each φ(ai) ∈ F contains no inverses aj
On the rose (1-vertex graph) with directed
edges labelled ai, one has a natural representative for any automorphism of F . The
key feature of positive automorphisms is the fact that the positive iterates of this
representative restrict to injections on each edge of the graph. Such maps are the
prototypes for train-track representatives.
This discussion suggests a strategy that one might follow in order to prove one
Main Theorem: first, one should prove it in the case of positive automorphisms,
relying on the simplifications afforded by the positivity hypothesis to confront the
web of large-scale cancellation phenomena that must be understood if one is to
have any chance of proving the theorem in general. Then, in the general case,
one should attempt to follow the architecture of the proof in the positive case,
using a suitably refined train-track description of the automorphism in place of the
positivity assumption. We shall implement the two stages of this plan in Parts 1
and 3 of this monograph, respectively. Ultimately, this strategy works. However, in
Part 3, in order to bring our plan to fruition we have to deal with myriad additional
complexities arising from intricate cancellations that do not arise in the positive
Roughly speaking, these additional complexities correspond to the fact that
most free group automorphisms do not have train track representatives, only relative
train track representatives. In Part 2 of this monograph, we refine the theory of
improved relative train track maps due Bestvina, Feighn and Handel , so as to
tease-out features that allow us to adapt the crucial arguments from Part 1. A
vital ingredient in this approach is the identification of basic units that will play
the role in the general case that single edges (letters) played in the positive case.
To this end, we develop a theory of beads, whose claim to the role is clinched by the
Beaded Decomposition Theorem 3.2.1. This theorem is the main objective of Part
2. Indeed we have gone to considerable lengths to distill the entire contribution of
Part 2 to Part 3 into this single statement and the important technical refinement
of it described in Addendum 2.0.1. We have done so in order that the reader who
is willing to accept it as an article of faith may proceed directly from Part 1 to Part
The introduction to each part of the book contains a more detailed explanation
of its contents.
The first author’s work was supported in part by Research Fellowships from the
EPSRC of Great Britain and by a Royal Society Wolfson Research Merit Award.
Much of this work was undertaken whilst he was a Professor at Imperial College
London, from which he was granted two terms of sabbatical leave. The second
author was supported in part by a Junior Research Fellowship at Merton College,
Oxford, by a Taussky-Todd Instructorship and a Senior Research Fellowship at the
California Institute of Technology and by NSF Grant DMS-0504251. We thank
these organisations for their support. We also thank the anonymous referee for his
careful reading and helpful comments.