Chapter 11)
Andris Ambainis (Chapter 39) Department of Computing Science
University of Latvia Umeå University
Faculty of Computing 90187 Umeå
Raina bulv. 19 Sweden
Rīga 1586 henrikb@cs.umu.se
Latvia
ambainis@lu.lv Johanna Björklund (Chapter 11)
Department of Computing Science
Laurent Bartholdi (Chapters 23, 24) Umeå University
Mathematisches Institut 90187 Umeå
Georg-August Universität zu Göttingen Sweden
Bunsenstraße 3–5 johanna@cs.umu.se
37073 Göttingen
Germany Luc Boasson (Chapter 10)
laurent.bartholdi@gmail.com IRIF, Université de Paris et CNRS
Bâtiment Sophie Germain
Marie-Pierre Béal (Chapter 27) Case courrier 7014
Laboratoire d’Informatique Gaspard-Monge 8 Place Aurélie Nemours
Université Paris-Est Marne-la-Vallée 75205 Paris Cedex 13
5, boulevard Descartes France
Champs-sur-Marne Luc.Boasson@irif.fr
77454 Marne-la-Vallée Cedex 2
France
beal@univ-mlv.fr
xviii List of contributors
Bernard Boigelot (Chapter 32) Thomas Colcombet (Chapter 18)

Institut Montefiore, B28 IRIF, Université de Paris et CNRS
Université de Liège Bâtiment Sophie Germain
10, Allée de la découverte Case courrier 7014
4000 Liège 8 Place Aurélie Nemours
Belgium 75205 Paris Cedex 13
Bernard.Boigelot@uliege.be France
thomas.colcombet@irif.fr
Mikołaj Bojańczyk (Chapter 22)
MIMUW Alfredo Costa (Chapter 17)
Banacha 2 CMUC, Department of Mathematics
02-097 Warszawa University of Coimbra
Poland Apartado 3008
EC Santa Cruz
bojan@mimuw.edu.pl
3001-501 Coimbra
Portugal
Patricia Bouyer (Chapter 34)
Université Paris-Saclay amgc@mat.uc.pt
CNRS, ENS Paris-Saclay
Laboratoire Méthodes Formelles Manfred Droste (Chapter 4)
91190 Gif-sur-Yvette Institut für Informatik
France Universität Leipzig
Augustusplatz 10-11
patricia.bouyer@ens-paris-saclay.fr 04109 Leipzig
Germany
Arnaud Carayol (Chapter 35)
LIGM, Université Gustave Eiffel droste@informatik.uni-leipzig.de
5, boulevard Descartes
Jacques Duparc (Chapter 19)
Champs-sur-Marne
Department of Operations
Faculty of Business and Economics
France
University of Lausanne
Arnaud.Carayol@univ-eiffel.fr 1015 Lausanne
Switzerland
Olivier Carton (Chapter 10)
jacques.duparc@unil.ch
IRIF, Université de Paris et CNRS
Bâtiment Sophie Germain Fabien Durand (Chapter 26)
Case courrier 7014 Université de Picardie Jules Verne
8 Place Aurélie Nemours CNRS UMR 6140
75205 Paris Cedex 13 33 rue Saint Leu
France 80039 Amiens Cedex 1
Olivier.Carton@irif.fr France
fabien.durand@u-picardie.fr
Stefano Crespi Reghizzi (Chapter 9)
Dipartimento di Elettronica Informazione Søren Eilers (Chapter 27)
e Bioingegneria Institut for Matematiske Fag
Politecnico di Milano Københavns Universitet
Piazza Leonardo da Vinci 32 Universitetsparken 5
20133 Milano 2100 København Ø
Italy Denmark
stefano.crespireghizzi@polimi.it eilers@math.ku.dk
List of contributors xix
Zoltán Ésik (Chapter 20) Tero Harju (Chapter 3)

Department of Mathematics and Statistics
Javier Esparza (Chapter 38) University of Turku
Institut für Informatik FI-20014 Turku
Technische Universität München Finland
Boltzmannstraße 3 harju@utu.fi
85748 Garching bei München
Germany Markus Holzer (Chapter 12)
esparza@in.tum.de Institut für Informatik
Universität Giessen
Kousha Etessami (Chapter 36) Arndtstraße 2
School of Informatics 35392 Giessen
University of Edinburgh Germany
10 Crichton Street holzer@informatik.uni-giessen.de
Edinburgh EH8 9AB
United Kingdom Juhani Karhumäki (Chapters 3, 30)
kousha@inf.ed.ac.uk Department of Mathematics and Statistics
University of Turku
Isabelle Fagnot (Chapter 10) FI-20014 Turku
Laboratoire d’Informatique Gaspard-Monge Finland
Université Paris-Est Marne-la-Vallée karhumak@utu.fi
Champs-sur-Marne Jarkko Kari (Chapters 15, 30)
77454 Marne-la-Vallée Cedex 2 Department of Mathematics and Statistics
France University of Turku
FI-20014 Turku
Isabelle.Fagnot@univ-eiffel.fr
Finland
Wouter Gelade (Chapter 29) jkari@utu.fi
Centre of Research in the Economics
of Development (CRED) Michal Koucký (Chapter 14)
University of Namur Computer Science Institute of Charles University
Rempart de la Vierge, 8 Malostranské nám 25
5000 Namur 118 00 Praha 1
Belgium Czech Republic
wouter.gelade@unamur.be koucky@iuuk.mff.cuni.cz
Dora Giammarresi (Chapter 9) Michal Kunc (Chapter 21)

Dipartimento di Matematica Department of Mathematics and Statistics
Università di Roma “Tor Vergata” Masaryk University
via della Ricerca Scientifica 1 Kotlářská 2
00133 Roma 611 37 Brno
Italy Czech Republic
giammarr@mat.uniroma2.it kunc@math.muni.cz
Orna Kupferman (Chapter 38)

Hermann Gruber (Chapters 12, 13)
School of Computer Science and Engineering
Knowledgepark GmbH
Hebrew University
Leonrodstr. 68
Jerusalem 91904
80636 München
Israel
Germany
orna@cs.huji.ac.il
hermann.gruber@kpark.de
xx List of contributors
Dietrich Kuske (Chapters 4, 31) Jean Mairesse (Chapter 5)

Institut für Theoretische Informatik LIP6 – Laboratoire d’Informatique de Paris 6
Fakultät Informatik und Automatisierung UMR 7606, CNRS
Technische Universtität Ilmenau Université Pierre et Marie Curie
Postfach 100565 Boîte courrier 169
98693 Ilmenau Tour 26, Couloir 26-00, 2è étage
Germany 4 place Jussieu
dietrich.kuske@tu-ilmenau.de 75252 Paris Cedex 05
France
Martin Kutrib (Chapter 12) jean.mairesse@lip6.fr
Institut für Informatik
Universität Giessen Wim Martens (Chapter 11)
Arndtstraße 2 Institut für Informatik
35392 Giessen Universität Bayreuth
Germany 95440 Bayreuth
kutrib@informatik.uni-giessen.de Germany
wim.martens@uni-bayreuth.de
Jonathan Lee (Chapter 13)
Department of Mathematics Anca Muscholl (Chapter 31)
Stanford University LaBRI, Université de Bordeaux et CNRS
Building 380, Sloan Hall 351 cours de la Libération
Stanford, CA 94305 33405 Talence Cedex
USA France
anca@labri.fr
Christof Löding (Chapters 7, 8)
Lehrstuhl Informatik 7
RWTH Aachen Mark-Jan Nederhof (Chapter 37)
52056 Aachen School of Computer Science
Germany University of St Andrews
North Haugh
loeding@cs.rwth-aachen.de St Andrews KY16 9SX
United Kingdom
Sylvain Lombardy (Chapter 5)
markjan.nederhof@gmail.com
LaBRI, Université de Bordeaux et CNRS
Institut Polytechnique de Bordeaux
351 cours de la Libération Alexander Okhotin (Chapter 21)
33405 Talence Cedex Department of Mathematics and Computer Science
France St. Petersburg State University
14th Line V.O., 29
sylvain.lombardy@labri.fr 199178 Saint Petersburg
Russian
Violetta Lonati (Chapter 9)
Dipartimento di Informatica alexander.okhotin@spbu.ru
Università degli Studi di Milano
via Celoria 18 Dominique Perrin (Chapter 27)
20100 Milano Laboratoire d’Informatique Gaspard-Monge
Italy Université Paris-Est Marne-la-Vallée
lonati@di.unimi.it
Champs-sur-Marne
France
perrin@univ-mlv.fr
List of contributors xxi
Jean-Éric Pin (Chapter 1) Thomas Schwentick (Chapter 29)

IRIF, Université de Paris et CNRS Fakultät für Informatik
Bâtiment Sophie Germain Technische Universität Dortmund
Case courrier 7014 Otto-Hahn-Straße 12
8 Place Aurélie Nemours 44227 Dortmund
75205 Paris Cedex 13 Germany
France thomas.schwentick@udo.edu
Jean-Eric.Pin@irif.fr
Olivier Serre (Chapter 25)
Michel Rigo (Chapter 26) IRIF, Université de Paris et CNRS
Université de Liège Bâtiment Sophie Germain
Institut de Mathématiques Case courrier 7014
12 Allée de la découverte (B37) 8 Place Aurélie Nemours
4000 Liège 75205 Paris Cedex 13
Belgium France
M.Rigo@ulg.ac.be Olivier.Serre@irif.fr
Sasha Rubin (Chapter 28) Jeffrey Shallit (Chapter 13)

School of Computer Science School of Computer Science
The University of Sydney University of Waterloo
Building J12/1, Cleveland St. Waterloo, ON N2L 3G1
Camperdown NSW 2006 Canada
Australia shallit@uwaterloo.ca
sasha.rubin@sydney.edu.au
Pedro V. Silva (Chapter 23, 24)
Jacques Sakarovitch (Chapter 2) Centro de Matemática
IRIF, Université de Paris et CNRS Faculdade de Ciências
Bâtiment Sophie Germain Universidade do Porto
Case courrier 7014 R. Campo Alegre 687
8 Place Aurélie Nemours 4169-007 Porto
75205 Paris Cedex 13 Portugal
France pvsilva@fc.up.pt
sakarovitch@telecom-paristech.fr
Howard Straubing (Chapter 16)
Giorgio Satta (Chapter 37) Computer Science Department
Department of Information Engineering Boston College
University of Padua Chestnut Hill, MA 02467
via Gradenigo 6/A USA
35131 Padova straubin@cs.bc.edu
Italy
satta@dei.unipd.it Wolfgang Thomas (Chapter 7)
Lehrstuhl Informatik 7
Sven Schewe (Chapter 6) RWTH Aachen
Department of Computer Science 52056 Aachen
University of Liverpool Germany
Ashton Building thomas@cs.rwth-aachen.de
Ashton Street
Liverpool L69 3BX
United Kingdom
sven.schewe@liverpool.ac.uk
xxii List of contributors
Moshe Y. Vardi (Chapter 38) Thomas Wilke (Chapter 6)

Department of Computer Science Department of Computer Science
Mail Stop 132 Christian-Albrechts-Universität zu Kiel
Rice University 24098 Kiel
6100 S. Main Street Germany
Houston, TX 77005-1892 thomas.wilke@email.uni-kiel.de
USA
vardi@cs.rice.edu Abuzer Yakaryılmaz (Chapter 39)
Faculty of Computing
Mikhail Volkov (Chapter 15) University of Latvia
Institute of Natural Sciences and Mathematics Raina bulv. 19
620000 Ural Federal University Rīga 1586
Ekaterinburg Latvia
Russia
abuzer@lu.lv
Mikhail.Volkov@usu.ru
Igor Walukiewicz (Chapter 33)

351 cours de la Libération
33405 Talence Cedex
France
igw@labri.fr
Pascal Weil (Chapter 16)

351 cours de la Libération
33405 Talence Cedex
France
ReLaX, CNRS IRL 2000
and Chennai Mathematical Institute
SIPCOT IT Park
603103 Siruseri, Chennai
India
pascal.weil@cnrs.fr
Part IV
Automata in mathematics
Chapter 23
Rational subsets of groups
Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 841
2. Finitely generated groups . . . . . . . . . . . . . . . . . . . . . . . . . 841
3. Inverse automata and Stallings’ construction . . . . . . . . . . . . . . . 843
4. Rational and recognisable subsets . . . . . . . . . . . . . . . . . . . . . 857
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 866
1. Introduction
Over the years, finite automata have been used effectively in the theory of infinite groups
to represent rational subsets. This includes the important particular case of finitely
generated subgroups (and the beautiful theory of Stallings automata for the free group
case), but goes far beyond that: certain inductive procedures need a more general setting
than mere subgroups, and rational subsets constitute the natural generalisation. The
connections between automata theory and group theory are rich and deep, and many
are portrayed in Sims’ book [60].
This chapter is divided into three parts: in § 2 we introduce basic concepts, ter-
minology and notation for finitely generated groups, devoting special attention to free
groups. These will also be used in Chapter 24.
§ 3 describes the use of finite inverse automata to study finitely generated subgroups
of free groups. The automaton recognises elements of a subgroup, represented as words
in the ambient free group.
§ 4 considers, more generally, rational subsets of groups, when good closure and
decidability properties of these subsets are satisfied.
The authors are grateful to Stuart Margolis, Benjamin Steinberg, and Pascal Weil
for their remarks on a preliminary version of this text.
2. Finitely generated groups

Let G be a group. Given A G , let hAi D .A [ A 1 / denote the subgroup of G
generated by A. We say that H 6 G is finitely generated and write H 6f:g: G if
H D hAi for some finite subset A of H .
842 Laurent Bartholdi and Pedro V. Silva
Given H 6 G , we let ŒG W H denote the index of H in G , that is, the number of

right cosets Hg for all g 2 G ; or, equivalently, the number of left cosets. If ŒG W H is
finite, we write H 6f:i: G . It is well known that every finite-index subgroup of a finitely
generated group is finitely generated (see Corollary 2.7.1 in [30] or Example 3.4).
We let 1 denote the identity of G . An element g 2 G has finite order if hgi is finite.
Elements g; h 2 G are conjugate if h D x 1 gx for some x 2 G . We use the notation
g h D h 1 gh and Œg; h D g 1 g h to denote, respectively, conjugates and commutators.
Given an alphabet A, we let A 1 denote a set of formal inverses of A, and write
Az D A [ A 1 . We say that Az is an involutive alphabet. We extend 1 W A ! A 1 ;
a 7! a 1 to an involution on Az through
.a 1 / 1 D a; .uv/ 1 D v 1 u 1 .a 2 A; u; v 2 Az /:
If G D hAi, we have a canonical epimorphism W Az G , mapping a˙1 2 Az to
˙1
a 2 G . We present next some classical decidability problems:
Definition 2.1. Let G D hAi be a finitely generated group.
Word problem: is there an algorithm that, on input a word u 2 Az , determines
whether or not .u/ D 1?
Conjugacy problem: is there an algorithm that, on input words u; v 2 Az , deter-
mines whether or not .u/ and .v/ are conjugate in G ?
Membership problem for K 2G : is there for every X 2 K an algorithm that, on
input a word u 2 Az , determines whether or not .u/ 2 X ?
Generalised word problem: is the membership problem for the class of finitely
generated subgroups of G solvable?
Order problem: is there an algorithm that, on input a word u 2 Az , determines
whether .u/ has finite or infinite order?
Isomorphism problem for a class G of groups: is there an algorithm that, on input
a description of groups G; H 2 G, decides whether or not G Š H ?
Typically, G may be a subclass of finitely presented groups (given by their presen-
tation), or automata groups (see Chapter 24) given by automata.
We can also require complexity bounds on the algorithms; more precisely, we may
ask with which complexity bound an answer to the problem may be obtained, and
also with which complexity bound a witness (a normal form for the word problem,
an element conjugating .u/ to .v/ in case they are conjugate, an expression of u in
the generators of X in the membership problem) may be constructed.
2.1. Free groups. We recall that an equivalence relation on a semigroup S is a
congruence if a b implies ac bc and ca cb for all a; b; c 2 S .
Definition 2.2. Given an alphabet A, let denote the congruence on Az generated by
the relation
¹.aa 1 ; 1/ j a 2 Az º: (1)
23. Rational subsets of groups 843
The quotient FA D Az = is the free group on A. We let W Az ! FA denote the

canonical morphism u 7! Œu .
Free groups admit the following universal property: for every map f W A ! G ,
there is a unique group morphism FA ! G that extends f .
Alternatively, we can view (1) as a confluent length-reducing rewriting system
on Az , where each word w 2 Az can be transformed into a unique reduced word
wN with no factor of the form aa 1 ; see [10]. As a consequence, the equivalence
u v () uN D vN .u; v 2 Az /
solves the word problem for FA .
We shall use the notation RA D Az . It is well known that FA is isomorphic to RA
under the binary operation
u ? v D uv .u; v 2 RA /:
We recall that the length jgj of g 2 FA is the length of the reduced form of g, also
denoted by gN .
The letters of A provide a natural basis for FA : they generate FA and satisfy no
nontrivial relations, that is, all reduced words on these generators represent distinct
elements of FA . A group is free if and only if it has a basis.
Throughout this chapter, we assume A to be a finite alphabet. It is well known that
free groups FA and FB are isomorphic if and only if Card A D Card B , see Theorem 2.4
in [30]. This leads to the concept of rank of a free group F : the cardinality of a basis
of F , denoted by rk F . It is common to use the notation Fn to denote a free group of
rank n.
We recall that a reduced word u is cyclically reduced if uu is also reduced. Any
reduced word u 2 RA admits a unique decomposition of the form u D vwv 1 with w
cyclically reduced. A solution for the conjugacy problem easily follows from this: first
reduce the words cyclically; then two cyclically reduced words in RA are conjugate
if and only if they are cyclic permutations of each other. On the other hand, the
order problem admits a trivial solution: only the identity has finite order. Finally, the
generalised word problem will be discussed in the following section.
3. Inverse automata and Stallings’ construction

The study of finitely generated subgroups of free groups entered a new era in the early
eighties when Stallings made explicit and effective a construction [61] that can be traced
back to the early part of the twentieth century in Schreier’s coset graphs (see [60]
and § 2 of Chapter 24) and to Serre’s work [52]. Stallings’ seminal paper was built on
immersions of finite graphs, but the alternative approach using finite inverse automata
became much more popular over the years; for more on their link, see [32]. An extensive
survey has been written by Kapovich and Miasnikov [24].
Stallings’ construction for H 6f:g: FA consists in taking a finite set of generators

for H in reduced form, building the so-called flower automaton and then proceeding to
make this automaton deterministic through the operation known as Stallings foldings.
This turns out to be a terminating procedure, but the key fact is that the construction is
independent from both the given finite generating set and the chosen folding sequence.
A short simple automata-theoretic proof of this claim will be given. The finite inverse
automaton S.H / thus obtained is usually called the Stallings automaton of H . Over
the years, Stallings automata became the standard representation for finitely generated
subgroups of free groups and are involved in many of the algorithmic results presently
obtained.
Several of these algorithms are implemented in computer software, see, e.g.,
CRAG [2], or the packages Automata and FGA in GAP [17].
3.1. Inverse automata. An automaton A is trim if every vertex lies in some successful
path (i.e. a path starting at some initial vertex and ending at some final vertex).
An automaton A over an involutive alphabet Az is involutive if, whenever .p; a; q/ is
an edge of A, so is .q; a 1 ; p/. Therefore it suffices to just depict the positively labelled
edges (having label in A) in their graphical representation.
Definition 3.1. An involutive automaton is inverse if it is deterministic, trim and has
a single final vertex.
If the latter happens to be the initial vertex, it is called the basepoint. It easily
follows from the computation of the Nerode equivalence (see Chapter 10, §2) that every
inverse automaton is a minimal automaton.
Finite inverse automata capture the idea of an action (of a finite inverse monoid,
their transition monoid) on a finite set (the vertex set) through partial bijections. We
recall that a monoid M is inverse if, for every x 2 M , there exists a unique y 2 M
such that xyx D x and y D yxy ; then M acts by partial bijections on itself. The next
result is easily proven, but is quite useful.
uvv 1w
Proposition 3.1. Let A be an inverse automaton and let p ! q be a path in A.
uw
Then there also exists a path p ! q in A.
Another important property relates languages to morphisms. For us, a morphism
between deterministic automata A and A0 is a mapping ' between their respective
vertex sets which preserves the initial vertex, final vertices, and edges, in the sense
that .'.p/; a; '.q// is an edge of A0 whenever .p; a; q/ is an edge of A.
Proposition 3.2. Given inverse automata A and A0 , then L.A/ L.A0 / if and only if
there exists a morphism 'W A ! A0 . Moreover, such a morphism is unique.
Proof. .H)/ Given a vertex q of A, take a successful path
u v
q0 ! q ! t
in A, for some u; v 2 Az . Since L.A/ L.A0 /, there exists a successful path
u v
q00 ! q 0 ! t 0
in A0 . We take '.q/ D q 0 .
To show that ' is well defined, suppose that

u0 v0
q0 ! q ! t
is an alternative successful path in A. Since u0 v 2 L.A/ L.A0 /, there exists a
successful path
u0 v
q00 ! q 00 ! t 0
in A0 and it follows that q 0 D q 00 since A0 is inverse. Thus ' is well defined.
It is now routine to check that ' is a morphism from A to A0 and that it is unique.
.(H/ It is immediate from the definition of morphism.
3.2. Stallings’ construction. Let X be a finite subset of RA . We build an involutive

automaton F.X / by fixing a basepoint q0 and gluing to it a petal labelled by every word
in X as follows: if x D a1 ak 2 X , with ai 2 Az, the petal consists of a closed path
of the form
a1 a2 ak
q0 ! ! ! q0
and the respective inverse edges. All such intermediate vertices are assumed to be
distinct in the automaton. For obvious reasons, F.X / is called the flower automaton
of X .
The automaton F.X / is almost an inverse automaton – except that it need not be
deterministic. We can fix it by performing a sequence of so-called Stallings foldings.
Assume that A is a trim involutive automaton with a basepoint, possessing two distinct
edges of the form
a a
p ! q; p ! r (2)
for a 2 Az. The folding is performed by identifying these two edges, as well as the two
respective inverse edges. In particular, the vertices q and r are also identified (if they
were distinct).
The number of edges is certain to decrease through foldings. Therefore, if we
perform enough of them, we are sure to turn F.X / into a finite inverse automaton.
Definition 3.2. The Stallings automaton of X is the finite inverse automaton S.X /
obtained through folding F.X /.
We shall see that S.X / depends only on the finitely generated subgroup hX i of FA
generated by X , being in particular independent from the choice of foldings taken to
reach it.
Since inverse automata are minimal, it suffices to characterise L.S.X // in terms of
H to prove uniqueness (up to isomorphism):
Proposition 3.3. Fix H 6f:g: FA and let X RA be a finite generating set for H .
Then
\
L.S.X // D ¹L Az j L is recognised by a finite inverse automaton
with a basepoint and Hx Lº:
Proof. . / Clearly, S.X / is a finite inverse automaton with a basepoint. Since

X [ X 1 L.F.X // L.S.X //, it easily follows from Proposition 3.1 that
x L.S.X //:
H (3)
. / Let L Az be recognised by a finite inverse automaton A with a basepoint,
with Hx L. Since X Hx , we have an automaton morphism from F.X / to A, hence
L.F.X // L. To prove that L.S.X // L, it suffices to show that inclusion in L is
preserved through foldings.
Indeed, assume that L.B/ L and B0 is obtained from B by folding the two
u
edges in (2). It is immediate that every successful path q0 ! t in B0 can be lifted to
v
a successful path q0 ! t in B by successively inserting the word a 1 a into u. Now
v 2 L D L.A/ implies u 2 L in view of Proposition 3.1.
Now, given H 6 FA finitely generated, we take a finite set X of generators.
Without loss of generality, we may assume that X consists of reduced words, and we
may define S.H / D S.X / to be the Stallings automaton of H .
Example 3.1. Stallings’ construction for X D ¹a 1 ba; ba2 º, where the next edges to
be identified are depicted by dotted lines, is
a a
F.X / D b q0 a
a b
b
b
a a
a a a D S.X /
q0 q0
b b
A simple, yet important example is given by applying the construction to Fn itself,
when we obtain the so-called bouquet of n circles:
c
a q0 a q0 b a q0 b
S.F1 / S.F2 / S.F3 /

In terms of complexity, the best known algorithm for the construction of S.X / is
due to Touikan [63]. Its time complexity is O.n log n/, where n is the sum of the
lengths of the elements of X .
3.3. Basic applications. The most fundamental application of Stallings’ construction
is an elegant and efficient solution to the generalised word problem:
Theorem 3.4. The generalised word problem in FA is solvable.
We will see many groups in Chapter 24 that have solvable word problem; however,
few of them have a solvable generalised word problem. The proof of Theorem 3.4 relies
on
Proposition 3.5. Consider H 6f:g: FA and u 2 FA . Then u 2 H if and only if
uN 2 L.S.H //.
Proof. .H)/ Follows from (3).
.(H/ It easily follows from the last paragraph of the proof of Proposition 3.3 that,
if B0 is obtained from B by performing Stallings foldings, then L.B0 / D L.B/. Hence,
if H D hX i, we get
L.S.H // D L.F.X // D .X [ X 1 / x
DH
and the implication follows.
It follows from our previous remark that the complexity of the generalised word
problem is O.n log n C m/, where n is the sum of the lengths of the elements of X and
m is the length of the input word. In particular, once the subgroup X has been fixed,
the complexity is linear in m.
Example 3.2. We may use the Stallings automaton constructed in Example 3.1 to
check that baba 1 b 1 2 H D ha 1 ba; ba2 i but ab … H .
Stallings automata also provide an effective construction for bases of finitely gen-
erated subgroups. Consider H 6f:g: FA , and let m be the number of vertices of S.H /.
A spanning tree T for S.H / consists of m 1 edges and their inverses which, together,
connect all the vertices of S.H /. Given a vertex p of S.H /, we denote by gp the T -ge-
gp
odesic connecting the basepoint q0 to p , that is, q0 ! p is the shortest path contained
in T connecting q0 to p .
Proposition 3.6. Let H 6f:g: FA and let T be a spanning tree for S.H /. Let EC be
the set of positively labelled edges of S.H /. Then H is free with basis
Y D ¹gp agq 1 j .p; a; q/ 2 EC n T º:
Proof. It follows from Proposition 3.5 that L.S.H // H , hence Y H . To show
that H D hY i, take h D a1 ak 2 H in reduced form .ai 2 Az /. By Proposition 3.5,
there exists a successful path
a1 a2 ak
q0 ! q1 ! ! qk D q0
in S.H /. For i D 1; : : : ; k , we have either gqi 1 ai gqi1 2 Y [ Y 1

or gqi 1
ai gqi1 D 1,
the latter occurring if .qi 1 ; ai ; qi / 2 T . In any case, we get
h D a1 ak D .gq0 a1 gq11 /.gq1 a2 gq21 / .gqk 1
ak gq01 / 2 hY i
and so H D hY i.
It remains to show that the elements of Y satisfy no nontrivial relations. Let
y1 ; : : : ; yk 2 Y [ Y 1 with yi ¤ yi 11 for i D 2; : : : ; k . Write yi D gpi ai gri 1 , where
ai 2 Az labels the edge not in T . It easily follows from yi ¤ yi 11 and the definition of
spanning tree that
y1 yk D gp1 a1 gr11 gp2 a2 ak 1 grk
1
1
gpk ak grk ;
a nonempty reduced word if k > 1. Therefore Y is a basis of H as claimed.
In the process, we also obtain a proof of the Nielsen–Schreier Theorem, in the case of
finitely generated subgroups. A simple topological proof may be found in [42]:
Theorem 3.7 (Nielsen and Schreier). Every subgroup of a free group is itself free.
Example 3.3. We use the Stallings automaton constructed in Example 3.1 to construct
a basis of H D ha 1 ba; ba2 i.
If we take the spanning tree T defined by the dotted lines in
a
a
q0
b
then Card EC n T D 2 and the corresponding basis is ¹ba2 ; baba 1 b 1 º. Another

choice of spanning tree actually proves that the original generating set is also a basis.
We remark that Proposition 3.6 can be extended to the case of infinitely generated
subgroups, proving the general case of Theorem 3.7. However, in this case there is
no effective construction such as Stallings’, and the (infinite) inverse automaton S.H /
remains a theoretical object, using appropriate cosets as vertices.
Example 3.4. For H 6f:i FA , the inverse automaton S.H / coincides with the Schreier
graph (see § 2 of Chapter 24) of H nFA , namely the graph with vertex set H nFA D
¹Hg j g 2 FA º, with an edge from Hg to Hga for each vertex Hg and each generator
a 2 Az. Since this graph is finite, this proves that H is finitely generated, and therefore
that finite-index subgroups of finitely generated groups are finitely generated.
Another classical application of Stallings’ construction regards the identification of
finite-index subgroups.
Proposition 3.8. Consider H 6f:g: FA .

i. H is a finite-index subgroup of FA if and only if S.H / is a complete automaton.
ii. If H is a finite-index subgroup of FA , then its index is the number of vertices
of S.H /.
Proof. (i, H) ) Suppose that S.H / is not complete. Then there exist some vertex q and
some a 2 Az such that q a is undefined. Let g be a geodesic connecting the basepoint
q0 to q in S.H /. We claim that
Hgam ¤ Hgan if m n > jgj: (4)
Indeed, Hgam D Hgan implies gam n g 1 2 H and so gam n g 1 2 L.S.H //

by Proposition 3.5. Since ga is reduced due to S.H / being inverse, it follows from
m n > jgj that gaam n 1 g 1 D gam n g 1 2 L.S.H //: indeed, g 1 is not long
enough to erase all the a’s. Since S.H / is deterministic, q a must be defined, a
contradiction. Therefore (4) holds and so H has infinite index.
gq
(i, (H ) Let Q be the vertex set of S.H / and fix a geodesic q0 ! q for each q 2 Q.
u
Take u 2 FA . Since S.H / is complete, we have a path q0S! q for some q 2 Q. Hence
ugq 1 2 H and so u D ugq 1 gq 2 Hgq . Therefore FA D q2Q Hgq and so H 6f:i: FA .
S
(ii) In view of FA D q2Q Hgq , it suffices to show that the cosets Hgq are
all distinct. Indeed, assume that Hgp D Hgq for some vertices p; q 2 Q. Then
gp gq 1 2 H and so gp gq 1 2 L.S.H // by Proposition 3.5. On the other hand, since
S.H / is complete, we have a path
gp gq 1
q0 !r
for some r 2 Q. In view of Proposition 3.1, and by determinism, we get r D q0 . Hence

we have paths
gq 1 gq 1
p !q0 ; q !q0 :
Since S.H / is inverse, we get p D q as required.
Example 3.5. Since the Stallings automaton constructed in Example 3.1 is not com-
plete, it follows that ha 1 ba; ba2 i is not a finite-index subgroup of F2 .
Corollary 3.9. If H 6 FA has index n, then rk H D 1 C n.Card A 1/.
Proof. By Proposition 3.8, the automaton S.H / has n vertices and n Card A positive
edges. A spanning tree has n 1 positive edges, so rk H D n Card A .n 1/ D
1 C n.Card A 1/ by Proposition 3.6.
Beautiful connections between finite-index subgroups and certain classes of bifix

codes – set of words none of which is a prefix or a suffix of another – have recently been
unveiled by Berstel, De Felice, Perrin, Reutenauer and Rindone [6].
3.4. Conjugacy. We start now a brief discussion of conjugacy. Recall that the outde-
gree of a vertex q is the number of edges starting at q ; for inverse automata, it equals
the indegree. The geodesic distance in a connected graph is the length of the shortest
undirected path connecting two vertices.
Since the original generating set is always taken in reduced form, it easily follows
that there is at most one vertex in a Stallings automaton having outdegree < 2: the
basepoint q0 . Assuming that H is nontrivial, S.H / must always be of the form

u
q0 q1

where q1 is either q0 or the closest vertex to q0 (in terms of geodesic distance) having
u
outdegree > 2. Note that q1 D q0 if q0 has outdegree > 2 itself. We call q0 ! the
tail (which is empty if q1 D q0 ) and the remaining subgraph the core of S.H /.
The Schreier graph of H nFA and S.H / are related as follows: the Schreier graph
consists of finitely many trees attached to the core of S.H /.
Theorem 3.10. There is an algorithm that decides whether or not two finitely gener-
ated subgroups of FA are conjugate.
Proof. Finitely generated subgroups G; H are conjugate if and only if the cores of S.G/
and S.H / are equal (up to change of basepoint).
The Stallings automata of the conjugates of H can be obtained in the following

alternative ways: (1) declaring a vertex in the core of S.H / to be the basepoint; (2)
gluing a tail to some vertex in the core and taking its other endpoint to be the basepoint.
Note that the tail must be glued in some way that keeps the automaton inverse, so
in particular this second type of operation can only be performed if the automaton is
not complete, or equivalently, if H has infinite index. An immediate consequence is
the following classical result.
Proposition 3.11. A finitely generated normal subgroup of a free group is either trivial
or has finite index.
Moreover, a finite-index subgroup H is normal if and only if its Stallings automaton

is vertex-transitive, that is, if all choices of basepoint yield isomorphic automata.
Example 3.6. Stallings automata of some conjugates of H D ha 1 ba; ba2 i:
b b
a 1 a
S.H / D a S.b H b/ D a
q0 q0
b b
2 a
S.b H b2/ D a
q0
b b
We can also use the previous discussion on the structure of (finite) Stallings automata
to provide them with an abstract characterisation.
Proposition 3.12. A finite inverse automaton with a basepoint is a Stallings automaton
if and only if it has at most one vertex of outdegree 1: the basepoint.
Proof. Indeed, for any such automaton we can take a spanning tree and use it to
construct a basis for the subgroup as in the proof of Proposition 3.6.
3.5. Further algebraic properties. The study of intersections of finitely generated

subgroups of FA provides further applications of Stallings automata. Howson’s classi-
cal theorem admits a simple proof using the direct product of two Stallings automata;
it is also an immediate consequence of Theorem 4.1 and Corollary 4.4 (ii).
Theorem 3.13 (Howson). If H; K 6f:g: FA , then also H \ K 6f:g: FA .
Stallings automata are also naturally related to the famous Hanna Neumann Con-
jecture, recently proved by Mineyev [38] and Friedman [16]: given H; K 6f:g: FA ,
then rk.H \ K/ 1 6 .rk H 1/.rk K 1/. The conjecture arose in a paper of Hanna
Neumann [40], where the inequality rk.H \ K/ 1 6 2.rk H 1/.rk K 1/ was also
proved. In one of the early applications of Stallings’ approach, Gersten provided an
alternative geometric proof of Hanna Neumann’s inequality [18].
A free factor of a free group FA can be defined as a subgroup H generated by a
subset of a basis of FA . This is equivalent to saying that there exists a free product
decomposition FA D H K for some K 6 FA .
Since the rank of a free factor never exceeds the rank of the ambient free group, it
is easy to construct examples of subgroups which are not free factors: it easily follows
from Proposition 3.6 that any free group of rank > 2 can have subgroups of arbitrary
finite rank (and even infinite countable).
The problem of identifying free factors has a simple solution based on Stallings
automata [57]: one must check whether or not a prescribed number of vertex identifi-
cations in the Stallings automaton can lead to a bouquet. However, the most efficient
solution, due to Roig, Ventura and Weil [46], involves Whitehead automorphisms and
will therefore be postponed to § 3.7.
Given a morphism 'W A ! B of inverse automata, let the morphic image '.A/ be
the subautomaton of B induced by the image under ' of all the successful paths of A.
The following classical result characterises the extensions of H 6f:g: FA contained
in FA . We present the proof from [37]:
Theorem 3.14 (Takahasi [62]). Given H 6f:g: FA , one can effectively compute
finitely many extensions K1 ; : : : ; Km 6f:g: FA of H such that the following conditions
are equivalent for every K 6f:g: FA :
i. H 6 K ;
ii. Ki is a free factor of K for some i 2 ¹1; : : : ; mº.
Proof. Let A1 ; : : : ; Am denote all the morphic images of S.H /, up to isomorphism.
Since a morphic image cannot have more vertices than the original automaton, there are
only finitely many isomorphism classes. Moreover, it follows from Proposition 3.12
that, for i D 1; : : : ; m, Ai D S.Ki / for some Ki 6f:g: FA . But then, since we have
L.S.H // L.Ai / D L.S.Ki //, it follows from Proposition 3.5 that H 6 Ki . Clearly,
we can construct all Ai and therefore all Ki .
(i) H) (ii). If H 6 K , it follows from Stallings’ construction that L.S.H //
L.S.K// and so there is a morphism 'W S.H / ! S.K/ by Proposition 3.2. Let Ai be,
up to isomorphism, the morphic image of S.H / through ' . Since Ai D S.Ki / is a
subautomaton of S.K/, it easily follows from Proposition 3.6 that Ki is a free factor of
K : it suffices to take a spanning tree for S.Ki /, extend it to a spanning tree for S.K/,
and the induced basis of Ki will be contained in the induced basis of K .
(ii) H) (i) is immediate.
An interesting research line related to this result is built on the concept of algebraic
extension, introduced by Kapovich and Miasnikov [24], and inspired by the homony-
mous field-theoretical classical notion. Given H 6 K 6 FA , we say that K is an
algebraic extension of H if no proper free factor of K contains H . Miasnikov, Ventura
and Weil [37] proved that the set of algebraic extensions of H is finite and effectively
computable, and it constitutes the minimum set of extensions K1 ; : : : ; Km satisfying the
conditions of Theorem 3.14.
Consider a subgroup H of a group G . The commensurator of H in G , is
CommG .H / D ¹g 2 G j H \ H g has finite index in H and H g º: (5)
For example, the commensurator of GLn .Z/ in GLn .R/ is GLn .Q/.
The special case of finite-index extensions, H 6f:i: K 6 FA is of special interest,

and can be interpreted in terms of commensurators. It can be proved (see Lemma 8.7
in [24] and [59]) that every H 6f:g: FA has a maximum finite-index extension inside FA ,
denoted by Hf:i: ; and Hf:i: D CommFA .H /. Silva and Weil [59] proved that S.Hf:i: /
can be constructed from S.H / using a simple automata-theoretic algorithm:
1. the standard minimisation algorithm is applied to the core of S.H /, taking all
vertices as final;
2. the original tail of S.H / is subsequently reinstated in this new automaton, at
the appropriate vertex.
We present now an application of different type, involving transition monoids.
It easily follows from the definitions that the transition monoid of a finite inverse
automaton is always a finite inverse monoid. Given a group G , we say that a subgroup
H 6 G is pure if the implication
g n 2 H H) g 2 H (6)
holds for all g 2 FA and n > 1. If p is a prime, we say that H is p -pure if (6) holds
whenever .n; p/ D 1.
The next result is due to Birget, Margolis, Meakin, and Weil, and relates these prop-
erties of subgroups to those of the transition monoid of their Stallings automaton [8];
they also show that these problems are PSPACE-complete.
Proposition 3.15. For every H 6f:g: FA ,
i. H is pure if and only if the transition monoid of S.H / is aperiodic;
ii. H is p -pure if and only if the transition monoid of S.H / has no subgroups of
order p .
Proof. Both conditions in (i) are easily proved to be equivalent to the nonexistence in
S.H / of a cycle of the form
u
q1 q2 .k > 1; q1 ¤ q2 /
uk
where u can be assumed to be cyclically reduced. The proof of (ii) runs similarly.
3.6. Topological properties. We require for this subsection some basic topological
concepts, which the reader can recover from Chapter 17.
For all u; v 2 FA , written in reduced form as elements of RA , let u ^ v denote
the longest common prefix of u and v . The prefix metric d on FA is defined, for all
u; v 2 FA , by
²
2 ju^vj if u ¤ v;
d.u; v/ D
0 if u D v:
It easily follows from the definition that d is an ultrametric on FA , satisfying in particular

the axiom
d.u; v/ 6 max¹d.u; w/; d.w; v/º:
The completion of this metric space is compact; its extra elements are infinite reduced
words a1 a2 a3 , with all ai 2 Az, and constitute the hyperbolic boundary @FA of FA ;
see Chapter 24, § 2.5. Extending the operator ^ to FA [@FA in the obvious way, it easily
follows from the definitions that, for every infinite reduced word ˛ and every sequence
.un /n in FA ,
˛ D lim un () lim j˛ ^ un j D C1: (7)

n!C1 n!C1
The next result shows that Stallings automata are given a new role in connection
with the prefix metric. We let cl H denote the closure of H in the completion of FA .
Proposition 3.16. If H 6f:g: FA , then cl H is the union of H with the set of all ˛ 2 @FA
that label paths in S.H / out of the basepoint.
Proof. Since the topology of FA is discrete, we have cl H \ FA D H .

. / If ˛ 2 @FA does not label a path in S.H / out of the basepoint, then ¹j˛^hjW h 2
H º is finite and so no sequence of H can converge to ˛ by (7).
. / Let ˛ D a1 a2 a3 2 @FA , with ai 2 Az, label a path in S.H / out of the

basepoint. Let m be the number of vertices of S.H /. For every n > 1, there exists some
word wn of length < m such that a1 an wn 2 H . Now ˛ D limn!C1 a1 an wn
by (7) and so ˛ 2 cl H .
The profinite topology on FA is defined in Chapter 17: for every u 2 FA , the

collection ¹Ku j K 6f:i: FA º constitutes a basis of clopen neighbourhoods of u. In his
seminal 1983 paper [61], Stallings gave an alternative proof of Marshall Hall’s theorem:
Theorem 3.17 (M. Hall). Every finitely generated subgroup of FA is closed for the
profinite topology.
Proof. Fix H 6f:g: FA and let u 2 FA n H be written in reduced form as an element of

RA . In view of Proposition 3.5, u does not label a loop at the basepoint q0 of S.H /. If
u
there is no path q0 ! in S.H /, we add new edges to S.H / to get a finite inverse
u
automaton A having a path q0 ! q ¤ q0 . Otherwise just take A D S.H /. Next add
new edges to A to get a finite complete inverse automaton B. In view of Propositions 3.8
and 3.12, we have B D S.K/ for some K 6f:i: FA . Hence Ku is open and contains u.
Since H \ Ku ¤ ; yields u 2 K 1 H D K , contradicting Proposition 3.5, it follows
that H \ Ku D ; and so H is closed as claimed.
Example 3.7. We consider the above construction for H D ha 1 ba; ba2 i and u D b 2 :
b b
a a
S.H / D a AD a
q0 q0
b b b
a
BD a
a
q0 b
a
b
b
If we take the spanning tree defined by the dotted lines in B, it follows from Proposi-
tion 3.6 that
K D hba 1 ; b 3 ; b 2 ab 2 ; ba2 ; baba 1 b 1 i
is a finite-index subgroup of F2 such that H \ Kb 2 D ;.
We recall that a group G is residually finite if its set of finite-index subgroups has
trivial intersection. Considering the trivial subgroup in Theorem 3.17, we deduce:
Corollary 3.18. FA is residually finite.
We remark that Ribes and Zalesskiı̆ extended Theorem 3.17 to products of finitely
many finitely generated subgroups of FA ; see [44]. This result is deeply connected to
the solution of Rhodes’ Type II conjecture; see Chapter 4 of [43].
If V denotes a pseudovariety of finite groups (see Chapter 16), the pro-V topology
on FA is defined by considering that each u 2 FA has
¹Ku j K Ef:i: FA ; FA =K 2 V º
as a basis of clopen neighbourhoods. The closure for the pro-V topology of H 6f:g FA
can be related to an extension property of S.H /, and Margolis, Sapir and Weil used
automata to prove that efficient computation can be achieved for the pseudovarieties of
finite p -groups and finite nilpotent groups [34]. The original computability proof for
the p -group case is due to Ribes and Zalesskiı̆ [45].
3.7. Dynamical properties. We briefly mention some examples of applications of

Stallings automata to the study of endomorphism dynamics, starting with Gersten’s
solution of the subgroup orbit problem [19].
The subgroup orbit problem consists of finding an algorithm to decide, for given
H; K 6f:g: FA , whether or not K D '.H / for some automorphism ' of FA . Equiv-
alently, this can be described as deciding whether or not the automorphic orbit of a
finitely generated subgroup is recursive.
Gersten’s solution adapts to the context of Stallings automata the strategy designed
by Whitehead in [66] for solving the orbit problem for words. Whitehead’s proof relies
on a suitable decomposition of automorphisms as products of elementary factors (which
became known as Whitehead automorphisms), and using these as a tool to compute the
elements of minimum length in the automorphic orbit of the word. In the subgroup
case, word length is replaced by the number of vertices of the Stallings automaton.
The most efficient solution to the problem of identifying free factors [46], men-
tioned in § 3.5, also relies on this approach: H 6f:g: FA is a free factor if and only if
the Stallings automaton of some automorphic image of H has a single vertex (that is, a
bouquet).
Another very nice application is given by the following theorem of Goldstein
and Turner [20], which holds more generally for homomorphisms H ! FA with
H 6f:i: FA .
Theorem 3.19. The fixed point subgroup of an endomorphism of FA is finitely gener-
ated.
Proof. Let ' be an endomorphism of FA . For every u 2 FA , define Q.u/ D '.u/ 1 u.
We define a potentially infinite automaton A by taking
¹Q.u/ j u 2 FA º FA
a
as the vertex set, all edges of the form Q.u/ ! Q.ua/ with u 2 FA ; a 2 Az, and
fixing 1 as the basepoint. Then A is a well-defined inverse automaton, with initial state
Q.1/ D 1.
Next we take B to be the subautomaton of A obtained by retaining only those
vertices and edges that lie in successful paths labelled by reduced words. Clearly, B is
still an inverse automaton, and it is easy to check that it must be the Stallings automaton
of the fixed point subgroup of ' .
It remains to be proved that B is finite. We define a subautomaton C of B by
removing exactly one edge among each inverse pair
a a 1
Q.u/ ! Q.ua/; Q.ua/ !Q.u/
a
with a 2 A as follows: if a is the last letter of Q.ua/, we remove Q.u/ ! Q.ua/;
a 1
otherwise, we remove Q.ua/ !Q.u/.
Let M denote the maximum length of the image of a letter by ' . We claim that,
whenever jQ.u/j > M C 1, the vertex Q.u/ has outdegree 1. Indeed, consider an edge
a
Q.u/ ! Q.ua/ in C with a 2 A. Then the final letter cancels in '.a/ 1 Q.u/a, and
'.a/ 1 Q.u/ ends with the last letter of Q.u/, so this letter must be a 1 . Similarly, the
a 1
edge Q.u/ !Q.ua 1
/ belongs to C only if a is the last letter of Q.u/.
Therefore, C consists of a bounded “kernel,” consisting of finitely many vertices

Q.u/ with jQ.u/j 6 M C 1, and finitely many paths from the kernel to the kernel. All
in all C is a finite automaton so the fixed point subgroup of ' is finitely generated.
Note that this proof is not constructive by any means. Bogopolski and Maslakova
give in [9] an algorithm that computes the fixed point subgroup of a free group auto-
morphism; it relies on the sophisticated train track theory of Bestvina and Handel [7]
and other algebraic geometry tools. The general endomorphism case remains open.
Stallings automata were also used by Ventura in the study of various properties of
fixed subgroups, considering in particular arbitrary families of endomorphisms, see [64]
and [36] (also see [65]).
Automata also play a part in the study of infinite fixed points. In [55], these are
considered for the continuous extension of a virtually injective endomorphism to the
hyperbolic boundary of a virtually free group.
4. Rational and recognisable subsets

Rational subsets generalise the notion of “finitely generated” from subgroups to arbi-
trary subsets of a group, and can be quite useful in establishing inductive procedures
that need to go beyond the territory of subgroups. Similarly, recognisable subsets ex-
tend the notion of finite-index subgroups. Basic properties and results can be found
in [5] or [49].
We consider a finitely generated group G D hAi, with the canonical map
W FA ! G . A subset of G is rational if it is the image by D of a rational
subset of Az , and is recognisable if its full preimage under is rational in Az .
For every group G , the classes Rat G and Rec G satisfy the following closure
properties:
Rat G is (effectively) closed under union, product, star, morphisms, inversion,
subgroup generating;
Rec G is (effectively) closed under boolean operations, translation, product,
star, inverse morphisms, inversion, subgroup generating.
Kleene’s Theorem is not valid for groups: Rat G D Rec G if and only if G is finite.
However, if the class of rational subsets of G possesses some extra algorithmic prop-
erties, then many decidability/constructibility results can be deduced for G . Two prop-
erties are particularly coveted for Rat G :
(effective) closure under complement (yielding closure under all the boolean
operations);
decidable membership problem for arbitrary rational subsets.
In these cases, one may often solve problems (e.g., equations, or systems of equations)
whose statement lies far out of the rational universe, by proving that the solution is a
rational set.
4.1. Rational and recognisable subgroups. We start by some basic, general facts.
The following result is essential to connect language theory to group theory.
Theorem 4.1 (Anisimov and Seifert). A subgroup H of a group G is rational if and
only if H is finitely generated.
Proof. (H) ) Let H be a rational subgroup of G and let W FA ! G denote a morphism.
Then there exists a finite Az-automaton A such that H D .L.A//. Assume that A has
m vertices and let X consist of all the words in 1 .H / of length < 2m. Since A is
finite, so is X . We claim that H D h.X /i. To prove it, it suffices to show that
u 2 L.A/ H) .u/ 2 h.X /i (8)
z
holds for every u 2 A . We use induction on juj. By the definition of X , implica-
tion (8) holds for words of length < 2m. Now assume that juj > 2m and (8) holds for
shorter words. Write u D vw with jwj D m. Then there exists a path
v z
! q0 ! q ! t !
in A with jzj < m. Thus vz 2 L.A/ and by the induction hypothesis .vz/ 2 h.X /i.
On the other hand, jz 1 wj < 2m and .z 1 w/ D .z 1 v 1 /.vw/ 2 H ; hence
z 1 w 2 X and so .u/ D .vz/.z 1 w/ 2 h.X /i, proving (8) as required.
((H ) It is trivial.
It is an easier task to characterise the smaller class of recognisable subgroups:
Proposition 4.2. A subgroup H of a group G is recognisable if and only if it has finite
index.
Proof. (H) ) In general, a recognisable subset of G is of the form NX , where N Ef:i: G
and X G is finite. If H D NX is a subgroup of G , then N H and so H has finite
index as well.
((H ) This follows from the well-known fact that every finite-index
T subgroup H of
G contains a finite-index normal subgroup N of G , namely N D g2G gHg 1 . Since
N has finite index, H must be of the form NX for some finite X G .
4.2. Benois’ theorem. The central result in this subsection is Benois’ theorem, the
cornerstone of the whole theory of rational subsets of free groups:
Theorem 4.3 (Benois). The following facts hold.
i. If L Az is rational, then L
x is also rational, and can be effectively construc-
ted from L.
ii. A subset of RA is a rational language as a subset of Az if and only if it is
rational as a subset of FA .
We illustrate this in the case of finitely generated subgroups: temporarily calling
those automata recognising rational subsets of RA “Benois automata,” we may convert
them into Stallings automata by adding inverse edges, identifying initial and terminal
vertices to enforce a basepoint and folding this new automaton. Given a Stallings
automaton, one intersects it with RA to obtain a Benois automaton.
Proof. (i) Let A D .Q; A;z E; I; T / be a finite automaton recognising L. We define

a sequence .An /n of finite automata with "-transitions as follows. Let A0 D A.
Assuming that An D .Q; A; z En ; I; T / is defined, we consider all instances of ordered
pairs .p; q/ 2 Q Q such that
aa 1 1
there exists a path p !q in An for some a 2 Az, but no path p ! q . (P)
Clearly, there are only finitely many instances of (P) in An . We define EnC1 to be the
union of En with all the new edges .p; 1; q/, where .p; q/ 2 Q Q is an instance of
(P). Finally, we define AnC1 D .Q; A; z EnC1 ; I; T /. In particular, note that An D AnCk
for every k > 1 if there are no instances of (P) in An .
Since Q is finite, the sequence .An /n is ultimately constant, say after reaching Am .
We claim that
x D L.Am / \ RA :
L (9)
Indeed, take u 2 L. There exists a sequence of words u D u0 ; u1 ; : : : ; uk 1 , uk D uN

where each term is obtained from the preceding one by erasing a factor of the form aa 1
for some a 2 Az. A straightforward induction shows that ui 2 L.Ai / for i D 0; : : : ; k ,
aa 1 1
since the existence of a path p !q in Ai implies the existence of a path p ! q in
Ai C1 . Hence uN D uk 2 L.Ak / L.Am / and it follows that Lx L.Am / \ RA .
u
For the opposite inclusion, we start by noting that any path p ! q in Ai C1 can
v
be lifted to a path p ! q in Ai , where v is obtained from u by inserting finitely many
factors of the form aa 1 . It follows that
L.Am / D L.Am 1/
x
D D L.A0 / D L;
x . Thus (9) holds.

and so L.Am / \ RA L.Am / D L
Since
[
RA D Az n Az aa 1 Az
a2Az
is obviously rational, and the class of rational languages is closed under intersection, it
x is rational. Moreover, we can effectively compute the automaton Am and
follows that L
a finite automaton recognising RA , hence the direct product construction can be used to
construct a finite automaton recognising the intersection L x D L.Am / \ RA .
(ii) Consider X RA . If X 2 Rat Az , then .X / 2 Rat FA and so X is rational as
a subset of FA .
Conversely, if X is rational as a subset of FA , then X D .L/ for some L 2 Rat Az .
Since X RA , we get X D L x . Now part (i) yields L x 2 Rat Az and so X 2 Rat Az as
required.
Example 4.1. Let A D A0 be depicted by
a
b b
a a 1
1
b
We get
a a
b b b b
1 1
A1 D a a 1 A2 D A3 D 1 a a 1
1 1
b b
x D L.A2 / \ R2 .
and we can then proceed to compute L
The following result summarises some of the most direct consequences of Benois’
Theorem:
Corollary 4.4. The following facts hold.
i. FA has decidable rational subset membership problem.
ii. Rat FA is closed under the boolean operations.
Proof. (i) Given X 2 Rat FA and u 2 FA , write X D .L/ for some L 2 Rat Az .
Then u 2 X if and only if uN 2 Xx D L
x . By Theorem 4.3 (i), we may construct a finite
x x.
automaton recognising L and therefore decide whether or not uN 2 L
(ii) Given X 2 Rat FA , we have FA n X D RA n Xx and so FA n X 2 Rat FA by
Theorem 4.3. Therefore Rat FA is closed under complement.
Since Rat FA is trivially closed under union, it follows from De Morgan’s laws that
it is closed under intersection as well.
Note that we can associate algorithms with these boolean closure properties of
Rat FA in a constructive way. We remark also that the proof of Theorem 4.3 can be
clearly adapted to more general classes of rewriting systems (see [10]). Theorem 4.3
and Corollary 4.4 have been generalised several times by Benois herself [4] and by
Sénizergues, who obtained the most general versions. Sénizergues’ results [50] hold
for rational length-reducing left basic confluent rewriting systems and remain valid for
the more general notion of controlled rewriting system.
Another random document with
no related content on Scribd:
LIST OF ILLUSTRATIONS
PAGE
1. My First Mechanical Drawing. Longitudinal Section of
my Stone-dressing Machine 7
2. The First Porter Governor 21
3. The Porter Marine Governor 37
4. Porter-Allen Engine. Diagram of Admission—Valve
Movements 48
5. Vertical Adjustment of Sustaining Pin for Trunnions of
the Allen Link 52
6. My Improvement in Cranks and Journal Boxes 54
7. My Improvement in Eccentrics 56
8. Diagram from the First Allen Engine taken with the
First Richards Indicator 59
9. Mr. Porter’s Exhibit at the London International
Exhibition, 1862 71
10. Diagram from Allen Engine in London Exhibition of
1862 73
11. Spring-testing Instrument Used in the Manufacture of
the Richards Indicator 86
12. Plan of Spring-testing Instrument 89
13. Diagrams from English Locomotives, taken with the
Richards Indicator 91
14. Engine Bed Designed by Mr. Porter 95
15. Cross-head Designed by Mr. Porter 96
16. Connecting-rod and Strap 99
17. Attaching a Steam-drum to a Lancashire Boiler 107
18. Diagrams from Engine of Evan Leigh, Son & Co. 114
19. Condenser and Air-pump Designed by Mr. Porter.
(Cross-section) 118
20. Diagrams from Engine Built for Mr. Adams 138
21. Exposition Universelle, Paris, 1867. Diagrams from the
“Allen” Engine Employed in Driving Machinery 142
22. Pair of Diagrams from 18×30 Allen Engine at South
Tyne Paper Mill, 108 Revolutions, Vacuum 28
Inches. Only Half Intended Load on Engine 160
23. Cross-section of Machine Shop Proposed by Mr.
Porter in 1868, after the Design of Smith & Coventry 168
24. Card from Allen Engine in Colt’s Armory 178
25. Sectional and Front Elevations of One of the Two
Pairs of Porter-Allen Engines in the Colt Armory,
Hartford, Conn. 180
26. Sectional and Side Elevations of One of the Two Pairs
of Porter-Allen Engines in the Colt Armory, Hartford,
Conn. 181
27. Porter-Allen Engines in the Colt Armory, Hartford,
Conn. Front View 181
28. Porter-Allen Engines in the Colt Armory, Hartford,
Conn. Rear View 181
29. Surface Plates Designed by Mr. Porter 182
30. Diagram from Allen Engine, Back End of Cylinder, at
Fair of American Institute, 1870 194
31. Friction Diagram from Allen Engine at Fair of American
Institute, 1870 196
32. Diagram from Allen Engine, Fair of American Institute,
1870, Cutting Off at ¹⁄₄ Stroke 196
33. Apparatus for Graphically Showing the Acceleration
and Retardation of the Reciprocating Parts of an
Engine 205
208
34. The Allen Boiler Facing
35. The Prototype of the Modern High-speed Engine, Fly-
wheel Side 223
36. Prototype of the Modern High-speed Engine, Crank
Side 224
37. Longitudinal Section of Cylinder and Valves 225
38. Cross-section of Cylinder and Valves 226
39. Connections of Admission Valves 226
40. First Arrangement of Exhaust Valves 228
41. Main Bearing 230
42. Eccentric and Cross-head and Crank-pin Lubricators 230
43. Surface Plate for Producing a True Plane 234
44. Mr. Porter’s Regulating Valve 244
45. The Corliss Engine Exhibited at the Centennial
Exhibition 249
46. Porter-Allen Engine Equal in Power to the Exhibited
Corliss Engine 250
47. Mr. Porter’s Fly-wheel 269
48. Connection of Arms and Rim in Mr. Fritz’ Fly-wheel 273
49. Mr. Allen’s Patent Pressure Plate 293
50. Diagrams from the Otis Engine 311
51. Otis Engine. Dash Pot for Governor 313
52. Diagrams from my First and Only Compound Engine 318
LIST OF FULL-PAGE HALF-TONE PORTRAITS
PAGE
1. Charles T. Porter Facing title page
2. My Father After dedication
3. My Mother After dedication
4. George T. Hope 6
5. Charles B. Richards, a.d. 1858 26
6. John F. Allen 48
7. Joseph E. Holmes 60
8. Alexander Gordon 62
9. Wellington Lee 66
10. Charles T. Porter, a.d. 1862 68
11. Frederick E. Sickels 78
12. W. H. Maw 92
13. William J. Hoyle 122
14. Sir Joseph Whitworth 124
15. Frederick J. Slade 154
16. Professor Charles B. Richards 178
17. President F. A. P. Barnard 198
18. Joseph Nason 204
19. Edwin F. Williams 206
20. Professor Robert H. Thurston 208
21. J. C. Hoadley 220
22. Alexander Lyman Holley 238
23. William R. Jones 244
24. Professor Francis Reuleaux 246
25. Colonel Alexis Petroff 252
26. James Moore 254
27. Emil Brugsch 256
28. Robert W. Hunt 262
29. Stephen W. Baldwin 264
30. Harris Tabor 266
31. Daniel N. Jones 272
32. John Fritz 274
33. E. D. Leavitt 308
34. Samuel T. Wellman 310
35. Charles A. Otis 312
36. Daniel J. Morrell 314
37. Benjamin F. Avery 324
38. James C. Brooks 332
ENGINEERING REMINISCENCES
CHAPTER I
Birth, Parentage and Education. Experience in the Practice of Law. Introduction to

Centrifugal Force. Invention and Operation of a Stone-dressing Machine.
was born in Auburn in the State of New York, January

18th, 1826. My parents were both of New England
descent. My father, John Porter, was born in Hadley,
Mass. His father, William Porter, was the son of
Eleazer Porter and his wife Susannah, one of the
daughters of Jonathan Edwards. My father’s mother
was Lois Eastman. My mother was born in Middletown, Conn. Her
maiden name was Abigail Phillips. Her ancestry in the maternal line
is traced back to Governors Saltonstall, Dudley and the two
Winthrops.
I graduated at Hamilton College, New York, in 1845, read law in
my father’s office, and in the fall of 1847 was admitted to the bar.
Practiced my profession for six or seven years, first in Rochester,
N.Y., afterwards in New York City.
My knowledge of mechanics may be illustrated by a story I once
heard in England of a man who had been prosecuted for selling
adulterated tobacco. He got off by proving that there was no tobacco
at all in the article that he sold. But this illustration hardly does the
case justice.
I had some mechanical ideas, but they were exactly wrong. For
example, I could not see any difficulty in perpetual motion. All one
had to do was to pump up water, which by its fall would furnish
power to run the pump. This, however, was no more absurd than
were two inventions which were brought out in England while I was
there. One of these was corrugating the faces of the piston, so as to
present more extended surfaces for the steam pressure to be
exerted upon. The other was a device for utilizing that half of the
force of the steam which had been wasted against the cylinder
heads. Both of these were published with commendatory remarks in
the Mechanics’ Magazine. The last, if I recollect rightly, was the
original bottom feature of the Wells balance-engine. My error was
that I made no account of friction, which must be overcome before
motion can take place. We shall see before long the same disregard
of friction by men who ought to have known better.
My utter ignorance of everything mechanical at that time is
capable of proof. I stepped right into one of those “springes to catch
woodcocks” which were being set in those days, and proved myself
to be about as green a gosling mechanically as ever was plucked.
I had a client by the name of Searle, who was a “dead-beat.” He
owed me about $100, which I could not collect. He finally called upon
me and told me frankly that he could not pay me one red cent,
because he had no money; but he could put me in the way of making
a fortune, and he was anxious in that way to discharge the great
obligation which he felt himself under to me.
A new invention had appeared, called the Gwynne & Sawyer
static-pressure engine, that was bound to revolutionize all
applications of power. It was, he told me, attracting great attention in
engineering circles, and there had been a hot discussion over its
theoretical principles, but its advocates had successfully vanquished
all their antagonists and now the invention was established on a
perfectly sound scientific basis. If I would give him a receipt in full for
the money that he owed me and put another $100 into this
enterprise, he was in a position to secure for me a number of rights
to use the machine. He kindly offered to introduce me to Mr. Sawyer.
Mr. Gwynne was unfortunately absent from home at the time. (I
learned afterwards that he was in jail.) Mr. Sawyer received me most
graciously. I think he had been told by Mr. Searle about how much
taffy I might be expected to swallow, but he must have ventured far
beyond his instructions. He told me that he was delighted to make
my acquaintance; he had frequently heard of me through our mutual
friend, Mr. Searle, and of my triumphs at the bar, and had come to
feel a great admiration for me, and was proud to show this great
invention to a man so eminently capable of appreciating it. He told
me that the invention was a practical method of utilizing that
wonderful power known as centrifugal force. This force could be
obtained in any amount. In fact, it was the force that kept the
universe in motion. It had lain unutilized for so long a time because
engineers had never been able to apply it practically. This difficulty
had been completely overcome in this great invention, and this
wonderful power was now to be made available for the world. He
gave me quite an oration on the subject, saying, “We do not
antagonize the forces of nature, we utilize them and apply them to
beneficial purposes; consequently all nature co-operates with us,”
and more to the same effect. He was able to show me a working
model of this great invention; was very sorry that he could not put it
in motion for me that day, as it happened to be a little out of order;
but I would be able to see the principle of its operation very distinctly.
I was flattered into believing that I saw the principle, with the result
that Mr. Sawyer saw the principal, and with the further result that
after that I never saw or heard of either principal or interest. Our
mutual friend, Mr. Searle, also disappeared.
This was my first lesson in mechanics, given to me by a master of
his art. I am not sure, on the whole, but that in one way and another
it has been worth the trifle it cost me.
Had any one at that time told me that the expression “centrifugal
force” is entirely misleading, that in reality there is no such force, that
what goes by this name is not a force at all, nothing but a resistance,
the resistance which a body revolving around an exterior point
opposes to being continually deflected from a straight line of motion,
and which ceases the instant the deflecting force ceases, when the
body merely moves on in a straight line tangent to the circle, and in
bodies revolving around their own axes or centers of gravity is the
same resistance of their atoms, he would probably have had about
the same success in making me see it that I long afterwards had with
some engineering friends.
It is difficult at the present day to conceive the confusion of thought
which then prevailed on this subject. The language of text-books was
vague in the extreme.
The coincidence is not without interest, that my first mechanical
experience, though in this ridiculous fashion, should have been with
what was to become so prominent a feature of the high speed
governors and engine.
I had for some time felt a growing disgust with the profession of
the law. The contrast between the glorious science of human rights
and the art of its practical application was very forcibly presented to
my mind. I realized the fitness of the protest of Bryant, who
described himself as being “forced to drudge for the dregs of men.” I
was a regular reader of the Evening Post, in which an article
appeared one day, written by John Bigelow, then the editor of the
Post, laudatory of a certain judge whose term on the bench had
lately closed, and who then retired from the profession. On this act
Mr. Bigelow warmly congratulated him. Among a number of pungent
expressions in the article I was particularly struck by this one: “The
association of lawyers is mostly with knaves and fools.” My own
experience bore witness to the truth of this statement. A few legal
successes, which cost me incredible labor, interspersed of course
with disappointments, weighed nothing compared with the daily
association which I seemed compelled to endure. I formed a scheme
for establishing a conciliation office for the amicable settlement of
disputes, but found every man prepared to compromise on the
extreme verge of his own position. So I gave that up.
I had another client, a Mr. Hastings, who had invented a stone-
dressing machine, which he had patented, and the patent for which
he wanted to dispose of. He had a working model of his invention,
which was operated for visitors in the shop where it was built. He
invited me to go and see it, which I did, and it certainly worked very
well indeed. I recalled afterwards that the stone was carefully
bedded on the table of the machine. I was quite fascinated with it
and took some friends to see it, who were equally captivated, and
the result was that we bought the patent. To make sure of its value,
however, I first called with Mr. Hastings on Mr. Munn, his patent
solicitor, and received Mr. Munn’s assurance that he had a very high
opinion of it.
I gradually abandoned my law business, and devoted myself to the
exploitation of this invention. I put into it all the money I had and all
that I could borrow. After a while a large working machine was
completed for us, the drawings for which I had made by a German
draftsman, and which was built under my direction at the works of
Mott & Ayers, near the foot of West Twenty-sixth Street. When this
machine was finished the parties in interest assembled at these
works to see it tried.
One experiment was enough. I had put into the machine a stone
that was quite a foot thick and which was supported at two points. At
the first cut made across this stone it broke in two in the middle. I
found myself, in the words of President Cleveland, “confronted not
by a theory but by a condition.” The machine was absurd. The patent
was worthless. The enterprise was a failure. Our money had all been
thrown into the sea. Nothing could be done unless I did it; and I knew
nothing of mechanics, of machine design or construction, or of
mechanical drawing, except the little that I had picked up in the
works of Mott & Ayers while this machine was in process of
construction. I should say, however, that the head draftsman in that
establishment had given me some instruction in mechanical drawing,
so that I knew the use of the instruments and what kind of ink to use.
I cannot recollect that I was in the least cast down or discouraged.
I cannot now account for my confidence. I believed that the
fundamental features of this machine were correct. These were:
cutting stone by a blow given by a hammer moving in an inclined
direction, and which was thrown up by a cam and thrown down by
springs. The more I reflected upon it the more I became convinced
that a successful stone-dressing machine could be made on those
general lines, and in no other way; and I also became impressed
with what seems the almost absurd conviction that I could make it.
The machine that broke the stone had a broad hammer—a cast-
iron plate with tongues on the sides running in grooves in a frame,
and to the end of which a long steel blade was bolted. My first idea
was to divide the single broad hammer into several hammers
working side by side and striking their blows successively; the
second was to separate the hammers from the tool-holders, the
third, to employ the same tools that were used by stone-cutters,
namely, the point, tooth-chisel and drove, and to give them as nearly
as possible the same blow that was given to them by the workman,
and the fourth, to give to the tools only the blow necessary to do their
work.
I infused my own enthusiasm into my associates to such a degree
that they agreed to put up the money and let me try the experiment.
That also is something that I now wonder at.
The most influential member of this devoted band was George T.
Hope, President of the Continental Fire Insurance Co., a gentleman
whom I shall have frequent occasion to mention, and who remained
my steadfast friend till his death, which occurred soon after the close
of my engineering career.
I set about my work in this manner. My house, on the south side of
Twenty-second Street west of Seventh Avenue, had been arranged
in its construction to use the extension room back of the parlor as a
dining-room. That left the front basement available for me. This I
equipped for a drawing-office, and set myself at work to learn
mechanical drawing, and at the same time to design this machine. I
bought a Scotch instruction book, and a sheet of “antiquarian”
drawing-paper. In those days all drawings were made on white linen
paper, and this was nearly the largest size that was made, and cost
75 cents a sheet. My principal drawing-implement was india-rubber.
As my plans grew in my mind I had to rub out my preceding
sketches. I spent a great deal of my time in visiting the large
engineering works on the East River—the Allaire Works, the Morgan
Works and the Novelty Works—and studying tools and machines
and principles and methods of construction. I tried to get my mind
saturated with mechanics. I finally succeeded in producing the
design, this vertical section of which I have sketched from memory
after fifty years.
It will be seen that this machine was massive in its construction.
This was required on account of the speed—300 rotations of the
shaft per minute—at which I had determined to run it. This was my
first employment of high speed.
George T. Hope
The original model of the machine made 60 strokes per minute. In
the machine that broke the stone I had increased the speed to 100
strokes per minute. In designing the successful machine I made the
great jump to 300 revolutions of the cam-shaft per minute. This was
done after much study of practical requirements. I observed carefully
the speed of planing-machines. I had also the opportunity of
witnessing the operation of the first wood-moulding machine, and
was much impressed by the speed of the rotary cutters and the
rapidity with which the work was turned out. I wanted a motion of 40
inches a minute for the stone table, which would make the output of
the machine satisfactory; 300 revolutions would give this motion, the
table advancing .133 of an inch at each blow.
Side frame not shown, except Channels
for Elevating Screws
My First Mechanical Drawing.

Longitudinal Section of my Stone-dressing Machine.
The machine contained six hammers, each 6 inches wide and

weighing about 200 pounds, which ran in a suspended frame. The
front member of this frame was a wrought-iron bar 6 inches square,
with a projection on the lower side, as shown. At the ends this bar
was first reduced to 5 inches square, the corners rounded to 1 inch
radius, and mortised into cast-iron side-bars 4 inches thick, one of
which is shown in the sectional view. Beyond these side-bars the
wrought-iron bar was turned down to journals 3¹⁄₂ inches in diameter,
which turned in the heads of large screws, one of which is
represented. Beyond those journals it was further reduced to 2
inches diameter, and the ends threaded. These projections extended
through slots in the main framing, and nuts on the outside provided
with long handles enabled the whole to be bound fast in its position,
when that had been determined.
The hammers had two faces; the upper faces struck on this 6-inch
square bar, the lower faces struck the backs of the heavy tool-
holders. These tool-holders were held in position in the manner
shown. At the extreme back end they rocked downward upon a
heavy cross-bar. At the front they rose against the 6-inch cross-bar.
They were made with a heavy hook at the back, which prevented
them from coming forward further than the projection at the bottom of
this cross-bar permitted. A curved spring held them up to the cross-
bar when the weight of the hammer was removed. Between the 6-
inch cross-bar and the tool-holders and the hammer faces I
introduced a sheet of heavy leather belting, which deadened the
force of the blow. A stone-cutter uses a wooden mallet to drive the
tooth-chisels and droves, because the impact of iron on iron has a
disintegrating effect upon the stone, which the stone-cutters call
“stunning the stone.” It produces a vibration in the body of the stone
to a depth of perhaps ¹⁄₈ inch, and, however well the surface of the
stone may appear when it is finished, after a while the outside will
flake off to the depth to which these vibrations have extended. This
leather buffer served the purpose of the wooden mallet, completely
avoiding this difficulty. Incidentally also it made the building
habitable, by transforming the blow into a dull thud, which at the rate
of 1800 blows per minute from the six hammers was itself quite
important to be done.
The large screws on each side of the machine at the front were
provided at the top with long nuts resting on a cross-bar and
combined with worm-wheels. A shaft carrying two worms engaging
with these wheels extended across the top of the machine, so that
the nuts were rotated identically, and the front of the suspended
frame was raised or lowered as the thickness of the stone or depth
of the cut required. The machine could cut stone from the thinnest
ashlar up to a thickness of about 3 feet. The hammers ran on rollers
as shown. At the back the frame and hammers were carried on
similar rollers on the same shaft. The ends of this shaft also turned in
square heads of screws, and by a mechanism similar to that already
described the back of the frame could be elevated or depressed to
the height required and be set at any desired angle.
The six tool-holders were made in the following manner: I got from
England a bar of steel long enough to make them all. This was
planed into the form shown in the section, and the sockets for the
shanks of the tools were finished to an equal depth and perfectly in
line. It was then parted, and the ends of each finished in a slotting-
machine.
The blows struck by the hammers were very effective. The cams
had a throw of 1¹⁄₄ inches, but they threw the hammers back against
the springs 1¹⁄₄ inches further, making their fall 2¹⁄₂ inches. This I
ascertained by holding a piece of thin board edgeways between the
upper end of a hammer and the cross-bar at the back, when the
hammer crushed it up to this height.
We never ran over the stone with the points but once. They made
everything before them fly. On the other hand, the droves merely
dusted the surface, to take out the marks of the tooth-chisels. All
surplus force in the blow was received on the 6-inch cross-bar. The
tools stood motionless unless pushed back by the stone, when they
received a sufficient portion of the blow to drive them forward to their
position.
The feed motion was powerful, being imparted by a worm
engaging in a worm-wheel 24 inches in diameter, while the run back
was swift, quite 100 feet in a minute.
The sides of the steel tool-holders, rubbing against each other,
became after a while badly abraded. I was obliged to plane them off
and dovetail thin strips of hardened steel into them. These prevented
any further trouble. The sides of the end tool-holders, however,
which rubbed against the cast-iron side-bars, I observed, were
polished without sensible wear.
This was a very important observation. These surfaces all rubbed
together dry. The pressure was only the side thrust, which was very
trifling. Under these conditions the molecules of the same material
interlocked, while those of the different materials did not. These two
materials were, however, extremely different in their constituent
features. Perhaps this point of freedom of some different materials
from interlocking was still better illustrated by the set-screws, where
this difference of molecular structure did not exist in the same
degree. These were made of Ulster iron, a superior quality of
American iron then largely used in New York City for bolts. They
were ⁵⁄₈-inch screws, and were also used dry, no oil being allowed
anywhere over the stones. Each tool-holder contained three of these
set-screws. The outside ones were tightened and loosened sixty
times every day. The middle ones, where only the points were used,
were tightened and loosened twenty times every day and at other
times stood loose in their threads. The tool-holders being massive,
and the blows of the hammers also coming on the leather cushion,
there was no vibration. At the end of the two years’ running the outer
bolts were all perfect fits. The middle ones were loose, but still held
the tools perfectly.
The rollers on which the hammers ran were hardened and turned
on hardened shafts. The hammers themselves had chilled faces,
and their surfaces running on the rollers were also chilled. The
surfaces of the tool-holders and of the bar on which these rocked
were provided with hardened strips to the extent that they came in
contact with each other. The cams and rollers and their pins were
also hardened.
When built this machine was found to require only a single
alteration. I had welded the cams onto the shaft, the welds being
guaranteed by the smith to be perfectly sound. No appearance of
unsoundness could be detected when the shaft was finished, but
after running a week or two the cams became loose. This also gave
me a useful lesson. I was obliged to send to England for blocks of
steel, which were bored, finished and keyed on the shaft in the
manner shown, and the working surfaces of the cams were
hardened. This required the substitution of new hammers, because
the cams could not be threaded through the old ones. The hubs of
these cams were 6 inches long, covering the shaft.
Our company, being satisfied from its design that the machine
when finished would prove a success, rented from Mr. Astor a large
lot on the south side of Fourteenth Street, west of Ninth Avenue,
extending through to Thirteenth Street, and erected and equipped a
building and established a stone-yard, where the machine ran
successfully for two seasons, principally employed in facing ashlar,
as the flat-faced stones of buildings are termed. It turned out with
ease 600 square feet of finished surface per day, which was the
work of thirty men, and it never broke a stone, however thin.
For facing in the machine the stones were set on bars 2 inches
thick and 4 inches high, cast on the surface of sliding tables. These
were both longitudinal and cross bars, and were provided with holes
³⁄₄ inch in diameter and about 3 inches apart. There were two tables,
each 16 feet in length.
Several pieces of ashlar were set upon each table and held by
dogs and wedges on these bars. They were wedged up very easily
by skilled workmen, so that they would finish at the same level. At
one side of the ways on which the tables moved, near each end, was
placed a swing-crane, which was double- and triple-geared, so that
by means of it any stone that the machine was adapted to cut could
be lifted by two men. The operations of cutting the stones on one
table and removing the stones and setting others on the other table
went on simultaneously, so that the cutting was never interrupted,
except to change the tools and the tables. This last was done as
follows: Each table, when the work on it was completed, was run
rapidly backward or forward to attach it to the other table. It was then
connected with this by a couple of hooks, and, the motion being
reversed, pulled it into place under the tools, and in doing this took
its own place under a crane, so that the work of removing the
finished stones and setting rough ones went on continuously at one
end or the other of the ways.
In addition to the machine I designed the building and the whole
plant and the plan of its operation, which moved like clockwork. I
made every drawing myself. The cranes I obtained in Rochester, N.
Y., of a pattern which the builders made for railroads for handling
heavy freight.
I bought from a stone-dressing company that had failed a rubbing
machine called the Jenny Lind rubber, from the fact that it was
started the same year in which that songstress was brought to the
United States by Mr. Barnum. This rubbing-machine was quite a
success. From a central vertical spindle a jointed arm extended in
three lengths, each about 12 feet long. The sections of this arm were
very deep, so that there was no sag at the end, where the rubbing-
plate was driven by belting and could be moved from stone to stone
around a circle of 36 feet radius. Half of this circle was sufficient for
our use. I made only one change in this machine. The pulleys, two
pairs on each joint, one at the top and one at the bottom, about two
feet in diameter by three inches face, were of course horizontal. The
makers were afraid the belts would fall off; so they made these
pulleys with two square grooves, ¹⁄₂ inch wide by ¹⁄₄ inch deep, in
their faces, and had corresponding strips of leather sewn on the
belts to run in these grooves. I threw all these away and substituted
ordinary pulleys with their faces slightly crowning. Never had the
least trouble. Indeed, these pulleys did better than I expected. I
supposed the belts would need to be taken up occasionally, on
account of becoming stretched, but they did not. Perhaps they would
have done so if the strain on them had been greater. This rubbing
machine resembled the stone-dressing machine in one respect:
everything about it was arranged for continuous operation and the
largest output.
The business was carried on the first season under the
management of Mr. John McClave, a master stone-cutter, and the
second season under the management of the firm of Brown &
Young, stone-cutters. Mr. Hugh Young, of this firm, has since been
prominent in the stone-cutting business in New York.
The machine was found to possess a remarkable advantage over
hand work. The sun was called by stone-cutters “the great revealer.”
When its rays fell at a small angle on a surface finished by hand they
showed very considerable irregularities. The same test showed work
in the machine to be true planes. It won a high reputation; stone-
cutters were anxious to get their surfaces done in the machine, and
we had more work offered us than we could do.
The following incident illustrates the favorable impression made by
the machine upon everyone who witnessed its operation:
At a meeting of the Directors of the Company at which I was
present Mr. Daniel S. Miller, a gentleman somewhat prominent in
financial New York, was late. He made the following explanation. “I
thought that before the meeting I would visit the stone yard and see

Ebook Handbook of Automata Theory Volume Ii Automata in Mathematics and Selected Applications First Edition Jean Eric Pin Online PDF All Chapter

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ebook Handbook of Automata Theory Volume Ii Automata in Mathematics and Selected Applications First Edition Jean Eric Pin Online PDF All Chapter

Uploaded by

Copyright:

Available Formats

Handbook of Automata Theory Volume

II Automata in Mathematics and

Handbook of Automata Theory Volume I Theoretical

Mathematical Theory of Switching Circuits and Automata

Automata Theory and Formal Languages Fundamental

FORMAL LANGUAGE AND AUTOMATA THEORY 2nd Edition Singh

Cellular Automata and Groups 2nd Edition Tullio

Introduction to Computation Haskell Logic and Automata

Linear Algebra and Optimization with Applications to

An Introduction to Formal Languages and Automata 6th ed

Edited by Jean-Éric Pin

Edited by Jean-Éric Pin

ISBN Vol. I 978-3-98547-002-0

Bibliographic information published by the Deutsche Nationalbibliothek

Published by EMS Press, an imprint of the

© 2021 EMS Press

Cover drawing of Jacques de Vaucanson’s digesting duck (canard digérant) published in

Volume I: Theoretical foundations

message-passing systems, symbolic methods, synthesis, timed automaton model, veri-

Acknowledgements. I would like to thank the European Science Foundation, and in

Boris Adamczewski (Chapter 25) Jason Bell (Chapter 25)

Bernard Boigelot (Chapter 32) Thomas Colcombet (Chapter 18)

Zoltán Ésik (Chapter 20) Tero Harju (Chapter 3)

Dora Giammarresi (Chapter 9) Michal Kunc (Chapter 21)

Orna Kupferman (Chapter 38)

Dietrich Kuske (Chapters 4, 31) Jean Mairesse (Chapter 5)

Jean-Éric Pin (Chapter 1) Thomas Schwentick (Chapter 29)

Sasha Rubin (Chapter 28) Jeﬀrey Shallit (Chapter 13)

Moshe Y. Vardi (Chapter 38) Thomas Wilke (Chapter 6)

Igor Walukiewicz (Chapter 33)

Pascal Weil (Chapter 16)

2. Finitely generated groups

Given H 6 G , we let ŒG W H  denote the index of H in G , that is, the number of

The quotient FA D Az = is the free group on A. We let W Az ! FA denote the

3. Inverse automata and Stallings’ construction

Stallings’ construction for H 6f:g: FA consists in taking a ﬁnite set of generators

To show that ' is well deﬁned, suppose that

3.2. Stallings’ construction. Let X be a ﬁnite subset of RA . We build an involutive

Proof. .  / Clearly, S.X / is a ﬁnite inverse automaton with a basepoint. Since

S.F1 / S.F2 / S.F3 /

in S.H /. For i D 1; : : : ; k , we have either gqi 1 ai gqi1 2 Y [ Y 1

then Card EC n T D 2 and the corresponding basis is ¹ba2 ; baba 1 b 1 º. Another

Proposition 3.8. Consider H 6f:g: FA .

Hgam ¤ Hgan if m n > jgj: (4)

Indeed, Hgam D Hgan implies gam n g 1 2 H and so gam n g 1 2 L.S.H //

for some r 2 Q. In view of Proposition 3.1, and by determinism, we get r D q0 . Hence

Beautiful connections between ﬁnite-index subgroups and certain classes of biﬁx

The Stallings automata of the conjugates of H can be obtained in the following

Moreover, a ﬁnite-index subgroup H is normal if and only if its Stallings automaton

Example 3.6. Stallings automata of some conjugates of H D ha 1 ba; ba2 i:

3.5. Further algebraic properties. The study of intersections of ﬁnitely generated

The special case of ﬁnite-index extensions, H 6f:i: K 6 FA is of special interest,

It easily follows from the deﬁnition that d is an ultrametric on FA , satisfying in particular

d.u; v/ 6 max¹d.u; w/; d.w; v/º:

˛ D lim un () lim j˛ ^ un j D C1: (7)

Proof. Since the topology of FA is discrete, we have cl H \ FA D H .

.  / Let ˛ D a1 a2 a3 2 @FA , with ai 2 Az, label a path in S.H / out of the

The proﬁnite topology on FA is deﬁned in Chapter 17: for every u 2 FA , the

Proof. Fix H 6f:g: FA and let u 2 FA n H be written in reduced form as an element of

3.7. Dynamical properties. We brieﬂy mention some examples of applications of

Therefore, C consists of a bounded “kernel,” consisting of ﬁnitely many vertices

4. Rational and recognisable subsets

Proof. (i) Let A D .Q; A;z E; I; T / be a ﬁnite automaton recognising L. We deﬁne

Given H 6 G , we let ŒG W H denote the index of H in G , that is, the number of

The quotient FA D Az = is the free group on A. We let W Az ! FA denote the

Proof. . / Clearly, S.X / is a ﬁnite inverse automaton with a basepoint. Since

. / Let ˛ D a1 a2 a3 2 @FA , with ai 2 Az, label a path in S.H / out of the