(Abhijit Champanerkar, Oliver Dasbach, Efstratia K (B-Ok - Xyz)

CONTEMPORARY
MATHEMATICS
541
Interactions Between
Hyperbolic Geometry,
Quantum Topology
and Number Theory
Workshop
June 3–13, 2009
Conference
June 15–19, 2009
Columbia University, New York, NY
Abhijit Champanerkar, Oliver Dasbach,

Efstratia Kalfagianni, Ilya Kofman,
Walter Neumann, Neal Stoltzfus
Editors
American Mathematical Society

Quantum Topology
and Number Theory
This page intentionally left blank
CONTEMPORARY
MATHEMATICS
541
Quantum Topology
and Number Theory
Workshop
June 3–13, 2009
Conference
June 15–19, 2009
Columbia University, New York, NY
Abhijit Champanerkar, Oliver Dasbach,

Efstratia Kalfagianni, Ilya Kofman,
Walter Neumann, Neal Stoltzfus
Editors
American Mathematical Society

Providence, Rhode Island
Editorial Board
Dennis DeTurck, managing editor
George Andrews Abel Klein Martin J. Strauss
2000 Mathematics Subject Classification. Primary 57Mxx, 32Qxx, 60Gxx, 16Txx, 17Bxx,
81Rxx, 81Txx, 11Sxx, 14Txx.
Library of Congress Cataloging-in-Publication Data

Workshop on Interactions between Hyperbolic Geometry, Quantum Topology, and Number Theory
(2009 : Columbia University)
Interactions between hyperbolic geometry, quantum topology, and number theory : Workshop
on Interactions between Hyperbolic Geometry, Quantum Topology, and Number Theory, June 3–
13, 2009, Columbia University, New York, New York / Abhijit Champanerkar . . . [et al.], editors.
p. cm.
Includes bibliographical references.
ISBN 978-0-8218-4960-6 (alk. paper)
1. Low-dimensional topology—Congresses. 2. Manifolds (Mathematics)—Congresses.
3. Quantum field theory—Congresses. 4. Number theory—Congresses. I. Champanerkar,
Abhijit, 1975– II. Title.
QA612.14.W67 2009
514.22—dc22
2010049884
Copying and reprinting. Material in this book may be reproduced by any means for edu-
cational and scientific purposes without fee or permission with the exception of reproduction by
services that collect fees for delivery of documents and provided that the customary acknowledg-
ment of the source is given. This consent does not extend to other kinds of copying for general
distribution, for advertising or promotional purposes, or for resale. Requests for permission for
commercial use of material should be addressed to the Acquisitions Department, American Math-
ematical Society, 201 Charles Street, Providence, Rhode Island 02904-2294, USA. Requests can
also be made by e-mail to reprint-permission@ams.org.
Excluded from these provisions is material in articles for which the author holds copyright. In
such cases, requests for permission to use or reprint should be addressed directly to the author(s).
(Copyright ownership is indicated in the notice in the lower right-hand corner of the first page of
each article.)

c 2011 by the American Mathematical Society. All rights reserved.
The American Mathematical Society retains all rights
except those granted to the United States Government.
Copyright of individual articles may revert to the public domain 28 years
after publication. Contact the AMS for copyright status of individual articles.
Printed in the United States of America.

∞ The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http://www.ams.org/
10 9 8 7 6 5 4 3 2 1 16 15 14 13 12 11
Dedicated to the memory of Xiao-Song Lin,
whose mathematics continues to inspire.
Contents
Preface ix
List of Participants xi
An Introduction to the Volume Conjecture
Hitoshi Murakami 1
Quantum Field Theory and the Volume Conjecture
Tudor Dimofte and Sergei Gukov 41
R-Matrix Knot Invariants and Triangulations
R. M. Kashaev 69
Knots and Tropical Curves
Stavros Garoufalidis 83
Quantum Coadjoint Action and the 6j-Symbols of Uq sl2
Stéphane Baseilhac 103
What is a Sequence of Nilsson Type?
Stavros Garoufalidis 145
From Angled Triangulations to Hyperbolic Structures
David Futer and François Guéritaud 159
Triangulated 3-Manifolds: From Haken’s Normal Surfaces to Thurston’s
Algebraic Equation
Feng Luo 183
An Introduction to Fully Augmented Links
Jessica S. Purcell 205
Orbifolds and Commensurability
Genevieve S. Walsh 221
Realizing Arithmetic Invariants of Hyperbolic 3-Manifolds
Walter D. Neumann 233
Fields of Definition of Canonical Curves
D. D. Long and A. W. Reid 247
vii
Preface
This book is based on a 10-day workshop of minicourses in hyperbolic geometry,

quantum topology and number theory, given by leading experts, in June 2009 at
Columbia University. The proceedings of this enormously successful workshop can
serve as an introduction to this active research area in a way that is expository and
broadly accessible to graduate students and researchers new to this area.
The twelve papers in this book present ideas and tackle problems that arose
from the confluence of several different currents of research mathematics over the
past thirty years. First, Thurston’s ground-breaking work established the impor-
tance of hyperbolic geometry in the study of 3-manifolds. Mostow-Prasad rigidity
implied that geometric invariants are also topological invariants, providing new
tools from geometry to study 3-manifolds. Second, rigidity also enabled the use
of arithmetic tools to study hyperbolic 3-manifolds giving rise to arithmetic topo-
logical invariants. Third, following the discovery and subsequent generalizations of
the Jones polynomial, new quantum diagrammatic invariants of 3-manifolds were
discovered. How these geometric, arithmetic and quantum invariants interact is
a fundamental open question in the study of knots, links and 3-manifolds that
motivates much of the research discussed in this book.
Quantum invariants such as the Jones polynomial are not understood in terms
of the geometry of the complementary space. But a recurrent theme is that geomet-
ric and quantum invariants can be related by quantization of geometric invariants.
Kashaev, Garoufalidis, Baseilhac and Dimofte & Gukov each discuss aspects of
geometric quantization. The big open problem in this area is the Volume Conjec-
ture, which relates hyperbolic volume to Jones polynomials. Murakami, Dimofte &
Gukov, and Kashaev each discuss different approaches to the Volume Conjecture.
A basic tool in the study of hyperbolic 3-manifolds is an ideal triangulation,
which can be described in terms of complex shape parameters that satisfy certain
polynomial gluing equations. These equations are in general difficult to solve. Futer
& Guéritaud and Luo discuss angle structures, which provide an alternative method
to solve the gluing equations and understand the geometry of the 3-manifold. Pur-
cell discusses the geometry of augmented link complements, which can be explicitly
described without having to solve gluing equations.
Rigidity implies that the shape parameters are algebraic numbers, which leads
to arithmetic invariants to study hyperbolic 3-manifolds. The resulting arithmetic
tools have been used to study geodesic surfaces, closed geodesics and volumes of
hyperbolic 3-manifolds. Neumann discusses arithmetic invariants like the invariant
trace field and Bloch invariants, and realization problems related to these invari-
ants. Deforming hyperbolic structures on 3-manifolds gives rise to algebraic curves
defined over number fields which are discussed by Long & Reid. Walsh discusses
ix
x PREFACE
applications of arithmetic invariants to study the problem of commensurability of

hyperbolic knots and 3-manifolds. Garoufalidis discusses how arithmetic invariants
arise in the study of the asymptotic behavior of quantum invariants.
Acknowledgements. We gratefully acknowledge the support of NSF for the
workshop and the subsequent conference. We thank the staff of mathematics de-
partments of Columbia University and Louisiana State University for their help
with various aspects of the workshop and conference. We thank the speakers and
the participants for making the workshop and conference a huge success. Finally,
we thank the authors for contributing to this volume.
Abhijit Champanerkar
Oliver Dasbach
Efstratia Kalfagianni
Ilya Kofman
Walter Neumann
Neal Stoltzfus
Department of Mathematics, College of Staten Island, CUNY, 2800 Victory

Boulevard, Staten Island, NY 10314
Department of Mathematics, Louisiana State University, Baton Rouge, LA

70803
Department of Mathematics, Michigan State University, East Lansing, MI

48824
Department of Mathematics, College of Staten Island, CUNY, 2800 Victory

Boulevard, Staten Island, NY 10314
Department of Mathematics, Barnard College, Columbia University, 2990 Broad-

way MC 4424, New York, NY 10027
Department of Mathematics, Louisiana State University, Baton Rouge, LA

70803
List of Participants
Abdelmalek Abdesselam Abhijit Champanerkar

University of Virginia CSI, CUNY
Ilesanmi Adeboye Lin Chen
University of California, Santa Barbara University of California, Los Angeles
Cody Armond Qingtao Chen
Louisiana State University University of California, Berkeley
Ken Baker David Cherney
University of Miami University of California, Davis
Maiia Bakhova David Clark
Louisiana State University Randolph-Macon College
Cheryl Balm Moshe Cohen
Michigan State University Louisiana State University
Yuanyuan Bao Benjamin J Cooper
Tokyo Institute of Technology University of California, Davis
Stéphane Baseilhac Christopher Cornwell
Université de Grenoble Michigan State University
Joan Birman Francesco Costantino
Barnard College, Columbia University IRMA, Strasbourg
Jonathan Bloom Marc Culler
Columbia University University of Illinois, Chicago
Jeffrey Boerner Jeffrey Danciger
University of Iowa Stanford University
Francis Bonahon Oliver Dasbach
University of Southern California Louisiana State University
Steve Boyer Tudor Dimofte
Université du Québec à Montréal Caltech
Xuanting Cai Giang Do
Louisiana State University Georgia Institute of Technology
Andrew Casson Andrew Elliott
Yale University Rice University
xi
xii PARTICIPANTS
Alexander Ellis Eric Katerman

Columbia University University of Texas, Austin
Lena Folwaczny Keiko Kawamuro
University of Illinois, Chicago IAS
Charles Frohman Mark Kidwell
University of Iowa U.S. Naval Academy
David Futer Lois Kilgore-Muhammad
Temple University Dowling College
Swiatoslaw Gal Ilya Kofman
Ohio State University CSI, CUNY
Stavros Garoufalidis Alex Kontorovich
Georgia Institute of Technology Brown University
Pat Gilmer Greg Kuperberg
Louisiana State University University of California, Davis
Elisenda Grigsby Grant Lakeland
Columbia University/Boston College University of Texas, Austin
Sergei Gukov Emily Landes
University of California, Santa Barbara University of Texas, Austin
Ren Guo Thang Le
University of Minnesota Georgia Institute of Technology
Mustafa Hajij Jonatan Lenells
Louisiana State University Cambridge University
Neil Hoffman Sam Lewallen
University of Texas, Austin Princeton University
Zheng Huang Adam Lowrance
CSI, CUNY Louisiana State University
William Jaco Feng Luo
Oklahoma State University Rutgers University
Thomas Jaeger Melissa Macasieb
Michigan State University Maryland
Yuichi Kabaya John Mangual
Tokyo Institute of Technology UC Santa Barbara
Aaron Kaestner Gregor Masbaum
University of Illinois, Chicago Université Paris Diderot
Effie Kalfagianni Donovan McFeron
Michigan State University Columbia University
Rinat Kashaev Madelyn Miller
Université de Genève NYU
PARTICIPANTS xiii
Heather Molle Neal Stoltzfus

University of Iowa Louisiana State University
Harriet Moser Lenny Tevlin
Columbia University Yeshiva University
Hitoshi Murakami Sean Thomas

Tokyo Institute of Technology Emory University
Kei Nakamura Rohit Thomas

Oklahoma State University University of California, Davis
Walter Neumann Dylan Thurston

Barnard College, Columbia University Barnard College, Columbia University
Anh Tran
Philip Ording
Georgia Institute of Technology
Medgar Evers College, CUNY
Roland van der Veen
Kathleen Petersen University of Amsterdam
Florida State University
Genevieve Walsh
Jessica Purcell Tufts University
Brigham Young University
Biao Wang
Yulan Qing Cornell University
Tufts University
Ben Webster
Himal Rathnakumara MIT
Brigham Young University
Stephan Wehrli
Alan Reid Université Paris Diderot
University of Texas, Austin
Erica Whitaker
Fernando Rodriguez Villegas Ohio State University
University of Texas, Austin Blake Winter
Julien Roger SUNY Buffalo
University of Southern California Helen Wong
Heather Russell Bowdoin College
University of Iowa Atakan Tugkan Yakut
Nigde University
Murat Savas
Gazi University Tian Yang
Rutgers University
Radmila Sazdanovic
George Washington University Christian Zickert
University of California, Berkeley
Adam Sikora
SUNY Buffalo
Jonah Sinick
UIUC
Noah Snyder
Columbia University
Contemporary Mathematics
Volume 541, 2011
An Introduction to the Volume Conjecture
Hitoshi Murakami
Abstract. This is an introduction to the Volume Conjecture and its gener-

alizations for nonexperts.
The Volume Conjecture states that a certain limit of the colored Jones
polynomial of a knot would give the volume of its complement. If we deform the
parameter of the colored Jones polynomial we also conjecture that it would also
give the volume and the Chern–Simons invariant of a three-manifold obtained
by Dehn surgery determined by the parameter.
I start with a definition of the colored Jones polynomial and include ele-
mentary examples and short description of elementary hyperbolic geometry.
1. Introduction
In 1995 Kashaev introduced a complex valued link invariant for an integer
N ≥ 2 by using the quantum dilogarithm [15] and then he observed that his invari-
ant grows exponentially with growth rate proportional to the volume of the knot
complement for several hyperbolic knots [16]. He also conjectured that this also
holds for any hyperbolic knot, where a knot in the three-sphere is called hyperbolic
if its complement possesses a complete hyperbolic structure with finite volume.
In 2001 J. Murakami and the author proved that Kashaev’s invariant turns
out to be a special case of the colored
√ Jones polynomial. More precisely Kashaev’s
invariant is equal to JN K; exp(2π −1/N ) , where JN (K; q) is the N -dimensional
colored Jones polynomial associated with the N -dimensional irreducible representa-
tion of the Lie algebra sl(2; C) and K is a knot (§ 2). We also generalized Kashaev’s
conjecture to any knot (Volume Conjecture) by using the Gromov norm, which can
be regarded as a natural generalization of the hyperbolic volume (§ 3). If it is
true it would give interesting relations between quantum topology and hyperbolic
geometry. So far the conjecture is proved only for several knots and some links but
we have supporting evidence which is described in § 4.
In the Volume Conjecture
√ we study the colored Jones polynomial
√ at the N -th
root of unity exp(2π −1/N ). What happens if we replace 2π −1 with another
complex number? Recalling that the complete hyperbolic structure of a hyperbolic
knot complement can be deformed by using a complex parameter [34], we √ expect
that we can also relate the colored Jones polynomial evaluated at exp (2π −1 +
2000 Mathematics Subject Classification. Primary 57M27 57M25 57M50.

Key words and phrases. volume conjecture, knot, hyperbolic knot, quantum invariant, col-
ored Jones polynomial, Chern–Simons invariant.
The author is supported by Grant-in-Aid for Challenging Exploratory Research (21654053).
1 2011
c American Mathematical Society
1
2 HITOSHI MURAKAMI

u)/N to the volume of the deformed hyperbolic structure. At least for the figure-
eight knot this is true if u is small [30]. It is also true (for the figure-eight knot)
that we can also get the Chern–Simons invariant, which can be regarded as the
imaginary part of the volume, from the colored Jones polynomial (§ 5).
In general we conjecture that this is also true, that is, for any knot the asymp-
totic behavior of the colored Jones polynomial would determine the volume of a
three-manifold obtained as the deformation associated with the parameter u.
The aim of this article is to give an elementary introduction to these conjectures
including many examples so that nonexperts can easily understand. I hope you will
join us.
Acknowledgments. The author would like to thank the organizers of the
workshop and conference “Interactions Between Hyperbolic Geometry, Quantum
Topology and Number Theory” held at Columbia University, New York in June
2009.
Thanks are also due to an immigration officer at J. F. Kennedy Airport, who
knows me by papers, for interesting and exciting discussion about quantum topol-
ogy.
2. Link invariant from a Yang–Baxter operator

In this section I describe how we can define a link invariant by using a Yang–
Baxter operator.
2.1. Braid presentation of a link. An n-braid is a collection of n strands
that go downwards monotonically from a set of fixed n points to another set of
fixed n points as shown in Figure 1. The set of all n-braids makes a group Bn with
Figure 1. braid
product of braids β1 and β2 given by putting β2 below β1 . It is well known (see

for example [2]) that Bn is generated by σ1 , σ2 , . . . , σn−1 (Figure 2) with relations
σi σj = σj σi (|i − j| > 1) and σk σk+1 σk = σk+1 σk σk+1 . See Figure 3 for the
latter relation, which is called the braid relation. So we have the following group
presentation of Bn .
(2.1) Bn = σ1 , σ2 , . . . , σn−1 | σi σj = σj σi (|i − j| > 1), σk σk+1 σk = σk+1 σk σk+1
It is known that any knot or link can be presented as the closure of a braid.
Theorem 2.1 (Alexander [1]). Any knot or link can be presented as the closure
of a braid.
AN INTRODUCTION TO THE VOLUME CONJECTURE 3
1 2 i i+1 n1 n
Figure 2. ith generator of the n-braid group Bn
Figure 3. braid relation
closure
−−−−→ =
Figure 4. the figure-eight knot is presented as the closure of a braid
Here the closure of an n-braid is obtained by connecting the n points on the

top with the n points on the bottom without entanglement as shown in the middle
picture of Figure 4.
There are many braids that present a knot or link but if two braids present
the same knot or link, they are related by a finite sequence of conjugations and
(de-)stabilizations. In fact we have the following theorem.
Theorem 2.2 (Markov [19]). If two braids β and β give equivalent links, then
they are related by a finite sequence of conjugations, stabilizations, and destabi-
lizations. Here a conjugation replaces αβ with βα, or equivalently β with α−1 βα
(Figure 5), a stabilization replaces β ∈ Bn with βσn±1 ∈ Bn+1 (Figure 6), a desta-
bilization replaces βσn±1 ∈ Bn+1 with β ∈ Bn .
2.2. Yang–Baxter operator. Alexander’s theorem (Theorem 2.1) and Markov’s
theorem (Theorem 2.2) can be used to define link invariants. I will follow Turaev
[36] to introduce a link invariant derived from a Yang-Baxter operator.
Let V be an N -dimensional vector space over C, R an isomorphism from V ⊗ V
to itself, m an isomorphism from V to itself, and a and b non-zero complex numbers.
Definition 2.3. A quadruple (R, μ, a, b) is called an enhanced Yang–Baxter
operator if it satisfies the following:
4 HITOSHI MURAKAMI
α β
conjugation
−−−−−−−→
β α
Figure 5. αβ is conjugate to β.
β stabilization
−−−−−−−−→ β
destabilization
←−−−−−−−−−
Figure 6. β (βσn± , respectively) is stabilized (destabilized, respec-

tively) to βσn± (β, respectively).
(1) (R ⊗ IdV )(IdV ⊗R)(R ⊗ IdV ) = (IdV ⊗R)(R ⊗ IdV )(IdV ⊗R),
(2) R(μ ⊗ μ) = (μ ⊗ μ)R,

(3) Tr2 R±1 (IdV ⊗μ) = a±1 b IdV .
Here Trk : End(V ⊗k ) → End(V ⊗(k−1) ) is defined by

N −1
j ,j ,...,j ,j
Trk (f )(ei1 ⊗ei2 · · ·⊗eik−1 ) := fi11,i22,...,ik−1
k−1
,j (ej1 ⊗ej2 ⊗· · ·⊗ejk−1 ⊗ej ),
j1 ,j2 ,...,jk−1 ,j=0
⊗k
where f ∈ End(V ) is given by

N −1
f (ei1 ⊗ ei2 ⊗ · · · ⊗ eik ) = fij11,i,j22,...,i
,...,jk
k
(ej1 ⊗ ej2 ⊗ · · · ⊗ ejk )
j1 ,j2 ,...,jk =0
and {e0 , e1 , . . . , eN −1 } is a basis of V .

Remark 2.4. The isomorphism R is often called an R-matrix, and the equation
(1) is known as the Yang–Baxter equation.
Given an n-braid β, we can construct a homomorphism Φ(β) : V ⊗n → V ⊗n by
⊗(i−1) ⊗(n−i−1)
replacing a generator σi with IdV ⊗R ⊗ IdV , and its inverse σi−1 with
⊗(i−1) ⊗(n−i−1)
IdV ⊗R−1 ⊗ IdV (Figure 7).
V
‫ۑ‬V V
‫ۑ‬V
⇒ R , ⇒ R

V
‫ۑ‬V V
‫ۑ‬V
Figure 7. Replace a generator with the R-matrix.

Example 2.5. For the braid σ1 σ2−1 σ1 σ2−1 the corresponding homomorphism is
give as follows (Figure 8):
Φ(σ1 σ2−1 σ1 σ2−1 ) = (R ⊗ IdV )(IdV ⊗R−1 )(R ⊗ IdV )(IdV ⊗R−1 ).
V
‫ۑ‬V
‫ۑ‬V
R

Φ R
−
→
R

R
V
‫ۑ‬V
‫ۑ‬V
Figure 8. A braid and the corresponding homomorhism
2.3. Invariant. Let (R, μ, a, b) be an enhanced Yang–Baxter operator on an

N -dimensional vector space V .
Definition 2.6. For an n-braid β, we define T(R,μ,a,b) (β) ∈ C by the following
formula.

T(R,μ,a,b) (β) := a−w(β) b−n Tr1 Tr2 · · · Trn Φ(β)μ⊗n · · · ,
where w(β) is the sum of the exponents in β. Note that Tr1 : End(V ) → C is the
usual trace.
Example 2.7. For the braid σ1 σ2−1 σ1 σ2−1 , we have
T(R,μ,a,b) (σ1 σ2−1 σ1 σ2−1 )

= b−2 Tr1 Tr2 (Tr3 ((R ⊗ IdV )(IdV ⊗R−1 )(R ⊗ IdV )(IdV ⊗R−1 )(μ ⊗ μ ⊗ μ)))
since w(σ1 σ2−1 σ1 σ2−1 ) = +1 − 1 + 1 − 1 = 0. See Figure 9 for a graphical inter-
pretation of the invariant. Note that Trk is regarded as closing the k-th strand.
We can show that T(R,μ,a,b) gives a link invariant.

Theorem 2.8 (Turaev [36]). If β and β present the same link, then TR,μ,a,b (β) =
TR,μ,a,b (β ).
Sketch of a proof. By Markov’s theorem (Theorem 2.2) it is sufficient to
prove that TR,μ,a,b is invariant under a braid relation, a conjugation and a stabi-
lization.
The invariance under a braid relation σi σi+1 σi = σi+1 σi σi+1 follows from Fig-
ure 10. Note that the left hand side depicts a braid relation (2.1) and the right
hand side depicts the corresponding Yang–Baxter equation (Definition 2.3 (1)).
The invariance under a conjugation follows from Figure 11. The first equality
follows since Tr1 ◦ Tr2 ◦ · · · ◦ Trn is invariant under a conjugation. The second
6 HITOSHI MURAKAMI
R

R
⇒ a−w(β) b−n × R

R
μ μ μ
Figure 9. A braid and its invariant
V
‫ۑ‬V ‫ۑ‬
V V
‫ۑ‬V
‫ۑ‬V
R R
Φ
= −
→ R = R
R R
V
‫ۑ‬V ‫ۑ‬
V V
‫ۑ‬V
‫ۑ‬V
braid relation Yang–Baxter equation
Figure 10. braid relation corresponds to the Yang–Baxter equation
equality follows since (μ ⊗ μ)R = R(μ ⊗ μ) (Definition 2.3 (2)). Note that the
equality (μ ⊗ μ)R = R(μ ⊗ μ) means that a pair μ ⊗ μ can pass through a crossing.
Φ(α) Φ(β) Φ(β)

= μ μ μ =
Φ(β) Φ(α)
μ μ μ Φ(α)
μ μ μ
Figure 11. invariance under a conjugation
To prove the invariance under a stabilization, we first note that if a homomor-

phism f : V ⊗n → V ⊗n given by

N −1
f (ei1 ⊗ · · · ⊗ ein ) = fij11,...,i
,...,jn
n
(ej1 ⊗ · · · ⊗ ejn ),
j1 ,...,jn =0
then its n-fold trace is given by

Tr1 (· · · (Trn (f )) · · · ) = fjj11,...,j
,...,jn
n
.
j1 ,...,jn
Therefore if g is a homomorphism g : V ⊗ V → V ⊗ V given by glk11,l,k2 2 , then we have

⊗(n−1)
Tr1 · · · Trn Trn+1 (f ⊗ IdV )(IdV ⊗g)
j ,...,jn−1 ,kn jn ,kn+1
= fj11,...,jn−1 ,jn gkn ,kn+1 ,
j1 ,...,jn ,kn ,kn+1
⊗(n−1)
which coincides with the n-fold trace of the homomorphism f IdV ⊗ Tr2 (g) : V ⊗n →
V ⊗n .
Therefore for β ∈ Bn we have

Tr1 Tr2 · · · Trn Trn+1 Φ(βσn±1 )μ⊗(n+1) ···

⊗n ⊗(n−1) ±1
= Tr1 Tr2 · · · Trn Trn+1 (μ Φ(β) ⊗ IdV )(IdV ⊗R (IdV ⊗μ)) ···

⊗(n−1)
= Tr1 Tr2 · · · Trn (μ⊗n Φ(β))(IdV ⊗ Tr2 (R±1 (IdV ⊗μ))) · · ·

±1
⊗n
=a b Tr1 Tr2 · · · Trn (μ Φ(β)) · · ·

=a±1 b Tr1 Tr2 · · · Trn (Φ(β)μ⊗n ) · · · ,

since Tr2 R±1 (IdV ⊗μ = a±1 b IdV (Definition 2.3 (3)) as depicted in Figure 12.
Since w(βσn±1 ) = w(β) ± 1, the invariance under a stabilization follows.
Φ(β) Φ(β)
_
R +1 = a±1 b
μ μ μ μ μ μ μ
Figure 12. invariance under a stabilization
Therefore we can define a link invariant TR,μ,a,b (L) to be TR,μ,a,b (β) if L is the
closure of β.
8 HITOSHI MURAKAMI
2.4. Quantum (g, V ) invariant. One of the important ways to construct an

enhanced Yang–Baxter operator is to use a quantum group, which is a deformation
of a Lie algebra.
Let g be a Lie algebra. Then one can define a quantum group Uq (g) as a
deformation of g with q a complex parameter ([4], [12]). Given a representation
ρ : g → gl(V ) of g one can construct an enhanced Yang–Baxter operator. The
corresponding invariant is called the quantum (g, V ) invariant. For details see [36].
To define the colored Jones polynomial we need the Lie algebra sl2 (C) and
its N -dimensional irreducible representation ρN : sl2 (C) → gl(VN ). The quan-
tum (sl2 (C), VN ) invariant is called the N -dimensional colored Jones polynomial
JN (L; q).
A precise definition is as follows.
Put V := CN and define the R-matrix R : V ⊗ V → V ⊗ V by

N −1
ij
R(ek ⊗ el ) := Rkl ei ⊗ ej ,
i,j=0
where
min(N −1−i,j)
ij
{l}!{N − 1 − k}!
Rkl := δl,i+m δk,j−m
(2.2) m=0
{i}!{m}!{N − 1 − j}!

× q i−(N −1)/2 j−(N −1)/2 −m(i−j)/2−m(m+1)/4 ,
with {e0 , e1 , . . . , eN −1 } is the standard basis of V , {m} := q m/2 −q −m/2 and {m}! :=
{1}{2} · · · {m}. Here q is a complex parameter. A homomorphism μ : V → V is
given by

N −1
μ(ej ) := μij ei
i=0
with
μij := δi,j q (2i−N +1)/2 .
2
−1)/4
Then it can be shown that (R, μ, q (N , 1) gives an enhanced Yang–Baxter
operator.
Definition 2.9 (colored Jones polynomial). For an integer N ≥ 1, put V :=
CN and define R and μ as above. The N -dimensional colored Jones polynomial
JN (L; q) for a link L is defined as
{1}
JN (L; q) := T(R,μ,q(N 2 −1)/4 ,1) (β) × ,
{N }
where β is a braid presenting the link L.
Remark 2.10. Note that JN (unknot; q) = 1 since

N −1
{N }
Tr1 (μ) = q (2i−N +1)/2 = .
i=0
{1}
The two-dimensional colored Jones polynomial J2 (L; q) is (a version) the orig-

inal Jones polynomial [14] as shown below.
L+ : L− : L0 :
, ,
Figure 13. skein triple
Lemma 2.11. Let L+ , L− , and L0 be a skein triple, that is, they are the same
links except for a small disk as shown in Figure 13. Then we have the following
skein relation:
qJ2 (L+ ; q) − q −1 J2 (L− ; q) = (q 1/2 − q −1/2 )J2 (L0 ; q).
Proof. By the definition, the R-matrix is given by
⎛ 1/4 ⎞
q 0 0 0
⎜ 0 q 1/4 − q −3/4 q −1/4 0 ⎟
R=⎜ ⎝ 0
⎟
q −1/4 0 0 ⎠
0 0 0 q 1/4
with respect to the basis {e0 ⊗ e0 , e0 ⊗ e1 , e1 ⊗ e0 , e1 ⊗ e1 } of V ⊗ V , and μ is given
by
−1/2
q 0
μ=
0 q 1/2
with respect to the basis {e0 , e1 } of V .
Therefore we can easily see that
(2.3) q 1/4 R − q −1/4 R−1 = (q 1/2 − q −1/2 ) IdV ⊗ IdV .
Since L+ , L− , and L0 can be presented by n-braids βσi β , βσi−1 β , and ββ
respectively, we have
{2}
qJ2 (L+ ; q) − q −1 J2 (L− ; q) ×
{1}

=q × q −3(w(ββ )+1)/4
Tr1 (Tr2 (· · · (Trn (Φ(βσi β )μ⊗n ))))

− q −1 × q −3(w(ββ )−1)/4
Tr1 (Tr2 (· · · (Trn (Φ(βσi−1 β )μ⊗n ))))

=q −3w(ββ )/4

× q 1/4 Tr1 (Tr2 (· · · (Trn (Φ(βσi β )μ⊗n )))) − q −1/4 Tr1 (Tr2 (· · · (Trn (Φ(βσi−1 β )μ⊗n ))))

=q −3w(ββ )/4

⊗(i−1) ⊗(n−i−1)
× Tr1 (Tr2 (· · · (Trn (Φ(β)(IdV ⊗q 1/4 R ⊗ IdV )Φ(β )μ⊗n ))))

⊗(i−1) ⊗(n−i−1)
− Tr1 (Tr2 (· · · (Trn (Φ(β)(IdV ⊗q −1/4 R−1 ⊗ IdV )Φ(β )μ⊗n ))))
(from (2.3))

=q −3w(ββ )/4
(q 1/2 − q −1/2 ) Tr1 (Tr2 (· · · (Trn (Φ(ββ )μ⊗n ))))
{2}
=(q 1/2 − q −1/2 )J2 (L0 ; q) × ,
{1}
completing the proof.
10 HITOSHI MURAKAMI
Remark 2.12. The original Jones polynomial V (L; q) satisfies

q −1 V (L+ ; q) − qV (L− ; q) = (q 1/2 − q −1/2 )V (L0 ; q)
[14, Theorem 12]. So we have J2 (L; q) = (−1)(L)−1 V (L; q −1 ), where (L) denotes
the number of components of L.
2.5. Example of calculation. Put β := σ1 σ2−1 σ1 σ2−1 . Its closure E is a knot
called the figure-eight knot (Figure 4). We will calculate JN (E; q).
Instead of calculating Tr1 (Tr2 (Tr3 (Φ(β)μ⊗3 ))) ∈ C, we will calculate Tr2 (Tr3 (Φ(β)
(Id ⊗μ ⊗ μ))) ∈ End(V ), which is a scalar multiple by Schur’s lemma (for a proof
see [18, Lemma 3.9]). See Figure 14.
β ⇒ Φ(β)
μ μ
Figure 14. We close all the strings except for the first one.
Then Tr1 (Tr2 (Tr3 (Φ(β)μ⊗3 ))) coincides with the trace of μ times the scalar S.
Since
T(R,μ,q(N 2 −1)/4 ,1) (β) = q −w(β)(N −1)/4
Tr1 (Tr2 (Tr3 (Φ(β)μ⊗3 )))
2
= q −w(β)(N −1)/4
2
Tr1 (Sμ)

N −1
= q −w(β)(N −1)/4
2
S q (2i−N +1)/2
i=0
−1)/4 {N }
= q −w(β)(N
2
S,
{1}
we have JN (L; q) = q −w(β)(N −1)/4 S = S.
2
We need an explicit formula for the inverse of the R-matrix, which is given by
min(N −1−i,j)
{k}!{N − 1 − l}!
(R−1 )ij
kl = δl,i−m δk,j+m
(2.4) m=0
{j}!{m}!{N − 1 − i}!

m − i−(N −1)/2 j−(N −1)/2 −m(i−j)/2+m(m+1)/4
× (−1) q .
To calculate the scalar S, draw a diagram for the braid β and close it except
for the first string (Figure 15). Fix a basis {e0 , e1 , . . . , eN −1 } of CN .
Label each arc with a non-negative integer i less than N , which corresponds to a
basis element ei , where our braid diagram is divided into arcs by crossings so that at
each crossing four arcs meet. Since the homomorphism Tr2 (Tr3 (Φ(β)(Id ⊗μ ⊗ μ)))
is a scalar multiple, we choose any basis for the first (top-left) arc of Figure 15 and
Figure 15. Draw the braid σ1 σ2−1 σ1 σ2−1 and close it except for
the left-most one.
N-1
Figure 16. Label the first (top-left) arc with N − 1.
calculate the scalar. For simplicity we choose eN −1 and so we label the first arc
with N − 1 (Figure 16).
Recall that we will associate the R-matrix or its inverse with each crossing as
follows.
i j i j
⇒ ij
Rkl , ⇒ (R−1 )ij
kl
k l k l
Therefore we will label the other arcs following the following two rules:
12 HITOSHI MURAKAMI
(i). At a positive crossing, the top-left label is less than or equal to the bottom-
right label, the top-right label is greater than or equal to the bottom-left
label, and their differences coincide (see (2.2)).
i j
: i + j = k + l, l ≥ i, k ≤ j,
k l
(ii). At a negative crossing, the top-left label is greater than or equal to the
bottm-right label, the top-right label is less than or equal to the bottm-left
label, and their differences coincide (see (2.4)).
i j
: i + j = k + l, l ≤ i, k ≥ j.
k l
From Rule (i), the next arc should be labeled with N − 1, and the difference at
the top crossing is 0 (Figure 17). This is why we chose N − 1 for the label of the
first arc.
N-1
0
N-1
Figure 17. The label of the next arc should be N − 1, and the
difference at the top crossing is 0
Label the top-right arc with i with 0 ≤ i ≤ N − 1 (Figure 18).

The label of the left-middle arc should be i since the difference at the top
crossing is 0.
Label the arcs indicated in Figure 20 with j and k with 0 ≤ j ≤ N − 1 and
0 ≤ k ≤ N − 1. Then the difference at the second top crossing is N − k − 1.
Therefore the arc between the second top crossing and the second bottom crossing
should be labeled with N − k + j − 1 from Rule (ii) (Figure 21). Since the label
should be less than N , we have j − k ≤ 0.
Look at the bottom-most crossing and apply Rule (ii). We see that i ≥ k and
the difference at the bottom-most crossing is i − k. So the label of the arc between
N-1
i
0
N-1
Figure 18. Label the top-right arc with i (0 ≤ i ≤ N − 1).
N-1
i
0
N-1
i
Figure 19. The label of the left-middle arc should be i since the
difference at the top crossing is 0.
the second bottom crossing and the bottom-most crossing is i + j − k (Figure 22).
We also see that the label of the bottom-most arc is N − 1 as we expected.
However from Rule (i) we have j − k ≥ 0 and so k = j. Therefore we finally
have the labeling as indicated in Figure 23.
14 HITOSHI MURAKAMI
N-1
i
0
N-1
N-k-1 j
i
k
Figure 20. Choose j and k. Then the difference at the second

top crossing is N − k − 1.
N-1
i
0
N-1
N-k-1 j
i
N-k+j-1
k
Figure 21. The label of the arc between the second top arc and
the second bottom arc should be N −k +j −1. Note that j −k ≤ 0.
Now we can calculate the colored Jones polynomial. We have

N −1,i −1,j i,N −1
JN (E; q) = Ri,N −1 (R−1 )N
N −1,j RN −1,i (R
−1 i,j j i
)i,j μj μi
i≥j
{N − 1}!{i}!{N − 1 − j}!
= (−1)N −1+i
({j}!)2 {i − j}!{N − 1 − i}!
i≥j
2
−2ij−2j 2 +3N +6N i+2N j−3N 2 )/4
× q (−i−i .
In this formula we need two summations. To get a formula involving only one
summation we regard the figure-eight knot E as the closure of a tangle as shown in
N-1
i
0
N-1
N-k-1 j
i
N-k+j-1
k
i+j-k
i-k
i j
N-1
Figure 22. The label of the arc between the second bottom cross-
ing and the bottom-most crossing should be i + j − k. Note that
j−k ≥0
N-1
i
0
N-1
N-j-1 j
i
0 N-1 j
i i-j
i j
N-1
Figure 23. The label k should coincide with j.
Figure 24. In this case we need to put μ at each local minimum where the arc goes
from left to right, and μ−1 at each local maximum where the arc goes from right
to left. See [18, Theorem 3.6] for details.
16 HITOSHI MURAKAMI
0
i
i
0
i j
j i+j 0
j 0
Figure 24. The figure-eight knot can also regarded as the closure
of a (1, 1)-tangle.
From Figure 14, we have

JN (E; q)

= i,0
R0,i (R−1 )i,j 0,i+j
i+j,0 Ri,j (R−1 )j,0
0,j (μ
−1 i j
) i μj
0≤i≤N −1,0≤j≤N −1
0≤i+j≤N −1
{i + j}!{N − 1}!
q −(N −1)i/2+(N −1)j/2−i /4+j /4−3i/4+3j/4 .
2 2
= (−1)i
0≤i≤N −1,0≤j≤N −1
{i}!{j}!{N − 1 − i − j}!
0≤i+j≤N −1
Putting k := i + j, we have
−1

N
{N − 1}! k2 /4+N k/2+k/4
k
{k}!
JN (E; q) = q (−1)i
q −N i−ik/2−i/2 .
{N − 1 − k}! i=0
{i}!{k − i}!
k=0
Using the formula (see [28, Lemma 3.2])

k
{k}!
k
(−1)i q li/2 = (1 − q (l+k+1)/2−g ),
i=0
{i}!{k − i}! g=1
we have the following simple formula with only one summand, which is due to
Habiro and Lê (I learned this method from Lê).
N −1
1 {N + k}!
(2.5) JN (E; q) = .
{N } {N − 1 − k}!
k=0
3. Volume conjecture
In this section we state the Volume Conjecture and then prove it for the figure-
eight knot. We also give supporting evidence for the conjecture.
3.1. Statement of the Volume Conjecture. In [15] Kashaev introduced

a link invariant LN ∈ C for an integer N greater than one and a link L by
using the quantum dilogarithm. Then he observed in [16] that for some examples
of hyperbolic knots the limit 2π limN →∞ log |KN |/N seems to be equal to the
hyperbolic volume of the knot complement. Here a knot in the three-sphere S 3 is
called hyperbolic if its complement possesses a complete hyperbolic structure with
finite volume. He also conjectured this would also hold for any hyperbolic knot.
In [28] J. Murakami and I proved that Kashaev’s invariant equals the N -
dimensional colored Jones √ polynomial evaluated at the N -th root of unity, that
is, LN = JN (L; exp(2π −1/N )) and proposed that Kashaev’s conjecture would
hold for any knot by using the simplicial volume.
Conjecture 3.1 (Volume Conjecture [16, 28]). The following equality would
hold for any knot K.
√
log |JN (K; exp(2π −1/N ))|
2π lim = Vol(S 3 \ K).
N →∞ N
To define the simplicial volume (or Gromov norm), we introduce the Jaco–
Shalen–Johannson (JSJ) decomposition (or the torus decomposition) of a knot
complement.
Definition 3.2 (Jaco–Shalen–Johannson decomposition [11, 13]). Let K be
a knot. Then its complement S 3 \ K can be uniquely decomposed into hyperbolic
pieces and Seifert fibered pieces by a system of tori:

S3 \ K = Hi Ej
with Hi hyperbolic and Ej Seifert-fibered.
Then the simplicial volume of the knot complement is defined to be the sum of
the hyperbolic volumes of the hyperbolic pieces.
Definition 3.3 (Simplicial volume (Gromov norm) [7]). If a knot complement
S 3 \ K is decomposed as above, then its simplicial volume Vol(S 3 \ K) is defined as

Vol(S 3 \ K) := Hyperbolic Volume of Hi .
Hi :hyperbolic piece
Example 3.4. Let us consider the (2, 1)-cable of the figure-eight knot as shown
in figure 25. Then its complement can be decomposed by a torus into two pieces
Figure 25. (2, 1)-cable of the figure-eight knot
(Figure 26), one hyperbolic and one Seifert fibered. Therefore we have
18 HITOSHI MURAKAMI
hyperbolic Seifert fibered
Figure 26. The JSJ decomposition of the (2, 1)-cable of the

figure-eight knot
⎛ ⎞ ⎛ ⎞
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
Vol ⎜ ⎟ = Vol ⎜ ⎟.
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎝ ⎠ ⎝ ⎠
3.2. Proof of the Volume Conjecture for the figure-eight knot. We

give a proof of the Volume Conjecture for the figure-eight knot due to T. Ekholm.
3.2.1. Calculation of the limit. We use the formula (2.5) of the colored Jones
polynomial for the figure-eight knot E due to Habiro and Lê. (See [8, 21] for
Habiro’s method.) By a simple calculation using it, we have

N −1 j

(3.1) JN (E; q) = q (N −k)/2 − q −(N −k)/2 q (N +k)/2 − q −(N +k)/2 .
j=0 k=1
√
Replacing q with exp(2π −1/N ), we have
−1
√ N j
JN E; exp(2π −1/N ) = f (N ; k)
j=0 k=1
where we put f (N ; k) := 4 sin2 (kπ/N ). The graph of f (N ; k) is depicted in Fig-

ure 27. j √ N −1
Put g(N ; j) := k=1 f (N ; k) so that JN E; exp(2π −1/N ) = j=0 g(N ; j).
Then g(N ; j) decreases when 0 < j < N/6 and 5N/6 < j, and increases when
N/6 < j < 5N/6. Therefore g(N ; j) takes its maximum at j = 5N/6. (To be
precise we need to take the integer part of 5N/6.) See Table 1. Since there are N
√ N −1
positive terms in JN E; exp(2π −1/N ) = j=0 g(N ; j) and g(N ; 5N/6) is the
maximum of these terms, we have
√
g(N ; 5N/6) ≤ JN E; exp(2π −1/N ) ≤ N × g(N ; 5N/6).
Figure 27. Graph of f (N ; k)
j 0 · · · N/6 ··· 5N/6 ··· 1

f (N ; k) <1 1 >1 1 <1
g(N ; j) 1 maximum
Table 1. table of f (N ; k) and g(N ; j)
Noting that each side is positive, we take their logarithms and divide them by N .
√
log g(N ; 5N/6) log JN E; exp(2π −1/N ) log N log g(N ; 5N/6)
≤ ≤ + .
N N N N
Since limN →∞ (log N )/N = 0, we have
√
log g(N ; 5N/6) log JN E; exp(2π −1/N ) log g(N ; 5N/6)
lim ≤ lim ≤ lim .
N →∞ N N →∞ N N →∞ N
Therefore we have
√
log JN E; exp(2π −1/N ) log g(N ; 5N/6)
lim = lim .
N →∞ N N →∞ N

We can calculate the limit limN →∞ log g(N ; 5N/6) /N by integration. Since
j
g(N ; j) = k=1 f (N ; k), we have
1
5N/6
log g(N ; 5N/6)
lim = lim log f (N ; k)
N →∞ N N →∞ N
k=1
1
5N/6
(3.2)
= 2 lim log 2 sin(kπ/N )
N →∞ N
k=1

2 5π/6
= log 2 sin x dx.
π 0
What does this mean? I will explain a geometric interpretation of this integral.
20 HITOSHI MURAKAMI
3.2.2. Lobachevsky function. The function defined by the integral in (3.2) is

known as the Lobachevsky function. More precisely, we define the Lobachevsky
function Λ(θ) as
θ
Λ(θ) := − log |2 sin x| dx
0
for θ ∈ R. By using this function, we can express the limit of the colored Jones
polynomial as
√
log JN E; exp(2π −1/N ) 2
(3.3) lim = − Λ(5π/6)
N →∞ N π
We show some properties of the Lobachevsky function (see for example [23]).
Lemma 3.5. The Lobachevsky function satisfies the following two properties.
(1) The Lobachevsky function is an odd function and has period π.
(2) We have Λ(2θ) = 2Λ(θ) + 2Λ(θ + π/2). In general we have Λ(nθ) =

n n−1
k=1 Λ(θ + kπ/n).
Proof. The first property is easily shown by the periodicity of the sine func-
tion.
To prove the second, we use the double angle formula of the sine function:
sin(2x) = 2 sin x cos x. We have
log |2 sin(2x)| = log |2 sin x| + log |2 cos x| = log |2 sin x| + log |2 sin(x + π/2)|,
completing the proof.
From (1) we have
Λ(5π/6) = Λ(−π/6) = −Λ(π/6).
From (2) and (1) we also have
Λ(π/3) = 2Λ(π/6) + 2Λ(2π/3) = 2Λ(π/6) − 2Λ(π/3).
Therefore we have
3
Λ(5π/6) = − Λ(π/3).
2
Returning to the limit of the Jones polynomial (3.3), we conclude that
√
log JN E; exp(2π −1/N )
2π lim = 6Λ(π/3).
N →∞ N
Next we show how the Lobachevsky function is related to hyperbolic geometry.
3.2.3. Hyperbolic geometry. It is well known that the complement of the figure-
eight knot can be decomposed into two ideal hyperbolic regular tetrahedra.
Theorem 3.6 (W. Thurston [34]). The complement of the figure-eight knot
can be obtained by gluing two ideal hyperbolic regular tetrahedra.
I will explain what is an ideal hyperbolic regular tetrahedron later. Topo-
logically, the theorem states that the complement of the figure-eight knot can be
obtained by gluing two truncated tetrahedra as in Figure 28 (see also [24]). In Fig-
ure 28 we identify A with A , B with B , C with C and D with D . Note that edges
with single arrows and edges with double arrows are also identified respectively.
Here I give a short introduction to hyperbolic geometry.
First consider
the upper half space {(x, y, z) ∈ R3 | z > 0} with hyperbolic
metric ds := dx + dy 2 + dz 2 /z and denote it by H3 . It is known that a geodesic
2
= ∪
A=A ,B=B ,
C=C ,D=D
Figure 28. The complement of the figure-eight knot is decom-

posed into two ideal hyperbolic regular tetrahedra. Shadowed tri-
angles make the boundary of the knot complement.
line in H3 is a semicircle or a straight line perpendicular to the xy-plane, and that

a geodesic plane is a hemisphere or a flat plane perpendicular to the xy-plane.
An ideal hyperbolic tetrahedron is a tetrahedron in H3 with geodesic faces
with four vertices at infinity, that is, on the xy-plain or at the point at infinity ∞.
By isometry we may assume that one vertex is ∞ and the other three are on the
xy-plane. So its faces consist of three perpendicular planes and a hemisphere as
shown in Figure 29. If we see the tetrahedron from the top, it is a (Euclidean)
Figure 29. An ideal hyperbolic tetrahedron is the part above the

hemisphere surrounded by three perpendicular planes.
triangle with angles α, β, and γ. It is known that an ideal hyperbolic tetrahedron
Figure 30. A top view (right) of the ideal hyperbolic tetrahedron (left)
is defined (up to isometry) by the similarity class of this triangle. Therefore we

22 HITOSHI MURAKAMI
can parametrize an ideal hyperbolic tetrahedron by a triple of positive numbers

(α, β, γ) with α + β + γ = π. We denote it by Δ(α, β, γ).
The hyperbolic volume Vol(Δ(α, β, γ)) can be expressed by using the Lobachevsky
function Λ(θ). In fact it can be shown that
Vol(Δ(α, β, γ)) = Λ(α) + Λ(β) + Λ(γ).
For a proof, see for example [34, Chapter 7].
Now we return to the decomposition of the figure-eight knot complement. Fig-
ure 28 shows that after identification we have two edges, edge with single arrow
and edge with double arrow, each of them is obtained by identifying six edges.
So if the ideal hyperbolic tetrahedra we are using are regular, that is, isometric
to Δ(π/3, π/3, π/3) then the sum of dihedral angles around each edge becomes
2π. This means that if we use two ideal hyperbolic regular tetrahedra, our glu-
ing is geometric, that is, the complement of the figure-eight knot is isometric
to the union of two copies of Δ(π/3, π/3, π/3). In particular its volume equals
2 Vol(Δ(π/3, π/3, π/3)) = 6Λ(π/3).
Thus we have proved
√
log JN E; exp(2π −1/N )
2π lim = 6Λ(π/3) = Vol S 3 \ E ,
N →∞ N
which is the statement of the Volume Conjecture for the figure-eight knot.
3.3. Knots and links for which the Volume Conjecture is proved. As
far as I know the Volume Conjecture is proved for
(1) figure-eight knot by Ekholm,
(2) 52 knot by Kashaev and Yokota,
(3) Whitehead doubles of torus knots by Zheng [43],
(4) torus knots by Kashaev and Tirkkonen [17],
(5) torus links of type (2, 2m) by Hikami [9],
(6) Borromean rings by Garoufalidis and Lê [6],
(7) twisted Whitehead links by Zheng [43],
(8) Whitehead chains by van der Veen [37],
(9) a satellite link around the figure-eight knot with pattern the Whitehead
link by Yamazaki and Yokota [39],
(10) any link is contained in a link L such that the Volume Conjecture is true
for L by van der Veen [38].
Note that (1) and (2) are for hyperbolic knots, (3) is for a knot whose JSJ decom-
position consists of a hyperbolic piece and a Seifert fibered piece, (4) and (5) are
for knots and links only with Seifert pieces, (6)–(8) are for hyperbolic links, and (9)
and (10) are for links whose JSJ decompositions consist of hyperbolic pieces and
Seifert fibered pieces.
4. Supporting evidence for the Volume Conjecture

The Volume Conjecture is proved only for several knots and links but I think
it is true possibly with some modification; for example we may need to replace the
limit with the limit superior (see [37, Conjecture 2]). In this section I will explain
why I think it is true.
Remark 4.1 (Caution!). Descriptions in this section are not rigorous.
4.1. Approximation of the colored Jones polynomial. Put ξN := exp

√
(2π −1/N ) and I will give an interpretation of the R-matrix used to define the
colored Jones polynomial. From (2.2) and (2.4) we have

ij
{l}!{N − 1 − k}!
Rkl = (a power of ξN ) × δl,i+m δk,j−m ,
q:=ξN
m
{i}!{m}!{N − 1 − j}!
(4.1)
±{k}!{N − 1 − l}!
(R−1 )ij (a power of ξN ) × δl,i−m δk,j+m
kl q:=ξq = .
m
{j}!{m}!{N − 1 − i}!
Since
√ k
{k}!q:=ξ = 2 −1 sin(π/N ) sin(2π/N ) · · · sin(kπ/N ),
N
we have

{k}!{N − k − 1}!q:=ξN
√ N −1
= 2 −1 sin(π/N ) sin(2π/N ) · · · sin(kπ/N )

× sin (N − (k + 1))π/N sin (N − (k + 2))π/N sin (N − (N − 1))π/N
√ N −1 N −1
=(−1)N −k−1 −1 2 sin(π/N ) sin(2π/N ) · · · sin((N − 1)π/N )
=εN
√ √
for ε ∈ {1, −1, −1, − −1}. For the last equality, see for example [28]. So if we
put
(ξN )k+ := (1 − ξN ) · · · (1 − ξN
k
),
N −1−k
(ξN )k− := (1 − ξN ) · · · (1 − ξN ),
we have

{k}!q:=ξ = (a power of ξN ) × (ξN )k+ ,
N
{N − 1 − k}!q:=ξ = (a power of ξN ) × (ξN )k−

N
and
(ξN )k+ (ξN )k− = (a power of ξN ) × N.

Therefore from (4.1) the R-matrix and its inverse can be written as

ij
(a power of ξN ) × N 2
Rkl q:=ξN
= δ l,i+m δ k,j−m ,
m
(ξN )m+ (ξN )i+ (ξN )k+ (ξN )j − (ξN )l−
(a power of ξN ) × N 2
(R−1 )ij
kl q:=ξN = δl,i−m δk,j+m .
m
(ξN )m+ (ξN )i− (ξN )k− (ξN )j + (ξN )l+
If we are given a knot K, we express it as a closed braid and calculate the
colored Jones polynomial as described in §2.5. Then we have
⎛ ⎞

JN (K; ξN ) = ⎝ (R±1 )ij ⎠ kl
labelings crossings
(4.2) ⎛ ⎞
(a power of ξ ) × N 2
= ⎝ N ⎠,
crossings
(ξ N ) m + (ξN )i± (ξN )k ± (ξN )j ∓ (ξN )l∓
labelings
24 HITOSHI MURAKAMI
where the summation is over all the labelings with {0, 1, . . . , N − 1} corresponding
to the basis {e0 , e1 , . . . , eN −1 } and for a fixed labeling the product is over all the
crossings, each of which corresponds to an entry Rkl ij
(or (R−1 )i,j
kl , respectively),
determined by the four labeled arcs around the vertex, of the R-matrix (or its
inverse, respectively).
We will approximate (ξN )k+ for large N . By taking the logarithm, we have

k
j
log(ξN )k+ = log(1 − ξN )
j=1

k
√
= log 1 − exp(2π −1j/N ) .
j=1
Putting x := j/N , we may replace the summation with the following integral for
large N (this is not rigorous!).
k/N √
log(ξN )k+ ≈ N log 1 − exp(2π −1x) dx
N →∞ 0
exp(2π√−1k/N )
N log(1 − y)
= √ dy,
2π −1 1 y
√
where we put y := exp(2π −1x) in the last equality and ≈ means a very rough
N →∞
approximation (which may be not true at all) for large N .
This integral is known as the dilogarithm function. We put
z
log(1 − y)
Li2 (z) := − dy
0 y
for z ∈ C \ [1, ∞). This is called dilogarithm since it has the Taylor expansion for
|z| < 1 as
∞
zn
.
n=1
n2
For more details about this function, see for example [42].
So by using the dilogarithm function, we have the following approximation:
N
log(ξN )k+ ≈ √ Li2 (1) − Li2 (ξN
k
) .
N →∞ 2π −1
Since Li2 (1), which equals ζ(2) = π 2 /6 (ζ(z) is the Riemann zeta function), can be
ignored for large N , we have

N
(ξN )k+ ≈ exp − √ k
Li2 (ξN ) .
N →∞ 2π −1
Similarly we have

N −k
(ξN )k− ≈ exp − √ Li2 (ξN ) .
N →∞ 2π −1
Therefore from (4.2) the colored Jones polynomial can be (roughly) approxi-
mated as follows.
⎡
N
(4.3) JN (K; ξN ) ≈ exp ⎣ √
N →∞
labelings
2π −1
⎛ ⎞⎤

×⎝ m
Li2 (ξN ) + Li2 (ξN±i ∓j
) + Li2 (ξN ±k
) + Li2 (ξN ∓l
) + Li2 (ξN ) + log terms ⎠⎦ ,
crossings
where the log terms come from powers of ξN and N .

Since the term
∓j

m ±i ±k ∓l
(4.4) Li2 (ξN ) + Li2 (ξN ) + Li2 (ξN ) + Li2 (ξN ) + Li2 (ξN ) + log terms
crossings
i1 i2 ic
can be regarded as a function of ξN , ξN , . . . , ξN with i1 , i2 , . . . , ic labelings of arcs,
i1 i1 ic
we can write it as V (ξN , ξN . . . , ξN ). Note that m can be expressed as a difference
of two such labelings. So we have

N
(4.5) JN (K; ξN ) ≈ exp √ V (ξN , ξN , . . . , ξN ) .
i1 i2 ic
N →∞
i ,i ,...,i
2π −1
1 2 c
We want to apply a method used in the proof for the figure-eight knot in §3.2.
Recall that the point of the proof is to find the maximum term of the summand
because the maximum dominates the asymptotic behavior. We will seek for the
“maximum” term of the summation in (4.5).
To do that we first approximate this with the following integral:

N
(4.6) JN (K; ξN ) ≈ ··· exp √ V (z1 , z2 . . . , zc ) dz1 dz2 · · · dzc ,
N →∞ J
1 J2 Jc 2π −1
id
where zd corresponds to ξN and J1 , J2 , . . . , Jc are certain contours. (The argument
here is not rigorous. In particular I do not know how to choose the contours.)
Then we will find the maximum of the absolute value of the integrand. To be
more precise we apply the steepest descent method. For a precise statement, see
for example [20, Theorem 7.2.9]).
Theorem 4.2 (steepest descent method). Under certain conditions for the
functions g, h, and a contour C, we have
√
exp(N h(z0 )) 2πg(z0 )
g(z) exp(N h(z)) dz ∼ √
C N →∞ N −h (z0 )
≈ exp(N h(z0 )),
N →∞

where h (z0 ) = 0 and Re(h(z)) takes its positive maximum at z0 .
Note that the symbol ∼ means that the ratio of both sides converges to 1 when
N →∞ √
N → ∞ and that we ignore the constant term and N in the rough approximation
≈ because exp(N h(z0 )) grows exponentially when N → ∞. (Recall that we want
N →∞
to know the limit of log |JN (K, ξN )|/N and so polynomial terms will not matter.)
Remark 4.3. In general, to apply the steepest descent method, we need to
change the contour C so that it passes through z0 .
26 HITOSHI MURAKAMI
Now we apply (a multidimensional version of) this method to (4.6) and we will
find the the maximum of {Im V (z1 , . . . , zc ) | (z1 , z2 , . . . , zc ) ∈ J1 × J2 · · · × Jc }. Let
(x1 , x2 , . . . , xc ) be such a point. Then we have

N
(4.7) JN (K; ξN ) ≈ exp √ V (x1 , x2 , . . . , xc )
N →∞ 2π −1
and so we finally have
√ log JN (K; ξN )
(4.8) 2π −1 lim = V (x1 , x2 , . . . , xc ).
N →∞ N
Note that the point (x1 , x2 , . . . , xc ) is a solution to the following equation:
∂V
(4.9) (z1 , z2 , . . . , zc ) = 0
∂ zd
for d = 1, 2, . . . , c.
Remember that our argument here is far from rigor! Especially I am cheating
on the following points:
• Replacing a summation with an integral in (4.6). Here we do not know
how to choose the multidimensional contour.
• Applying the steepest descent method in (4.7). In general, we have many
solutions to the system of equations (4.9) but we do not know which
solution gives the maximum. Moreover we may need to change the contour
so that it passes through the solution that gives the maximum but we do
not know whether this is possible or not.
4.2. Geometric interpretation of the limit. In this subsection I will give

a geometric interpretation of the limit (4.8).
Since V (z1 , z2 , . . . , zc ) is the sum of the terms as in (4.4), we first describe a
±i
geometric meaning of Li2 (ζN ).
Recall that an ideal hyperbolic tetrahedron can be put in H3 . Regarding the
xy-plane as the complex plain, we can assume that the three of the four (ideal)
vertices are at 0, 1 and z ∈ C (Im z > 0), respectively (Figure 31). Thus the set
z
⇒
0 1
Figure 31. Parametrization of ideal hyperbolic tetrahedra
{z ∈ C | Im z > 0} gives a parametrization of ideal hyperbolic tetrahedra. We

denote by Δ(z) the hyperbolic tetrahedron parametrized by z.
The volume of Δ(z) is given as follows (see for example [31, p. 324]):
(4.10) Vol(Δ(z)) = Im Li2 (z) + log |z| arg(1 − z).
So we expect V (x1 , x2 , . . . , xc ) gives the sum of the volumes of certain tetrahe-
dra related to the knot.
In fact we can express the volume of the knot complement in terms of
V (z1 , z2 , . . . , zc ) [33, 40].
We follow [33] to describe this.
We decompose the knot complement into topological, truncated tetrahedra. To
do this we put an octahedron at each positive crossing as in Figure 32, where i,
j, k and l are labeling of the four arcs around the vertex. Then decompose the
i j
k l
Figure 32. An octahedron put at a crossing
octahedron into five tetrahedra as in Figure 33, where the four of them are decorated
i −j k −l
with ξN , ξN , ξN and ξN , respectively and the one in the center is decorated with
m
ξN , where m := l − i = j − k. Here each truncated tetrahedron is just a topological
one with some decoration.
Figure 33. decomposition of the octahedron into five tetrahedra

28 HITOSHI MURAKAMI
Now only two of the vertices are attached to the knot. We pull the two of the
remaining four vertices to the top (+∞) and the other two to the bottom (−∞)
as shown in Figure 34. We attach five tetrahedra to every crossing (if the crossing
Figure 34. Pull the vertices to +∞ and −∞.
is negative, we change them appropriately) in this way. At each arc two faces
meet, and we paste them together. Thus we have a decomposition of S 3 \ (K ∪
{+∞, −∞}). By deforming this decomposition a little we get a decomposition of
±ik
S 3 \K by (topological) truncated tetrahedra, decorated with complex numbers ξN
(k=1,2,. . . ,c).
Next we want to regard each tetrahedron as an ideal hyperbolic one.
Recall that when we approximate the summation in (4.5) by the integral in (4.6)
ik
we replace ξN with a complex variable zk . Following this we replace the decoration
ik
ξN for a tetrahedron with a complex number zk . Then regard the tetrahedron
decorated with zk as an ideal hyperbolic tetrahedron parametrized by zk .
So far this is just formal parametrizations. We need to choose appropriate
values for parameters so that the tetrahedra fit together to provide a complete
hyperbolic structure to S 3 \ K. To do this we choose z1 , z2 , . . . , zc so that:
• Around each edge several tetrahedra meet. To make the knot complement
hyperbolic, the sum of these dihedral angles should be 2π,
• Even if the knot complement is hyperbolic, the structure may not be com-
plete. To make it complete, the parameters should be chosen as follows.
Since we truncate the vertices of the tetrahedra, four small triangles
appear at the places where the vertices were (see Figure 31 for the triangle
associated with the vertex at infinity). After pasting these triangles make
a torus which can be regarded as the boundary of the regular neighborhood
of the knot K. Each triangle has a similarity structure provided by the
parameter zk . We need to make this boundary torus Euclidean.
See [34, Chapter 4],[31, § 2] for more details.
Surprisingly these conditions are the same as the system of equations (4.9) that
we used in the steepest descent method! Therefore we can expect that a solution
(x1 , x2 , . . . , xc ) to (4.9) gives the complete hyperbolic
√ structure.

Then, what does V (x1 , x2 , . . . , xc ) = 2π −1 limN →∞ log JN (K, ξN ) /N mean?
Recall the formula (4.10) and that V (x1 , x2 , . . . , xc ) is a sum of dilogarithm
functions and logarithm functions. Using these facts we can prove
Im V (x1 , x2 , . . . , xc ) = Vol(S 3 \ K),
that is,
√ log JN (K, ξN )
Im 2π −1 lim = Vol(S 3 \ K).
N →∞ N
So we have proved
log |JN (K, ξN )|
2π lim = Vol(S 3 \ K),
N →∞ N
which is the Volume Conjecture (Conjecture 3.1).
5. Generalizations of the Volume Conjecture

In this section we consider generalizations of the Volume Conjecture.
5.1. Complexification. In [35] W. Thurston pointed out that the Chern–
Simons invariant [3] can be regarded as an imaginary part of the volume. Neumann
and Zagier gave a precise conjecture [31, Conjecture, p. 309] which was proved to be
true by Yoshida [41]. For combinatorial approaches to the Chern–Simons invariant,
see [32] and [44].
So it would be natural to drop the absolute value sign of the left hand side of
the Volume Conjecture and add the Chern–Simons invariant to the right hand side.
Conjecture 5.1 (Complexification of the Volume Conjecture [29]). If a knot
K is hyperbolic, that is, its complement possesses a complete hyperbolic structure,
then
√
log JN (K; exp(2π −1/N )) √
2π lim ≡ Vol(S 3 \ K) + −1 CS(S 3 \ K)
N →∞ N
√
(mod π 2 −1Z),
where CS is the Chern–Simons invariant defined for a three-manifold with torus
boundary by Meyerhoff [22].
Remark 5.2. We may regard the left hand side as a definition of the Chern–
Simons invariant for non-hyperbolic knots provided that the limit of Conjecture 5.1
exists.
5.2. Deformation of the parameter. In the Volume Conjecture (Conjec-
ture 3.1) and its complexification (Conjecture 5.1), the (possible) limit corresponds
to the complete hyperbolic structure of S 3 \ K for a hyperbolic knot K. As de-
scribed in [34, Chapter 4] the complete structure can be deformed to incomplete
ones.
How can we perform√this deformation in the colored Jones polynomial? If we
deform the parameter 2π −1 in the Volume Conjecture, is the corresponding limits
related to incomplete hyperbolic structures?
30 HITOSHI MURAKAMI
Let us consider the limit

√
log JN K; exp (u + 2π −1)/N
lim .
N →∞ N
Note that when u = 0, this limit is considered in the (complexified) Volume Con-
jecture.
5.2.1. Figure-eight knot. Before stating a conjecture for general knots, I will
explain what happens in the case of the figure-eight knot.
Theorem 5.3 ([30]). Let E be the figure-eight
√ knot. There exists a neighbor-
hood O ⊂ C of 0 such that if u ∈ (O \ π −1Q) ∪ {0}, then the following limit
exists: √
log JN (E; exp (u + 2π −1)/N )
lim .
N →∞ N
Moreover if we put
√
H(u) := (u + 2π −1) × (the limit above),
and
d H(u) √
v := 2 − 2π −1
du
then we have
√
Vol(Eu ) + −1 CS(Eu )
√ √ √
≡ − −1H(u) − πu + u v −1/4 − πκ(γu )/2 (mod π 2 −1Z).
Here Eu is the closed hyperbolic
three-manifold associated with the following
representation of π1 S 3 \ E → SL(2; C):

exp(u/2) ∗
μ → ,
0 exp(−u/2)
(5.1)

exp(v/2) ∗
λ → .
0 exp(−v/2)
Here μ is the meridian of E (a loop in S 3 \ E that goes around the knot, which
generates H1 (S 3 \ E) ∼ = Z), λ is the longitude (a loop in S 3 \ E that goes along the
knot such that it is homologous to 0 in H1 (S 3 \ E)), and γu is the loop attached to
S 3 \E when we √ complete the hyperbolic structure defined by u. We also put κ(γu ) :=
length(γu ) + −1 torsion(γu ), where length(γu ) is the length of the attached loop
γu , and torsion is its torsion, which is defined modulo 2π as the rotation angle when
one travels along γu . See [31] for details (see also [24]).
We will give a sketch of the proof in the following two subsections.
5.2.2. Calculation of the limit. First we calculate the limit. Note that here I
give just a sketch of the calculation but it can be done rigorously. For details see
[30].
From (3.1) we have
−1
j
N

JN (E; q) = q jN 1 − q −N −k 1 − q −N +k .
j=0 k=1
√ √
Put q := exp(θ/N
√ ) for θ near 2π −1. If θ is not a rational multiple of π −1 (but
it can be 2π −1), we have
j

−N ±k
log 1−q
k=1

j

= log 1 − exp(±kθ/N − θ)
k=1
j/N
≈ N log(1 − exp(±θs − θ)) ds
N →∞ 0

±N exp(±jθ/N −θ) log(1 − t)
= dt
θ exp(−θ) t
±N
= Li2 exp(−θ) − Li2 exp(±jθ/N − θ) .
θ
So we have

JN E; exp(θ/N )
−1
N
N
≈ exp(jθ) exp Li2 exp(−jθ/N − θ) − Li2 exp(jθ/N − θ)
N →∞ θ
j=0
−1
N
N
= exp H exp(jθ/N ), exp(θ)
j=0
θ

N
≈ exp H x, exp(θ) dx
N →∞ C θ
for a suitable contour C. Here we put
(5.2) H(ζ, η) := Li2 (1/(ζη)) − Li2 (ζ/η) + log ζ log η.
To apply the steepest

descent method (Theorem 4.2), we find the maximum of
Re H x, exp(θ)
/θ over x. To do that we will find a solution y to the equation
d H x, exp(θ) /d x = 0, which is

log exp(θ) + exp(−θ) − x − x−1
= 0.
x
We can show that appropriately chosen y gives the maximum and from the steepest
descent method we have

N
JN E; exp(θ/N ) ≈ exp H y, exp(θ) ,
N →∞ θ
that is,

JN E; exp(θ/N )
(5.3) θ lim = H y, exp(θ) ,
N →∞ N
where y satisfies
y + y −1 = exp(θ) + exp(−θ) − 1.
32 HITOSHI MURAKAMI
5.2.3. Calculation of the volume. Next we will relate the limit to the volume
of a three-manifold obtained by S 3 \ E.
As described in § 3.2.3, S 3 \ E is obtained by gluing two ideal hyperbolic
tetrahedra as in Figure 28. Here we assume √ that they are parametrized by complex
numbers z and w. When z = w = (1 + −3)/2, S 3 \ E has a complete hyperbolic
structure as described in § 3.2.3. We assume that the left tetrahedron (with faces
labeled with A, B, C and D) and the right tetrahedron (with faces labeled with
A , B , C and D ) in Figure 28 are Δ(z) and Δ(w) respectively.
The boundary torus, which is obtained from the shadowed triangles in Fig-
ure 28, looks like Figure 35. Here the leftmost triangle is the one in the center
of the picture of Δ(z) and the second leftmost one is the one in the center of the
picture of Δ(w). Let α, β and γ be the dihedral angle between B and C, A and B,
and C and A respectively. Let α , β and γ be the dihedral angle between B and
D , A and B , and D and A respectively. As described in § 3.2.3, the sum of the
Figure 35. Identifying the sides as indicated by the circles, we

get a triangulation of the boundary torus. Here the single circles
denote the arrow head of the single arrow, the double circles denote
the arrow head of the double arrow, the circles with − denote the
arrow tail of the single arrow, and the circles with + denote the
arrow tail of the double arrow in Figure 28. Note that we view this
torus from outside of S 3 \ E.
dihedral angles around each edge should be 2π. So from Figure 35, we have
β + 2γ + β + 2γ = 2π from the single circles,

2α + β + 2α + β = 2π from the double circles,

β + 2γ + β + 2γ = 2π from the circles with −,

2α + β + 2α + β = 2π from the circles with +,
which is equivalent to a single equation
2α + β + 2α + β = 2π
since α + β + γ = α + β + γ = π.
Assume that α = arg z and α = arg w. Since β = arg(1 − 1/z) and β =
arg(1 − 1/w) (see Figures 30 and 31) this turns out to be
2 arg z + arg(1 − 1/z) + 2 arg w + arg(1 − 1/w) = 2π.
So we have
(5.4) zw(z − 1)(w − 1) = 1.
Remark 5.4. This is just a condition that S 3 \ E is hyperbolic. To make the

metric complete we need to add the condition that the upper side and the lower
side of the parallelogram in Figure 35 are parallel.
Now we introduce parameters x and y as
x := w(1 − z),
(5.5)
y := −zw.
Note that the following equality holds from (5.4):
(5.6) y + y −1 = x + x−1 − 1.
Since Δ(z) and Δ(w) can also be parametrized as Δ(1 − 1/z) and Δ(1 − 1/w),
we have
Vol(Δ(z)) + Vol(Δ(w))
= Im Li2 (w ) + Im Li2 (z ) + log |w | arg(1 − w ) + log |z | arg(1 − z ),
from (4.10), where we put z := 1 − 1/z and w := 1 − 1/w. Using the equation
−1 π2 1 2
Li2 (z ) + Li2 (z ) = − − log(−z )
6 2
(see for example [42, § 1.2]), we have
(5.7)
Vol(Δ(z)) + Vol(Δ(w))
−1 −1 −1
= Im Li2 (w ) − Im Li2 (z ) + log |w | arg(1 − w ) − log |z | arg(1 − z )
−1
(since w = 1/(yx) and z = y/x)
= Im Li2 (1/(yx)) − Im Li2 (y/x) + log |1/(yx)| arg(1 − 1/(yx)) − log |y/x| arg(1 − y/x)
(from (5.6))
1 − y/x
= Im Li2 (1/(yx)) − Im Li2 (y/x) + log |y| arg x + log |x| arg
1 − 1/(yx)
(from (5.5) and (5.4))
y
= Im Li2 (1/(yx)) − Im Li2 (y/x) + log |y| arg x + log |x| arg
z(z − 1)
(from (5.2))
= Im H(y, x) − log |x| arg(z(z − 1)).
Putting
u := log x = log(w(1 − z)),
v := 2 log(z(1 − z)),
H(u) := H(y, x),
we have
1
(5.8) Vol(Δ(z)) + Vol(Δ(w)) = Im H(u) − π Re u − Re u Im v.
2
Moreover from (5.3), we have
√
√ log JN E; exp((u + 2π −1)/N )
(u + 2π −1) lim = H(u)
N →∞ N
34 HITOSHI MURAKAMI
√
if we put θ = u + 2π −1 since x = exp(u) = exp(θ). Note that z, w, x, y and v
are functions of u. Note also that v is given as
d H(u) √
v=2 − 2π −1
du
d H(u)
since = log(z(z − 1)).
du
Remark 5.5. We need to be more careful about the arguments of variables.
For details see [30].
I will give geometrical interpretation of u and v to relate the term Re u Im v in
(5.8) to the length of γu .
We first calculate H1 (S 3 \ E) = H1 (Δ(z) ∪ Δ(w)). Since the interiors of three-
simplices do not matter to the first homology, we can calculate it from the boundary
torus, the edges of Δ(z) and Δ(w), and the faces A = A , B = B , C = C , and
D = D (see Figures 28 and 35). From Figures 28, 35 and 36 one reads
∂A = ∂A = − e7 + + e10 − − e5 ,
→

∂B = ∂B = − − e11 − + e6 + − e9 ,
→

∂C = ∂C = + e4 + + e8 − + e1 ,
→
→
∂D = ∂D =
+ e3 − → →
+ e2 − + e12 ,
where and mean the single arrowed edge and the double arrowed edge in
→
Figure 28 respectively, and the ei are the edges of the boundary torus as indicated
in Figure 36. Since
e7 = e6 + e2 ,
e10 = e6 ,
e5 = e1 + e6 ,
e11 = e4 + e12 = e4 + e6 ,
e9 = e3 + e10 = e3 + e6 ,
e8 = e6 ,
e12 = e6
in the first homology group, we have
− e1 − e2 − e6 = 0,
→ →
− − e3 − e4 − e6 = 0,
+ e1 + e4 + e6 = 0,
− + e2 + e3 + e6 = 0.
So if we put μ := e6 , λ := e1 + e2 + e3 + e4 + 2μ, then we see that the first homology
group of the boundary torus is generated by μ and λ, that H1 (S 3 \ E) ∼ = Z is
generated by μ, and that λ = 0 in H1 (S 3 \ E). Therefore μ is the meridian and λ
is the longitude.
Now let us consider the universal cover S 3 \ E of S 3 \ E = Δ(z) ∪ Δ(w) which
is H . We can construct it by developing Δ(z) and Δ(w) isometrically in H3 .

3
Then each loop in S 3 \ E is regarded as a covering translation of S 3 \ E and it
defines an isometric translation of H . This defines a representation (holonomy

3
e1 e2 e3 e4
e5 e6 e7 e8 e9 e10 e11 e12 e5
e2 e3 e4 e1
Figure 36. The cycle e6 is the meridian and the cycle e1 + e6 +
e2 + e8 + e3 + e4 is the longitude.

representation) of π1 S 3 \ E)
at P SL(2; C). Taking a lift to SL(2; C), we can
define a representation ρ : π1 S 3 \ E) → SL(2; C).
We consider how ρ(μ) and ρ(λ) act on ∂H3 = S 2 = C ∪ {∞}. The image of
the meridian ρ(μ) sends the top side to the bottom side. So it is the composition
of a −α-rotation around the circle with + in the top (between e1 and e2 ) and a γ -
rotation around the single circle in the bottom (between e2 and e3 ), which means
a multiplication by 1/z × 1/(1 − w)) = w(1 − z) plus a translation from (5.4).
Similarly ρ(λ) acts as a multiplication by z 2 (1 − z)2 plus a translation.
Therefore u = log(w(1 − z)) and v = 2 log(z(1 − z)) can be regarded as the
logarithms of the actions by the meridian μ and the longitude λ, respectively.
Since the meridian and the longitude commute in π1 (S 3 \ E), their images can
be simultaneously triangularizable. Recalling that μ and λ define multiplications
by exp(u) and exp(v) plus translations on ∂H3 , we may assume

exp(u/2) ∗
ρ(μ) = ,
0 exp(−u/2)

exp(v/2) ∗
ρ(λ) = ,
0 exp(−v/2)
which is (5.1). This is a geometric interpretation of u and v.
Since u determines z and w, it defines a hyperbolic structure of S 3 \ E as the
union of Δ(z) and Δ(w). When u = 0 this hyperbolic structure is incomplete. We
can complete this incomplete structure by attaching either a point or a circle.
Since v is not a real√multiple of u when u is small, there exists a pair (p, q) ∈ R2
such that pu + qv = 2π −1. The pair (p, q) is called the generalized Dehn surgery
coefficient [34]. If p and q are coprime integers, then the completion is given by
attaching a circle γu and the result is a closed hyperbolic three-manifold which we
denote by Eu . (For other cases the completion is given by adding either a point
or a circle. In the former case the regular neighborhood of the attached point is a
cone over a torus, and in the latter case the regular neighborhood of the attached
circle is topologically a solid torus but geometrically the angle around the core is
not 2π.)
If p and q are coprime integers, the completion is nothing but the (p, q)-Dehn
surgery along the knot, that is, we attach a solid torus D to S 3 \ Int(N (E)) so
that the meridian of D coincides with the loop on the boundary of the regular
neighborhood N (E) of E ⊂ S 3 presenting pμ + qλ ∈ H1 (S 3 \ Int(N (E))), where
36 HITOSHI MURAKAMI
Int denotes the interior (Figure 37). Then the circle γu can be regarded as the core
Figure 37. (2, 1)-Dehn surgery along the figure-eight knot
of D.
To complete the proof of Theorem 5.3, we want to describe the length of the
attached circle γu in terms of u and v. We will show
1
(5.9) length γu = − Im (uv) .
2π
When u is small and non-zero, we can assume that exp(u) = 1 and exp(v) = 1.
So we can also assume that ρ(μ) and ρ(λ) are both diagonal. This means that the
image of μ is a multiplication by exp(u) and that the image of λ is a multiplication
by exp(v) (with no translations). Note that now S 3 \ E is identified with H3 minus
the z-axis, and the completion is given by adding the z-axis.

Since p and q are coprime, we can choose integers r and s so that ps − qr = 1.
We push γu ∈ D to the boundary of the solid torus ∂D and denote the resulting
circle by γ̃u . Then we see that [γ̃u ] = rμ + sλ ∈ H1 (∂(S 3 \ Int(N (E))); Z) since the
meridian of D is identified with pμ + qλ, and the images of the meridian and γ̃u
make a basis of H1 (∂(S 3 \ Int(N (E))); Z).
Remark 5.6. Even if we use another pair (r , s ) such that ps − qr = 1, we

get the same manifold. This is because changing (r, s) corresponds to changing of
γ̃u ∈ ∂D. Observe that ambiguity of the choice of γ̃u is given by a twist of D and
that it does not matter to the resulting manifold.
Therefore ρ(γu ) corresponds to a multiplication by exp(±(ru+sv)). This means

that if we identify the completion of S3 \ E with H3 , a fundamental domain of the

lift of γu is identified with the segment [1, exp ± Re(ru + sv) ] in the z-axis. Since

the metric is given by dx2 + dy 2 + dz 2 /z, the length of γu is given by

exp ± Re(ru+sv)
dz
length(γu ) =
1 z
= ± Re(ru + sv)

ps − 1
=± Re(u) + s Re(v)
q

s 1
=± (p Re(u) + q Re(v)) − Re(u)
q q
Re(u)
=∓
q
∓1
= (Re u Im v − Im u Re v)
2π
±1
= Im(uv),
2π
where the fourth and the sixth equalities follow from

Re(u) Re(v) p 0
= .
Im(u) Im(v) q 2π
|v|2
Since v = u × and the orientation of (u, v) should be positive on C, we see that
uv
Im(uv) is negative (see [31] for details) and so (5.9) follows.
Therefore from (5.8) we finally have

√ 1 √ π
Vol(Eu ) = Re − −1H(u) − πu + uv −1 − κ(γu ) .
4 2
The Chern–Simons invariant is obtained by Yoshida’s formula [41]. See [30]
for details.
5.2.4. General knots. Here I propose a generalization of the Volume Conjecture
for general knots.
Conjecture 5.7 ([26]). For any knot K, there exists an open set U ∈ C such
that if u ∈ U , then the following limit exists:
√
log JN (K; exp (u + 2π −1)/N )
lim .
N →∞ N
Moreover if we put
√
H(K; u) := (u + 2π −1) × (the limit above),
and
d H(K; u) √
v := 2 − 2π −1,
du
then we have
1
Vol(K; u) = Im H(K; u) − π Re u − Re u Im v.
2
Here Vol(K; u) is the volume function corresponding to the representation of π1 (S 3 \
K) to SL(2; C) as in Theorem 5.3.
38 HITOSHI MURAKAMI
Remark 5.8. In the case of a hyperbolic knot, we can also propose a similar
conjecture with the imaginary part as in the case of the figure-eight knot. For a
general knot, a relation to the Chern–Simons invariant is also expected by using a
combinatorial description of the Chern–Simons invariant by Zickert [44].
Remark 5.9. Conjecture 5.7 is known to be true for the figure-eight knot
[30] and for torus knots [26]. See also [27] and [10] for a possible relation to the
Chern–Simons invariant.
Finally note that Garoufalidis and Lê proved the following result, which should
be compared with Conjecture 5.7. (See also [25] for the case of the figure-eight
knot.)
Theorem 5.10 (S. Garoufalidis and T. Lê [5]). For any K, there exists ε > 0
such that if |θ| < ε, we have
1
lim JN (K; exp(θ/N )) = ,
N →∞ Δ(K; exp θ)
where Δ(K; t) is the Alexander polynomial of K.
References
[1] J. W. Alexander, A lemma on systems of knotted curves., Proc. Nat. Acad. Sci. U.S.A. 9
(1923), no. 3, 93–95.
[2] J. S. Birman, Braids, links, and mapping class groups, Princeton University Press, Princeton,
N.J., 1974. MR0375281 (51 #11477)
[3] S.-S. Chern and J. Simons, Characteristic forms and geometric invariants, Ann. of Math.
(2) 99 (1974), 48–69. MR50#5811
[4] V. G. Drinfel d, Quantum groups, Proceedings of the International Congress of Mathemati-
cians, Vol. 1, 2 (Berkeley, Calif., 1986) (Providence, RI), Amer. Math. Soc., 1987, pp. 798–820.
[5] S. Garoufalidis and T. T. Q. Le, An analytic version of the Melvin-Morton-Rozansky Con-
jecture, arXiv:math.GT/0503641.
[6] , On the volume conjecture for small angles, arXiv:math.GT/0502163.
[7] M. Gromov, Volume and bounded cohomology, Inst. Hautes Études Sci. Publ. Math. (1982),
no. 56, 5–99 (1983). MR84h:53053
[8] K. Habiro, On the colored Jones polynomials of some simple links, Sūrikaisekikenkyūsho
Kōkyūroku (2000), no. 1172, 34–43. MR1805 727
[9] K. Hikami, Quantum invariant for torus link and modular forms, Comm. Math. Phys. 246
(2004), no. 2, 403–426. MR2048 564
[10] K. Hikami and H. Murakami, Representations and the colored jones polynomial of a torus
knot, arXiv:1001.2680, 2010.
[11] W. H. Jaco and P. B. Shalen, Seifert fibered spaces in 3-manifolds, Mem. Amer. Math. Soc.
21 (1979), no. 220, viii+192. MR81c:57010
[12] M. Jimbo, A q-difference analogue of U (g) and the Yang-Baxter equation, Lett. Math. Phys.
10 (1985), no. 1, 63–69. MR797001 (86k:17008)
[13] K. Johannson, Homotopy equivalences of 3-manifolds with boundaries, Lecture Notes in
Mathematics, vol. 761, Springer, Berlin, 1979. MR82c:57005
[14] V. F. R. Jones, A polynomial invariant for knots via von Neumann algebras, Bull. Amer.
Math. Soc. (N.S.) 12 (1985), no. 1, 103–111. MR86e:57006
[15] R. M. Kashaev, A link invariant from quantum dilogarithm, Modern Phys. Lett. A 10 (1995),
no. 19, 1409–1418. MR96j:81060
[16] , The hyperbolic volume of knots from the quantum dilogarithm, Lett. Math. Phys.
39 (1997), no. 3, 269–275. MR98b:57012
[17] R. M. Kashaev and O. Tirkkonen, A proof of the volume conjecture on torus knots, Zap.
Nauchn. Sem. S.-Peterburg. Otdel. Mat. Inst. Steklov. (POMI) 269 (2000), no. Vopr. Kvant.
Teor. Polya i Stat. Fiz. 16, 262–268, 370. MR1805 865
[18] R. Kirby and P. Melvin, The 3-manifold invariants of Witten and Reshetikhin-Turaev for
sl(2, C), Invent. Math. 105 (1991), no. 3, 473–545. MR92e:57011
[19] Markoff, A. (Markov, A. A.), Über die freie Äquivalenz der geschlossenen Zöpfe. (German),
Rec. Math. Moscou, n. Ser. 1 (1936), 73–78.
[20] J. E. Marsden and M. J. Hoffman, Basic complex analysis, W. H. Freeman and Company,
New York, 1987. MR88m:30001
[21] G. Masbaum, Skein-theoretical derivation of some formulas of Habiro, Algebr. Geom. Topol.
3 (2003), 537–556 (electronic). MR1997328 (2004f:57013)
[22] R. Meyerhoff, Density of the Chern–Simons invariant for hyperbolic 3-manifolds, Low-
dimensional topology and Kleinian groups (Coventry/Durham, 1984), Cambridge Univ. Press,
Cambridge, 1986, pp. 217–239. MR88k:57033a
[23] J. Milnor, Hyperbolic geometry: the first 150 years, Bull. Amer. Math. Soc. (N.S.) 6 (1982),
no. 1, 9–24. MR82m:57005
[24] H. Murakami, The colored Jones polynomials of the figure-eight knot and the volumes of
three-manifolds obtained by Dehn surgeries, Fund. Math. 184 (2004), 269–289.
[25] , The colored Jones polynomials and the Alexander polynomial of the figure-eight
knot, JP J. Geom. Topol. 7 (2007), no. 2, 249–269. MR2349300 (2008g:57014)
[26] , A version of the volume conjecture, Adv. Math. 211 (2007), no. 2, 678–683.
MR2323541
[27] , An introduction to the volume conjecture and its generalizations, Acta Math. Viet-
nam. 33 (2008), no. 3, 219–253. MR2501844
[28] H. Murakami and J. Murakami, The colored Jones polynomials and the simplicial volume of
a knot, Acta Math. 186 (2001), no. 1, 85–104. MR2002b:57005
[29] H. Murakami, J. Murakami, M. Okamoto, T. Takata, and Y. Yokota, Kashaev’s conjecture
and the Chern-Simons invariants of knots and links, Experiment. Math. 11 (2002), no. 3,
427–435. MR1959 752
[30] H. Murakami and Y. Yokota, The colored Jones polynomials of the figure-eight knot and its
Dehn surgery spaces, J. Reine Angew. Math. 607 (2007), 47–68. MR2338120
[31] W. D. Neumann and D. Zagier, Volumes of hyperbolic three-manifolds, Topology 24 (1985),
no. 3, 307–332. MR87j:57008
[32] W.D. Neumann, Combinatorics of triangulations and the Chern-Simons invariant for hyper-
bolic 3-manifolds, Topology ’90 (Columbus, OH, 1990), de Gruyter, Berlin, 1992, pp. 243–271.
MR93i:57020
[33] D. Thurston, Hyperbolic volume and the Jones polynomial, Lecture notes, École d’été
de Mathématiques ‘Invariants de nœuds et de variétés de dimension 3’, Institut Fourier
- UMR 5582 du CNRS et de l’UJF Grenoble (France) du 21 juin au 9 juillet 1999,
http://www.math.columbia.edu/ dpt/speaking/Grenoble.pdf.
[34] W. P. Thurston, The Geometry and Topology of Three-Manifolds, Electronic version 1.1 -
March 2002, http://www.msri.org/publications/books/gt3m/.
[35] W. P. Thurston, Three-dimensional manifolds, Kleinian groups and hyperbolic geometry,
Bull. Amer. Math. Soc. (N.S.) 6 (1982), no. 3, 357–381.
[36] V. G. Turaev, The Yang-Baxter equation and invariants of links, Invent. Math. 92 (1988),
527–553.
[37] R. van der Veen, Proof of the volume conjecture for Whitehead chains, Acta Math. Vietnam.
33 (2008), no. 3, 421–431. MR2501851
[38] , The volume conjecture for augmented knotted trivalent graphs, Algebr. Geom. Topol.
9 (2009), no. 2, 691–722. MR2496886
[39] M. Yamazaki and Y. Yokota, On the limit of the colored Jones polynomial of a non-simple
link, preprint, Tokyo Metropolitan University, 2009.
[40] Y. Yokota, On the potential functions for the hyperbolic structures of a knot complement,
Invariants of knots and 3-manifolds (Kyoto, 2001), Geom. Topol. Monogr., vol. 4, Geom.
Topol. Publ., Coventry, 2002, pp. 303–311 (electronic). MR2002618 (2004g:57037)
[41] T. Yoshida, The η-invariant of hyperbolic 3-manifolds, Invent. Math. 81 (1985), no. 3, 473–
514. MR87f:58153
[42] D. Zagier, The dilogarithm function, Frontiers in number theory, physics, and geometry. II,
Springer, Berlin, 2007, pp. 3–65. MR2290758 (2008h:33005)
[43] H. Zheng, Proof of the volume conjecture for Whitehead doubles of a family of torus knots,
Chin. Ann. Math. Ser. B 28 (2007), no. 4, 375–388. MR2348452
40 HITOSHI MURAKAMI
[44] C. K. Zickert, The volume and Chern-Simons invariant of a representation, Duke Math. J.
150 (2009), no. 3, 489–532.
Department of Mathematics, Tokyo Institute of Technology, Oh-okayama, Me-

guro, Tokyo 152-8551, Japan
E-mail address: starshea@tky3.3web.ne.jp
Volume 541, 2011
Quantum Field Theory and the Volume Conjecture
Tudor Dimofte and Sergei Gukov
Abstract. The volume conjecture states that for a hyperbolic knot K in

the three-sphere S 3 the asymptotic growth of the colored Jones polynomial
of K is governed by the hyperbolic volume of the knot complement S 3 \K.
The conjecture relates two topological invariants, one combinatorial and one
geometric, in a very nonobvious, nontrivial manner. The goal of the present
lectures is to review the original statement of the volume conjecture and its
recent extensions and generalizations, and to show how, in the most general
context, the conjecture can be understood in terms of topological quantum field
theory. In particular, we consider: a) generalization of the volume conjecture
to families of incomplete hyperbolic metrics; b) generalization that involves
not only the leading (volume) term, but the entire asymptotic expansion in
1/N ; c) generalization to quantum group invariants for groups of higher rank;
and d) generalization to arbitrary links in arbitrary three-manifolds.
1. Preliminaries
Let K be an oriented knot (or link) in the three-sphere S 3 . The original volume
conjecture [21, 26] relates the N -colored Jones polynomial of K to the hyperbolic
volume of the knot complement S 3 \K:
N -colored Jones poly of K ←→ hyperbolic volume of S 3 \K

(1.1)
(combinatorial, rep. theory) (geometric) .
We begin by reviewing some of the definitions and ingredients that enter on the two
sides here in order to make this statement more precise, and to serve as a precursor
for its subsequent generalization.
Jones polynomials. The (non-colored) Jones polynomial J(K; q) of a knot or
link can be defined combinatorially via the skein relation
) − q −1 J( ) = (q 2 − q − 2 ) J(
1 1
(1.2) q J( ),
2010 Mathematics Subject Classification. Primary 58J28, 81T45, 57M27.

The work of SG is supported in part by DOE Grant DE-FG03-92-ER40701, in part by NSF
Grant PHY-0757647, and in part by the Alfred P. Sloan Foundation. Opinions and conclusions
expressed here are those of the authors and do not necessarily reflect the views of funding agencies.
2011
c 0000
c Mathematical
American (copyright Society
holder)
1
41
42
2 TUDOR DIMOFTE AND SERGEI GUKOV
along with the normalization1
) = q 2 + q− 2
1 1
(1.3) J( for = unknot ,
and the rule
(1.4) J(K1 K2 ) = J(K1 ) J(K2 )
for any disjoint union of knots or links. Thus, for example, the (right-handed)
trefoil and figure-eight knots have Jones polynomials
J(31 ) = q − 2 + q − 2 + q − 2 − q − 2 ,
1 3 5 9
J(41 ) = q 2 + q − 2 .
5 5
In general, J(K, q) is a Laurent polynomial, J(K, q) ∈ Z[q 2 , q − 2 ].

1 1
The combinatorial construction of the Jones polynomial is intimately related

to representation theory of SU (2) — or the closely related representation theories
In particular, the
of the quantum group Uq (su(2)) or the affine Lie algebra su(2).
classical Jones polynomial above is obtained by “coloring” the knot (or link) K in
S 3 with the 2-dimensional representation of SU (2). More generally, such a knot or
link can be colored with any finite-dimensional representation R of SU (2), leading
to a colored Jones polynomial JR (K, q). The N-colored Jones polynomial JN (K, q)
takes R to be the irreducible N -dimensional representation [38, 33, 22]. The col-
ored Jones polynomial can again be computed in a purely algebraic/combinatorial
manner, by using the two rules

(1.5a) J⊕i Ri (K; q) = JRi (K; q)
and i
(1.5b) JR (K n ; q) = JR⊗n (K; q) ,

together with
(1.6) JR (K1 K2 ) = JR (K1 )JR (K2 ) ,
and the fact that J1 (K; q) = JR= (K; q) ≡ 1. The first rule says that if R is
reducible, then JR splits as a sum over irreducible components. The second rule
says that the R-colored Jones polynomial for the n-cabling of a knot (formed by
taking n copies of the knot or link, slightly displaced away from one another2) is
equal to the colored Jones polynomial of the original knot but in representation
R⊗n .
For example, from (1.5a-b) and the fact that JN =2 (K, q) = J(K, q), it is easy
to see that
q 2 − q− 2
N N
(1.7) JN ( )= .
q 2 − q− 2
1 1
1The most common normalization for the unknot seen in the mathematics literature is
J( ) = 1. For the connection with topological quantum field theory, however, (1.3) is much
more natural.
2This displacement must be done in a way that produces zero linking number between the
various copies.
QFT AND VOLUME CONJECTURE 43
3
More generally, for any knot K, relations (1.5) can be used to reduce JN (K; q) to
ordinary Jones polynomials of K and its cablings. We have
J1 (K; q) = 1 ,
J2 (K; q) = J(K; q) ,
J3 (K; q) = J(K 2 ; q) − 1 ,
J4 (K; q) = J(K 3 ; q) − 2J(K; q) ,
...,
where the expressions for J3 , J4 , etc. follow from the rules for decomposing rep-
resentations of SU (2): 2⊗2 = 1 ⊕ 3 , 2⊗3 = (1 ⊕ 3) ⊗ 2 = 2 ⊕ 2 ⊕ 4 , etc. Since
J(K; q) ∈ Z[q 2 , q − 2 ] for any K, it is clear that the colored Jones polynomials
1 1
JN (K; q) will also be elements of Z[q 2 , q − 2 ] .

1 1
We have explained the left side of (1.1), completely, if somewhat tersely, in

terms of algebra and combinatorics. The right side has a very different interpreta-
tion.
Hyperbolic volumes. It was conjectured by Thurston [34] (and is now proved
[31]) that every three-manifold may be decomposed into pieces that admit exactly
one of eight different geometric structures. The most common structure by far is
hyperbolic. Indeed, in the case of knot complements in S 3 this statement can be
made exact: a knot complement has a hyperbolic structure if and only if it is not
a torus or satellite knot [34]. By definition, a “hyperbolic structure” refers to a
geodesically-complete metric of constant curvature −1. If a hyperbolic structure
does exist on a manifold M , then it is unique, and the corresponding hyperbolic
volume Vol(M ) is a well-defined topological invariant.
In fact, there also exists a natural complexification of the hyperbolic volume of
a three-manifold M , obtained as
(1.8) Vol(M ) + iCS(M ) ,
where CS(M ) is the so-called Chern-Simons invariant of M . To understand this,
suppose that M allows a spin structure (as all knot/link complements in S 3 do)
and consider flat SL(2, C) connections on M in place of hyperbolic metrics.3 There
exists a flat connection A whose real and imaginary parts can be interpreted, re-
spectively, as the vielbein and spin connection of the hyperbolic metric. The real
part of the quantity

i i 2
(1.9) ICS (A) = Tr A ∧ dA + A ∧ A ∧ A
2 2 M 3
then reproduces Vol(M ), while the imaginary part defines CS(M ). The expression
ICS (A) is the so-called Chern-Simons functional of A. Further details can be found
e.g. in [36, 37] or [16, 8]. Under gauge transformations acting on A, the functional
ICS (A) is only well-defined up to shifts of 8π 2 , leading to an ambiguity of 4π 2 in
the definition of CS(M ). Because of this, it is often convenient to exponentiate the
3Recall that a “G-connection” on a principal G-bundle E → M can be written locally as

a g-valued one-form A. The bundle E is typically taken to be trivial in the present context,
E = G × M . A gauge transformation (a change of coordinates on E) induced by an element
g ∈ Γ(E) acts locally on the connection as A → g −1 Ag + g −1 dg.
44
complexified volume (1.8), writing it in the unambiguous form

i 1
(1.10) Z(M ) = e 4π ICS (A) = e 2π Vol(M )+iCS(M ) .
For hyperbolic knot complements, the full complexified volume Z(M ) can be
efficiently computed in terms of ideal hyperbolic triangulations, cf. [7, 30, 42].
The Volume Conjecture. We have not said much yet about the variable q ap-
pearing in the Jones polynomials. Strictly speaking, this variable should be a root
of unity4
2πi
(1.11) q=e k , k ∈ Z+ .
At the special value k = N , all Jones polynomials JN (K; q) vanish, but the ratio
JN (K; q)
(1.12) VN (K, q) =
JN ( ; q)
remains finite. The original volume conjecture [21, 26] then states that
2πi
2π log VN (K; q = e N )
(1.13) lim = Vol(M ) .
N →∞ N
It is also possible to remove the absolute value and exponentiate to obtain the
complexified generalization (cf. [27])

2πi N →∞ N
(1.14) VN (K; q = e N ) ∼ Z(M )N = e 2π Vol(M )+iCS(M ) .
Figure 1. The figure-eight knot, 41
As an example, consider the figure-eight knot (Figure 1), the simplest hyper-
bolic knot. The colored Jones polynomial (see e.g. [21] or [19]) is
(1.15)

N −1
(q)m (q −1 )m ,
2πi
VN (41 ; q = e N ) = (x)m := (1 − x)(1 − x2 ) · · · (1 − xm ) .
m=0
The hyperbolic volume of the figure-eight knot complement is

(1.16) Vol(S 3 \41 ) = 2 Vol(Δ) = 2.02988... ,
4In terms of representation theory, the integer k is identified as the level of the affine Lie alge-

bra su(2)k . The representation theory of the quantum group Uq (su(2)) also (crucially) simplifies
greatly when q is a root of unity, becoming essentially equivalent to the representation theory of
. See also Sections 3.3.1-3.3.2.
su(2) k
5
π
where Vol(Δ) = Im Li2 (ei 3 ) denotes the volume of a regular hyperbolic ideal tetra-
hedron. The Chern-Simons invariant CS(S 3 \41 ) vanishes. It is fairly straightfor-
ward (and an informative exercise5) to show that in the limit N → ∞ one has, as
expected,
2πi
2π log VN (41 ; e N )
(1.17) lim = Vol(S 3 \41 ) .
N →∞ N
2. The many dimensions of the volume conjecture
There are several natural ways in which one might try to generalize the basic
volume conjecture (1.14). One possibility is to consider not just k = N (or q =
2πi
e N ), but arbitrary values of k (or q). Another option would be to ask what happens
to subleading terms in the asymptotic expansion of VN (K; q) as N → ∞. It might
also be interesting to consider not just hyperbolic knots in S 3 but arbitrary links in
more complicated three-manifolds. It turns out that all these generalizations make
sense, and can be nicely combined and interpreted in terms of Chern-Simons theory
with complex gauge group [16]. In this section, we detail each of them (and one
additional generalization) in turn, and begin to explain what kind of new objects
one should expect on the right-hand-side of (1.14). Then, in section 3, our goal will
be to explain where such generalizations come from.
2.1. Parametrized VC. The original volume conjecture only held for a spe-
2πi 2πi
cial root of unity q = e N . In order to generalize to arbitrary q = e k , the
appropriate limit to consider is
N
(2.1) k → ∞, N → ∞, u := iπ fixed
k
(or q → 1, q N = e2u fixed). The question, then, is how to understand
(2.2) lim JN (K; q)1/k ?
k,N →∞
The answer, described in [16], uses the fact that in correspondence with the
“deformation” in the colored Jones polynomial, there exists a one-parameter defor-
mation of the hyperbolic structure on a knot complement S 3 \K. To understand
this, let μ be a small loop linking the excised knot K, as in Figure 2a. In terms of flat
SL(2, C) connections, the geodesically complete hyperbolic metric has a parabolic
SL(2, C) holonomy around μ,

1 1
(2.3) Hol(μ, complete) = ± ,
0 1
whereas the incomplete, u-deformed hyperbolic metric/SL(2, C) connection is de-
fined to have a holonomy conjugate to
u
e 1
(2.4) Hol(μ, u) = .
0 e−u
(As long as eu = e−u , this deformed holonomy is also conjugate to the purely
diagonal matrix diag(eu , e−u ).) The resulting metric is not complete. For example,
when u is purely imaginary, the u-deformed metric has a conical cusp of angle
2Im(u) at the knot K.
5One method involves analytically continuing the summand as a ratio of quantum dilogarithm
functions (cf. [11, 8]), approximating the sum by an integral, and evaluating it at its saddle point.
46
The complexified hyperbolic volume for this one-parameter family of metrics

can again be defined in terms of the Chern-Simons functional ICS (A) appearing
in (1.9). Now, however, A = A(u) should be a flat SL(2, C) connection with
prescribed holonomy (2.4). The ”parametrized” volume conjecture then takes the
form [16]
k,N →∞
e− 4πi ICS (A(u)) .
k
(2.5) JN (K; q) ∼
Figure 2. a) The “longitude” λ and “meridian” μ holonomy paths

in the knot complement S 3 \K. b) Integration on the A-polynomial
curve to find the deformed complex volume.
The quantity ICS (A(u)) can be described very explicitly. Indeed, suppose that
we require a hyperbolic metric (expressed in terms of a flat SL(2, C) connection) to
have holonomies conjugate to diag(eu , e−u ) and diag(ev , e−v ), respectively, along
the meridian and longitude loops depicted in Figure 2a. Such a metric exists if and
only if the so-called A-polynomial of K vanishes [6],
(2.6) A(, m) = 0 for = ev , m = eu .
Given a fixed eu ∈ C∗ , exactly one of the solutions v = v hyp (u) of this equation
corresponds to the u-deformed hyperbolic metric. The Chern-Simons functional
evaluated at the flat connection A(u) can then be written as [16]

(2.7) ICS (A(u)) = ICS (A(iπ)) + 4 θ ,
γ
where A(iπ) is the non-deformed hyperbolic flat connection,

(2.8) θ = −(v + iπ) du

is a one-form on the curve A(ev , eu ) = 0, and γ is a path on this curve that connects

the complete hyperbolic structure at (ev , eu ) = (−1, ±1) to the u-deformed metric
hyp
at (ev , eu ) = (ev (u) , eu ), as in Figure 2b.6
6The actual complexified volume that appears in the literature on hyperbolic geometry (cf.
[29, 41, 20]) is related to ICS (A(u)) as
i
(2.9) Vol(S 3 \K; u) + iCS(S 3 \K; u) = ICS (A(u)) + 2iv(u)Re(u) − 2πu + 2π 2 i .
2
Note that ICS (A(u)) is analytic in u, whereas Vol(u) + iCS(u) is not.
7
As our recurrent example, consider again the figure-eight knot. The complete
colored Jones polynomial, cf. [19], is
N N −1
q 2 − q− 2
N j
(2.10) JN (41 ; q) = 1
− 12
qN j (1 − q k−N )(1 − q −k−N ) .
q −q
2
j=0 k=1
The A-polynomial of the figure-eight knot is

(2.11) A(, m) = ( − 1)(m4 2 − (1 − m2 − 2m4 − m6 + m8 ) + m4 2 ) ,
and from (2.7) and (1.16), it results that the Chern-Simons functional can be written
as
(2.12) ICS (A(u)) = 2Li2 (e−p−u ) − 2Li2 (ep−u ) + 8(p − iπ)(u − iπ) ,
where x = ep is the solution to m3 x2 + (1 − m2 − m4 )x + m3 = 0 with smallest
negative imaginary part. (Some algebra is required to arrive at this form of the
ICS (A(u)), cf. [16, 8]; it is easiest to check the result by direct differentiation.)
For irrational u/iπ in a neighborhood of u = iπ it can then be shown (cf. [24, 28])
that the proposed asymptotics (2.5) indeed hold.
The necessity for taking u/iπ irrational here may appear a little strange at
first glance. It stems fundamentally from the fact that the Jones polynomials
2πi
JN (K; q = e k ) are really only defined for N, k ∈ Z. A subtle analytic continuation
in either N or k is necessary to achieve u/iπ = N/k ∈ / Q. As anticipated in [16]
and explained recently in [39], it is this continuation that causes the growth of
the colored Jones polynomial to be exponential. We will remark on this further in
Section 3.3.2.
In light of this argument, one might ask now why the original volume conjecture
at the rational value k = N or u = iπ held in the first place. Recall that JN (K; q)
actually vanished at k = N , so it was necessary to divide by JN ( ; q) to obtain
the non-vanishing ratio VN (K; q). Examining VN (K; q) at u → 0 is equivalent to
considering the derivative of JN (K; q) at u = iπ, which of course knows about
analytic continuation.7
2.2. Quantum VC. The second option for generalizing the volume conjecture
(1.14) is to ask for higher-order terms in the asymptotic expansion of the colored
Jones polynomial. Let us define a new “quantum” parameter as
iπ
(2.13) = ,
k
so that
(2.14) q = e2 .
The two parameters N and k of the colored Jones polynomial can be traded for
and u, and the limit (2.1) is simply → 0. At u = iπ, higher-order asymptotics
are then predicted [16, 8] to have the form
(2.15)
∞
2πi N →∞ 1 3 1 −iπTK
VN (K; q = e N ) ∼ exp (Vol + iCS) − log + log + S̃n n−1 .
2 2 2 4 n=2
7We thank E. Witten for useful observations on this subject.

48
Here, for example, TK is the Ray-Singer torsion of the knot complement S 3 \K. It
can be defined after putting any background metric on S 3 \K [32] as

1 (det Δ0 ) 2
3 3

(2.16) T (M ) = exp − n
n(−1) log det Δn = 1 ,
2 n=0 (det Δ1 ) 2
where Δn is the Laplacian acting on n-forms.

It is fairly straightforward to combine the present quantum deformation with
the parametrization of the volume conjecture in u. The expectation is that
(2.17)
∞

N,k→∞ 1 3 1 iTK (u)
JN (K; q) ∼ exp − ICS (A(u)) − log + log + Sn (u) n−1
.
4 2 2 4π n=2
Here, TK (u) is a u-deformed torsion, and is related to the Alexander polynomial of

K [25]. The higher-order coefficients in (2.15) are related8 to those in (2.17) as
sinh
(2.18) S̃n n−1 = Sn (iπ)n−1 − log .

n≥2 n≥2
For the figure-eight knot, the quantum volume conjecture (2.17) was tested to
first subleading order in [17], using the Ray-Singer torsion
4π 2
(2.19) T41 (u) = √ .
−m−4 + 2m−2 + 1 + 2m2 − m4
Higher-order coefficients Sn (u) can also be computed [8]. For example,
−i(T41 )3
(2.20) S2 (u) = 1 − m2 − 2m4 + 15m6 − 2m8 − m10 + m12 ,
12(4π 2 )3 m6
−2(T41 )6 1
(2.21) S3 (u) = 1 − m2 − 2m4 + 5m6 − 2m8 − m10 + m12 − .
(4π 2 )6 m6 6
These expressions appear to be intersting, unexplored knot invariants with distinc-
tive number-theoretic properties [8]. Needless to say, it would be interesting to test
the quantum volume conjecture (2.17) for other hyperbolic knots and/or to higher
order in the -expansion.
Just as the generalization of the volume conjecture to u = 0 was interpreted in

terms of the SL(2, C) Chern-Simons functional, there is also a Chern-Simons inter-
pretation of the quantum volume conjecture. One must consider how the functional
ICS (A) behaves when the connection A undergoes “quantum fluctuations” away
from the flat connection A(u). This is accomplished in physics via perturbative
quantum field theory. Symbolically, we can write A = A(u)+A , where A contains
8It is more natural to give this expansion in terms of the unnormalized Jones polynomial
JN (K; q), rather than the normalized VN (K; q). Strictly speaking, JN (K; q) vanishes at u = iπ.
However, as explained in the last paragraph of Section 2.1, in greater detail in [39] and in Section
3.3.2 of these lectures, removing a combination of phase factors from JN (K; q), or simply using
higher order coefficients Sn (u) that have been analytically continued in a neighborhood of u = iπ
allows for a consistent definition of Sn (iπ).
9
the fluctuations away from flatness, and define a perturbative “partition function”
via the path integral

Z(S 3 \K; u; )pert = DA e− 4 ICS (A(u)+A ) .
1
(2.22)
The exponent in the integrand has a critical point at A = 0, and a saddle point
expansion around this point yields the right-hand-side of (2.17).
(To be very pre-
cise, JN (K; q) ∼ Z(S 3 \K; ; u)/Z(S 3 ; ), where Z(S 3 ; ) = 2/k sin(π/k) is the
partition function of the three-sphere S 3 .)
2.3. Groups and representations. So far, we have considered two contin-
uous deformations of the volume conjecture, in u and , as drawn schematically in
Figure 3. In addition, there are two discrete generalizations that we can make.
Figure 3. Continuous and discrete generalizations of the volume conjecture.
The first such generalization involves the “gauge groups” and representations
that define colored Jones polynomials. Recall from Section 1 that the N -colored
Jones polynomial is a quantum SU (2) invariant that corresponds to coloring a knot
with the N -dimensional representation of SU (2). More generally, one can consider
“quantum SU (n) invariants,” or in fact invariants for any compact Lie group G.
Knots or links should then be colored by finite-dimensional representations R of G.
For semisimple G and irreducible R, the representation can be labelled by a highest
weight λ in the weight lattice Λwt ⊂ t∗ , where t is the dual of the Cartan subalgebra
t of the Lie algebra g = Lie(G). The resulting quantum polynomial invariant of a
knot in S 3 may be denoted
(2.23) PRGλ (K; q) .
Just like the colored Jones polynomial, PRG (K; q) depends on a root of unity
2πi
q = e k . Also like the colored Jones, these invariants satisfy

(2.24) P⊕Gi Ri (K; q) = PRGi (K; q) , and PRG⊗n (K; q) = PRG (K n ; q) .
i
More general tensor products can also be produced by cabling a knot or link and
coloring each component of the cable with a different representation. When G =
SU (n) and R is the fundamental representation (or any of its conjugates), the
polynomial PRG (K; q) satisfies a skein relation similar to (1.2).
50
Using the positive nondegenerate trace form −Tr : g×g → R, the weight λ can
be identified with its dual element λ∗ in t. Let us also define ρ to be half the sum
of positive roots, and ρ∗ ∈ t ⊂ g its dual. Then the interesting limit to consider for
PRGλ (K, q) is
iπ ∗
(2.25) k → ∞, λ∗ → ∞ , u := (λ + ρ∗ ) fixed ,
k
or
∗
2πi
+ρ∗
(2.26) q = e2 = e k →1 ( → 0) , qλ = e2u fixed .
The parameter u has now become a diagonal matrix, an element of tC . Coming

back to the case of SU (2) and an N -dimensional representation, in this notation
we have
(2.27)

N
∗ N −1 0 ∗ 1 0 k 0
λ = , ρ = , u = iπ .
0 −(N − 1) 0 −1 0 −N k
The asymptotics of the invariant PRG (K; q) should look very similar to those of
the colored Jones polynomial, namely
(2.28)
∞

→0 1 δ 1 iT (u)
PRλ (K; q) ∼ exp − ICS (u) − log + log
G
+ Sn (u) n−1
.
4 2 2 4π n=2
The leading term ICS (u) is now the Chern-Simons functional (1.9) evaluated at
a flat GC connection A(u) — in other words, a connection taking values in the
complexified Lie algebra gC — whose holonomy around the meridian of the knot as
in Figure 2a is
(2.29) Hol(μ) = m = eu .
For generic u, this holonomy is an element of the complexified maximal torus TC ⊂

GC . Again, ICS (u) can be expressed as

(2.30) ICS (u) = const. + 4 θ,
γ(u)
r
where θ ∼ − i=1 vi dui + exact is a differential on an r-dimensional complex
variety cut out by r equations Aj (ev , eu ) = 0, with r = rank(G). The equations
Aj (ev , eu ) = 0 describe the moduli space of flat GC connections on S 3 \K. (These
equations will be discussed in further detail in Section 3.3.1.)
Subleading terms on the right side of (2.28) also have a geometric interpreta-
tion. The function T (u) is the Ray–Singer torsion of the knot complement twisted
by the flat connection A(u), and the number δ is a fixed integer which can be
computed in terms of cohomology of S 3 \K with coefficients in the associated flat
bundle, with structure group GC and connection A(u) (cf. [3, 8]). More gener-
ally, the full asymptotic expansion can be written as a perturbative path integral
just like (2.22), which takes into account the quantum fluctuations of a flat GC
connection.
11
2.4. Links and 3-manifolds. The final generalization of the volume conjec-
ture that we consider is to arbitrary links in arbitrary three-manifolds. Here we
really begin to require a true TQFT description of the “quantum G-invariants” of
knots and links. This was supplied by quantum Chern-Simons theory with compact
gauge group G in [38], and reinterpreted via quantum groups and R-matrices in
[33]. Using either of these approaches, one may define a quantum partition function
(2.31) Z G (M, L; {Ra }; )
for a link L in any three-manifold M , where each component of the link is colored
with a different representation Ra . The “polynomial” PRG is obtained from this
after normalizing by the partition function of an empty manifold,
G Z G (M, L; {Ra }; )
(2.32) P{R a}
(M, L; q) = , (q = e2 ) .
Z G (M ; )
Thus, in the case of the colored Jones polynomial,
Z SU(2) (S 3 , K; RN ; )
(2.33) JN (K; q) = .
Z SU(2) (S 3 ; )
2πi
The integer k (appearing in q = e2 = e k ) is identified with the “level” or coupling
constant of the compact Chern-Simons theory.9
The partition function (2.31) supplies the left-hand-side of the volume conjec-
ture. We then want to understand the asymptotics of Z G (M, L; {Ra }; ) in the
limit → 0, with a parameter ua = (λ∗a + ρ∗ ) held fixed for each separate link
component. The answer should be given by perturbative, quantum Chern-Simons
theory with complex gauge group GC , evaluated on the link complement M \L, in
the background of a flat connection with fixed holonomy
(2.34) ma = exp(ua )
around the meridian of each excised link component. Denoting this perturbative
Chern-Simons partition function by

1 δ
(2.35) Zpert (M \L; {ua }; ) = exp − ICS ({ua }) − log + . . . ,
GC
4 2
we expect that
→0
(2.36) Z G (M, L; {Ra }; ) ∼ Zpert
GC
(M \L; {ua }; ) .
This discussion can also be rephrased in a somewhat more symmetric manner,

using link complements on both sides of the volume conjecture. It turns out that
in compact Chern-Simons theory the partition function of a knot (or link) K ⊂ M
colored by representation Rλ is equivalent to the partition function of the knot
complement M \K with fixed meridian holonomy

λ∗ + ρ∗
(2.37) m = exp iπ = exp ((λ∗ + ρ∗ )) = exp(u) .
k
9To be completely precise, the integer k used throughout these lectures is the sum of the
Chern-Simons level and the dual Coxeter number of G. In what follows, note that exponential
growth of (2.31) is only observed when k is analytically continued from integer values, just as in
the case of the colored Jones polynomial for links in S 3 .
52
For the compact G theory to make sense, the eigenvalues of the matrix u/iπ must
clearly be rational. However, interesting asymptotics — potentially with exponen-
tial growth as in (2.36) — occur when u is analytically continued away from such
rational values. This process of analytic continuation naturally lands one in the
regime of Chern-Simons theory with complex gauge group GC [8].
After so many generalizations, it may be unclear that the volume conjecture

has anything to do with volumes anymore. Indeed, for higher-rank gauge groups G,
“volume” should not be a hyperbolic volume but rather the “volume” of a holonomy
representation
(2.38) : π1 (M \K) → GC .
Even in the case of G = SU (2) and knots in the three-sphere, one may run across
cases of non-hyperbolic knot complements. It was clear from the initial days of
the volume conjecture [26] that even in these cases the asymptotics of JN (K; q)
could still be given by an appropriate flat (but non-hyperbolic/non-metric) SL(2, C)
structure.
3. TQFT
We have just seen that the volume conjecture admits a multitude of generaliza-
tions, all of which seem to be related to Chern-Simons quantum field theory. The
most complete statement of the volume conjecture (2.36) involves Chern-Simons
theory with compact gauge group G on the left-hand side and Chern-Simons the-
ory with complex gauge group GC on the right:
combinatorics/rep. theory geometry
quantum G-invariants volumes of representations
JN (K; q) , PRGλ (K; q) , ←→ : π1 (M \K) → GC ,
(3.1)
Z G (M, K; u; ) , etc. GC
Zpert (M \K; u; ) , etc.
∗
2πi +ρ∗
q=e k = e2 , u = iπ λ k .
Chern-Simons theory is a topological quantum field theory (TQFT). In addition to

the basic implication that partition functions such as Z G (M, K; u; ) or Zpert
GC
(M \K; u; )
are topological invariants of colored knots and links in three-manifolds, the struc-
ture of TQFT provides powerful methods for actually computing them in multiple
ways. It also shows why a general correspondence like (3.1) might be expected to
hold.
3.1. Cutting and gluing. In its more mathematical incarnation, a 3-dimensional

TQFT can be thought of as a functor Z that assigns
closed 3-manifold M number Z(M )
closed 2-manifold Σ vector space Z(Σ)
(3.2)
closed 1-manifold S 1 category Z(S 1 )
point p 2-category Z(p) .
For our applications to Chern-Simons theory, we will really only need the top
two levels Z(M ) and Z(Σ). The finer structure of categories and 2-categories has
recently been explored in e.g. [12].
13
Figure 4. Hilbert space assigned to a surface Σ and partition

function assigned to a three-manifold M in TQFT.
If a 3-manifold M has a boundary Σ = ∂M , the object Z(M ) is no longer

a number, but an element of the vector space Z(Σ) assigned to the boundary, as
shown in Figure 4. This vector space is in fact a Hilbert space, so let us denote it
as HΣ = Z(Σ). At the top two levels, the TQFT must then satisfy the following
axioms of Atiyah and Segal (cf. [1]).
∗
(1) A change of orientation Σ → −Σ dualizes the Hilbert space, H−Σ = HΣ .
(2) For a boundary consisting of multiplet disjoint components, HΣ1
Σ2 =
HΣ1 ⊗ HΣ2 .
(3) Using the first two axioms, we see that for a manifold M with ∂M =
(−Σ1 ) Σ2 one obtains a map Z(M ) : HΣ1 → HΣ2 . Then, given a
3-manifold N that can be written as N = M1 ∪Σ2 M2 , with ∂M1 =
(−Σ1 ) Σ2 and ∂M2 = (−Σ2 ) Σ3 as illustrated below, the functoriality
property Z(N ) = Z(M2 ) ◦ Z(M1 ) must hold.
(4) For the empty boundary, HΣ= = C .

id
(5) For M = Σ × I, the map Z(M ) : HΣ → HΣ is just the identity.
Using these axioms, the partition function Z(M ) of any three-manifold, with
or without boundary, may be constructed by cutting the manifold into pieces and
taking inner products in boundary Hilbert spaces to glue the pieces back together.
For this purpose, it is often convenient to know how the mapping class group of a
surface Σ acts on HΣ , in order to properly identify the Hilbert spaces on two sides
of a gluing.
There are many examples of three-dimensional TQFT, differing essentially in
the definitions of the boundary Hilbert spaces H(Σ), as well as the action of the
mapping class groups on these spaces. In the case of Chern-Simons theory with
gauge group G (whether compact or complex), HΣ is a quantization of the space
54
Mflat (G; Σ) of flat G-connections on Σ:

(3.3)
connections A on principal
Mflat (G; Σ) = FA = 0 gauge equivalence .
G-bundle over Σ
(Recall that a connection is flat if the curvature FA = dA + A ∧ A vanishes.) The
precise meaning of the quantization used to obtain HΣ from Mflat (G; Σ) will be
the subject of Section 3.2. It depends on the level k = iπ−1 (or coupling constant)
of Chern-Simons theory, the only adjustable parameter in the TQFT.
In Chern-Simons theory, one is also interested in colored knots or links embed-
ded in 3-manifolds. Suppose for the moment that we have compact Chern-Simons
theory with gauge group G and level k ∈ Z. The intersection of a knot and a
boundary surface Σ shows up as a puncture on Σ and TQFT would assign the
boundary S 1 surrounding this puncture in Σ the category of representations of the
affine Lie algebra gk ,
(3.4) Z(S 1 ) ∼ reps of
gk .
The definition of the Hilbert space H(Σ) of a multiple-punctured Σ would then have
to be altered to include the space of homomorphisms between such representations.
For our purposes, however, the complication of knots can be conveniently avoided
by excising the knots and trading representations that color the knots for boundary
conditions on knot complements.
This trick was already mentioned in Section 2.4. In the language of TQFT,
it can be described the following way. Suppose that we have a knot K colored by
representation Rλ inside the closed manifold M . We cut out a tubular neighborhood
NK of the knot, so that
(3.5) M = (M \NK ) ∪T 2 NK , NK D2 × S 1 .
Of course, M \NK is just the knot complement, and NK is topologically a 2-disk
times S 1 that contains the knot running through its center. The partition functions
Z(M \NK ; u; ) — which by a slight abuse of notation we will write as Z(M \K; u; )
— and Z(NK ; Rλ ; ) are both vectors in the boundary Hilbert space HT 2 ; therefore,
by TQFT,
(3.6) Z(K ⊂ M ; Rλ ; ) = Z(M \K; u; ) , Z(NK ; Rλ ; ) HT 2 ∈ C .
As we will see in the next section, the Hilbert space HT 2 can be understood as a
space of functions of the variable u that describes the holonomy of flat connections
around the meridian of T 2 (as in Figure 2a). The crucial fact, then, is that the
vector Z(N ; Rλ ; ) ∈ HT 2 is only supported on the part of this space with

λ ∗ + ρ∗
(3.7) eu = Hol(μ) = exp iπ .
k
∗ ∗
In other words, Z(N ; Rλ ; ) acts like a delta-function δ(u−iπ λ +ρ
k ). Therefore, col-
oring by Rλ is equivalent to restricting Z(M \K) to an appropriate one-dimensional
subspace of HT 2 :

(3.8) Z(K ⊂ M ; Rλ ; ) = Z(M \K; u; ) u=iπ λ∗ +ρ∗ ∈ C .
k
Our plan now is to give a complete description of HT 2 and to explain how the
elements Z(M \K) ∈ HT 2 may be calculated for knot complements, in the case of
15
Chern-Simons theory with both compact and complex gauge groups. (The exten-
sion to links is straightforward and will not be mentioned explicitly hereafter.) This
will first require a brief discussion of quantization.
3.2. Quantization. The basic problem of quantization begins with a pair
(M, ω), where M is a manifold with symplectic structure ω, called a classical
“phase space.” Quantization takes this pair and constructs a quantum Hilbert
space H. Moreover, quantization should map the algebra of functions on M to an
algebra A of operators on H:
(M, ω) H (= Hilbert space)
(3.9)
alg. of functions on M alg. of operators on H
f → Of : H → H .
The functions on M form a Poisson algebra with respect to the usual pointwise mul-
tiplication of functions and a Lie algebra structure {•, •} induced by the symplectic
structure. Quantization must map this algebra to an associative but noncommuta-
tive algebra A , such that
(3.10) [Of , Og ] = −i O{f,g} + . . . ,
where [•, •] is the commutator of operators. Here is a parameter that is involved
in the determination of H itself as well as the algebra of operators.
Very roughly, the Hilbert space H consists of L2 sections of a complex line
bundle over M with curvature 1 ω. Locally, these sections are only allowed to
depend on half of the coordinates of M. In a standard physical setup, M can be
thought of as the space of all possible positions xi and momenta pi of particles; thus
the elements of H are functions (“wavefunctions”) that depend on either positions
or moments, but not both.
In addition to the construction of H, the process of quantization must also
explain how classical motions or trajectories of a physical system are associated to
quantum states in H. A classical trajectory (or “semiclassical state”) is described
by a Lagrangian submanifold L ⊂ M. Being Lagrangian means that L is middle-
dimensional and ω|L = 0. Let θ be 1-form (called a Liouville 1-form10) that satisfies
ω = dθ. Notice that θ|L is closed. Then the Lagrangian L is called quantizable if

(3.11) θ ∈ 2π Z
γ
for any closed cycle γ ⊂ L. The vector (or wavefunction) Z ∈ H corresponding to

L can be written as

i
(3.12) Z = Z(x) = exp S0 (x) + . . . ,

with
x
(3.13) S0 (x) = θ
x0
for some fixed x0 and varying x ∈ L. Due to the quantization (3.11), the expression
(3.12) is completely well-defined.
10There is an ambiguity in choosing θ, directly related to the choice of coordinates of M
(positions versus momenta) that elements of H are to depend on.
56
Expression (3.12) only defines Z to leading order in . To find subleading

corrections, it is useful to employ a complementary approach. Suppose that the
Lagrangian submanifold L is cut out by certain equations fi = 0 on M. Quantiza-
tion promotes these functions to operators Ofi acting on H, and the vector Z can
also be defined as a solution to the equations
(3.14) Of i · Z = 0 ∀ i.
If the Ofi are properly quantized, then the solution to these equations will be the
exact wavefunction.
3.2.1. Methods. The problem of quantization can be approached in many dif-
ferent ways. Each approach has its advantages and disadvantages, but in the end all
methods are expected to yield the same result. The classic approach of geometric
quantization (cf. [40]) starts by defining a prequantum line bundle L → M with a
unitary connection of curvature 1 ω. Note that such a line bundle only exists for
1
(3.15) ω ∈ H 2 (M; Z) ,
2π
which can lead to a quantization of −1 (i.e. a restriction of to a discrete set of
values in C∗ ). The local choice of “position” versus “momentum” coordinates is
encoded in the choice of a set of 12 dimR M vector fields Pj , called a polarization,
and the Hilbert space H is then defined as the set of square-integrable, Pj -invariant
sections of L. This gives a very concrete definition of H, although it can be very
hard to show that the construction is independent of the choice of polarization.
(The problem becomes more manageable if M is Kähler.) Moreover, it is often dif-
ficult in geometric quantization to find the full quantum expressions for operators
Ofi .
An alternative, deformation quantization [4] partially solves this latter prob-

lem. It describes a formal -deformation of the ring of functions on M, using a
noncommutative product of the type
2 ij kl
f g = f g + αij ∂i (f )∂j (g) + α α ∂i ∂k (f )∂j ∂l (g)
i,j
2
i,j,k,l
⎛ ⎞
2 ⎝ ij
(3.16) + α ∂j (αkl )(∂i ∂k (f )∂l (g) − ∂k (f )∂i ∂l (g))⎠ + . . . ,
3
i,j,k,l
−1
where α = ω is the Poisson structure corresponding to the symplectic form ω. In
local coordinates {f, g} = αij ∂i (f )∂j (g). One important advantage of deformation
quantization is that it is completely canonical and does not require any auxiliary
choices. In particular, there is an explicit formula for the -product (3.16) due to
Kontsevich [23], that allows one to express it as a sum over admissible graphs,
∞

(3.17) f g := n w(Γ)BΓ (f, g) ,
n=0 graphs Γ
of order n
where w(Γ) is a weight (number) assigned to Γ, and BΓ (f, g) are bilinear differential
operators whose coefficients are differential polynomials, homogeneous of degree n
in the components of the bivector field α on M. By definition, an admissible graph
of order n is an ordered pair of maps i, j : {1, . . . , n} → {1, . . . , n, L, R} where
17
neither map has fixed points and both maps are distinct at every point. There are
nn (n + 1)n such graphs.
For example, the graph of order 2 corresponding to the first term in the second
line of eq. (3.16) has 4 vertices and 4 edges:
1 i1
2
i1 = (1, 2)
j1 j2 j1 = (1, L)
(3.18) Γ = i2 i2 = (2, L)
j2 = (2, R)
L R
An example of a more complicated admissible graph (of order 4) is shown on Figure
5. The corresponding bidifferential operator is

(3.19) BΓ (f, g) = αi4 j4 (∂i3 αi1 j1 )(∂j1 ∂j4 αi2 j2 )(∂i2 ∂i4 αi3 j3 )(∂i1 ∂j3 f )(∂j2 g) .
When the Poisson structure is flat, a graph with an edge ending in a vertex
other than L or R will have zero contribution to the sum (3.17), since it will involve
derivatives of α. In this case the -product (3.17) becomes the usual Moyal product

ij ∂ ∂
(3.20) f g(x) = exp α f (x)g(y)|y=x
∂xi ∂y j
i4 4
3 j4
i2
i3 2
j1
j3 1 j2
i1
L R
Figure 5. An example of an admissible graph of order 4.
Deformation quantization is a powerful method for finding the operators Ofi .

It is important to stress, however, that, by itself, it does not explain how to con-
struct the space H (it is not an honest quantization), and can not capture the fact
that −1 should ever be discretized.
A third option, brane quantization [18], is a marriage of geometric and deforma-

tion quantizations in a physical context. It approaches the problem of quantization
by complexifying M and ω, and constructing a certain (secondary) topological
quantum field theory on the resulting space MC . It has the advantage of easily
characterizing the various choices that one must make in quantization, and provides
simple geometric criteria that describe quantizable (M, ω; ). In this approach, the
Hilbert space H is constructed as the space of morphisms (space of open strings),
(3.21) H = Hom(Bcc , B ) ,
58
where Bcc and B are objects (branes) of a certain category associated to the sym-
plectic manifold MC . Moreover, in this approach, independence of H on various
choices can be reformulated as a problem of constructing a flat connection on the
space of such choices, which identifies the space of ground states in the secondary
TQFT. In a closely related context, this problem has been studied in the mathemat-
ical physics literature [5, 9], and leads to a beautiful story that involves integrable
systems and tt∗ equations.
3.2.2. Simple examples. Let us now adapt the general statements here to some
specific examples.
Harmonic oscillator. The quintessential simplest nontrivial problem of quan-
tization is the harmonic oscillator. Consider a classical system that consists of
a particle moving on a line (with coordinate x = x(t)) with a potential energy
V = 12 x2 . This is depicted in Figure 6. The total (potential + kinetic) energy of
the particle at any moment of time is given by the Hamiltonian
1 2 1 2
(3.22) H= x + p ,
2 2
where classically p = ẋ = dx
dt is the momentum. This total energy H is conserved.
The classical phase space M is just R2 = {(x, p)}, endowed with a symplectic
√ ω = dp ∧ dx. A classical trajectory with energy H = E is just a circle of
structure
radius 2E in phase space. This defines a Lagrangian submanifold L(E) S 1 .
Figure 6. The harmonic oscillator: potential V (x) = 12 x2 in phys-

ical space, phase space M, a classical trajectory L(E) in phase
space, and the ground state quantum wavefunction Z(x).
Now let us quantize the system. Since H 2 (M; Z) = 0, there is no restriction or

quantization of −1 . On the other hand, there is a restriction on L which quantizes
the energy. Namely, according to (3.11), for a Liouville 1-form θ such that ω = dθ,
the integral

(3.23) θ= θ = 2πE
L S1
must be an element of 2π Z, implying that E = n for positive n ∈ Z. In fact,

this equation is corrected by quantum effects — a Maslov correction in geometric
19
quantization — to
1
(3.24) E = n+ .
2
This leads to the famous result that the lowest possible energy of a quantum har-
monic oscillator (at n = 0) is nonzero.
Suppose we choose a polarization ∂/∂p = 0, and a corresponding Liouville
1-form θ = p dx. The Hilbert space H can simply be identified as L2 (R) ∼
{functions of x}, on which the functions x and p act as operators
d
(3.25) x̂ := Ox = x , p̂ := Op = −i .
dx
In this case, the exact quantum expression for the Hamiltonian is
1
(3.26) OH = (x̂2 + p̂2 ) .
2
It is then easy to find the quantum wavefunctions corresponding to classical states
L(E) . From (3.12), we find a leading contribution
(3.27) x
x

i i 1
Z(x) exp θ = exp 2E − x2 dx exp − x2 + . . . .
0 0 2
Since the Lagrangian L(E) is defined classically by H − E = 0, the complete expres-
sion for Z(x) can be obtained by solving the operator equation (OH − E)Z = 0.
This eigenvalue equation has square-integrable solutions only for the quantized en-
ergies (3.24); for
example,
at the ground state energy E = /2, the exact solution
is Z(x) = exp − 2 1 2
x .
Representations of Lie groups. Another famous application of quantization is
the construction of unitary representations of Lie groups by quantization of coad-
joint orbits. A basic premise of this approach, also known as the orbit method,
is that coadjoint orbits come equipped with a natural symplectic structure (the
Kostant-Kirillov-Souriau symplectic structure), therefore providing interesting ex-
amples for quantization.
Continuing with our default notations in these notes, we use G for a compact
Lie group (that we usually assume to be simple), GC for its complexification, and
GR for some real form of the complex group GC (that may be equal to G). We
denote by gR the Lie algebra of GR , and similarly for G and GC . Given an element
λ ∈ g∗R (the highest weight of the desired unitary representation Rλ ) one constructs
M = GR · λ as the coadjoint orbit of GR in g∗R passing through λ.
In the case of compact groups, the phase space M is compact and its quan-
tization leads to a finite-dimensional Hilbert space H as the space of the unitary
representation Rλ . This is the statement of the Borel-Bott-Weil theorem. More-
1
over, the condition 2π ω ∈ H 2 (M; Z) that ensures the existence of a prequantum
line bundle becomes equivalent to the condition that λ be an element of the weight
lattice Λw ⊂ g∗ .
As a very simple illustration, consider the group SU (2). In this case, a non-
trivial coadjoint orbit is topologically equivalent to the flag manifold
(3.28) SU (2)/U (1) P1 .
Letting ω be the unit volume form on P1 , we see that (M, ω) is quantizable for
(3.29) −1 = 2πλ , λ ∈ Z(+) .
60
The prequantum line bundle with curvature −1 ω is simply O(λ) → P1 . Choosing a
holomorphic polarization, so that H is defined as the space of holomorphic sections
of O(λ), we see that dim H = λ + 1. The Hilbert space is precisely the space of the
(λ + 1)-dimensional representation of SU (2).
Similarly, some infinite-dimensional representations, such as unitary principal
series representations of SL(n, C) or SL(n, R), can be described as quantized orbits.
Nevertheless, there remain some outstanding puzzles: there exist unitary represen-
tations that don’t appear to correspond to orbits, and, conversely, there are real
orbits that don’t seem to correspond to unitary representations. An example of
first kind occurs even in the basic case of the real group GR = SL(2, R) and the
complementary series representations. To illustrate the second phenomenon, one
can take GR to be a real group of Cartan type BN , i.e. GR = SO(p, q) with
p + q = 2N + 1. The minimal orbit Omin of BN is a nice symplectic manifold of
(real) dimension 4N − 4, for any values of p and q. On the other hand, the minimal
representation of SO(p, q) exists only if p ≤ 3 or q ≤ 3 [35]. Both of these issues can
be resolved in the brane quantization approach [18], at the cost of replacing clas-
sical geometric objects (namely, coadjoint orbits) with their quantum or “stringy”
analogs (branes). In particular, in the case of BN one finds that, while the min-
imal orbit exists for any values of p and q, the corresponding brane exists only if
p ≤ 3 or q ≤ 3. (In general, the condition is that the second Stieffel-Whitney class
w2 (M) ∈ H 2 (M; Z2 ) must be a mod 2 reduction of a torsion class in the integral
cohomology of M.)
3.3. Chern-Simons theory. Finally, we arrive at our goal, Chern-Simons

theory. Let us recall for a second why we began discussing quantization in the
first place. In Section 3.1, we reviewed how partition functions in TQFT could be
obtained by cutting and gluing three-manifolds. We explained that the partition
function of a manifold with a knot is equivalent to the partition function of the
corresponding knot complement, projected onto appropriate boundary conditions
in H(T 2 ) as in (3.8). To make complete sense of this, however, and to actually
calculate partition functions, we must understand what H(T 2 ) really is. Using
Section 3.2 we are finally in a position to do so.
3.3.1. Quantization of Chern-Simons theory. Consider Chern-Simons theory
with gauge group G — either compact or complex — on a knot complement M =
\K, with ∂M = T 2 . The phase space M associated to T 2 is simply the space
M
of flat G-connections on T 2 , modulo gauge equivalence. Since a flat connection is
completely determined by the conjugacy classes of its holonomies, we have
(3.30) M = Mflat (G; T 2 )

(3.31) = representations : π1 (T 2 ) → G /conjugation .
The fundamental group π1 (T 2 ) Z ⊕ Z is abelian, generated by the meridian

and longitude of the torus. The holonomies along these loops can therefore be
simultaneously diagonalized11 into the maximal torus T ⊂ G. Coordinates on M
are then given by the 2r independent eigenvalues (m1 , ..., mr ) and (1 , ..., r ) of the
meridian and longitude holonomies, where r is the rank of G. We must also divide
11If G is not compact, there may be elements that are not so diagonalizable, but they form
lower-dimensional components of M which should not be considered in the quantization.
21
by the Weyl group W of G, which simultaneously permutes both sets of eigenvalues,

to obtain
(3.32) M (Tr × Tr )/W = T2r /W .
For example, for a compact group G = SU (n) the phase space is M = (S 1 )2(n−1) /Sn ,
where Sn is the symmetric group on n elements. Similarly, for its complexification
GC = SL(n, C), the phase space is M = (C∗ )2(n−1) /Sn . In general, ignoring
subtleties in high codimension that are not pertinent to quantization, the relation
between compact and complex phase spaces can be described as
(3.33) Mflat (GC ; Σ) = [Mflat (G; Σ)]C T ∗ Mflat (G; Σ) .
(In particular, the last relation is only a birational equivalence.)
Compact theory. In order to quantize M, we need a symplectic structure. In
compact Chern-Simons theory, it is given by

1
(3.34) ω= Tr δA ∧ δA .
4 T2
This can be expressed more concretely in coordinates {mi , i } = {eui , evi } as

(3.35) ω= d log mi ∧ d log i = dui ∧ dvi .
i i
The holonomy variables ui and vi function as “positions” and “momenta,” respec-

tively. Now, the parameter = iπ/k that appeared naturally in the discussion of
the volume conjecture in Section 2 is rescaled from the standard geometric quanti-
zation parameter of Section 3.2 by a factor of i. In terms of k, the quantization
condition (3.15) simply takes the form k ∈ Z. The integer k is identified as the
Chern-Simons level, modulo the shift mentioned in Footnote 9.
The last ingredient we need to describe the Hilbert space H is a choice of
polarization. For clarity, let us take G = SU (2) to be of rank one, and let us
choose the polarization ∂/∂v = 0, so that HT 2 essentially consists of periodic and
Weyl-invariant functions of u, f (u) = f (u + 2πi) = f (−u). Being somewhat more
careful, and thinking of these not as functions but as sections of the line bundle
with curvature πk ω, one finds that the simultaneous periodicity in the momentum
k Z. Therefore, a function f (u)
v and the position u restricts u to take values in iπ
only takes nonzero values at k + 1 distinct points u = 0, iπ 2iπ
k , k ..., iπ, and the space
HT 2 is finite-dimensional. For general compact semi-simple G, the Hilbert space H
takes the form [10, 2]
Λw
(3.36) HT 2 .
W kΛr
where Λw , Λr are the weight and root lattices of G. In other words, HT 2 is the set
of weights (hence representations) in a level-k affine Weyl chamber.
Given a “wavefunction” Z(M ; u; ) ∈ HT 2 associated to the knot comple-
ment M = M \K (with ∂M = T 2 ), the partition function Z G (M , K; Rλ ; ) for

K ∈ M colored by representation Rλ is simply given by evaluating Z(M ; u; ) at
∗ ∗
u = iπ λ +ρk as in (3.8). For example, in the case of SU (2) theory, we evaluate
Z(M ; u; ) at u = iπN/k (and normalize by the partition function of S 3 ) to find
the colored Jones polynomial JN (K; q). The single wavefunction Z(M ; u; ) in HT 2
comprises the entire family of colored Jones polynomials JN (K; q), N ∈ Z.
62
How is such a wavefunction obtained in the compact theory? For any three-
manifold M , there is a Lagrangian submanifold L ⊂ M corresponding to the semi-
classical “state” M . This manifold L is simply defined as the set of flat connections
on T 2 that can extend to a flat connection on all of M . It is the so-called G-
character variety of M and can be described by a set of polynomial equations in
the eigenvalues i and mi :
(3.37) L : Aj (, m) = 0 .
Depending on whether we restrict to , m ∈ S 1 or , m ∈ C∗ , these same equations
describe flat G or GC connections. In the rank-one case, there is just a single equa-
tion, the A-polynomial of the knot complement. Upon quantization, the functions
Aj get mapped to quantum operators
(3.38) A m,
j (, 2πi
q = e ) := OAj ,
where i := Oi = evi and m
i := Omi = eui act on HT 2 as
(3.39) i Z(u) = Z(ui + ) (shifting only ui ) , i Z(u) = eui Z(u) .
m
N (K, q) = JN +1 (K, q)
In terms of the colored Jones polynomial JN (K, q), this means J
N (K, q) = q
and mJ N/2
JN (K, q). The wavefunction Z(M ; u) must satisfy [16, 8]
(3.40) j Z(M ; u) = 0
A ∀j ,
which leads to a set of recursion relations on polynomial invariants of the knot
K. In the mathematical literature, such a recursion relation for the colored Jones
polynomial (i.e. in the case of G = SU (2)) is known as the AJ conjecture [13, 14]
(also cf. [15]).
Complex theory. Now, we would like to relate partition functions in Chern-
Simons theory with compact gauge group G to Chern-Simons theory with complex
gauge group GC . In the case of complex gauge group, the phase space is M =
Mflat (GC ; T 2 ) = (C∗ )r × (C∗ )r /W, and the full symplectic structure induced by
Chern-Simons theory is
τ τ̃
(3.41) ω = ω0 + ω0 ,
2 2

with ω0 = 4 T 2 Tr(δA∧δA) as in (3.34). The connection A now takes values in gC ,
1
and a priori there are two independent coupling constants τ and τ̃ . These are the
analog of the level k in the compact theory; we include them here in the definition
of ω. Since M is noncompact, the quantization condition (3.15) is less restrictive,
only fixing τ + τ̃ ∈ Z.
The noncompactness of M changes the nature of the Hilbert space H — as in
the case of the harmonic oscillator, it is no longer finite-dimensional. Choosing a
polarization ∂/∂v = 0, we can effectively take H to consist of Weyl-invariant square-
integrable functions f (u, ū) ∈ L2 ((C∗ )r ). However, the fact that (3.41) is a simple
sum of holomorphic and antiholomorphic pieces means that at a perturbative level
any wavefunction Z(M ; u) ∈ H will factorize into holomorphic and antiholomorphic
components. Put more concretely, the exact wavefunction Z(M ; u, ū; = 2πi ˜
τ , =
2πi
τ̃ ) corresponding to complex Chern-Simons theory on the knot complement M
can be written as [8, 39]
ᾱ
(3.42) ˜ =
Z(M ; u, ū; , ) α
nα,ᾱ Zpert ˜ ,
(M ; u; ) Z pert (M ; ū; )
α,ᾱ
23
for some coefficients nα,ᾱ , where, as → 0, each component Zpert α

(M ; u) can be
expressed as a perturbative series
(3.43)
1 δ 2πi
α
Zpert (M ; u) = exp − S0 (u) + log + S1 (u) + S1 (u) + . . . , = .
2 τ
α
Each partition function Zpert (M ; u; ) corresponds to complex Chern-Simons
theory in the background of a fixed flat connection on M that has meridian holo-
nomy eigenvalues m = eu . The set of such connections, labelled by α, is nothing
but the (finite) set of solutions {v α (u)} (mod 2πi) to the equations
(3.44) Aj (, m) = 0
at fixed m = e . In the case of SL(2, C) theory, one of these flat connections is the
u
geometric one, corresponding to a hyperbolic metric on M .

Since the complex phase space M is just the complexification of the phase
space of the compact theory, the quantization of the functions Aj (, m) is formally
identical to the quantization in the compact case. In other words, the operators
OA j = A j (,
ˆ m̂, q = e2 ) are identical to those of the compact theory. Every
component Zpert α
(M ; u; ) must therefore satisfy [16, 8]
(3.45) A m,
j (, e2 ) Zpert
α
(M ; u; ) = 0 ∀ j, α ,
with i Z(M ; u; ) = Z(M ; ui + ; ) (in other words vi = ∂ui ) and m
i Z(M ; u; ) =
uui Z(M ; u; ). In particular, at leading order in , we can write

1
(3.46) Zpert (M ; u; ) = exp −
α
θ + ... ,
γα

where θ ∼ − i vi dui is a Liouville 1-form and γα is a path on the complex variety
α
L = {Aj = 0} ending at the point (ev (u) , eu ), as in Figure 2b. Now that u
is a continuous parameter in the complex theory, this integral expression makes
complete sense.
3.3.2. Synthesis. It is fairly clear from the above discussion of quantization that
there should be a relation between the partition function for Chern-Simons theory
with compact gauge group G and the partition function for Chern-Simons theory
with complex gauge group GC . Essentially the same equations (3.40) and (3.45)
define the two partition functions — though in one case they are difference equations
and in the other they are differential equations. This relation was developed in
[16, 8], and was recently explained very concretely in [39] in terms of analytic
continuation.
Algebraically, there may be several solutions to the difference equations (3.40)
α
of the compact theory. Let us label them as ZG (M ; u). The exact partition function
of the compact theory (i.e. the colored Jones polynomial for G = SU (2)) is given
as a linear combination

(3.47) Z G (M ; u; ) = nα ZGα
(M ; u; ) .
α
The → 0 asymptotics of each component in this sum are then governed by the
α
corresponding solution Zpert (M ; u) to the differential equation (3.45), written in
the form (3.43). These are holomorphic pieces of the GC partition function. The
physical statement of the volume conjecture for SU (2) is that the component of
the sum (3.47) with the dominant leading asymptotics corresponds to the SL(2, C)
64
α=hyp
partition function Zpert (M ; u) around the hyperbolic SL(2, C) flat connection.
Of all the flat SL(2, C) connections, this has the largest volume in a neigborhood
hyp
of the complete hyperbolic point u = iπ. Therefore, if the solution ZG (M ; u) of
the difference equations contributes to the colored Jones polynomial in (3.47), it
will have the dominant asymptotic. One must simply assure that
(3.48) Physical volume conjecture : nhyp = 0 .
For higher-rank groups, it is again clear that the overall asymptotics of Z G (M ; u)
will be controlled by the flat GC connection with the largest volume that makes
a corresponding contribution to (3.47). One may expect by comparison to SU (2)
theory that the connection with the largest overall volume (the analog of the hyper-
bolic flat connection) in fact contributes and dominates. This has yet to be explored.
The expansions (3.47) and (3.42) for compact and complex Chern-Simons the-
ory, and the relation between them, were explained in [39] using analytic contin-
uation of the Chern-Simons path integral. The path integral provides yet another
method for quantizing a topological quantum field theory, with its own inherent
advantages. Let us finish by saying a few words about this.
The path integral for compact Chern-Simons theory takes the form

ik
(3.49) Z (M ; u; ) = DA(u) exp
G
ICS (A)
4π

1
= DA(u) exp − ICS (A) ,
4

where ICS (A) = M Tr AdA + 3 A is the Chern-Simons action as in (1.9) and
2 3
k ∈ Z. The integral is over all G-connections on M = M \K, modulo gauge

u
equivalence, with fixed holonomy eigenvalues e at the meridian of K. (In order to
obtain a nonzero answer, u/ ∼ λ∗ + ρ∗ must be integral.) For g-values connections
A, the action ICS (A) is real. Therefore, for k ∈ Z, the integral (3.49) is oscillatory
and can be calculated by appropriately regulating the oscillations as A → ∞. In
[39], however, the problem was posed of analytically continuing to k ∈ C. Roughly
speaking, to accomplish this one must also complexify the gauge connection A
so that it is gC -valued. When k ∈ R, the integral (3.49) is then interpreted as
a holomorphic contour integral along the real subspace in the space of complex
connections. As k is pushed away from the real line, this integration contour must
also move. In general, the appropriate integration contour for k ∈ C is a sum of
contours going through the various saddle points of the complexified action ICS (A).
Each saddle point is a flat GC connection, and an expression of the form (3.47)
results.
For complex Chern-Simons theory, the procedure is quite similar. The path
integral is

(3.50) ˜ = DA(u) DA(ū) exp iτ ICS (A) + iτ̃ ICS (A)

Z GC (M ; u; , )
8π 8π

1 1
= DA(u) DA(ū) exp − ICS (A) − ICS (A) ,
4 ˜
4
for a gC -valued GC -connection A. The integrand is initially well-defined when
τ + τ̃ ∈ Z, and leads to a convergent oscillatory integral when the exponent is
25
imaginary — i.e. for τ̃ = τ̄ . In order to analytically continue to independent

τ, τ̃ ∈ C, one must treat A and A as independent connections and again complexify
each of them. One then deforms the contour of integration away from the “real”
subspace when τ̃ = τ̄ , and writes the resulting contour as a sum over pairs of saddle
points for A and A. Since (gC )C gC × gC , however, these are just pairs of saddle
points of flat GC -connections. An expression of the form (3.42) results:

(3.51) ˜ =
Z GC (M ; u; , ) α
nα,ᾱ ZG (M ; u; )ZG ˜ .
α (M ; ū; )
C C
α,ᾱ
The functions α
ZG C
(M ; u; ) and ZGα
(M ; u; ) here and in (3.47) should be identical,
since they both correspond to GC connections.
In [39], it is explained how the coefficients nα and nα,ᾱ may be calculated for
specific examples, like the trefoil and figure-eight knot complements. As expected,
the coefficient of the hyperbolic component “α = hyp” of the SU (2) partition
function is nonzero, leading to another demonstration of the volume conjecture.
The careful reader may still be wondering why it is only the growth of the
colored Jones polynomial at nonrational N/k that shows exponential behavior.
The answer comes from a final subtlety in the analytic continuation of the path
integral: for k ∈ / Z, the sum (3.47) can have multiple contributions from the same
flat connection, differing by a multiplicative factor e2πik . (If analytically continuing
2πiu
in N as well,
ik factors of e
may also arise.) This behavior originates from the fact
that exp 4π ICS ) is not completely gauge-invariant when k ∈ / Z. For example, in
the case of the figure-eight knot, the actual hyperbolic contribution to (3.47) goes
like
iπk
(3.52) e − e−iπk )ZG hyp
C
(M ; u; ) ,
which vanishes at k ∈ Z, leading to polynomial rather than exponential growth of
JN (K; q) for u/iπ ∼ N/k ∈ Q. It is expected that this feature is fairly generic for
hyperbolic knots.
Acknowledgements
We would like to thank Edward Witten, Don Zagier, and Jonatan Lenells for en-
lightening discussions on subjects considered in these notes. We would also like to
thank the organizers of the workshop Interactions Between Hyperbolic Geometry,
Quantum Topology, and Knot Theory and Columbia University for their generous
support, accommodations, and collaborative working environment.
References
1. Michael Atiyah, The geometry and physics of knots, Lezioni Lincee. [Lincei Lectures], Cam-
bridge University Press, Cambridge, 1990. MR 1078014 (92b:57008)
2. Scott Axelrod, Steve Della Pietra, and Edward Witten, Geometric quantization of Chern-
Simons gauge theory, J. Diff. Geom. 33 (1991), 787–902.
3. D Bar-Natan and E Witten, Perturbative expansion of Chern-Simons theory with non-compact
gauge group, Comm. Math. Phys. 141 (1991), no. 2, 423–440.
4. F Bayen, M Flato, C Fronsdal, A Lichnerowicz, and D Sternheimer, Deformation theory
and quantization. I. deformations of symplectic structures, Ann. Physics 111 (1978), no. 1,
61–110.
5. Sergio Cecotti and Cumrun Vafa, Topological antitopological fusion, Nucl. Phys. B367 (1991),
359–461.
66
6. D Cooper, M Culler, H Gillet, and D Long, Plane curves associated to character varieties of
3-manifolds, Invent. Math. 118 (1994), no. 1, 47–84.
7. D Coulson, OA Goodman, CD Hodgson, and WD Neumann, Computing arithmetic invariants
of 3-manifolds, Experimental Mathematics 9 (2000), no. 1, 127–152.
8. Tudor Dimofte, Sergei Gukov, Jonatan Lenells, and Don Zagier, Exact results for perturbative
Chern-Simons theory with complex gauge group, Comm. Num. Thy. and Phys. 3 (2009), no. 2,
363–443, arXiv: 0903.2472v1 [math.QA].
9. B. Dubrovin, Integrable systems in topological field theory, Nucl. Phys. B379 (1992), 627–689.
10. S Elitzur, G Moore, A Schwimmer, and N Seiberg, Remarks on the canonical quantization of
the Chern-Simons-Witten theory, Nucl. Phys. B326 (1989), 108.
11. L. D Faddeev and R. M Kashaev, Quantum dilogarithm, Modern Phys. Lett. A9 (1994), no. 5,
427–434, arXiv: hep-th/9310070v1.
12. Daniel S Freed, Michael J Hopkins, Jacob Lurie, and Constantin Teleman, Topological quan-
tum field theories from compact lie groups, ”A Celebration of the Mathematical Legacy of
Raoul Bott” (2009), 367–403, arXiv: 0905.0731v2 [math.AT].
13. Stavros Garoufalidis, On the characteristic and deformation varieties of a knot, Geometry
and Topology Monographs 7 (2004), 291–304, arXiv: math/0306230v4.
14. Stavros Garoufalidis and Thang TQ Le, The colored jones function is q-holonomic, Geom.
Topol. 9 (2005), 1253–1293, arXiv: math/0309214v3.
15. Stavros Garoufalidis and Xinyu Sun, The non-commutative a-polynomial of twist knots,
(2008), arXiv: 0802.4074v1 [math.GT].
16. Sergei Gukov, Three-dimensional quantum gravity, Chern-Simons theory, and the a-
polynomial, Commun. Math. Phys. 255 (2005), 577–627, arXiv: hep-th/0306165v1.
17. Sergei Gukov and Hitoshi Murakami, SL(2,c) Chern-Simons theory and the asymptotic be-
havior of the colored Jones polynomial, (2006), arXiv: math/0608324v2.
18. Sergei Gukov and Edward Witten, Branes and quantization, (2008), arXiv: 0809.0305v2 [hep-
th].
19. Kazuo Habiro, On the colored jones polynomials of some simple links, Surikaisekikenkyusho
Kokyuroku 1172 (2000), 34–43.
20. Hugh M Hilden, Maria Teresa Lozano, and Jose Maria Montesinos-Amilibia, On volumes and
Chern-Simons invariants of geometric 3-manifolds, J. Math. Sci. Univ. Tokyo 1996 (3), no. 3,
723–744.
21. R. M Kashaev, The hyperbolic volume of knots from quantum dilogarithm, Lett. Math. Phys.
39 (1997), 269–265, arXiv: q-alg/9601025v2.
22. R Kirby and P Melvin, The 3-manifold invariants of Witten and Reshetikhin-Turaev for
sl(2,c), Invent. Math. 105 (1991), no. 1, 1432–1297.
23. Maxim Kontsevich, Deformation quantization of Poisson manifolds, Lett. Math. Phys. 66
(2003), no. 3, 157–216, arXiv: q-alg/9709040v1.
24. Hitoshi Murakami, Some limits of the colored Jones polynomials of the figure-eight knot,
Kyungpook Math. J. 44 (2004), no. 3, 369–383, arXiv: math/0308002v1.
25. , The colored jones polynomials and the alexander polynomial of the figure-eight knot,
J. Geom. Topol. 7 (2007), no. 2, 249–269, arXiv: math/0502428v1.
26. Hitoshi Murakami and Jun Murakami, The colored Jones polynomials and the simplicial
volume of a knot, Acta Math. 186 (2001), no. 1, 85–104, arXiv: math/9905075v2.
27. Hitoshi Murakami, Jun Murakami, Miyuki Okamoto, Toshie Takata, and Yoshiyuki Yokota,
Kashaev’s conjecture and the Chern-Simons invariants of knots and links, Experiment. Math.
11 (2002), no. 3, 427–435, arXiv: math/0203119v2.
28. Hitoshi Murakami and Yoshiyuki Yokota, The colored Jones polynomials of the figure-
eight knot and its dehn surgery spaces, J. Reine Angew. Math 607 (2007), 47–68, arXiv:
math/0401084v3.
29. W D Neumann and Don Zagier, Volumes of hyperbolic three-manifolds, Topology 24 (1985),
no. 3, 307–332.
30. Walter D Neumann, Extended Bloch group and the Cheeger-Chern-Simons class, Geom.
Topol. 8 (2004), 413–474, arXiv: math/0307092v2.
31. Grisha Perelman, Ricci flow with surgery on three-manifolds, (2003), arXiv: math/0303109v1.
32. Joan Porti, Torsion de Reidemeister pour les variétés hyperboliques, Mem. Amer. Math. Soc.
128 (1997), no. 612, x+139. MR 1396960 (98g:57034)
27
33. N Reshetikhin and V Turaev, Invariants of 3-manifolds via link polynomials and quantum
groups, Invent. Math. 103 (1991), no. 3, 547–597.
34. William P Thurston, Three dimensional manifolds, kleinian groups, and hyperbolic geometry,
Bull. AMS 6 (1982), no. 3, 357–381.
35. David A. Vogan, Jr., Singular unitary representations, Noncommutative harmonic analysis
and Lie groups (Marseille, 1980) (J. Carmona and M. Vergne, eds.), Lecture Notes in Math.,
vol. 880, Springer, Berlin, 1981, pp. 506–535. MR MR644845 (83k:22036)
36. E Witten, 2+1 dimensional gravity as an exactly soluble system, Nucl. Phys. B311 (1988),
46.
37. , Quantization of Chern-Simons gauge theory with complex gauge group, Commun.
Math. Phys. 137 (1991), 29.
38. Edward Witten, Quantum field theory and the jones polynomial, Comm. Math. Phys. 121
(1989), no. 3, 351–399.
39. , Analytic continuation of Chern-Simons theory, (2010), arXiv: 1001.2933 [hep-th].
40. N. M. J. Woodhouse, Geometric quantization, second ed., Oxford Mathematical Monographs,
The Clarendon Press Oxford University Press, New York, 1992, Oxford Science Publications.
MR 1183739 (94a:58082)
41. T Yoshida, The eta-invariant of hyperbolic 3-manifolds, Invent. Math. 81 (1985), 473–514.
42. Christian K Zickert, The Chern-Simons invariant of a representation, (2007), arXiv:
0710.2049v1 [math.GT].
California Institute of Technology 452-48, Pasadena, CA 91125

E-mail address: tdd@theory.caltech.edu
California Institute of Technology 452-48, Pasadena, CA 91125

E-mail address: gukov@theory.caltech.edu
Volume 541, 2011
R-matrix knot invariants and triangulations
R. M. Kashaev
Abstract. The construction of quantum knot invariants from solutions of the

Yang–Baxter equation (R-matrices) is reviewed with the emphasis on a class
of R-matrices admitting an interpretation in intrinsically three-dimensional
terms.
Introduction
The Yang–Baxter equation [Y, B] is the central tool for studying exactly solv-
able models of quantum field theory and statistical mechanics [B, F] and in the
theory of quantum groups [D]. Its significance in knot theory and topology of
three-manifolds comes from the fact that its solutions, called R-matrices, are the
principal ingredients in the constructions of quantum knot invariants [J, T1].
In this paper, we review the construction of quantum knot invariants by using
R-matrices, putting the emphasis on a particular class of R-matrices which admit an
interpretation in intrinsically three-dimensional terms. We show that for any finite
dimensional Hopf algebra, there exists a canonical R-matrix such that the associated
knot invariant can be given an interpretation in terms of the combinatorics of
ideal triangulations of knot complements. On the algebraic level, our construction
corresponds to a canonical algebra homomorphism of the Drinfeld double of a Hopf
algebra into the tensor product of the Heisenberg doubles of the same Hopf algebra
and its dual Hopf algebra [Kas].
The organization of the paper is as follows. In Section 1, we review the con-
struction of knot invariants by using R-matrices. From one hand side, following the
modern tendency, instead of finite dimensional vector spaces, we work in a more
general framework of monoidal categories. On the other hand side, unlike the com-
mon practice of starting from a quasi-triangular Hopf algebra or (and) a braided
monoidal category of its representations, we start from a particular R-matrix with-
out precising its algebraic origin. In order to be able to construct a knot invariant,
we need only one extra condition on the R-matrix which we call rigidity. In the case
of an R-matrix acting in a finite dimensional vector space, the rigidity is nothing
else but the invertibility of the partially transposed matrix. Such approach first
was suggested by N. Reshetikhin in [R]. In our mind, it is a direct and economical
way of constructing a knot invariant from a given R-matrix, without preliminary
1991 Mathematics Subject Classification. Primary 57M27, 16T05; Secondary 57R56, 46L85.
Key words and phrases. Low-dimensional topology, Quantum topology.
The work is supported in part by the Swiss National Science Foundation.
1
2011
c 0000
c Mathematical
holder)
69
70
2 R. M. KASHAEV
study of the underlying quasi-triangular Hopf algebra and its representation theory.
This approach is useful in the case of R-matrices constructed without using rep-
resentation theory of quantum groups, for example, some of the two-dimensional
R-matrices in the classification list of J. Hietarinta [Hi]. In Section 2, we intro-
duce the notion of a rigid T -matrix and describe a canonical construction of a
rigid R-matrix from a rigid T -matrix. Together with the result of the previous
section, we are able to associate a knot invariant to any rigid T -matrix. Finally,
in Section 3, we associate to any tangle diagram a 3-dimensional Δ-complex in the
sense of A. Hatcher [Ha], and interpret the knot invariants associated with finite
dimensional Hopf algebras in terms of state sums over 3-dimensional Δ-complexes.
Acknowledgements. The author would like to thank N. Reshetikhin and
A. Virelizier for interesting discussions.
1. R-matrices and knot invariants

1.1. Notation. In this paper, the term monoidal category means strict monoidal
category. For notational simplification, we shall suppress the tensor product symbol
but always keep the usual composition symbol ◦, and to avoid writing many paren-
theses, we shall assume that the tensor product takes precedence over the usual
composition of morphisms. We shall also write X instead of idX , provided the
context permits to avoid confusion. The price paid for this notation is that powers
of an endomorphism f : X → X are now preceded by the composition symbol:
f ◦k = f ◦ f ◦ · · · ◦ f , k ∈ Z≥0 .

k times
1.2. Dual objects in monoidal categories. In a monoidal category, a du-

ality is a quadruple X, Y, η, consisting of two objects X, Y and two morphisms
η : I → Y X, : XY → I
such that
(1.1) X ◦ Xη = X, Y ◦ ηY = Y.
In this situation, X is called left dual of Y , and Y is called right dual of X. An
object X is called rigid if it is a right and a left dual. A rigid object is called
strongly rigid (or s-rigid ) if its left dual is isomorphic to its right dual.
For two dualities p = X, Y, η, , q = U, V, η , , we define their product
pq = XU, V Y, V ηU ◦ η , ◦ X Y .
For a given duality p = X, Y, η, and any objects U and V , there are canonical
bijections
ap : Hom(XU, V ) → Hom(U, Y V ), bp : Hom(U Y, V ) → Hom(U, V X)
defined by the formulae
ap (f ) = Y f ◦ ηU, bp (g) = gX ◦ U η, f ∈ Hom(XU, V ), g ∈ Hom(U Y, V ),
their inverses being
a−1
p (f ) = V ◦ Xf, b−1
p (g) = V ◦ gY, f ∈ Hom(U, Y V ), g ∈ Hom(U, V X).
We shall use particular compositions of these bijections
lp = ap ◦ b−1
p : Hom(XU, V X) → Hom(U Y, Y V )
R-MATRIX KNOT INVARIANTS AND TRIANGULATIONS 71
3
and
rp = lp−1 = bp ◦ a−1
p : Hom(U Y, Y V ) → Hom(XU, V X).
Definition 1.1. A morphism f : XU → V X (respectively f : U X → XV )

of a monoidal category is called X-left (respectively X-right) isomorphism, if X
is a left (respectively right) dual, and for a duality p = X, Y, η, (respectively
p = Y, X, η, ) the morphism lp (f ) (respectively rp (f )) is an isomorphism.
Lemma 1.2. Let X be a left dual and Y a right dual objects of a monoidal
category. Then, a morphism f : XY → Y X is X-left isomorphism if and only if it
is Y -right isomorphism.
Proof. Choose dualities p = X, U, η, and q = V, Y, η , . If f : XY → Y X

is a X-left isomorphism, i.e. lp (f ) is invertible, then rpq ((lp (f ))◦−1 ) is inverse of
rq (f ), i.e. f is a Y -right isomorphism. Similarly, if f is a Y -right isomorphism,
i.e. rq (f ) is invertible, then lqp ((rq (f ))◦−1 ) is inverse of lp (f ), i.e. f is a X-left
isomorphism.
1.3. Rigid R-matrices.
Definition 1.3. An R-matrix is a triple (C, X, ρ) given by a monoidal cate-

gory C, an object X in C, and an isomorphism ρ : X 2 → X 2 satisfying the Yang–
Baxter equation
ρX ◦ Xρ ◦ ρX = Xρ ◦ ρX ◦ Xρ.
The morphism ρ itself will also be informally called R-matrix, if the underlying
triple is clear from the context.
Lemma 1.4. Let (C, X, ρ) be an R-matrix. If ρ◦±1 are X-left or X-right iso-
morphisms, then X is a s-rigid object.
Proof. Assume that ρ◦±1 are X-left isomorphisms. Choosing a duality p =

X, Y, η, , the morphisms
η̄ = ρ◦−1
− ◦ η : I → XY, ¯ = ◦ ρ◦−1
+ : Y X → I, ρ± = lp (ρ◦±1 ),
satisfy the identities

¯Y ◦ Y η̄ = Y, ◦ η̄X = X
X¯
which mean that we have another duality
p̄ = Y, X, η̄, ¯,
i.e. X is a right dual of Y , and thus X is s-rigid. The case when ρ◦±1 are X-right
isomorphisms is similar.
Definition 1.5. An R-matrix (C, X, ρ) is called rigid if ρ◦±1 are X-left iso-
morphisms (and by Lemmas 1.2 and 1.4 they are also X-right isomorphisms).
Remark 1.6. The rigidity of an R-matrix (C, X, ρ), where C is the category
of finite dimensional vector spaces, is equivalent to its non-degenerateness in the
sense of [R].
72
4 R. M. KASHAEV
1.4. The category of directed planar ribbon tangles. A directed tangle

is a smooth embedding f : (M, ∂M ) → (R2 × [0, 1], R2 × ∂[0, 1]) where M is a
compact oriented one-dimensional manifold. A directed tangle diagram is the image
of a directed tangle under the projection R2 × [0, 1] → R × [0, 1], (x, y, t) → (x, t),
provided the tangle is in general position with respect to the projection, and with
the additional information of over- and under-crossings as with knot diagrams. A
connected component of a directed tangle diagram is the image of a connected
component of the corresponding directed tangle. A directed planar ribbon tangle,
to be referred later as dpr-tangle, is an equivalence class of directed tangle diagrams
with respect to the equivalence relation generated by isotopies of the strip R × [0, 1]
and the oriented Reidemeister moves of types II and III. In other words, a dpr-tangle
is a directed tangle diagram considered up to regular isotopies [Kau].
The set of dpr-tangles forms the set of morphisms of a monoidal category Tdpr
defined in a similar way as for ordinary directed tangles [T2]. The set of objects
Ob Tdpr are finite sequences (including the empty one) of pluses and minuses. For
a dpr-tangle x, the elements of the sequence dom(x) (respectively cod(x)) are in
bijective correspondence with the points of the intersection set x∩(R×{0}) (respec-
tively x∩(R×{1})) with the order induced from that of R, and for each intersection
point the associated sign is plus, if the point is the image of the initial (respectively
terminal) point of the corresponding oriented segment of the directed tangle, and
minus otherwise. For two composable dpr-tangles x and y (i.e. satisfying the
condition cod(x) = dom(y)) their composition is obtained by choosing representa-
tives of x and y so that the projection images in the real line of the intersections
x ∩ (R × {1}) and y ∩ (R × {0}) coincide, and then by identifying the boundary
component R × {1} of the strip containing x with the boundary component R × {0}
of the strip containing y, thus forming a new strip containing a representative of
the composed dpr-tangle y ◦ x. The tensor product xy of two dpr-tangles x and y
is obtained by choosing representatives of x and y so that the representative of x
is located to the left of the representative of y, and then by taking their union as a
representative of xy.
As a monoidal category, the category Tdpr is generated by the following dpr-
tangles:
+ − + + + +
∅= , ↑= , ↓= , = , = ,
+ − + + + +
+ − − +
= , = , = , = .
− + + −
For any dpr-tangle x, there are two functions on the set of its connected components
wr : π0 (x) → Z, wn : π0 (x) → 2−1 Z,
called respectively writhe and winding number. For a connected dpr-tangle x, these
functions are calculated according to the formulae
wr(x) = #() − #(), 2 wn(x) = #( ) + #() − #() − #(),
5
where #(y) denotes the number of generating elements of type y in a given decom-
position of a diagram representing x.
Remark 1.7. The usual category of framed directed tangles is a quotient cat-
egory of Tdpr obtained by adding the Reidemeister move of type I .
1.5. Knot invariants from rigid R-matrices. The following theorem is
essentially a reformulation of a result of N. Reshetikhin (Theorem 1.1 of [R]).
Theorem 1.8. Let (C, X, ρ) be a rigid R-matrix. Fix a duality p = Y, X, η, .
Then, for any integer k there exists a unique functor Φρ,k : Tdpr → C such that
→ ρ, → η, → , → η̄, → ¯,
where
η̄ = Y ω ◦k ◦ ρ◦−1
− ◦ η, ¯ = ◦ ρ◦−1
+ ◦ ω ◦−k Y, ρ± = rp (ρ◦±1 ),
ω = ( ◦ ρ◦−1 ◦−1
− )X ◦ X(ρ− ◦ η).
If, furthermore, there exists a morphism ν : X → X such that
Xν ◦ ρ◦±1 = ρ◦±1 ◦ νX, ω = ν ◦2 ,
then the parameter k can take half-integer values, and the functor Φ−1/2,ρ factors
through the category of framed directed tangles.
Proof. The proof goes along a similar line of arguments as the proof of The-
orem 1.1 of [R].
Theorem 1.9. Let x be a connected dpr-tangle with dom(x) = cod(x) = (+).

Then, the morphism
(1.2) Ψρ (x) = ω ◦−(wr(x)+(1+2k) wn(x))/2 ◦ Φρ,k (x) : X → X
is independent of k and invariant under all Reidemeister moves of type I.
Proof. The statement directly follows from Theorem 1.8 and the formulae
⎛ ⎞ ⎛ ⎞
Φρ,k ⎝ ⎠ = ω ◦−k , Φρ,k ⎝ ⎠ = ω ◦−1−k ,
⎛ ⎞ ⎛ ⎞
Φρ,k ⎝ ⎠ = ω ◦1+k , Φρ,k ⎝ ⎠ = ω ◦k .
Notice that for such x the sum wr(x) + wn(x) is always even, so that the power of
ω in (1.2) is an integer.
If we define a knot as a (1, 1)-tangle, then Theorem 1.9 implies that function
Ψρ (x) is an invariant of the knot represented by x.
2. R-matrices from T -matrices

In this section, we define rigid T -matrices and, following the ideas of [Kas],
describe a canonical construction of rigid R-matrices from rigid T -matrices.
74
6 R. M. KASHAEV
2.1. Rigid T -matrices.

Definition 2.1. A T -matrix is a quadruple (C, X, σ, τ ) given by an involutive
R-matrix (C, X, σ), i.e. satisfying the condition σ = σ ◦−1 , and an isomorphism
τ : X 2 → X 2 satisfying the equations
(2.1) τ X ◦ Xσ ◦ σX = Xσ ◦ σX ◦ Xτ,
(2.2) τ X ◦ Xσ ◦ τ X = Xτ ◦ τ X ◦ Xτ.
As in the case of R-matrices, the morphism τ itself will informally be called
T -matrix, if the underlying quadruple is clear from the context.
Remark 2.2. It is easily verified that if (C, X, σ, τ ) is a T -matrix, then (C, X, σ, τ ◦−1)
is also a T -matrix. This symmetry will allow us to simplify some proofs below.
Definition 2.3. A T -matrix (C, X, σ, τ ) is called rigid if (C, X, σ) is a rigid
R-matrix and τ ◦±1 are X-left isomorphisms (and, by Lemma 1.2, also X-right
isomorphisms).
Lemma 2.4. For a rigid T -matrix (C, X, σ, τ ) and a duality p = Y, X, η, ,
the morphisms (rp (τ ◦±1 ))◦−1 are X-left and Y -right isomorphisms.
Proof. Due to Lemma 1.4 and Remark 2.2, it is enough to prove that τ+ =
(rp (τ ))◦−1 is a Y -right isomorphism. Defining another duality
q = X, Y, η̄ = (rp (σ))◦−1 ◦ η, ¯ = ◦ (rp (σ))◦−1 ,
the morphisms
α = a−1
q (lq (τ )X ◦ Xτ+ ◦ σY ◦ Xrpq (τ+ ) ◦ (rp (τ ) ◦ η̄)X)
and
β = ap (( ◦ lq (τ ))X ◦ Xτ+ ◦ σY ◦ Xrpq (τ+ ) ◦ rp (τ )X)
are such that
β ◦ rq (τ+ ) = rq (τ+ ) ◦ α = X 2
which, by considering the product β ◦ rq (τ+ ) ◦ α, imply that α = β = (rq (τ+ ))◦−1 .
Thus, τ+ is a Y -right isomorphism.
2.2. Rigid R-matrices from rigid T -matrices.
Theorem 2.5. For a rigid T -matrix (C, X, σ, τ ) and a duality p = Y, X, η, ,
the triple
(C, XY, ρτ ), ρτ = X(rp (τ ))◦−1 Y ◦ τ rp (rp (τ )) ◦ Xrp (τ ◦−1 )Y,
is a rigid R-matrix with self-dual object XY .
Proof. For the inverse of ρτ , we have the formula
ρ◦−1
τ = ρτ ◦−1 ,
while the Yang–Baxter equation
ρτ XY ◦ XY ρτ ◦ ρτ XY = XY ρτ ◦ ρτ XY ◦ XY ρτ
is a consequence of eight Pentagon relations
τ X ◦ Xσ ◦ τ X = Xτ ◦ τ X ◦ Xτ, τ Y ◦ Y τ ◦ τ Y = Y τ ◦ σ Y ◦ Y τ ,
τ Y ◦ X σ̂ ◦ τ̂ X = X τ̂ ◦ τ̂ X ◦ Y τ, τ X ◦ Y τ̌ ◦ τ̌ Y = Y τ̌ ◦ σ̌Y ◦ Xτ ,
7
τ̂ X ◦ Y τ ◦ τ̌ X = X τ̌ ◦ σY ◦ X τ̂ , τ̌ Y ◦ Xσ ◦ τ̂ Y = Y τ̂ ◦ τ X ◦ Y τ̌ ,
τ̌ X ◦ X σ̌ ◦ τ Y = Y τ ◦ τ̌ X ◦ X τ̌ , τ̂ Y ◦ Y τ̂ ◦ τ X = Xτ ◦ σ̂Y ◦ Y τ̂
where we use the notation
τ̂ = rp (τ ◦−1 ), τ̌ = (rp (τ ))◦−1 , τ = rp (rp (τ )),
σ̂ = rp (σ), σ̌ = σ̂ ◦−1 , σ = rp (rp (σ)).

In fact, all of these Pentagon relations are equivalent to each other. If we define
another duality q = X, Y, σ̌ ◦ η, ◦ σ̌, then the object XY is selfdual as it enters
the duality
qp = XY, XY, X(σ̌ ◦ η)Y ◦ η, ◦ σ̌ ◦ XY .
Rigidity of ρτ is equivalent to invertibility of the morphisms rqp (ρ◦±1
τ ). We have
explicitly
rqp (ρτ ) = Xlq (τ )Y ◦ rq (τ̌ )(τ )◦−1 ◦ Xrp (τ )Y,
rqp (ρ◦−1
τ ) = Xlq (τ ◦−1 )Y ◦ rq (τ̂ ◦−1 )τ ◦ X τ̂ Y
which are invertible by Lemma 2.4.
2.3. T -matrices from Hopf objects. Recall that a symmetric monoidal cat-
egory is a monoidal category C with a natural isomorphism s : ⊗ → ⊗op satisfying
the equations
sX,Y Z = Y sX,Z ◦ sX,Y Z, sXY,Z = sX,Z Y ◦ XsY,Z , sY,X sX,Y = XY.
Definition 2.6. A Hopf object in a symmetric monoidal category C is a six-
tuple (X, ∇, Δ, η, , γ) consisting of an object X of C and five morphisms: the
product ∇ : X 2 → X, the co-product Δ : X → X 2 , the unit η : I → X, the co-unit
: X → I, and the antipode γ : X → X such that
∇ ◦ ∇X = ∇ ◦ X∇, ∇ ◦ Xη = ∇ ◦ ηX = X,
i.e. the triple (X, ∇, η) is a monoid in C,
XΔ ◦ Δ = ΔX ◦ Δ, X ◦ Δ = X ◦ Δ = X,
i.e. the triple (X, Δ, ) is a co-monoid in C,
◦ ∇ = 2 , Δ ◦ η = η2 , ◦ η = I,
∇ ◦ Xγ ◦ Δ = ∇ ◦ γX ◦ Δ = η ◦ ,
and
Δ ◦ ∇ = ∇2 ◦ XsX,X X ◦ Δ2 .
Proposition 2.7. Let (X, ∇, Δ, η, , γ) be a Hopf object in a symmetric monoidal
category C. Then, the quadruple
(2.3) (C, X, σ, τ ), σ = sX,X , τ = σ ◦ X∇ ◦ ΔX
is a T -matrix. Moreover, if X is a left or right dual, and the antipode is invertible,
then this T -matrix is rigid.
76
8 R. M. KASHAEV
Proof. Verification of the equations (2.1) and (2.2) is straightforward. As-

sume that X is a left or right dual, and the antipode is invertible. Then, it is easily
seen that the involutive R-matrix (C, X, σ) is rigid so that X is s-rigid. Fixing a
duality p = Y, X, η, , we also have another duality
q = X, Y, sX,Y ◦ η, ◦ sX,Y .
◦±1
Now, τ are X-right isomorphisms due to the identities
(rp (τ ))◦−1 = lq (τ−1 ◦ σ), (rp (τ ◦−1 ))◦−1 = lq (σ ◦ τ2 )
where
τk = X∇ ◦ Xγ ◦k X ◦ ΔX, k ∈ Z.

3. Knot invariants and Δ-complexes

The results of the previous sections permit us to lift certain knot invariants,
associated to finite dimensional Hopf algebras, into the combinatorial setting of
ideal triangulations of knot complements. Such a lifting is based on the geometrical
interpretation of the T -matrices in terms of tetrahedra.
3.1. Δ-complexes. Let

Δn = {(t0 , t1 , . . . , tn ) ∈ [0, 1]n+1 | t0 + t1 + · · · + tn = 1}, n ≥ 0,
be the standard n-simplex with the face inclusion maps
δm : Δn → Δn+1 , 0 ≤ m ≤ n + 1,
defined by the formula
δm (t0 , . . . , tn ) = (t0 , . . . , tm−1 , 0, tm , . . . , tn ).
A (simplicial) cell in a topological space X is a continuous map
f : Δn → X
such that the restriction of f to the interior of Δn is an embedding. On the set of
cells Σ(X) we have the dimension function
d : Σ(X) → Z, (f : Δn → X) → n.
Following [Ha], we define a Δ-complex structure on a topological space X as a pair
(Δ(X), ∂), where Δ(X) ⊂ Σ(X) and ∂ is a set of maps

∂ = ∂n : d|−1
Δ(X) (Z ≥max(1,n) ) → d| −1
Δ(X) (Z ≥n−1 ) n ≥ 0
such that:
(i) each point of X is in the image of exactly one restriction, α|int(Δn ) for
α ∈ Δ(X);
(ii) α ◦ δm = ∂m (α);
(iii) a set A ⊂ X is open iff α−1 (A) is open for each α ∈ Δ(X).
Notice that any Δ-complex is a CW-complex. We denote by Δn (X) = Δ(X) ∩
d−1 (n) the set of n-dimensional cells of a Δ-complex X.
9
Definition 3.1. A combinatorial Δ-complex is a triple (S, d, ∂) consisting of

a set S, a map d : S → Z≥0 , and a family of maps
∂ = {∂n : d−1 (Z≥max(1,n) ) → d−1 (Z≥n−1 )| n ∈ Z≥0 }
such that
∂i (d−1 (n)) ⊂ d−1 (n − 1), 0 ≤ i ≤ n, n ≥ 1,
and
∂i ◦ ∂j+1 = ∂j ◦ ∂i , i≤j.
A morphism between two Δ-complexes (S, d, ∂) and (S , d , ∂ ) is a map of sets
f : S → S such that d ◦ f = d and f ◦ ∂i = ∂i ◦ f , i ≥ 0.
For any Δ-complex X, the triple (Δ(X), d|Δ(X) , ∂) is a combinatorial Δ-complex,

and it is clear that any combinatorial Δ-complex (S, d, ∂) can be realized this way.
Indeed, we define a topological space X as the quotient space X̃/R where

X̃ = Δn × d−1 (n),
n≥0
and R is the equivalence relation generated by the relations

R
(δi (x), α) ∼ (x, ∂i (α)), 0 ≤ i ≤ n, x ∈ Δn−1 , α ∈ d−1 (n), n ≥ 1.
Any element α ∈ d−1 (n) ⊂ S corresponds to an n-cell
fα : Δn → X, fα (x) = [x, α],
and, defining ∂i (fα ) = f∂i (α) , we verify the property (ii):
fα ◦ δi (x) = [δi (x), α] = [x, ∂i (α)] = f∂i (α) (x) = ∂i (fα )(x).
The same space can be constructed as a CW-complex by noting that
∪ni=0 δi (Δn−1 ) = ∂(Δn ), n ≥ 1.
We define the 0-skeleton X 0 to be the discrete space d−1 (0), and for n ≥ 1, we
define recursively the n-skeleton X n as the space obtained by attaching the space
Δn × d−1 (n) on the n − 1-skeleton X n−1 along the boundary
∂(Δn × d−1 (n)) = ∂(Δn ) × d−1 (n)
via the map
φn : ∂(Δn × d−1 (n)) → X n−1 , φn (δi (x), α) = Φn−1 (x, ∂i (α)),
where x ∈ Δn−1 , 0 ≤ i ≤ n, and
Φn−1 : Δn−1 × d−1 (n − 1) → X n−1
is the characteristic map of the n − 1-cells.

78
10 R. M. KASHAEV
3.2. Presentations of finite dimensional Δ-complexes. A combinatorial

Δ-complex (S, d, ∂) is called finite dimensional if there exists an integer n ≥ 0 such
that d−1 (k) = ∅ for all k > n. In this case, the dimension of the complex is defined
as the unique integer dim(S, d, ∂) such that d−1 (dim(S, d, ∂)) = ∅, and d−1 (k) = ∅
for all k > dim(S, d, ∂).
Proposition 3.2. Let A, B be two sets and {fi : A → B| 0 ≤ i ≤ n} a set of

n + 1 maps from A to B. Then, there exists a unique, up to a unique isomorphism
of Δ-complexes, n-dimensional combinatorial Δ-complex (S, d, ∂) such that
(3.1) d−1 (n) = A, d−1 (n − 1) = B, ∂i |d−1 (n) = fi , 0 ≤ i ≤ n,
and it is universal in the sense that for any other n-dimensional combinatorial
Δ-complex (S , d , ∂ ), having the above properties, there exists a unique morphism
f : (S, d, ∂) → (S , d , ∂ )
which is identity on d−1 {n, n − 1}.
Proof. We define d−1 (m) for m ∈ {n, n − 1} and the maps ∂i |d−1 (n) by using
equations (3.1). Then, for m < n − 1, we recursively define
d−1 (m) = (d−1 (m + 1) × {0, . . . , m + 1})/Rm
where the equivalence relation Rm is generated by the relations

R
(∂j+1 (α), i) ∼m (∂i (α), j), 0 ≤ i ≤ j ≤ m + 1, α ∈ d−1 (m + 2).
Denoting by [α, i] the Rm -equivalence class of (α, i), we define
∂i (α) = [α, i], 0 ≤ i ≤ m + 1, α ∈ d−1 (m + 1).
Then, for any α ∈ d−1 (m + 2) and 0 ≤ i ≤ j ≤ m + 1, we verify that
∂i ◦ ∂j+1 (α) = [∂j+1 (α), i] = [∂i (α), j] = ∂j ◦ ∂i (α).
In this way, we come to an n-dimensional combinatorial Δ-complex (S, d, ∂), where

S = nk=0 d−1 (k), which has the required properties.
Let (S , d , ∂ ) be another n-dimensional combinatorial Δ-complex satisfying
the conditions (3.1). Then, we define a map f : S → S by recursion so that the
restriction of f on the subset d−1 {n, n − 1} is the identity, for any 0 ≤ m < n − 1,
we set
f ([α, i]) = ∂i (f (α)), α ∈ d−1 (m + 1), 0 ≤ i ≤ m + 1.
It is clear that f is a morphism of combinatorial Δ-complexes, and that it is a
unique morphism with the required property.
By using this proposition, we will call the triple
A, B, {fi : A → B}0≤i≤n
a presentation for the associated universal n-dimensional combinatorial Δ-complex.

11
3.3. Combinatorial Δ-complexes associated with tangle diagrams.

For an oriented tangle diagram Γ, we denote by Γ04 the set of crossings, Γ01 the
set of boundary points, and Γ1 the set of edges of Γ. We have two maps
dom, cod : Γ1 → Γ01 Γ04 ,
where dom(e) (respectively cod(e)) is the crossing or the boundary point where the
oriented edge e goes out (respectively goes in). We define another pair of maps
ss, st : Γ1 → {0, ±1}
where ss(e) (respectively st(e)) takes the value +1, if the edge e is over-passing at
the crossing dom(e) (respectively at the crossing cod(e)), the value −1, if the edge e
is under-passing at the crossing dom(e) (respectively at the crossing cod(e)), and the
value 0, if dom(e) (respectively cod(e)) is a boundary point. The function ε : Γ04 →
{±1} takes the value +1 on positive crossings and −1 on negative crossings.
To any tangle diagram Γ, we associate a 3-dimensional combinatorial Δ-complex
DΓ by the presentation
Γ04 , Γ1 , {fi : Γ04 → Γ1 }0≤i≤3
where the maps fi are given by the pictures
1 3 2 0
and
2 0 1 3
which mean that if, for example, ε(v) = +1, then the left picture tells that f0 (v) is
the unique edge e such that cod(e) = v and st(e) = −1.
A Z-coloring of a 3-dimensional combinatorial Δ-complex (S, d, ∂) is a map
c : d−1 (3) → Z.
3.4. Lifting R-matrix knot invariants into the combinatorial frame-

work of Δ-complexes. In the rest of the paper we restrict ourselves to the cate-
gory VectFinK of finite dimensional vector spaces over a field K with the symmetric
monoidal structure given by the tensor product over K, and the standard permuta-
tion of the tensor components. The unit object is identified with the base field K.
A Hopf object in this category is nothing else but a finite dimensional Hopf algebra
over K.
Let (X, ∇, Δ, η, , γ) be a finite dimensional Hopf algebra over K. Denote d =
dim(X), and fix a linear basis {vi }1≤i≤d in X. Let {wi }1≤i≤d be the corresponding
dual basis of the dual vector space X ∗ . We have the canonical dualities

d
(3.2) p = X ∗ , X, coev, eval, eval(f v) = f (v), coev(1) = vi wi ,
i=1
and

d
(3.3) q = X, X ∗ , coev , eval , eval (vf ) = f (v), coev (1) = wi vi .
i=1
With respect to the chosen basis, the structural maps of our Hopf algebra are given
in terms of the corresponding structural constants {∇ki,j }, {Δi,j j
k }, {η }, {i }, {γi }
i
80
12 R. M. KASHAEV
according to the formulae

d
d
(3.4) ∇(vi vj ) = ∇ki,j vk , Δ(vi ) = Δi,j
k vi vj ,
k=1 i,j=1

d
d
η(1) = η i vi , (vi ) = i , γ(vi ) = γij vj .
i=1 j=1
The action on the basis elements of the maps

τr = X∇ ◦ Xγ ◦r X ◦ ΔX, r ∈ Z,
has the form

d
τr vi vj = X∇ ◦ Xγ ◦r X ◦ ΔX(vi vj ) = Δk,l ◦r
i X∇ ◦ Xγ X(vk vl vj )
k,l=1

d
d
◦r m ◦r m n
= Δk,l
i (γ )l X∇(vk vm vj ) = Δk,l
i (γ )l ∇m,j vk vn
k,l,m=1 k,l,m,n=1

d
= (τr )k,n
i,j vk vn ,
k,n=1
where

d
◦r m n
(τr )k,n
i,j = Δk,l
i (γ )l ∇m,j .
l,m=1
Let Γ be a tangle diagram, DΓ , the associated 3-dimensional combinatorial
Δ-complex, and c : Γ04 → Z, a Z-coloring. For any map
f : Γ1 → {1, . . . , d}
we associate a weight function of Wf (DΓ , c) ∈ K given by the product

Wf (DΓ , c) = Wf (v, c)
v∈Γ04
where Wf (v, c) is given by the number (τc(v) )k,l

i,j for the configuration
l k
i j
and by the number (τc(v) )k,l

i,j for the configuration
k l
j i
The results of the previous sections imply the following proposition.

Proposition 3.3. Let (X, ∇, Δ, η, , γ) be a finite dimensional Hopf algebra
over K with invertible antipode, and ρτ , the associated rigid R-matrix corresponding
to the canonical dualities (3.2), (3.3). Let Γ be a tangle diagram, and Γ̃ its 2-cable
where the second component is taken with the reversed orientation. Then, there
13
exists a representative Γ of the dpr-tangle defined by Γ̃, and a Z-coloring c of the

3-dimensional combinatorial Δ-complex DΓ such that the matrix elements of the
operator Φρτ ,k (Γ) with respect to a chosen basis in X are given by a state sum of
the form
Wf (DΓ , c),
f,f |∂Γ =g
where g : Γ01→ {1, . . . , dim X} is a multi-index parameterizing the matrix elements
of the operator Φρτ ,k (Γ).
References
[B] R.J. Baxter, Exactly Solved Models in Statistical Mechanics. London: Acad. Press, 1982. 486
p.
[D] V.G. Drinfeld, Quantum groups. Proc. I.C.M. Berkeley 1986, Amer. Math. Soc, Providence,
RI, vol. 1 (1987), 798–820.
[F] L.D. Faddeev, Integrable Models in 1 + 1 Dimensional Quantum Field Theory. Les Houches
Lectures 1982. Berlin: Elsevier Sci. Publ., 1984. P. 21–157.
[Ha] Allen Hatcher, Algebraic Topology. Cambridge University Press, Cambridge, 2002.
[Hi] J. Hietarinta, Solving the two-dimensional constant Yang–Baxter equation. J. Math. Phys.
34 (1993), 1725–1756.
[J] V.F.R. Jones, On knot invariants related to some statistical mechanical models. Pacific J.
Math. Volume 137, Number 2 (1989), 311–334.
[Kas] R.M. Kashaev, The Heisenberg double and the pentagon relation. Algebra i Analiz, 8:4
(1996), 63–74.
[Kau] L.H. Kauffman: An invariant of regular isotopy. Transactions of the American Mathemat-
ical Society 318(2) (1990), 417–471.
[R] N. Reshetikhin, Quasitriangular Hopf algebras and invariants of links. Algebra i Analiz, 1:2
(1989), 169–188.
[T1] V.G. Turaev, The Yang–Baxter equation and invariants of links. Inven. Math. 92 (1988),
527–553.
[T2] V.G. Turaev, The Conway and Kauffman modules of the solid torus with an Appendix on
the operator invariants of tangles. LOMI-preprint. E-6-88. Leningrad. 1988.
[Y] C.N. Yang, S-matrix for the one-dimensional N-body problem with repulsive or attractive
δ-function interaction. Phys. Rev. Vol. 168, N 5 (1968), 1920–1923.
Université de Genève, Section de mathématiques, 2-4, rue du Lièvre, CP 64, 1211

Genève 4, Suisse
E-mail address: rinat.kashaev@unige.ch
Volume 541, 2011
Knots and tropical curves
Stavros Garoufalidis
Abstract. A sequence of rational functions in a variable q is q-holonomic

if it satisfies a linear recursion with coefficients polynomials in q and q n . In
the paper, we assign a tropical curve to every q-holonomic sequence, which is
closely related to the degree of the sequence with respect to q. In particular,
we assign a tropical curve to every knot which is determined by the Jones
polynomial of the knot and its parallels. The topical curve explains the relation
between the AJ Conjecture and the Slope Conjecture (which relate the Jones
polynomial of a knot and its parallels to the SL(2, C) character variety and to
slopes of incompressible surfaces). Our discussion predicts that the tropical
curve is dual to a Newton subdivision of the A-polynomial of the knot. We
compute explicitly the tropical curve for the 41 , 52 and 61 knots and verify
the above prediction.
1. Introduction
1.1. What is a q-holonomic sequence? A sequence of rational functions
fn (q) ∈ Q(q) in a variable q is q-holonomic if it satisfies a linear recursion with
coefficients polynomials in q and q n . In other words, we have

d
(1) ai (q n , q)fn+i (q) = 0
i=0
where the coefficients ai (M, q) ∈ Z[M, q] are polynomials for i = 0, . . . , d where
ad (M, q) = 0. The term was coined by Zeilberger in [Z] and further studied in
[WZ]. q-holonomic sequences appear in abundance in Enumerative Combinatorics;
[PWZ, St]. The fundamental theorem of Wilf-Zeilberger states that a multi-
dimensional finite sum of a (proper) q-hyper-geometric term is always q-holonomic;
see [WZ, Z, PWZ]. Given this result, one can easily construct q-holonomic se-
quences. Combining this fundamental theorem with the fact that many state-sum
invariants in Quantum Topology are multi-dimensional sums of the above shape,
The author was supported in part by NSF.
1991 Mathematics Classification. Primary 57N10. Secondary 57M25.

Key words and phrases: Knot, Jones polynomial, AJ Conjecture, Slope Conjecture, A-polynomial,
non-commutative A-polynomial, Jones slope, tropicalization, tropical curve, tropical geometry,
Newton polygon, Quantization, BPS states, twist knots.
2011
c 0000
c Mathematical
holder)
1
83
84
2 STAVROS GAROUFALIDIS
it follows that Quantum Topology provides us with a plethora of q-holonomic se-

quences of natural origin; [GL]. For example, the sequence of Jones polynomials of
a knot and its parallels which we will study below (technically, the colored Jones
function) is q-holonomic.
The goal of our paper is to assign a tropical curve to a q-holonomic sequence.
To motivate the connection between q-holonomic sequences and tropical curves, we
will write Equation (1) in operator form using the operators M, L which act on a
sequence fn (q) ∈ Q(q) by
(M f )n (q) = q n fn (q), (Lf )n (q) = fn+1 (q).
It is easy to see that LM = qM L generate the q-Weyl algebra
(2) W = Z[q ±1 ]M, L/(LM − qM L)
Equation (1) becomes
(3) Pf = 0
where

d
(4) P = ai (M, q)Li ∈ W.
i=0
In other words, Equation (4) says that P annihilates f . Although a q-holonomic
sequence f is annihilated by many operators P ∈ W , it was observed in [Ga2] that it
is possible to canonically choose an operator Pf with coefficients ai (M, q) ∈ Z[M, q].
Likewise, there is a unique non-homogeneous linear recursion relation of the form
Pf f = bf where bf ∈ Z[M, q]. For a detailed definition, see Section 2 below.
Definition 1.1. We call Pf and (Pfnh , bf ) the homogeneous and the non-
homogeneous annihilator of the q-holonomic sequence f .
1.2. What is a tropical curve? In this section we will recall the definition
of a tropical curve. For a survey on tropical curves, see [RGST, SS]. With those
conventions, a tropical polynomial P : R2 −→ R is a function of the form:
(5) P (x, y) = min{a1 x + b1 y + c1 , . . . , ar x + br y + cr }
where ai , b, ci are rational numbers for i = 1, . . . , r. P is convex and piecewise
linear. The tropical curve T (P ) of the tropical polynomial P is the set of points
(x, y) ∈ R2 such that P is not linear at (x, y). Equivalently, T (P ) is the set of
points where the minimum is attained at two or more linear functions. A rational
graph Γ is a finite union of rays and segments whose endpoints and directions
are rational numbers, and each ray has a positive integer multiplicity. A balanced
rational graph is defined in [RGST, Eqn.10]: at every vertex the sum of the slope
vectors with multiplicities adds to zero. Every tropical curve is a balanced rational
graph and vice-versa; see [RGST, Thm.3.6]. Tropical curves are very computable
objects. For example, the vertices of a rational curve are the points (x, y) where
the minimum in (5) is attained at least three times. The coordinates of such points
can be solved by solving a system of linear equations. An explicit algorithm to
compute the vertices and the slopes of a tropical curve is given in [RGST, Sec.3],
and a computer implementation in Singular is available from [Ma]. This allows
us to compute the tropical curves of the 41 , 52 and 61 knots in Sections 5.1-5.4
below. In the case of the 61 knot, the non-homogeneous tropical curve is defined
by an explicit polynomial with r = 346 terms.
KNOTS AND TROPICAL CURVES 85
3
Tropical curves arise from 2-variable polynomials Pt (x, y) whose coefficients

depend on an additional parameter t as follows. Consider

r
(6) Pt (x, y) = γi (t)xai y bi
i=1
where γi (t) are algebraic functions of t with order at t = 0 equal to ci . Then, the
corresponding tropical polynomial is given by (5). Pt (x, y) gives rise to two Newton
polytopes:
• The 3-dimensional Newton polytope NP , i.e., the convex hull of the ex-
ponents of (x, y, t) in Pt (x, y).
• The 2-dimensional Newton polygon NP,0 , i.e., the convex hull of the ex-
ponents of (x, y) in Pt (x, y).
In fact, NP,0 is the image of NP under the projection map (x, y, t) −→ (x, y). The
lower faces of NP give rise to a Newton subdivision of NP,0 which is combinatorially
dual to the tropical curve T (P ); see [RGST].
The polynomials Pt (x, y) appear frequently in numerical problems of Path Ho-
motopy Continuation where one is interested to connect P0 (x, y) to P1 (x, y). They
also appear in Quantization problems in Physics, where t (or log t) plays the role
of Planck’s constant. We will explain below that they also appear in Quantum
Topology, and they are a natural companion of the AJ and the Slope Conjecture.
1.3. The tropical curve of a q-holonomic sequence. In this section we

associate a tropical surve to a q-holonomic sequence. The main observation is that
an element of the q-Weyl algebra is a polynomial in 3 variables M, L, q. Two of
those q-commute (i.e., satisfy LM = qM L) but we can always sort the powers of
L to the right and the powers of M to the left. In other words, there is an additive
map
(7) Z[q ±1 ]M, L/(LM − qM L) −→ Z[M, L, q ±1 ]
Let us change variables (x, y, 1/t) = (L, M, q) and ignore the coefficients of the
monomials of xi y j tk , and record only their exponents. They give rise to a tropical
curve. Explicitly, let

(8) P = ai,j,k q k M j Li ∈ W
(i,j,k)∈A
denote an element of the q-Weyl elgebra, where A is a finite set and ai,j,k ∈ Z \ {0}
for all (i, j, k) ∈ A.
Definition 1.2. There is a map
(9) W −→ {Tropical Curves in R2 }, P → ΓP
which assigns to P in (9) the tropical polynomial Pt (x, y) given by:
Pt (x, y) = min {ix + jy − k}
(i,j,k)∈A
ΓP is the tropical curve of Pt (x, y).

Combining Definitions 1.1 and 1.2 allows us to assign a tropical curve to a
q-holonomic sequence f .
86
Definition 1.3. (a) If f is a q-holonomic sequence, let Γf and Γnh f denote the
tropical curves of Pf (y, x, 1/t) and Pfnh (y, x, 1/t) respectively, where Pf (M, L, q)
and Pfnh (M, L, q) are given in Definition 1.1.
The tropical curve Γf of a q-holonomic sequence f is closely related to the
degree (with respect to q) of the sequence of rational functions fn (q). If δn =
degq (fn (q)) denotes this degree, then it was shown in [Ga4] that for large enough
n, δn is a quadratic quasi-polynomial with slope recorded by the rays of the tropical
curve Γf .
1.4. 3 polytopes of a q-holonomic sequence. In this section we assign 3
polytopes to a q-holonomic sequence.
Definition 1.4. (a) If P ∈ W is given by Equation (8), it defines 3 polytopes:
• NP is the convex hull of the exponents of the polynomial P (M, L, q) with
respect to the variables (M, L, q).
• NP,0 is the projection of NP under the projection map (M, L, q) −→
(L, M ).
• NP,1 is the convex hull of the exponents of the polynomial P (L, M, 1).
(b) If f is a q-holonomic sequence, its annihilator Pf gives rise to the polytopes
NPf , NPf ,0 and NPf ,1 .
Note that NP is a 3-dimensional convex lattice polytope, and NP,0 , NP,1 are
2-dimensional convex lattice polygons. Since every exponent of P (M, L, 1) comes
from some exponents of P (M, L, q), it follows that
(10) NP,1 ⊂ NP,0
Remark 1.1. It follows by [RGST] that the tropical curve ΓP is dual to a
Newton subdivision of NP,0 .
We will say that P (M, L, q) is good if NP,1 = NP,0 . It is easy to see that
goodness is a generic property.
1.5. The slopes of a q-holonomic sequence. In this section we discuss the
slopes of a q-holonomic sequence and their relation with its tropical curve. The
proof of the following theorem uses differential Galois theory and the Lech-Mahler-
Skolem theorem from number theory.
Theorem 1.5. [Ga4] The degree with respect to q of a q-holonomic sequence
fn (q) ∈ Q(q) is given (for large values of n) by a quadratic quasi-polynomial.
Recall that a quadratic quasi-polynomial is a function of the form:

n
(11) p : N −→ N, p(n) = γ2 (n) + γ1 (n)n + γ0 (n)
2
where γj (n) are rational-valued periodic functions of n. Quasi-polynomials appear
in lattice point counting problems, and also in Enumerative Combinatorics; see
[BP, BR, Eh, St] and references therein.
The set of slopes s(p) of a quadratic quasi-polynomial is the finite set of values
of the periodic function γ2 (n). These are essentially the quadratic growth rates
of the quasi-polynomial. More precisely, recall that x ∈ R is a cluster point of a
sequence (xn ) of real numbers if for every > 0 there are infinitely many indices
5
n ∈ N such that |x−xn | < . Let {xn } denote the set of cluster points of a sequence
(xn ). It is easy to show that for every quadratic quasi-polynomial p we have:
2
(12) s(p) = { p(n) |n ∈ N} ⊂ Q
n2
Given a q-holonomic sequence fn (q) ∈ Q(q), let s(f ) denote the slopes of the
quadratic quasi-polynomial degq fn (q). Let s(N ) denote the set of slopes of the
edges of a convex polygon N in the plane. The next proposition relates the slopes
of a q-holonomic sequence with its tropical curve. See also [Ga4, Prop.4.4].
Proposition 1.2. If f is q-holonomic, then s(f ) ⊂ −s(NPf ,0 ).
Proof. Let δ(n) = degq fn (q) denote the degree of fn (q) with respect to q,
and let P denote the annihilator of f . We expand P in terms of monomials as in
Equation (8). For every monomial q k M j Li and every n we have
degq ((q k M j Li )fn (q)) = k + jn + δ(n + i).
Since P annihilates f , for every n the following maximum is attained at least twice
(from now on, twice will mean at least twice as is common in Tropical Geometry):
(13) max {jn + k + δ(n + i)}

(i,j,k)
Subtracting δ(n), it follows that the maximum is obtained twice:
(14) max {jn + k + δ(n + i) − δ(n)}

(i,j,k)
Now δ(n) is a quadratic quasi-polynomial given by

n
δ(n) = γ2 (n) + γ1 (n)n + γ0 (n)
2
Theorem 1.5 implies that for large enough n, in a fixed arithmetic progression, we
i for i = 1, 2, thus
have γi (n) = γ

i
δ(n + i) − δ(n) = γ
2 i n + γ
2 1 i
+γ
2
Substituting into (14), it follows that for large enough n in an arithmetic progres-
sion, the max is obtained twice:

i
(15) max {jn + k + γ 2 i n + γ
2 +γ1 i}
(i,j,k) 2
It follows that there exists (i , j ) = (i, j) such that
j − j
(16) 2 = −
γ .
i − i
This proves Proposition 1.2.
88
2. The q-Weyl alegbra and its localization

In this section we will discuss some algebraic properties of the q-Weyl algebra
and its localization, which will justify Definition 1.1.
Recall the q-Weyl algebra from (2). We will say that an element P of W is
reduced if it has the form (4) where ai (M, q) ∈ Z[M, q] for all i, and the greatest
common divisor of ai (M, q) ∈ Z[M, q] is 1.
Consider the localized q-Weyl algebra Wloc given by
(17) Wloc = Q(M, q)L/(Lf (M, q) − f (M q, q)L)
It was observed in [Ga2] that W is not a principal left-ideal domain, but becomes
so after localization; see [Cou]. If f is a sequence of rational functions, consider
the left ideal Mf
Mf = {P ∈ Wloc |P f = 0}
Mf is a principal ideal, which is nonzero if f is q-holonomic. Let P denote the
monic generator of Mf . Left multiply it by a polynomial in M, q so as to obtain a
reduced annihilator Pf of f .
Now, we discuss non-homogeneous recursion relations of the form

d
ai (q n , q)fn+i (q) = b(q n , q)
i=0
where ai (M, q), b(M, q) ∈ Q(M, q) for all i. In operator form, we can write the
above recursion as
P f = b.
Consider the left ideal
(18) Mfnh = {P ∈ Wloc |∃b ∈ Q(M, q) : P f = b}
It is easy to see that Mfnh is a left ideal. If f is q-holonomic, Mfnh = 0. Let P
denote the monic generator of Mfnh . There exists b ∈ Q(M, q) such that
P f = b
There are two cases: b = 0 or b = 0. If b = 0, then dividing by b we obtain
that 1/b · P f = 1. We left multiply both sides by a polynomial in M, q so as to
obtain Pfnh f = bf where P nh is reduced. If b = 0 then multiply by a polynomial
in M, q so as to obtain Pfnh f = 0 and define bf = 0 in tha case. This concludes
Definition 1.1.
The next lemma relates the homogeneous and the non-homogeneous annihi-
lator of a q-holonomic sequence. It is well-known that one can convert an non-
homogeneous recursion relation P f = b where b = 0 into a homogeneous recursion
relation of order one more. Indeed, P f = b where b = 0 is equivalent to
(L − 1)b−1 P f = 0
This implies the following conversion between (Pfnh , bf ) and Pf . Fix a q-holonomic
sequence fn (q) ∈ Q(q).
Lemma 2.1. (a) If bf = 0 then Pfnh = Pf . If bf = 0, then Pfnh is obtained by
clearing denominators of (L − 1)b−1 f Pf by putting the powers of L on the right and
the elements of Q(M, q) on the left.
(b) If Pf is not left divisible by L − 1 in W , then Pf = Pfnh and bf = 0. If Pf is left
7
divisible by L − 1 in W , then Pf = (L − 1)Qf and if d is the common denominator

of Qf , then (dQf , d) = (Pfnh , bf ).
Definition 2.2. We say that a q-holonomic sequence f is homogeneous if
bf = 0–else f is non-homogeneous.
In other words, a q-holonomic sequence f is homogeneous if and only if Pf is
left-divisible by L − 1 in W .
3. Quantum Topology
3.1. The tropical curve of a knot. Quantum Topology is a source of q-
holonomic sequences attached to knotted 3-dimensional objects. Let JK,n (q) ∈
Z[q ±1 ] denote the colored Jones polynomial of a knot K in 3-space, colored by the
(n + 1)-dimensional irreducible representation of sl2 and normalized to be 1 at the
unknot; [Jo, Tu]. The sequence JK,n (q) for n = 0, 1, . . . essentially encodes the
Jones polynomial of a knot and all of its parallels; see [Tu]. In [GL, Thm.1] it
was shown that the sequence JK,n (q) of colored Jones polynomials of a knot K is
q-holonomic.
Definition 3.1. (a) If K is a knot, we denote by AK (M, L, q) and (Anh
K (M, L, q),
BK (M, q)) the homogeneous and the non-homogeneous annihilator of the q-holono-
mic sequence JK,n (q). These are the non-commutative and the non-homogeneous
non-commutative A-polynomials of the knot.
(b) If K is a knot, let ΓK and Γnh K denote the tropical curves of AK and AK
nh
respectively.
The non-homogeneous non-commutative A-polynomial of a knot appeared first
in [GS].
3.2. The AJ Conjecture. The AJ Conjecture (resp. the Slope Conjecture)
relates the Jones polynomial of a knot and its parallels to the SL(2, C) character
variety (resp. to slopes of incompressible surfaces) of the knot complement. We
will relate the two conjectures using elementary ideas from Tropical Geometry.
The A-polynomial of a knot is a polynomial in two commuting variables M
and L that essentially encodes the image of the SL(2, C) character variety of K,
projected in C∗ × C∗ by the eigenvalues of a meridian and longitude of K. It was
defined in [CCGLS].
Conjecture 3.1. [Ga2] The AJ Conjecture states that
(19) AK (M, L, 1) = BK (M )AK (M 1/2 , L)
where AK (M, L) is the A-polynomial of K and BK (M ) ∈ Z[M ] is a polynomial
that depends on M and of course K.
The AJ Conjecture is known for infinitely many 2-bridge knots; see [Le].
It is natural to ask whether the q-holonomic sequence JK,n (q) is of non-homoge-
neous type or not. Based on geometric information (the so-called loop expansion of
the colored Jones polynomial, see [Ga1]), as well as experimental evidence for all
knots whose non-commutative A-polynomial is known (these are the torus knots in
[Hi] and the twist knots in [GS]) we propose the following conjecture.
Conjecture 3.2. For every knot K, JK,n (q) is non-homogeneous.
90
The above conjecture implies that BK (M, q) ∈ Z[M, q] \ {0} is an invariant

which is independent and invisible from the classical A-polynomial of the knot.
There is a close connection between the BK (M, q) invariant of a knot and the
torsion polynomial of the knot introduced in [DbG]. We will discuss this in a
future publication.
3.3. The Slope Conjecture. The Slope Conjecture of [Ga3] relates the
degree of the colored Jones polynomial of a knot and its parallels to slopes of
incompressible surfaces in the knot complement. To recall the conjecture, let
δK (n) = degq JK,n (q) (resp. δK ∗
(n) = deg∗q JK,n (q)) denote the maximum (resp.
minimum) degree of the polynomial JK,n (q) ∈ Z[q ±1 ] (or more generally, of a ratio-
nal function) with respect to q.
For a knot K, define the Jones slopes jsK by:
2
(20) δK (n) |n ∈ N}
jsK = {
n2
(b) Let bsK ⊂ Q ∪ {1/0} denote the set of boundary slopes of incompressible
surfaces of K; [Ha, HO].
Conjecture 3.3. [Ga3] The Slope Conjecture states that for every knot K
we have
(21) 2jsK ⊂ bsK .
Note that the Slope Conjecture applied to the mirror of K implies that 2js∗K ⊂
bsK . The Slope Conjecture is known for alternating knots and torus knots (see
[Ga3]), for adequate knots (which include all alternating knots; see [FKP]), for
(−2, 3, n) pretzel knots (see [Ga3]), and for 2-fusion knots; see [DnG]. A general
method for verifying the Slope Conjecture is discussed in [Ga5, DnG].
3.4. The AJ Conjecture and the Slope Conjecture. In this section we

will see how the AJ Conjecture relates to the Slope Conjecture, expanding a com-
ment of [Ga3, Sec.2]. We will specialize Definition 1.4 to knot theory when P = AK
is the non-commutative A-polynomial of a knot K, and we will denote by NK , NK,0
and NK,1 the three polytopes associated to AK . Proposition 1.2 implies that
(22) jsK ⊂ −NK,0
Let bsA
denote the slopes of the A-polynomial of K. The AJ Conjecture implies
K
that up to possibly excluding the slope 1/0 from 2NK,1 , we have:
(23) 2NK,1 = bsA
K.
For a careful proof, see Proposition 3.4 and Remark 3.5 below. Culler and Shalen
show that edges of the Newton polygon of the A-polynomial of K give rise to ideal
points of the SL(2, C) character variety of K; see [CS, CGLS, CCGLS]. For every
ideal point, Culler and Shalen construct an incompressible surface whose slope is a
boundary slope of K; see [CS, CCGLS]. bsA K is the set of the so-called strongly
detected boundary slopes of K, and satisfies the inclusion:
(24) K ⊂ bsK .
bsA
If AK (M, L, q) is good, then
(25) NK,0 = NK,1 .
9
If K ∗ denotes the mirror of K, then JK ∗ ,n (q) = KK,n (q −1 ) which implies that

−NK,0 = NK ∗ ,0 . Combining Equations (22)-(25), it follows that
2jsK ⊂ bsK ∗
which is the Slope Conjecture, up to a harmless mirror image. This derivation also
explains two independent factors of 2, one in Equation (20) and another one in
Equation (19).
Proposition 3.4. If the non-commutative A-polynomial of K is good, and
if the AJ Conjecture holds, then ΓK is dual to the Newton subdivision of the A-
polynomial of K (multiplied by a polynomial in M ).
Proof. Let P denote the non-commutative A-polynomial of a knot K. ΓK is
dual to NP,0 . If P is good, then NP,0 = NP,1 . With the notation of Conjecture 3.1,
the AJ Conjecture implies that
P (M, L, 1) = AK (M 1/2 , L)BK (M )
where BK (M ) is a polynomial of M , and AK is the A-polynomial of A. The Newton
polygon of of the product of two polynomials is the Minkowski sum of their Newton
polygons. Moreover, the Newton polygon of BK (M ) is a vertical line segment in
the (L, M )-plane. It follows that the Newton polygon of AK (M 1/2 , L)BK (M ) is
the Newton polygon of the A-polynomial of K and its translation by a vertical
segment. On the other hand, the Newton polygon of P (M, L, 1) is NP,1 . The result
follows.
Remark 3.5. Note that the Newton polygon of AK (M 1/2 , L)BK (M ) is the
Newton polygon of AK (M 1/2 , L) and its shift by a vertical line segment. It follows
that the slopes of the Newton polygon of AK (M 1/2 , L)BK (M ) are the slopes of
AK (M 1/2 , L) plus the slope of a vertical segment (i.e., 1/0). For concrete examples,
see Section 5 where the Newton polygons of the non-homogeneous A-polynomials of
41 , 52 , 61 , 81 is shown and it differs from the Newton polygon of the A-polynomial
by a shift by a vertical segment.
The only knots with explicitly known non-commutative A-polynomials (ho-
mogeneous and non-homogeneous) are the handful of twist knots Kp of [GS] for
p = −8, . . . , 11. An explicit check shows that these non-commutative A-polynomials
(both the homogeneous and the non-homogeneous) are good. For details, see Sec-
tion 5.
4. Quantization and Tropicalization

Quantization is the process of producing the non-commutative A-polynomial
of a knot from the usual A-polynomial. In other words, Quantization starts with
P1 (x, y) and produces Pt (x, y) as in Equation (6). On the other hand, Tropical
Geometry expands Pt (x, y) at t = 0 (or equivalently at q = ∞) and produces a
tropical curve. Schematically, we have a diagram:
⎛ ⎞
non-commutative
A-polynomial Classical limit
⎝ A-polynomial ⎠ Tropicalization Tropical curve
←− −→
q=1 q=∞
q
Quantization is a map reverse to the Classical Limit map in the above diagram.
Both sides of the above diagram (i.e., the limits at q = 1 and q = ∞) are classical
92
dual invariants of the knot. Indeed, the tropical curve ought to be dual to a Newton
subdivision of the A-polynomial of K. This duality is highly nontrivial, even for
the simple case of the 41 knot, computed in Section 5.1 below.
This conjectured duality may be related to the duality between Chern-Simons
theory (i.e., colored U(N ) polynomials of a knot) and Enumerative Geometry (i.e.,
BPS states) of the corresponding Calabi-Yau 3-fold. For a discussion of the latter
duality, see [ADKMV, DGKV, LMV, DV] and references therein.
Physics principles concerning Quantization of complex Lagrangians in Chern-
Simons theory suggest that the A-polynomial of a knot should determine the non-
commutative A-polynomial. In particular, it should determine the polynomial in-
variant BK (M, q) of Definition 1.1, and it should determine the tropical curves ΓK
and Γhn
K .
Aside from duality conjectures, let us concentrate on a concrete question. It is
well-known that the A-polynomial of a knot is a triangulated curve in the sense of
algebraic K-theory. In other words, if X is the curve of zeros AK (M, L) = 0 of the
A-polynomial then there exist nonzero rational functions z1 , . . . , zr ∈ C(X)∗ in X
such that

r
(26) M ∧L=2 zi ∧ (1 − zi ) ∈ ∧2Z (C(X)∗ )
i=1
where C(X) is the field of rational functions of X and M, L ∈ C(X)∗ are the
eigenvalues of the meridian and the longitude. For a proof of (26) (which uses the
symplectic nature of the so-called Neumann-Zagier matrices), see [Ch, Lem.10.1].
For an excellent discussion of triangulated curves X and for a plethora of examples
and computations, see [BRVD]. Geometrically, a triangulation of X comes from
an ideal triangulation of the knot complement with r ideal tetrahedra with shape
parameters z1 , . . . , zr which satisfy some gluing equations. The symplectic nature
of these gluing equations, introduced and studied by Neumann and Zagier in [NZ],
implies (26). The triangulation of X has important arithmetic consequences re-
garding the volume of the knot complement and its Dehn fillings, and it is closely
related to the Bloch group of the complex numbers. It is important to realize that
X has infinitely many triangulations, and in general it is not possible to choose a
canonical one. In addition, triangulations tend to work well with hyperbolic knots.
On the contrary, the non-commutative A-polynomial and its corresponding tropical
curve exist for every knot in 3-space, hyperbolic or not. Let us end with some ques-
tions, which aside from its theoretical interest, may play a role in the Quantization
of the A-polynomial.
Question 4.1. Is the tropical curve ΓK of a hyperbolic knot K related to a

triangulation of its A-polynomial curve?
To formulate our next question, recall that the tropical curve ΓK is dual to
a Newton subdivision of the 2-dimensional Newton polytope of the polynomial
AK (M, L, q) with respect to the variables L and M . Assuming that AK (M, L, q)
is good, and assuming the AJ Conjecture, it follows that ΓK is dual to the Newton
polygon of the A-polynomial of K. ΓK is a balanced rational graph that consists
or edges and rays, and the above assumptions imply that the slopes of the rays
11
are negative inverses of the slopes of the A-polynomial of K. Consequently, Culler-

Shalen theory (see [CS]) implies that the slopes of the rays of ΓK are negative
inverses of boundary slopes of K, appropriately normalized by a factor of 2.
Question 4.2. What is the geometric meaning of the vertices of ΓK (those are
points in Q2 ) and of the slopes of the edges of ΓK ?
5. Computations of tropical curves of knots

5.1. The homogeneous tropical curve of the 41 knot. The non-commutative
A-polynomial A41 (M, L, q) of 41 was computed in [GL, Sec.6.2] and also [Ga2,
Sec.3.2] using the WZ method of [WZ, Z] implemented by [PR] in Mathematica.
The non-commutative A-polynomial is given by

x3 t2 − y t3 − y y 2 (t + y) t − y 2 t3 − y 2
A41 (y, x, 1/t) = 14
2 t
t − y (−1 + y)y 2 t2 + y t3 − y 2 t5 − y 2
+
t15
1 2 2 2 2
− 18
x t −y t + y t − y 2 t3 − y 2
t

· t8 − 2t6 y + t7 y − t3 y 2 + t4 y 2 − t5 y 2 + ty 3 − 2t2 y 3 + y 4
1
+ 17
x(t − y) t2 − y (t + y) t3 − y 2 t5 − y 2
t

· t4 + y 4 − t3 y(2 + y) − ty 2 (1 + 2y) + t2 y 1 + y + y 2
Notice that
A41 (x, y, 1) = (−1 + x)(−1 + y)4 (1 + y)3 (−x + xy + y 2 + 2xy 2 + x2 y 2 + xy 3 − xy 4 )
confirms the AJ Conjecture, since the last factor is the geometric component of the
A-polynomial of 41 , the first term is the abelian component of the A-polynomial,
and the remaining second and third terms depend only on y = M . Expanding out
the terms, we obtain that:
·x2 y 11 + t−1 + t−1 + −1+t+t
2
14 ·x y + t17 ·x y 17 ·xy ·x3 y 8 +
1 3 9 1−3·t 2 10 11
A41 (y, x, 1/t) = t18 t13
−1−3·t2 +2·t3 −t4 1+2·t2 +t3 −t4 −1+3·t−t2 +3·t3 +2·t5
t17 · x y + t16 · xy +
2 9 2 10
t13 ·x y +
3 7
t16 · x2 y 8 +
· xy 9 + 1−t−tt12−2·t · x3 y 6 + 3−2·t+3·t · x2 y 7 + −1−t−tt15−t −2·t · xy 8 + t15
3 4 3 4 2 4 2 3 4
1+t +t
t16 t14
+t 1
·
−2−t−t +t 3 4
1−3·t−2·t −t −2·t
3 4 6
−1−t −t −t −t
2 3 4 6
−1
9
y + t10 ·x y +
3 5
t13 ·x y +
2 6
t14 · xy + t15 · y +
7 8
−1+t+2·t2 +t4 −2−t2 −2·t3 −3·t5 +t6 1+t2 +t3 +2·t4 +t5 +t6
xy 6 + −1−t−t
2
t9 · x 3 4
y + t11 · x 2 5
y + t13 · t12 ·
2
1+3·t −2·t +3·t
2 3 4 2 3 4 6 2
−1
y + t7 · x y +
7 1+t−t 3 3
t8 ·x y +
2 4 1+t+2·t +t +t +t
t11 · xy + t12 · y + t4 ·
5 1+t+t 6
2+3·t2 −t3 +3·t4 −t5 −1−t2 −t3 −t4 −t6

· xy + t8 · y 5 + −1+2·t−3·t
1+t+t2 −t4
2
3 2
x y + t7 ·x y +
2 3
t10
4
t5 ·
x2 y 2 + −2−t−tt7−t −t · xy 3 + −1−t−t −3+t −1
2 3 4 2 4
t8 · y 4
+ t2 · x 2
y + 1+t+t
t6 · xy 2
+ t3 · y 3
+ x 2
+
−1
t2 · xy + t3 · y + t · x
2 1 2
Inspection of the above formula shows that A41 (y, x, 1/t) is good. Using the
drawing polymake program of [Ma] implemented in Singular one can compute
the vertices of the tropical curve:
(3, −1/2), (−1, −1/3), (−3/4, −1/2), (−2, 0), (2, −1), (−1/2, −1), (1, −3/2),
(0, −3/2), (−1/2, −5/4), (1/2, −7/4), (−1, −3/2), (1/2, −2), (2, −3), (3/4, −5/2),
(1, −8/3), (−2, −2), (−3, −5/2)
94
The tropical curve (with the convention that unmarked edges or rays have
multiplicity 1) is:
2
2
2 2 2
2 2 2
2 2 2
2
The Newton subdivision of the Newton polygon is:
The reader may observe that the above Newton polygon is the Minkowski sum of
the Newton polygon of the A-polynomial of 41 with a vertical segment.
5.2. The non-homogeneous tropical curve of the 41 knot. The non-

homogeneous A-polynomial of the 41 knot was computed in Theorem 1 of [GS]
(with the notation A−1 (E, Q, q) where E = L and Q = M ). It has 22 terms and it
is given by:

Anh
41 (M, L, q) = L2 M 2 q 2 −1 + M 2 q −1 + M q 2 + (−1 + M )M 2 q 2 −1 + M 2 q 3

−L(−1 + M q)2 (1 + M q) 1 − M q − M 2 q − M 2 q 3 − M 3 q 3 + M 4 q 4

B41 (M, L) = M q(1 + M q) −1 + M 2 q −1 + M 2 q 3
It follows that:
−1 −1 1+t2 −1
41 (y, x, 1/t) = t7 · xy + t5 · x y + t6 · xy + t3 · x y + t6 · xy + t4 ·
1 2
Anh 7 2 5 6 2 4 5
x2 y 3 + −1−t−t · xy 4 + t15 · y 5 + t12 · x2 y 2 + −1−t−t · xy 3 + −1 −1

2 2 2
t5 · y + t3 · xy + t2 ·
4 1+t 2
t5 t4
y + t · xy + t2 · y − x
3 2 1 2
13
It is easy to see that the above polynomial is good. The vertices of the corresponding
tropical curve are:
(1, −1/2), (−1/2, −1/2), (−2, 0), (0, −1), (2, −2), (1/2, −3/2), (−1, −3/2)
The tropical curve is:
The Newton subdivision of the Newton polygon is:
This example exhibits that the non-homogeneous tropical curve is much simpler
than the homogeneous one.
homogeneous non-commutative A-polynomial Anh 52 (M, L, q) has 98 terms, and it is
given by [GS] (with the notation Anh 2 (E, Q, q) where E = L, Q = M ):
−1
52 (y, x, 1/t) = t19 · xy
Anh 12
+ t171
· x2 y 10 + t18 3
· xy 11 + t18 1
· y 12 + t−215 · x y
2 9
+
3 4
−1 −1−t+t +t −t
2 3 4
−2−2·t+t −2·t −3·t
2 3 4
−1−t
1−t+t +t
t18 · xy + t18 · y +
10 11
t16 ·x y +
2 8
t17 · xy + t14 ·
9
y 10 + 2+2·t−tt14+t +2·t ·x2 y 7 + 1−2·t−tt15+t −t ·xy 8 + 1+t 3 5 1−t−t2 +t3 +t4 −2·t5
2 3 4 2 3 5
t14 ·y 9
+ 1
t6 ·x y + t14 ·
2−t+t2 +4·t3 +2·t4 −t5 +2·t6 −1 −2+t−t2 −4·t3 −2·t4 +t5 −2·t6
2 6
x y + t15 · xy + t9 · y + t3 · x y +
7 1 8 3 4
t12 ·
x2 y 5 + −1+t+t −t −t4 +2·t5 −1 −1−t −1+2·t+t2 −t3 +t5
2 3
t14 · xy 6
+ t9 · y 7
+ t5 · x 3 3
y + t9 · x 2 4
y +
−2−2·t+t2 −t3 −2·t4 2+2·t−t2 +2·t3 +3·t4 1+t−t2 −t3 +t4
t11 · xy 5
+ 1+t
t2 · x 3 2
y + t8 · x 2 3
y + t10 · xy 4 +
−1+t−t −t
· x2 y 2 + t26 · xy 3 − x3 + −3 −1
3 4
t3 · x y + t3 · x y + t5 · xy + t · x
1 3 2 2 1 2
t6
The vertices of the tropical curve are:
(1, −1/2), (−1, 0), (−1/2, −1/2), (17/2, −1/2), (−1, −1), (0, −1), (−6, −2),
96
(6, −1), (−17/2, −5/2), (0, −2), (1, −2), (−1, −5/2), (1/2, −5/2), (1, −3)
The Newton subdivision of the tropical curve is:
2
2 2 2
4 2
2
2 2 4
2 2

homogeneous non-commutative A-polynomial Anh 61 (M, L, q) has 346 terms, and it
is given by [GS] (with the notation Anh −2 (E, Q, q) where E = L, Q = M ):
A61 (y, x, 1/t) = t31 · x y + t28 · x y + t30 · x2 y 14 + t−1
nh 1 2 15 −1−t 3 13 −1−3·t
30 · xy
15
+ t221
·
2
−1−t+2·t +2·t −t −t −t
2 3 4 5 6
−1
x4 y 11 + 1+3·t+t
t26 · x 3 12
y + t30 · x 2 13
y + 2
t29 · xy 14
+ t18 · x 4 10
y +
1+2·t+2·t2 +t3 −t4 −2·t5 +2·t6 +t7 1+4·t+3·t2 −t3 +2·t5 +4·t6 +3·t7
t27 ·x y +
3 11
t29 ·x y +
2 12
1−t−t2 +t4 +t5 +t6 −1−t−t2 −1−4·t−4·t2 −3·t3 −2·t6 −3·t7 −t8
t29 · xy + t21
13
·x y +
4 9
t25 · x3 y 10 +
1−2·t−3·t2 +3·t4 +3·t5 +t6 −4·t7 −t8 +t9 +t10 −2+2·t 2
−2·t 4
−2·t 5
−2·t6 2
t28 · x2 y 11 + t28 · xy 12 + 1+t+t t17 ·

−1−2·t−2·t +3·t −2·t −4·t −t +3·t +t −t
2 4 6 7 8 9 10 11
4 8
x y + t25 ·x y +
3 9
−1−3·t+t2 −3·t4 −6·t5 −6·t6 −4·t7 +t8 −t9 −2·t10 −3·t11 1+2·t−t3 +t5 +2·t6 +t7 −t8 −t9 −t10
t27 ·x y +
2 10
t27 ·
2
1+3·t+3·t2 +2·t3 +2·t5 +5·t6 +5·t7 +3·t8 −t10 +t11 +t12
xy 11 + 1+t+tt19 · x 4 7
y + t23 · x 3 8
y +
1+t−2·t2 −2·t3 −2·t4 +2·t5 −t7 −3·t8 −3·t9 +2·t11 −t12
t 25 · x2 y 9 +
15
1−3·t−t2 +3·t3 +3·t4 +t5 −2·t6 −2·t7 +2·t8 +2·t9 +2·t10

· y 11 + −1−t−t
2
t26 · xy 10 + t26 1
t15 · x4 y 6 +
1+t−t −2·t +t +3·t +4·t −2·t −4·t −2·t +t +2·t −2·t
2 3 4 5 6 7 8 9 10 11 13
t22 · x3 y 7 +
1+2·t+2·t2 +3·t3 +t4 +3·t5 +3·t6 +3·t7 +3·t8 +t9 +t11
t22 ·x y +
2 8
−2−2·t3 −4·t4 −4·t5 −2·t6 +t7 −t9 −2·t10 −t11 +t13

t25 · xy + t26 · y 10 + t−1
9 −1
16 · x y +
4 5
−2−2·t −4·t −4·t −2·t +t −t −2·t −t +t

3 4 5 6 7 9 10 11 13
t19 · x3 y 6 +
1+2·t+2·t2 +3·t3 +t4 +3·t5 +3·t6 +3·t7 +3·t8 +t9 +t11
t20 · x2 y 7 +
1+t−t −2·t +t +3·t +4·t −2·t −4·t −2·t +t +2·t −2·t
· xy 8 + −1−t−t
2 3 4 5 6 7 8 9 10 11 13 2
t24 t21 · y 9 + t121

·
1−3·t−t +3·t +3·t +t −2·t −2·t +2·t +2·t +2·t
2 3 4 5 6 7 8 9 10
4 4
x y + t16 ·x y +
3 5
1+t−2·t2 −2·t3 −2·t4 +2·t5 −t7 −3·t8 −3·t9 +2·t11 −t12

t19 · x 2 6
y +
1+3·t+3·t2 +2·t3 +2·t5 +5·t6 +5·t7 +3·t8 −t10 +t11 +t12 1+t+t2
t21 · xy + t21 · y 8 +
7
1+2·t−t +t +2·t +t −t −t −t
· x3 y 4 + −1−3·t+t −3·t −6·t −6·tt17−4·t +t −t −2·t −3·t ·
3 5 6 7 8 9 10 2 4 5 6 7 8 9 10 11
t13
x2 y 5 + −1−2·t−2·t +3·t −2·tt19 −4·t −t +3·t +t −t
2 4 6 7 8 9 10 11 2
· xy 6 + 1+t+t t15 · y7 +
−2+2·t −2·t −2·t −2·t
· x3 y 3 + 1−2·t−3·t +3·t +3·tt14+t −4·t −t +t +t · x2 y 4 +
2 4 5 6 2 4 5 6 7 8 9 10
t10
−1−4·t−4·t2 −3·t3 −2·t6 −3·t7 −t8
· xy 5 + −1−t−t
2 2 4
+t5 +t6
t15 t15 · y 6 + 1−t−t +t t7 · x3 y 2 +
1+4·t+3·t −t +2·t +4·t +3·t
2 3 5 6 7
1+2·t+2·t +t −t −2·t +2·t +t
2 3 4 5 6 7
−1
t11 ·x y +
2 3
t13 · xy + t8 · y 5 + t23 ·
4
x3 y + −1−t+2·t +2·t −t −t −t
· xy 3 + t18 · y 4 − x3 + −1−3·t
2 3 4 5 6 2
t8 · x2 y 2 + 1+3·t+t t8 t4 · x2 y +
−1−t
t6 · xy + t · x
2 1 2
The vertices of the tropical curve are:

(2, −1/2), (−1, −1/2), (5, −1/2), (−3/2, −1/2), (−4, 0), (1, −1), (−1/2, −1),
(−1, −2/3), (4, −1), (1/2, −3/2), (3, −3/2), (1/5, −8/5), (−1/2, −5/4),
(1/2, −11/4), (−1/5, −12/5), (−3, −5/2), (4, −4), (1/2, −3), (1, −10/3),
(3/2, −7/2), (−1/2, −5/2), (−4, −3), (−1, −3),
(−5, −7/2), (1, −7/2), (−2, −7/2)
2
2
2 2 22
2 3 2
2 2
2
2 2
2 3 2
2 2 2
98

homogeneous non-commutative A-polynomial Anh 81 (M, L, q) has 2112 terms, which
we not present here. The vertices of the tropical curve are:
(3, −1/2), (−1, −1/2), (6, −1/2), (−2, −1/2), (9, −1/2), (2, −1), (−1, −1),
(−5/2, −1/2), (−6, 0), (5, −1), (−2, −3/5), (8, −1), (3/2, −3/2), (4, −3/2),
(−1/2, −3/2), (−3/4, −11/8), (7, −3/2), (1, −2), (3, −2), (0, −2), (6, −2),
(0, −5/2), (5/2, −5/2), (5, −5/2), (1, −3), (0, −3), (−5, −7/2), (0, −7/2),
(−1, −3), (−5/2, −7/2), (3/4, −37/8), (0, −4), (1/2, −9/2), (−6, −4), (6, −6),
(1, −5), (2, −27/5), (5/2, −11/2), (−3, −4), (−1, −4), (−7, −9/2), (−4, −9/2),
(−3/2, −9/2), (−8, −5), (−5, −5), (−2, −5), (−9, −11/2), (2, −11/2),
(−6, −11/2), (1, −11/2), (−3, −11/2)
2
2
2
22 2 2 2
22 32 2
2
2
2
2 22 3 2 2
2 2
2 2
2 2
2 2 3 2 22
2 2 2 2
2 2 3
2 2 2 22
17
5.6. The number of terms of the non-homogeneous A-polynomial of

twist knots. In [GS] we explicitly computed the non-homogeneous A-polynomial
Kp , BKp ) of the twist knots Kp for p = −15, . . . , 15. Kp is the knot obtained by
(Anh
1/p surgery on one component of the Whitehead link. This includes the following
knots in the Rolfsen notation:
K1 = 31 K2 = 52 K3 = 72 K4 = 92
K−1 = 41 K−2 = 61 K−3 = 81 K−4 = 101
The computations reveal that for p = 1, . . . , 15, Anh

Kp has (L, M, q) degree equal to

17
2p − 1, 8p − 4, p(p − 1) + 2
2
The total number of terms of the 3-variable polynomial Anh

Kp is given by
139976, 80252, 41996, 19402, 7406, 2112, 346, 22
for p = −8, . . . , −1, and by
4, 98, 908, 4100, 12236, 28978, 58668, 106800, 179814, 284998, 430652
for p = 1, . . . , 11. Using the data from [GS], the author has computed the tropical
curves (homogeneous or not) of all twist knots Kp with p = −15, . . . , 15. Needless
to say, the output of the computations it too large to be displayed in the paper.
100
5.7. Acknowledgment. The idea of the present paper was conceived during
the New York Conference on Interactions between Hyperbolic Geometry, Quantum
Topology and Number Theory in New York in the summer of 2009. An early ver-
sion of the present paper appeared in the New Zealand Conference on Topological
Quantum Field Theory and Knot Homology Theory in January 2010. The author
wishes to thank the organizers of the New York Conference, A. Champanerkar,
O. Dasbach, E. Kalfagianni, I. Kofman, W. Neumann and N. Stoltzfus and the
New Zealand Conference R. Fenn, D. Gauld and V. Jones for their hospitality and
for creating a stimulating atmosphere. The author also wishes to thank J. Yu for
many enlightening conversations and T. Markwig for the drawing implementation
of polymake.
References
[ADKMV] M. Aganagic, R. Dijkgraaf, A. Klemm, M. Mariño and C. Vafa, Topological strings
and integrable hierarchies, Comm. Math. Phys. 261 (2006) 451–516.
[BP] A. Barvinok and J.E. Pommersheim, An algorithmic theory of lattice points in polyhedra,
New perspectives in algebraic combinatorics, Math. Sci. Res. Inst. Publ., 38 (1999)
Cambridge Univ. Press 91–147.
[BR] M. Beck and S. Robins, Computing the continuous discretely, Integer-point enumeration
in polyhedra, Undergraduate Texts in Mathematics. Springer, (2007).
[BRVD] D.W. Boyd, F. Rodriguez-Villegas and N.M. Dunfield. Mahler’s Measure and the Dilog-
arithm II, preprint 2008 math.NT/0308041.
[Ch] A. Champanerkar, A-polynomial and Bloch invariants of hyperbolic 3-manifolds, Thesis,
Columbia University 2003.
[CCGLS] D. Cooper, D, M. Culler, H. Gillet, D. Long and P. Shalen, Plane curves associated to
character varieties of 3-manifolds, Invent. Math. 118 (1994) 47–84.
[Cou] S. Coutinho, A primer of algebraic D-modules, London Math. Soc. Student Texts, 33
Cambridge University Press 1995.
[CS] M. Culler and P.B. Shalen, Bounded, separating, incompressible surfaces in knot man-
ifolds, Invent. Math. 75 (1984) 537–545.
[CGLS] , C.McA. Gordon, J. Luecke and P.B. Shalen, Dehn surgery on knots, Ann. of
Math. 125 (1987) 237–300.
[DV] R. Dijkgraaf and C. Vafa, Matrix models, topological strings, and supersymmetric gauge
theories, Nuclear Phys. B 644 (2002) 3–20.
[DGKV] , S. Gukov, V.A. Kazakov and C. Vafa, Perturbative analysis of gauged matrix
models, Phys. Rev. D (3) 68 (2003), no. 4 045007, 16 pp.
[DbG] J. Dubois and S. Garoufalidis, Rationality of the SL(2,C)-Reidemeister torsion in di-
mension 3, preprint 2009 arXiv:0908.1690.
[DnG] N.M. Dunfield and S. Garoufalidis, Boundary slopes of 1-cusped manifolds, preprint
2010.
[Eh] E. Ehrhart, Sur les polyèdres homothétiques bordés n dimensions, C. R. Acad. Sci.
Paris 254 (1962) 988–990.
[FKP] D. Futer, E. Kalfagianni and J.S. Purcell, Slopes and colored Jones polynomials of
adequate knots, preprint 2010 arXiv:1002.0256.
[GL] S. Garoufalidis and T.T.Q. Le, The colored Jones function is q-holonomic, Geom. and
Topology, 9 (2005) 1253–1293.
[GS] and X. Sun, The non-commutative A-polynomial of twist knots, Journal of Knot
Theory and Its Ramifications, in press.
[Ga1] , Difference and differential equations for the colored Jones function, J. Knot
Theory Ramifications 17 (2008) 495–510.
[Ga2] , On the characteristic and deformation varieties of a knot, Proceedings of the
CassonFest, Geometry and Topology Monographs 7 (2004) 291–309.
[Ga3] , The Jones slopes of a knot, preprint 2009 arXiv:0911.3627, to appear in the
Journal of Quantum Topology.
19
[Ga4] , The degree of a q-holonomic sequence is a quadratic quasi-polynomial, preprint

2010 arXiv:1005.4580.
[Ga5] , Tropicalization and the Slope Conjecture for 2-fusion knots, preprint 2010.
[Ha] A.E. Hatcher, On the boundary curves of incompressible surfaces, Pacific J. Math. 99
(1982) 373–377.
[HO] and U. Oertel, Boundary slopes for Montesinos knots, Topology 28 (1989) 453–
480.
[Hi] K. Hikami, Note on the colored Jones polynomials of torus knots, preprint 2007.
[Jo] V.F.R. Jones, Hecke algebra representations of braid groups and link polynomials, Ann.
of Math. 126 (1987) 335–388.
[LMV] J.M.F. Labastida, M. Mariño and C. Vafa Knots, links and branes at large N , J. High
Energy Phys. 2000 no. 11, Paper 7, 42 pp.
[Le] T.T.Q. Le, The Colored Jones Polynomial and the A-Polynomial of Two-Bridge Knots,
Advances in Math. 207 (2006) 782–804.
[Ma] T. Markwig, polymake and tropical.lib, software available from
www.mathematik.uni-kl.de/~keilen/en/tropical.html
[NZ] W.D. Neumann and D. Zagier, Volumes of hyperbolic three-manifolds, Topology 24
(1985) 307–332.
[PR] P. Paule and A. Riese, qZeil, Mathematica software available from RISC
www.risc.uni-linz.ac.at
[PWZ] M. Petkovšek, H.S. Wilf and D.Zeilberger, A = B, A.K. Peters, Ltd., Wellesley, MA
1996.
[RGST] J. Richter-Gebert, B. Sturmfels and T. Theobald, First steps in tropical geometry, Con-
temp. Math., 377, Amer. Math. Soc., (2005) 289–317.
[SS] D. Speyer and B. Sturmfels, Tropical mathematics Math. Mag. 82 (2009) 163–173.
[St] R.P. Stanley, Enumerative Combinatorics, Volume 1, Cambridge University Press
(1997).
[Tu] V. Turaev, The Yang-Baxter equation and invariants of links, Inventiones Math. 92
(1988) 527–553.
[WZ] H. Wilf and D. Zeilberger, An algorithmic proof theory for hypergeometric (ordinary
and q) multisum/integral identities, Inventiones Math. 108 (1992) 575–633.
[Z] D. Zeilberger, A holonomic systems approach to special functions identities, J. Comput.
Appl. Math. 32 (1990) 321–368.
School of Mathematics, Georgia Institute of Technology, Atlanta, GA 30332-0160,

USA, http://www.math.gatech.edu/∼stavros
Volume 541, 2011
Quantum coadjoint action and the 6j-symbols of Uq sl2
Stéphane Baseilhac
Dedicated to my advisor, Claude Hayat-Legrand, on the occasion of his 65th birthday.
Abstract. We review the representation theory of the quantum group Uε sl2 C

at a root of unity ε of odd order, focusing on geometric aspects related to the
3-dimensional quantum hyperbolic field theories (QHFT). Our analysis relies
on the quantum coadjoint action of De Concini-Kac-Procesi, and the theory
of Heisenberg doubles of Poisson-Lie groups and Hopf algebras. We identify
the 6j-symbols of generic representations of Uε sl2 C, the main ingredients of
QHFT, with a bundle morphism defined over a finite cover of the algebraic
quotient P SL2 C//P SL2 C, of degree two times the order of ε. It is charac-
terized by a non Abelian 3-cocycloid identity deforming the fundamental five
term relation satisfied by the classical dilogarithm functions, that relates the
volume of hyperbolic 3-polyhedra under retriangulation, and more generally,
the simplicial formulas of Chern-Simons invariants of 3-manifolds with flat
sl2 C-connections.
1. Introduction
After more than twenty years of outstanding efforts, the geometry of quantum
groups, especially the relationships between their representation theories at roots of
unity and the underlying Lie groups, remains a prominent matter of quantum topol-
ogy. It stands, for instance, in the background of the geometric realization prob-
lem of the combinatorially defined state spaces of the Reshetikhin-Turaev TQFT
([BHMV], [T]), the deformation quantization of character varieties of 3-manifolds
via skein modules [BFK, PS], or the asymptotic expansion of quantum invariants
([Prob, Ch. 7], [MN]).
In recent years, new and completely unexpected interactions between the non
restricted quantum group Uq = Uq sl2 C and 3-dimensional hyperbolic geometry have
been revealed by the volume conjecture [Ka], and the subsequent development of
the quantum hyperbolic field theories (QHFT) [BB1, BB2, BB3] and quantum
1991 Mathematics Subject Classification. Primary 17B37, 57R56; Secondary 14L24.

Key words and phrases. quantum groups, Poisson-Lie groups, coadjoint action, geometric
invariant theory, dilogarithm functions, TQFT, invariants of 3-manifolds.
This work was supported by the grant ANR-08-JCJC-0114-01 of the French Agence Nationale
de la Recherche.
2011
c 0000
c Mathematical
holder)
1
103
104
2 STÉPHANE BASEILHAC
Teichmüller theory [BL]. The global picture is that many of the fundamental invari-
ants of hyperbolic (ie. P SL2 C) geometry, like the volume of hyperbolic manifolds,
and more generally (some of the) Chern-Simons invariants of 3-manifolds with flat
sl2 C-connections, should be determined by the semi-classical limits of these quan-
tum algebraic objects.
Like any topological quantum field theory, the QHFT are symmetric tensor
functors defined on a category of 3-dimensional bordisms. For QHFT, these are
equipped with additional structures given by refinements of holonomy representa-
tions in P SL2 C, and links up to isotopy. The QHFT are built from ”local” basic
data, universally encoded by the moduli space of isometry classes of “flattened” (a
kind a framing) hyperbolic ideal 3-simplices, and the associativity constraint, or
6j-symbols, of the cyclic representations of a Borel subalgebra Uε b of Uε at a prim-
itive root of unity ε of odd order. Working with such a moduli space allows one
to define the QHFT also for 3-dimensional bordisms equipped with holonomies
having singularities along the links, like cusped hyperbolic manifolds (the link
being “at infinity”), and to get surgery formulas. For cylindrical bordisms the
QHFT are related by this way to the local version of quantum Teichmüller theory
([BB3],[Bai, BBL]).
Similarly, the Chern-Simons gauge theory for flat sl2 C-connections, which was
originally derived from the integral complex-valued invariant “volume” 3-form of
SL2 C via secondary characteristic class theory, defines a functor by means of the
same moduli space as the QHFT, using Neumann’s simplicial formulas [N] in place
of quantum state sums, where a dilogarithm function formally corresponds to the
cyclic 6j-symbols of Uε b ([BB2], see also [Ma]). In fact, both maps are 3-cocycloids
in a natural way, the latter on the category of cyclic Uε b-modules with (partially
defined) tensor product, and the former on the group P SL2 C with discrete topology,
via Neumann’s isomorphism of H3 (BP SL2 Cδ ; Z) with a certain extension of the
Bloch group.
In this paper we consider this interplay of Abelian vs. non Abelian coho-
mological structures, which certainly concentrates a key part of the quantization
procedure relating the Chern-Simons theory for P SL2 C to the QHFT. We describe
in detail the non Abelian part of the story, that is, how the simple Uε -modules
“fiber” over P SL2 C, and the 6j-symbols of regular Uε -modules (rather than the
cyclic Uε b-modules). By the way we indicate common features and discrepancies
with the 6j-symbols of the color modules of the restricted quantum group Ūε .
We point out also that the cyclic 6j-symbols of Uε b, or (basic) matrix dilog-
arithms, coincide with the regular 6j-symbols of Uε . More precisely, we define a
bundle Ξ(2) of regular Uε -modules over a covering of degree n2 (n being the order
of ε) of a smooth subset of a Poisson-Lie group H 2 dual to P SL2 C2 , endowed with
an action of an infinite dimensional Lie group derived from the quantum coadjoint
action of De Concini-Kac-Procesi, originally defined for Uε . We have (see Theorem
6.13, 6.14 and 6.16 for precise statements):
Theorem 1.1. The regular 6j-symbols of Uε and the matrix dilogarithms coin-
cide and define a bundle morphism R : Ξ(2) −→ Ξ(2) equivariant under the quantum
coadjoint action.
Since the quantum coadjoint action lifts the adjoint action of P SL2 C via an
unramified 2-fold covering H → P SL2 C0 of the big cell of P SL2 C, it will follow
QUANTUM COADJOINT ACTION AND THE 6j-SYMBOLS OF Uq sl2 105
3
that R descends to a morphism of a vector bundle of rank n2 over a 2n-fold covering

of the algebraic quotient P SL2 C//P SL2 C.
The remarkable dependence of the matrix dilogarithms on cross-ratios of 4-
tuples of points on P1 , which was previously known by direct computation and
allowed the QHFT to be defined on the moduli space of flattened hyperbolic ideal
tetrahedra, is a consequence of Theorem 1.1. This alternative description is pre-
sented in Section 6.4.
In order to achieve our goals we have to review quite a lot of material, starting
from the basic properties of Uε (Section 2), and developing in detail its represen-
tation theory and quantum coadjoint action (Section 4 and 3, respectively). The
6j-symbols of the color modules of Ūε and of the regular modules of Uε are defined
in Section 5. Theorem 1.1 is proved in Section 6. There we make a crucial use of
fundamental results of Semenov-Tian-Shansky [STS], Weinstein-Xu [WX], and Lu
[Lu1] on Poisson-Lie groups and their doubles and quantizations.
There should be no obstruction to extend Theorem 1.1 to the quantum groups
Uε g of arbitrary complex simple Lie algebras g. The corresponding quantum coad-
joint action theory is described in [DCK], [DCKP], [DCP], and [DCPRR]. Re-
cent works of Geer and Patureau-Mirand [GP] show that the QHFT setup extends
to ”relative homotopy quantum field theories”, including TQFT associated to the
categories of finite dimensional weight Uε g-modules. A challenging problem is to
relate them to the three-dimensional Chern-Simons theory for flat g-connections.
Finally, let us note that a similar approach can be used to describe “holonomy”
R-matrices for the regular Uε g-modules, in the spirit of [KaRe].
We will often meet notions from the theory of Poisson-Lie groups. The reader
will find the needed material in standard textbooks, like [CP], [ES] and [KS].
This paper is based on my lectures at the workshop “Interactions Between
Hyperbolic Geometry, Quantum Topology and Number Theory” held at Columbia
University (June 3rd - 19th, 2009). I am grateful to the organizers for their hospi-
tality and wonderful working conditions during my stay.
2. The quantum group Uq sl2

2.1. Definition & PBW basis. We fix our ground ring to be C, and denote
by q a complex number such that q = −1, 0, 1. When q is constrained to be a root
of unity we denote it by ε, and we assume that ε has odd order n ≥ 3.
Definition 2.1. The quantum group Uq = Uq sl2 is the algebra generated over
C by elements E, F , K and K −1 , with defining relations KK −1 = K −1 K = 1 and
K − K −1
(2.1) KEK −1 = q 2 E , KF K −1 = q −2 F , [E, F ] = .
q − q −1
The algebra Uq is a Hopf algebra with coproduct Δ : Uq → Uq ⊗ Uq , antipode
S : Uq → Uq and counit η : Uq → C defined on generators by
Δ(K) = K ⊗ K
Δ(E) = E ⊗ 1 + K ⊗ E, Δ(F ) = 1 ⊗ F + F ⊗ K −1
(2.2)
S(K) = K −1 , S(E) = −K −1 E, S(F ) = −F K
η(K) = 1, η(E) = η(F ) = 0.
To make sense of this definition, let us just recall here that being a Hopf algebra
means that Δ and η are morphisms of algebras satisfying the coassociativity and
106
counitality constraints
(2.3) (Δ ⊗ id) ◦ Δ = (id ⊗ Δ) ◦ Δ , (η ⊗ id) ◦ Δ = (id ⊗ η) ◦ Δ = id.
Here the algebra structure of Uq ⊗ Uq is by componentwise multiplication, and Uq
is identified with Uq ⊗ C in the canonical way. The antipode S is the inverse of the
identity for the convolution product, that is, it satisfies
(2.4) μ ◦ (id ⊗ S) ◦ Δ = μ ◦ (S ⊗ id) ◦ Δ = η1,
where μ : Uq ⊗ Uq → Uq is the product. In particular, this implies S(1) = 1,
η ◦ S = η, S(xy) = S(y)S(x) for all x, y ∈ Uq , and
(2.5) (S ⊗ S)Δ = τ ◦ Δ ◦ S,
where τ (u ⊗ v) = v ⊗ u is the flip map ([Ks, Ch. III]).
Remark 2.2. A simply-connected version of Uq is often considered in the liter-
ature (see eg. [DCP, §9]). It is obtained by adding to the generators a square root
of K acting by conjugation on E and F by multiplication by q and q −1 , respectively.
Exercise 2.3. Define an algebra Uq with generators E, F , K and K −1 satis-
fying the same relations as in Uq except the last one in (2.1), and with a further
generator L such that
[E, F ] = L , (q − q −1 )L = K − K −1
(2.6)
[L, E] = q(EK + K −1 E) , [L, F ] = −q −1 (F K + K −1 F ).
Show that we have isomorphisms Uq ∼ = Uq , U1 ∼
= U [K]/(K 2 − 1), U ∼
= U1 /(K − 1),
where U = U sl2 is the universal enveloping algebra of sl2 .
This exercise shows that, as suggested by the notation, Uq is a genuine defor-
mation depending on the complex parameter q of the universal enveloping algebra
of sl2 . Like the latter, we can think of Uq as a ring of polynomials (in non com-
muting variables). In particular, we have the following fundamental result (see [J,
Th. 1.5-1.8] or [Ks, Th. VI.1.4] for a proof):
Theorem 2.4. 1) (PBW basis) The monomials F t K s E r , where t, r ∈ N and
s ∈ Z, make a linear basis of Uq .
2) The algebra Uq has no zero divisors and is given a grading by stipulating
that each monomial F t K s E r is homogeneous of degree r − t.
Note that the relations (2.1) are homogeneous of degree 1, −1 and 0, respec-
tively. Products of monomials can be written in the basis {F t K s E r }r,t∈N,s∈Z by
using the two first commutation relations in (2.1), together with

min(r,s)
(2.7) ErF s = F s−i hi E r−i , r, s ∈ N,
i=0
where the hi are Laurent polynomials in C[K, K −1 ] given by

i
r s
hi = [i]! [K; i + j − (r + s)] .
i i
j=1
Here we assume that the product is 1 for i = 0, and we put

Kq l − K −1 q −l
(2.8) [K; l] = .
q − q −1
5
We use also the standard notations for q-integers, q-factorials, and q-binomial co-
efficients

q l − q −l l [l]!
(2.9) [l] = , [l]! = [l] [l − 1] . . . [1] , = , l ∈ Z,
q − q −1 m [m]! [l − m]!
with [0]! = 1 by convention. Note that [l] = q l−1 + . . . + q 1−l , so [l] ∈ Z[q, q −1 ].
Also,

l l l
(2.10) = = 1 and [l] divides if 1 < m < l and l is odd.
0 l m
The elements (2.8) appear in all computations involving commutators. In the sequel
we will use (2.7) when r = 1 and s is arbitrary, or r is arbitrary and s = 1:
(2.11) EF s = F s E + [s]F s−1 [K; 1 − s] , F E r = E r F − [r]E r−1 [K; r − 1].
Another important identity is the q-binomial formula, which holds for any u, v such
that vu = q 2 uv [Ks, Ch. VI (1.9)]:
r
r j(r−j) j r−j
(2.12) (u + v)r = q u v .
j
j=0

l
In particular, it implies that ∈ Z[q, q −1 ].
m
Exercise 2.5. (a) Prove (2.11) by induction, using the relations (2.1).
(b) Show that the second relation in (2.11) follows from the first one by applying
the Cartan automorphism of the algebra Uq , defined by
(2.13) ω(E) = F , ω(F ) = E , w(K) = K −1 .
2.2. Center. Theorem 2.4 gives a lot of information on the center Zq of Uq .
Indeed, denote by Ud the degree d piece of Uq . For any u ∈ Ud we have
(2.14) KuK −1 = q 2d u.
Hence, when q is not a root of unity, Ud is the eigenspace of adK with eigenvalue
q 2d . In particular, Zq ⊂ U0 .
When q = ε is a root of unity of odd order n ≥ 3, Zε is much bigger. Let
us show that it is generated by E n , F n and Zε ∩ U0 . Since the relations (2.1) are
homogeneous, the homogeneous parts of a central element are central. Hence it is
enough to prove our claim for a central element u ∈ Ud . By (2.14), if u ∈ Ud is
central then n divides d. Now, if d = ln for some l ∈ Z \ {0} with l > 0 (resp.
l < 0), Ud is spanned by the monomials F t K s E t+ln (resp. F t+ln K s E t ). Hence
u = u E ln (resp. u = F ln u ) for some u ∈ U0 . The relations
K l E = q 2l EK l , K l F = q −2l F K l
(2.15) , l∈N
KE l = q 2l E l K , KF l = q −2l F l K
imply that K ±n ⊂ Zε . Together with [n] = 0 and (2.11) for s = n and r = n, they
give also E n , F n ⊂ Zε . Since Uε has no zero divisors, we get that any u ∈ Ud ∩ Zε
can be decomposed as a product of central elements E ln and F ln with a central
element in U0 .
Definition 2.6. The subalgebra Z0 of the center Zε of Uε is generated by E n ,
F n
and K ±n .
108
The algebra Z0 plays a key role in the structure of Uε sl2 . In particular, the
next theorem shows that any element of Zε is a root of a monic polynomial with
coefficients in Z0 .
Theorem 2.7. 1) The algebra Uε is a free Z0 -module of rank n3 , with basis
the monomials F t K s E r , 0 ≤ t, s, r ≤ n − 1.
2) The center Zε of Uε is a finitely generated algebra, and is integrally closed
over Z0 .
Proof. The first claim follows directly from Theorem 2.4 (note in particular
that E n , F n and K ±n are independent). It implies that Uε is a noetherian Z0 -
module, and so the Z0 -submodule Zε is finitely generated. Hence Zε is integrally
closed over Z0 [AMcD, Th. 5.3]. Then, by Hilbert’s basis theorem Zε is finitely
generated as an algebra.
In another direction, we will see in the next section that Z0 is the basic link
between Uε and the Lie group P SL2 C. This depends on the fact that it is a
(commutative) Hopf algebra. To prove this, let us introduce for future reference a
preferred set {x, y, z} of generators of Z0 , given by
(2.16) x = −(ε − ε−1 )n E n K −n , y = (ε − ε−1 )n F n , z = K n .
We also put
(2.17) e = (ε − ε−1 )n E n , f = −(ε − ε−1 )n F n K n .
Note that
x = T (y) , f = T (e), x = S(e) , y = S(f )
where S is the antipode of Uε , and T the braid group automorphism of the algebra
Uε , defined by
(2.18) T (E) = −F K , T (F ) = −K −1 E , T (K) = K −1 , T (K −1 ) = K.
Lemma 2.8. The subalgebra Z0 of Uε is a Hopf subalgebra.
Proof. We have to show that Δ(Z0 ) ⊂ Z0 ⊗ Z0 and S(Z0 ) ⊂ Z0 . Because
these maps are morphisms of algebras, it is enough to prove this on the generators
x, y and z of Z0 . We claim that
Δ(z) = z ⊗ z
Δ(x) = x ⊗ z −1 + 1 ⊗ x, Δ(y) = 1 ⊗ y + y ⊗ z −1
(2.19)
S(z) = z −1 , S(x) = −zx, S(y) = −yz
η(z) = 1, η(x) = η(y) = 0.
The proof of the formulas for S is immediate. As for Δ, they are consequences of
(2.10), the q-binomial identity (2.12), and the fact that [n] = 0 when q = ε.
So far, we have only considered in (2.14) the invariance of the center of Uq

under conjugation by K. In order to describe the effect of the conjugation action
by E and F , let us introduce the automorphisms γl of C[K, K −1 ] given by
(2.20) γl (K) = q l K, l ∈ Z.
By (2.1), for all h ∈ C[K, K −1 ] we have
hE = Eγ2 (h) and hF = F γ−2 (h).
7

It is easy to check that u = i≥0 F i hi E i ∈ U0 satisfies uE = Eu and uF = F u in
Uq if and only if for all i,
(2.21) hi − γ−2 (hi ) = [i + 1][K; −i]hi+1 .
Since C[K, K −1 ] is an integral domain, hi+1 is thus determined inductively by h0
if [j + 1] = 0 for all 0 ≤ j ≤ i. This condition is satisfied for all i when q is not a
root of unity, and for i < n − 1 when q = ε. Hence the projection map
π : U0 −→ C[K, K −1 ]
(2.22) ∞ r
r=0 F hr E
r
−→ h0
is injective over Zq ⊂ U0 when q is not a root of unity, and injective over Zε ∩ U0
when q = ε, where
n−1

−1
(2.23) U0 = F hi E | h0 , . . . , hn−1 ∈ C[K, K ] ⊂ U0 .
i i
i=0
In fact, Ker(π) = F U0 E is a two-sided ideal in U0 , so π is a homomorphism

of algebras, the Harish-Chandra homomorphism. It is easy to check that Zε is
generated by F n , E n and Zε ∩ U0 . On another hand, (2.21) is solved by
qK + q −1 K −1
h0 = , h1 = 1 , hr = 0 if r ≥ 2,
(q − q −1 )2
which defines the Casimir element
qK + q −1 K −1
(2.24) Ω = FE + ∈ Zq .
(q − q −1 )2
Note that (see (2.20))
K + K −1
γ−1 ◦ π(Ω) = .
(q − q −1 )2
We will see in Section 4 that Zq is actually generated by Ω, and Zε ∩ U0 by Ω
and K ±n (see Theorem 4.9; for an alternative approach based on induction on i in
(2.21), see [J, Prop. 2.20]). More precisely:
Theorem 2.9. (1) If q is not a root of unity, Zq is a polynomial algebra over
C generated by Ω, and γ−1 ◦ π : Zq → C[K + K −1 ] is an isomorphism.
(2) If q = ε is a root of unity of odd order n ≥ 3, then Zε is the algebra over C
generated by E n , F n , K ±n and Ω with relation

n−1
K n + K −n − 2
(2.25) (Ω − cj ) = E n F n +
j=0
(ε − ε−1 )2n
where
εj+1 + ε−j−1
cj = .
(ε − ε−1 )2
Statement (2) is part of [DCK, Th. 4.2]. It shows that Zε is a finite extension
of Z0 built on the image C[Ω] of the injective homomorphism
(2.26) (γ−1 ◦ π)−1 : C[K + K −1 ] −→ Zε .
The domain C[K + K −1 ] is the fixed point set of C[K, K −1 ] under the involution
s(K ±1 ) = K ∓1 , which can be identified with a generator of the Weyl group of
P SL2 C.
110
3. The quantum coadjoint action for Uε

Consider the set Spec(Zε ) of algebra homomorphisms from the center Zε of Uε
to C. An element of Spec(Zε ) is called a central character of Uε . The set Spec(Zε )
has a very rich Poisson geometry related to the adjoint action of P SL2 C, that we
are going to describe. It will be used in Section 4.
Since Zε is finitely generated, by Hilbert’s Nullstellensatz Spec(Zε ) is an affine
algebraic set. The inclusion Z0 ⊂ Zε induces a regular (restriction) map
(3.1) τ : Spec(Zε ) −→ Spec(Z0 ).
We have seen in Proposition 2.7 that Zε is integrally closed over Z0 . Hence τ is
finite: any y ∈ Im(τ ) has a finite number of preimages. From this property it
follows that τ is surjective [Sh, §5.3].
3.1. A Poisson-Lie group structure on Spec(Z0 ). We need some explicit
formulas. Denote by T , U± and P B± the subgroups of P SL2 C = SL2 C/(±1) of
diagonal, upper/lower unipotent and upper/lower triangular matrices up to sign,
with their natural structures of affine algebraic groups induced by the adjoint action
on sl2 C. The action

a b aw + b
: w → , w ∈ P1
c d cw + d
identifies P SL2 C with the group Aut(P1 ) of automorphisms of the Riemann sphere,
and T with the subgroup of maps w → zw, z ∈ C∗ . Consider the group
H = {(tu+ , t−1 u− ) |t ∈ T, u± ∈ U± } ⊂ P B+ × P B− .
By evaluating on the generators x, y and z of Z0 (see (2.16)) we get a map
ψ : Spec(Z0 ) −→ C2 × C∗
(3.2)
g −→ (xg , yg , zg ).
For all g ∈ Spec(Z0 ), consider the automorphisms of P1 given in ψ -coordinates by
ψ1 (g) : w → zg w − xg zg , ψ2 (g) : w → w/(−zg yg w + zg ), w ∈ P1 .
Under the above isomorphism of Aut(P1 ) with P SL2 C, they correspond respec-
tively to
√ √
√
zg −xg zg 1/ zg 0
± √ ∈ P B+ , ± √ √ ∈ P B−
0 1/ zg −yg zg zg
√
for any choice of square root zg . Define
ψ : Spec(Z0 ) −→ H
g −→ (ψ1 (g), ψ2 (g)).
Clearly, ψ is an isomorphism of algebraic varieties. It gives H a Poisson-Lie bracket
as follows. By using the PBW basis {F t K s E r }r,t∈N,s∈Z we can identify Uε as
a linear subspace of Uq , considered as a family of algebras over C with varying
parameter q. Hence, for any given element u ∈ Uε we can specify a lift ũ ∈ Uq , such
that
u = ũ mod(q n − q −n )Uq .
We can write this as u = limq→ε ũ. Then, for all a ∈ Z0 and u ∈ Uε , put
[ã, ũ]
Da (u) = lim .
q→ε n(q n − q −n )
9
Note that

1 ã 1 ã
(3.3) Da (u) = lim , ũ = lim , ũ
(ε − ε−1 ) q→ε n[n] (ε − ε−1 )n q→ε [n]!
where we use the formula [n − 1]! = n(ε − ε−1 )1−n .
Proposition 3.1. (i) The map ψ : Spec(Z0 ) → H is an isomorphism of alge-
braic groups.
(ii) The maps Da : Uε → Uε are well-defined derivations of Uε preserving Z0
and Zε . At the generators e = −xz and z in (2.17) we have:
(ε − ε−1 )n−1 (ε − ε−1 )n ±1 n
(3.4) De (F ) = [K; 1]E n−1 , De (K ±1 ) = ∓ K E
n n
1 n 1
(3.5) K E , Dz (F ) = − F K n .
Dz (E) =
n n
where [K; l] is given by (2.8) with q = ε.
Proof. (i) This follows from classical duality arguments in Hopf algebra theory
(see [Mo, § 9]). Because Z0 is a commutative Hopf algebra (Lemma 2.8), the
algebraic set Spec(Z0 ) has a canonical group structure identifying Z0 with the
algebra of regular functions on Spec(Z0 ). It is defined dually by stipulating that
for any u ∈ Z0 and g, h ∈ Spec(Z0 ),
(3.6) u(gh) = μ ◦ Δ(u)(f, g) , u(g −1 ) = S(u)(g) , u(11) = η(u).
Here, 11 ∈ Spec(Z0 ) is the identity and μ the product of Z0 . The associativity of the
product of Spec(Z0 ) follows from the coassociativity of Δ (see (2.3)). The inverse
is well-defined by u(11) = η(u) and the computation
u(gg −1 ) = μ ◦ (id ⊗ S) ◦ Δ(u)(g, g) = η(u)1(g) = η(u),
where 1 ∈ Z0 is identified with the constant function on Spec(Z0 ) with value 1, and
we use (2.4) in the second equality. By (2.19) and (3.6) we get in the ψ -coordinates
(xg , yg , zg ) of the point g of Spec(Z0 ):
xgh = xh + xg zh−1 , ygh = yh + yg zh−1 , zgh = zg zh
(3.7) xg−1 = −zg xg , yg−1 = −yg zg , zg−1 = zg−1
x1 = 0 , y1 = 0 , z1 = 1.
Now we have
ψ1 (g)ψ1 (h)(w) =zg (zh w − zh xh ) − zg xg
=zg zh w − zg zh (xh + xg zh−1 ),
which is equal to ψ1 (gh) by (3.7). A similar computation gives ψ2 (g)ψ2 (h)(w) =
ψ2 (gh). Hence ψ is a group homomorphism.
(ii) The maps Da are well-defined because a ∈ Zε implies [ã, ũ] ∈ (q n − q −n )Uq .
They are derivations of Uε because the commutator [ , ] of Uq satisfies the Leibniz
rule. Clearly, they map Zε to Zε , and also Z0 to Z0 since for all u ∈ Z0 , Da (u) is
a sum of monomials in Z0 .
All this can also be check by direct computation, as follows. By (2.7), the
products of basis elements of Uq read as

min(r,c)
t s r c
(F K E )(F K E ) = b a
q −2((r−i)b+s(c−i)) F c+t−i hi K s+b E r+a−i .
i=0
112
Suppose that r, s and t are (possibly 0) multiples of n, say r = r n, s = s n

and t = t n. Then, all the hi except h0 = 1 are divided by [ln] for some l, and
so vanish at q = ε. Since the coefficient of F c+t K s+b E r+a in the commutator
[F t K s E r , F c K b E a ] is

q −2n(r b+s c) − q −2n(s a+bt ) = q −n(b(r +t )+s (a+c) [n(s (a − c) + b(t − r ))],
and limq→ε [nl]/[n] = l for any integer l, the limit (3.3) is well-defined.
The formula for De (F ) follows from a straightforward computation using (2.11)
and (2.15). Also we have
[(q − q −1 )n E n , K ±1 ]
De (K ±1 ) = lim
q→ε n(q n − q −n )
(q − q −1 )n q ∓2n − 1 ±1 n
= lim K E
q→ε n q n − q −n
(ε − ε−1 )n ±1 n
=∓ K E .
n
We get Dz by similar computations.
For all u, v ∈ Z0 and f ∈ Spec(Z0 ), put

{u, v}(f ) = Du (v)(f ).
By Proposition 3.1, for all a ∈ Z0 , Da : Z0 → Z0 defines an algebraic vector field
on Spec(Z0 ), and { , } is a bivector satisfying the Jacobi identity. Hence:
Corollary 3.2. The bivector { , } is a Poisson bracket on Spec(Z0 ).

Exercise 3.3. (a) Check that
{e, x} = −zx2 , {e, y} = z − z −1 , {e, z} = xz 2
{f, x} = z −1 − z , {f, y} = y 2 z , {f, z} = −yz 2
{y, x} = 1 − xy − z −2 , {z, x} = zx , {z, y} = −yz.
(b) Deduce that the Poisson bracket ψ∗ { , } on H does not depend on the order n
of ε.
(c) Show that for any algebra automorphism φ of Uε we have
(3.8) φDa φ−1 (u) = Dφ(a) (u).
Deduce that the group of algebra automorphisms of Uε induces a group of Poisson
automorphisms of (Spec(Z0 ), { , }). By using the braid group automorphism (2.18),
compute the value of the derivation Df on E, F and K ±1 .
(d) Show that the derivations Dã obtained from the Da by allowing the lift ã
to be arbitrary generate a Lie algebra L. Show that L fits into an exact sequence
of Lie algebras 0 −→ L0 −→ L −→ L −→ 0, where L0 (resp. L ) is the Lie algebra
of inner derivations of Uε (resp. of derivations of Z0 induced by { , }) .
Remark 3.4. (Semi-classical geometry, I) By Exercise 3.3 the Poisson bracket
ψ∗ { , } is canonically associated to Uq . Conversely, (H, ψ∗ { , }) is dual to the
standard Poisson-Lie structure on P SL2 C, and the rational quantization of the
latter is Uq ([DCKP, §7], [DCP, §11, 14 & 19]).
11
Denote by P SL2 C0 = U− T U+ the big cell of P SL2 C; it is the open subset

consisting of matrices up to sign with non vanishing upper left entry. We have an
unramified 2-fold covering
σ: H −→ P SL2 C0
(3.9)
(t −1
u− , tu+ ) −→ u−1 2
− t u+ .
Consider the matrices

0 1 0 0 1 0
(3.10) , , .
0 0 1 0 0 −1
They correspond to the Chevalley generators of sl2 C in its fundamental represen-
tation. Since σ is an unramified covering, the associated (complex) left invariant
vector fields on P SL2 C lift to vector fields e, f and h on H. By using Proposition
3.1 we consider them as vector fields on Spec(Z0 ).
Recall the generators e, f and z of Z0 , which define functions on Spec(Z0 ).
Denote by g (resp. g̃) the Lie algebras of vector fields on Spec(Z0 ) generated by e
and f (resp. by De and Df ).
Proposition 3.5. We have
(3.11) De = zf , Df = −ze , Dz = zh/2
and
(3.12) [De , Df ] = z 2 h − z 2 xe + z 2 yf .
Hence the linear spans of g and g̃ coincide at every point of Spec(Z0 ).
Proof. Since all vector fields are algebraic, it is enough to check (3.11) on
functions on Spec(Z0 ) ∼
= H lifting functions on P SL2 C0 , and we can restrict to
coordinates of the map σ. Put

√
1 0 1 −x z 0√
u− = ± , u+ = ± , t=±
−y 1 0 1 0 1/ z
and
z −zx
M = σ(tu+ , t−1 u− ) = u−1
− t u+ = ±
2
.
zy −zxy + z −1
By Exercise 3.3 (a) we have

z2x 0
De M = ± .
z xy + z 2 − 1
2
−z 2 x
Let a ∈ sl2 C. Denote by a∗ the left invariant vector field on P SL2 C associated to
a, and a its pull-back to Spec(Z0 ) via σ. By definition, a(M ) = a∗ (M ), and for all
g ∈ P SL(2, C)0 ,
d d
a∗ (M )(g) = M (geat ) t=0 = (R∗at M )t=0 (g),
dt dt e
where Reat denotes right translation by eat . On another hand, for any vector field
X on P SL2 C we have Rh∗ X = Adh−1 (X). Since the adjoint representation is
faithful, dually (Rh∗ ρ)(g) = Adh (ρ(g)) for any linear representation ρ of P SL2 C.
Hence
a∗ (M )(g) = ada (M )(g) = [a, M (g)].
114
It is immediate to check that

0 0
[ , M ] = De M.
z 0
Hence De = zf . The formulas for Df and Dz can be proved in the same way. We
obtain (3.12) by using [e, f ] = h, the formulas in Exercise 3.3 (a), and the fact that
[f U, gV ] = f g[U, V ] + f U (g)V − gV (f )U
for any vector fields U , V and functions f , g.
3.2. Flows on Uε . The infinitesimal action of the Lie algebra g̃ generated by
De and Df can be integrated to an infinite dimensional group of automorphisms of
the algebra Uε . In fact, in order to make sense of this we have to allow holomorphic
series in the generators of Z0 as coefficients.
More precisely, denote by Ẑ0 the algebra of power series in the generators x, y,
z and z −1 of Z0 which converge to a holomorphic function for all values of (x, y, z)
in C2 × C∗ . Set
(3.13) Ûε = Uε ⊗Z0 Ẑ0 , Ẑε = Zε ⊗Z0 Ẑ0 .
Proposition 3.6. For all values of t ∈ C the series exp(tDe ), exp(tDf ) and
exp(tDz ) converge to automorphisms of Ûε preserving Ẑ0 and Ẑε .
Proof. By (3.5) the statement is clearly true for exp(tDz ). To conclude it is
enough to check that the series exp(tDe ) when applied to the generators K and F
converge to an element of Ûe ; the result for exp(tDf ) follows from this by using
T (e) = f (see (2.17)) and Exercise 3.3 (c). Now, by (3.4)-(3.5) we have
exp(tDe )K = exp(−te/n)K
and
(ε − ε−1 )n−2 l−1
(De )l (F ) = − e ((−1)l Kε + K −1 ε−1 )E n−1 .
nl
Hence

e−te/n − 1 ete/n − 1 −1 −1
exp(tDe )F = F − (ε − ε−1 )n−2 Kε + K ε E n−1 .
e e
The stability of Ẑ0 and Ẑε follows from that of Z0 and Zε under the derivations
Da (see Proposition 3.1).
Definition 3.7. The subgroup G of Aut(Ûε ) is generated by the 1-parameter

groups exp(tDe ) and exp(tDf ), t ∈ C.
Since G leaves Ẑ0 and Ẑε invariant, it acts dually by holomorphic transforma-
tions on the varieties Spec(Z0 ) and Spec(Zε ):
∀g ∈ G, u ∈ Ẑε , χ ∈ Spec(Zε ), (g.u)(χ) = u(g −1 .χ).
By Proposition 3.5, the G-action on Spec(Z0 ) lifts the conjugation action of P SL2 C.
Also, it follows from (3.12) that the G-orbits are both open and closed in the preim-
age of any conjugacy class, and so are connected. Since any element of P SL2 C is
conjugate to one in a given Borel subgroup, any conjugacy class in P SL2 C inter-
sects P SL2 C0 in a non empty smooth connected variety. Hence we have [DCKP,
Prop. 6.1]:
13
Theorem 3.8. (i) For any conjugacy class Γ in P SL2 C, the connected com-
ponents of the variety σ −1 Γ are orbits of G in Spec(Z0 ).
(ii) The set σ −1 ({id}) coincides with the fixed point set of G in Spec(Z0 ).
Remark 3.9. (Semi-classical geometry, II) (i) By construction, the orbits of
the G-action on Spec(Z0 ) are symplectic leaves for the Poisson bracket { , }. A result
of M. Semenov-Tian-Shansky implies that they coincide under ψ with the orbits of
the dressing action of P SL2 C on H ([KS, Ch. 1.3], [CP, Ch. 1.5]) (see Remark
3.4). However there does not seem to have any homomorphism G → P SL2 C relating
these two actions on H.
(ii) The action of G on Spec(Z0 ) is called the quantum coadjoint action [DCK],
because it coincides at the tangent space of fixed points with the coadjoint action
of P SL2 C on its dual Lie algebra.
4. Representation theory of Uε
In this section we use the quantum coadjoint action of G on Spec(Z0 ), and its
extension to Ûe , to describe the representation theory of Uε in geometric terms.
The main results are Theorem 4.11 and Theorem 4.12.
To begin with, recall that two Uε -modules V and W are isomorphic if there
exists a linear isomorphism θ : V → W commuting with the action of Uε . A Uε -
module is simple if it has no proper non trivial submodule. In what follows, all
modules are left modules.
It is classical that any simple Uε -module is a finite dimensional vector space
[CP, p. 339], and that the action of the center Zε is by scalars. More precisely:
Lemma 4.1. Any central element u ∈ Zq acts on a finite dimensional simple
Uq -module V by multiplication by a scalar.
Proof. Since C is algebraically closed and V is finite dimensional, the linear
operator uV ∈ End(V ) associated to u has an eigenvalue λ. Because u is central,
uV commutes with the action of Uq . Hence the kernel of uV − λ.id must be the
whole of V , for otherwise it would be a proper non zero submodule.
By the lemma we can associate to any simple Uε -module V the central character
χV : Zε → C in Spec(Zε ) such that uv = χV (u)v for all u ∈ Zε and v ∈ V . Clearly
any two isomorphic Uε -modules define the same central character. Hence, denoting
by Rep(Uε ) the set of isomorphism classes of simple Uε -modules, we have a central
character map
Ξ: Rep(Uε ) −→ Spec(Zε )
(4.1)
V −→ χV .
For any χ ∈ Spec(Zε ), consider the two-sided ideal I χ of Uε generated by its kernel,
Ker(χ) = {z − χ(z).id|z ∈ Zε }. Define an algebra
(4.2) Uεχ = Uε /I χ .
By Theorem 2.7 and the fact that the degree of a homogeneous element of Zε is
always a multiple of n, the algebra Uεχ is finite dimensional over C and non zero.
It is also a Uε -module by the left regular representation on cosets, and clearly any
simple submodule of Uεχ is in Ξ−1 (χ). This proves:
Lemma 4.2. The central character map Ξ is surjective.
116
In order to clarify the properties of Ξ we are going to describe the set Rep(Uε )
explicitly.
4.1. The simple Uε -modules. Consider the following two families of Uε -
modules:
• the modules Vr± of dimension r + 1, where 0 ≤ r ≤ n − 1, with a basis
v0 , . . . , vr such that

vj+1 , if j < r
Kvj = ±ε r−2j
vj , F vj =
0, if j = r

±[j][r − j + 1]vj−1 , if j > 0
Evj =
0, if j = 0.
• the modules V (λ, a, b) of dimension n, where λ ∈ C∗ and a, b ∈ C, with a
basis v0 , . . . , vn−1 such that

vj+1 , if j < n − 1
Kvj = λε−2j vj , F vj =
bv0 , if j = n − 1

avn−1 ,
if j = 0
E · vj = 1−j
−λ−1 εj−1
ab + [j] λε ε−ε −1 vj−1 , if j > 0.
That these formulas actually define actions of Uε is a direct consequence of (2.1)
and (2.11). We will see in Section 5 that the modules Vr− and Vr+ are essentially
equivalent (we have an isomorphism Vr− = V0− ⊗ Vr+ ). The two families of modules
±
Vr± and V (λ, a, b) intersect exactly at Vn−1 = V (±ε−1 , 0, 0). The action of Uε on
±
the modules Vr is often expressed in the literature in a different form, by using
the “balanced” basis mj = vj /[j]! rather than the vj .
4.1.1. Highest weight Uε -modules. The modules Vr± and V (λ, 0, b) have the
important common property to be highest weight modules.
Recall that a Uε -module is a highest weight module of highest weight λ ∈ C∗ if
it is generated by a non zero vector v such that Ev = 0 and Kv = λv; v is called
a highest weight vector. It is canonical up to a scalar factor, since K is diagonal
in the basis F i v. In fact, there is a universal Uε -module of highest weight λ, the
(infinite-dimensional) Verma module M (λ) defined by
M (λ) = Uε /(Uε · E + Uε · (K − λ)).
Equivalently, M (λ) has for highest weight vector the coset v0 of 1 ∈ Uε , and has
the basis v0 , v1 , v2 , . . . where vi is the coset of F i and the action of Uε is given by
Kvj = λε−2j vj , F vj = vj+1
1−j
−λ−1 εj−1
[j] λε ε−ε −1 vj−1 , if j > 0
Evj =
0, if j = 0.
The linear independence of the vi follows from Theorem 2.4. From the formulas we
see that the modules Vr± and V (λ, 0, b) are highest weight Uε -modules of highest
weights ±εr and λ, respectively, and that
Vr± = M (±εr )/Mr , V (λ, 0, b) = M (λ)/Uε · (vn − bv0 ),
where Mr is the submodule of M (±εr ) spanned by the vi with i > r. Note that
Uε · (vn − bv0 ) is spanned by all F i (vn − bv0 ) = vi+n − bvi , since [n] = 0 implies
Evn = Ev0 = 0, and K(vn − bv0 ) = λ(vn − bv0 ).
15
The modules Vr± and M (λ) extend to modules Vr,q ±

and M (λ)q over Uq for
all values of q such that q = 1, by replacing ε by q in all formulas. The Uq -
2
±
modules Vr,q are defined in any dimension, hence with no bound on r, and still
±
satisfy Vr,q = M (±q r )q /Mr . They are simple modules, and any simple Uq -module
±
is isomorphic to some Vr,q ([J, Prop. 2.6], [Ks, Th. VI.3.5]). In fact, M (λ)q is
simple if and only if λ = ±q l for all integers l ≥ 0. When λ = ±q l , the vi for i > l
span a submodule of M (λ)q isomorphic to M (q −2(l+1) λ), and this is the only non
trivial proper submodule of M (λ)q [J, Prop. 2.5].
4.1.2. Symmetries. We have just claimed that the modules M (λ)q are not sim-
ple when λ = ±q l for some integer l ≥ 0. This follows from the fact that the vector
vl+1 satisfies
q l q 1−(l+1) − q −l q (l+1)−1
Evl+1 = ±[l + 1] vl = 0
q − q −1
and
Kvl+1 = ±q l q −2(l+1) vl+1 = ±q −l−2 vl+1 .
Hence we have an injection
(4.3) M (±q −l−2 )q −→ M (±q l )q
mapping the highest weight vector v0 of M (±q −l−2 )q , which is a simple Uq -module
since −l − 2 ≤ 0, to vl+1 .
In particular, by specializing at q = ε and taking the quotient by Uε · (vn − bv0 ),
we get that V (±εr , 0, 0) is simple if and only if r = n − 1, and there is for every
0 ≤ r ≤ n − 2 a (non split) short exact sequence of Uε -modules:
±
(4.4) 0 −→ Vn−r−2 −→ V (±εr , 0, 0) −→ Vr± −→ 0.
± ±
The inclusion Vn−r−2 → V (±εr , 0, 0) maps the highest weight vector v0 ∈ Vn−r−2
to vr+1 ∈ V (±ε , 0, 0).
r
Exercise 4.3. (Application: the image of the Harish-Chandra homomorphism.)

Recall that π maps a central element z = i≥0 F i hi E i of degree 0 to the Laurent
polynomial π(z) = h0 ∈ C[K, K −1 ] (see (2.22)).
(a) Prove a statement analogous to Lemma 4.1 for finite dimensional highest weight
Uq -modules V of highest weight λ ∈ C∗ (possibly, not simple): show that when q
is not a root of unity (resp. when q = ε), for all v ∈ V and z ∈ Zq (resp. Zε ∩ U0 )
we have
(4.5) zv = π(z)(λ) v.
(b) Deduce from (4.3) and (4.5) that when q is not a root of unity, for all z ∈ Zq
and integer l ≥ 0 we have
π(z)(±q l−1 ) = π(z)(±q −l−1 ).
Similarly, deduce from (4.4) and (4.5) that for all z ∈ Zε ∩ U0 and 1 ≤ r ≤ n − 1
we have
π(z)(±εr−1 ) = π(z)(±ε−r−1 ).
(c) Show that the monomials K in (K + K −1 )j and K l , where i ∈ Z, 0 ≤ j < n, and
0 < l < n, form a linear basis of C[K, K −1 ].
(d) Deduce from (b) and (c) that when q is not a root of unity (resp. when q = ε),
the homomorphism γ−1 ◦ π maps Zq (resp. Zε ∩ U0 ) to C[K + K −1 ] (resp. the
polynomial subalgebra C[K n , K −n , K + K −1 ] of C[K, K −1 ]).
118
By the discussion preceding Theorem 2.9, we know that γ−1 ◦ π is injective over
Zq (resp. Zε ∩ U0 ) when q is not a root of unity (resp. when q = ε), and maps the
Casimir element Ω to a scalar multiple of K + K −1 . Also, Zε is generated by E n ,
F n and Zε ∩ U0 , and γ−1 ◦ π(Zε ∩ U0 ) = γ−1 ◦ π(Zε ∩ U0 ). Then, by Exercise 4.3
we have:
Proposition 4.4. (i) When q is not a root of unity, we have an isomorphism
∼
γ−1 ◦ π : Zq −→ C[K + K −1 ].
=
(ii) When q = ε, we have an isomorphism

∼
γ−1 ◦ π : Zε ∩ U0 −→ C[K n , K −n , K + K −1 ]
=
mapping C[Ω] to C[K + K −1 ]. Moreover, Zε is generated by E n , F n , K ±n and Ω.

4.1.3. Classification. The following result can be proved by elementary meth-
ods [J, §2.11-2.13].
Theorem 4.5. The Uε -modules Vr± , 0 ≤ r ≤ n − 1, and V (λ, a, b) are simple,
and any non zero simple Uε -module is isomorphic to one of them.
Remark 4.6. When b = 0 the module V (λ, a, b) uniquely determines b but not
λ and a. In fact, we get an isomorphic module by replacing v0 by any other vi ,
that is, λ by λε−2i and a by a + [i](λε1−i − λ−1 εi−1 )(ε − ε−1 )−1 b−1 .
By Proposition 4.4, any central character χ = χV ∈ Spec(Zε ) is determined by
its values at the generators x, y, z and Ω of Zε . Put
(4.6) (xχ , yχ , zχ , cχ ) = (χ(x), χ(y), χ(z), χ(Ω)) ∈ C4 .
We have:
• if V = Vr± , then
εr+1 + ε−1−r
(4.7) xχ = 0 , yχ = 0 , zχ = ±1 , cχ = ± ;
(ε − ε−1 )2
• If V = V (λ, a, b), then

n−1
λε1−j − λ−1 εj−1

−1 n −n
(4.8) xχ = −(ε − ε ) λ a ab + [j]
j=1
ε − ε−1
λε + λ−1 ε−1
(4.9) yχ = (ε − ε−1 )n b , zχ = λn , cχ = ab + .
(ε − ε−1 )2
Set
εr+1 + ε−1−r
c±
r = ± .
(ε − ε−1 )2
From (4.4) or by a direct computation, we get
±
(4.10) c±
r = cn−r−2 , 0 ≤ r ≤ n − 2.
Hence there are n − 1 distinct values c±
of the Casimir element, achieved at r =
r
0, 1, . . . , (n − 3)/2.
It will be useful to distinguish among central characters and Uε -modules by
using the map
Ξ τ σ
(4.11) ϕ : Rep(Uε ) −→ Spec(Zε ) −→ Spec(Z0 ) −→ P SL2 C0 .
17
Recall the big cell decomposition P SL2 C0 = U− T U+ . Set

D = {χ±
r ∈ Spec(Zε ) | (xχ±
r
, yχ±
r
, zχ±
r
, c χ±
r
) = (0, 0, ±1, c±
r ), 0 ≤ r ≤ (n − 3)/2}.
By (4.7)-(4.10) we have
±
(4.12) Ξ−1 (χ± ±
r ) = {Vr , Vn−r−2 } , ϕ
−1
(Id) = D.
Definition 4.7. A simple Uε -module V and its central characters Ξ(V ) and
τ ◦ Ξ(V ) are called:
• diagonal (resp. triangular ) if ϕ(V ) ∈ T (resp. ϕ(V ) ∈ P B± = U− T or
T U+ );
• regular (resp. singular ) if ϕ(V ) = Id (resp. ϕ(V ) = Id);
• regular semisimple if the conjugacy class of ϕ(V ) intersects T \ Id.
• cyclic if E n and F n act as non zero scalars.
So, a point χ ∈ Spec(Zε ) is regular if χ ∈
/ D. Regular semisimple characters
are sent by π to loxodromic elements of P SL2 C; the regular diagonal Uε -modules
are the V (λ, 0, 0)s with λn = ±1.
The cyclic simple Uε -modules have the property that any two eigenvectors of
K can be obtained one from each other by applying some power of E or F ; matrix
realizations are immediately derived from the formulas of V (λ, a, b) when ab = 0.
They are complementary to the highest weight modules in Rep(Uε ), since a module
on which F n acts as zero and E n does not is isomorphic to some V (λ, 0, b) by
applying the Cartan automorphism (2.13). In particular, the cyclic Uε -modules
map under ϕ to a dense subset of P SL2 C, and together with the diagonal modules
they cover a neighborhood of the identity.
4.2. Quantum coadjoint action and the bundle ΞM . We are going to

show that the central character map Ξ : Rep(Uε ) → Spec(Zε ) can be used to define
a bundle ΞM of Uε -modules over Spec(Zε ), endowed with an action of the subgroup
G of Aut(Ûε ).
Let V ∈ Rep(Uε ), and
ρV : Uε → End(V )
be the corresponding linear representation of Uε . Recall (3.13) and Definition (3.7).
Since Z0 acts by scalar operators on V , the same is true for Ẑ0 , which consists of
holomorphic functions of x, y, z ±1 ∈ Z0 . Hence V is naturally a Ûε -module. Since
any Ûε -module is uniquely determined by the action of Uε on it, we will henceforth
identify Uε -modules and Ûε -modules.
Definition 4.8. Given g ∈ G, the twisted Uε -module g V is defined by
ρgV (u) = ρ(gu), u ∈ Uε .
Note that χgV = χV ◦ g for all g ∈ G and χ ∈ Spec(Zε ), so the action of G on
simple Uε -modules lifts the opposite of the action of G on Spec(Zε ).
The basic properties of the central character map Ξ are given by the following
result. It is a special case of [DCK, §3.7-3.8 & Th. 4.2] (see also [DCP, §20]), which
applies to the quantum groups of an arbitrary complex semi-simple Lie algebra.
Recall the algebras Uεχ in (4.2).
120
Theorem 4.9. (i) If χ ∈ Spec(Zε ) \ D, then Uεχ ∼ = Mn (C). Hence there is up

to isomorphism a unique simple Uε -module Vχ with central character χ; we have
Vχ ∼
= V (λ, a, b) for any (λ, a, b) ∈ C∗ × C2 satisfying (4.8)-(4.9).
(ii) If χ ∈ D, there are exactly two simple Uε -modules Vχ with central character
±
χ: if (xχ , yχ , zχ , cχ ) = (0, 0, ±1, c±
r ) we have Ξ
−1
(χ) = {Vr± , Vn−r−2 }.
(iii) The restriction map τ : Spec(Zε ) → Spec(Z0 ) has degree n, and the coordi-
nate ring Zε is generated by Z0 and Ω subject to any one of the equivalent relations
(R± ) given by

n−1
K n + K −n ∓ 2
(4.13) Ω − c± = E n n
F + .
j=0
j
(ε − ε−1 )2n
−1−j
c±
j+1
where, as usual, j = ± ε (ε−ε
+ε
−1 )2 .
Proof. (i) Consider a cyclic central character χ (so χ ∈ Spec(Zε ) \ D). For
any V ∈ Ξ−1 (χ), ρV induces an irreducible representation
ρ̄V : Uεχ → End(V ).
Because χ is cyclic, dim(V ) = n, and Uεχ has no non trivial proper ideal (by
Theorem 2.4 the unit necessarily belongs to any non zero ideal). Hence ρ̄V is
faithful. Then, Wedderburn’s theorem implies that ρ̄V is an isomorphism, that is,
Uεχ ∼= Mn (C) [L, §XVII.3]. Conjugation with G yields a similar isomorphism for
any module in the G-orbit of V . Since any non trivial conjugacy class of P SL2 C
contains an element whose entries are all non zero, by Theorem 3.8 the G-orbit of
any character χ ∈ Spec(Zε ) \ D contains one which is cyclic. Hence Uεχ ∼ = Mn (C).
Since Mn (C) is a simple ring [L, §XVII.4 & 5], this implies the uniqueness of
Vχ ∈ Ξ−1 (χ) when χ ∈ / D.
(ii) is the content of (4.12).
(iii) Because of (i), in order to prove that deg(τ ) = n it is enough to find an
open subset O of Spec(Z0 ) (in the complex topology) such that for all τ (χ) ∈ O,
(τ ◦ Ξ)−1 (τ (χ)) consists of n simple Uε -modules. In fact, this is true for any regular
diagonal Z0 -character τ (χ), for (τ ◦ Ξ)−1 (τ (χ)) consists of the V (λ, 0, 0) with λn =
zχ (= ±1). Since the G-action on Rep(Uε ) preserves the isomorphism type of the
fibers of Ξ, and in particular their dimension and number of components, it is also
true for any Z0 -character in the G-orbit of a regular diagonal one. Hence we can
take O to be the set of regular semisimple Z0 -characters. Theorem 3.8 implies that
O is Zariski open and dense.
By Proposition 4.4 we know that Zε is generated by E n , F n , K ±n and Ω. On
another hand, (2.11) gives by an easy induction on r the relation

r−1
ε2j+1 K + ε−2j−1 K −1

(4.14) Ω− −1 )2
= F rEr.
j=0
(ε − ε
When r = n this relation makes sense in Zε . To see this, let us expand the left-
n
hand side as k=0 (−1)k σk Ωn−k . The coefficients σk are the elementary symmetric
2j+1
K+ε−2j−1 K −1
functions of the variables xj = ε (ε−ε−1 )2 , 0 ≤ j ≤ n − 1. It is classical
that σk can beexpressed as a polynomial with rationaln−1coefficient of the power sum
n−1
functions ti = j=0 xij , 0 ≤ i ≤ k. Then, by using i=0 ε2i = 0 it is immediate to
n −n
check that all σk are complex numbers not involving K, except σn = − (ε−ε
K +K
−1 )2n .
19
Hence, for r = n we can rewrite (4.14) as

n−1
K n + K −n
(Ω − αj ) = F n E n + .
j=0
(ε − ε−1 )2n
for some αj ∈ C. The relation (4.13) follows from this by noting that the left-hand
side (resp. right-hand side) acts as 0 (resp. ±2(ε − ε−1 )−2n ) on all the simple
Uε -modules Vr± if and only if αj = c±
j for all 0 ≤ j ≤ n − 1.
Remark 4.10. (i) We have seen in Theorem 2.7 that rkZ0 Uε = n3 . From this
one deduces that deg(τ ) = n and V ∈ Ξ−1 (χ) is unique for generic χ, by using that
dim(V ) = n for regular diagonal characters χ, as in the first part of the proof of
Theorem 4.9 (iii).
(ii) It follows directly from (4.8)-(4.9) and Remark 4.6 that χ ∈ Spec(Zε ) \ D
determines uniquely V (λ, a, b) up to isomorphism. This implies the uniqueness of
Vχ ∈ Ξ−1 (χ), but depends on the classification of the simple Uε -modules, which is
provided by Theorem 4.5.
Following [DCKP, §6] and [DCP, §11 & 21], we use Theorem 4.9 to provide
the collection of simple regular Uε -modules a structure of vector bundle over the
smooth part of Spec(Zε ).
To simplify notations, let
G = P SL2 C.
It will be useful to distinguish between the Cartan subgroup T = C∗ of H, and its
image T̄ = T /(±1) under the 2-fold covering σ : H → G0 .
Recall that G is an affine algebraic group, since the adjoint representation
Ad : G → Aut(g) identifies G with a subgroup of the complex orthogonal group
SO3 C for the Killing form of g. Denote by G//G the affine variety with coordinate
ring C[G]G , the ring of regular functions on G invariant under conjugation. The
points of G//G are in one to one correspondence with the conjugacy classes of
elements of G having distinct traces up to sign, and we have isomorphisms
(4.15) = C[T̄ ]W , G//G ∼
C[G]G ∼ = C∗ /(z ∼ ±z −1 ).
Here W ∼ = Z/2 is the Weyl group of G acting on the torus T̄ ⊂ G by inversion, and
z is the coordinate function of T ⊂ H. Consider the maps
(4.16) p1 = p ◦ σ : H −→ G//G , p2 : G//G −→ G//G,
where p : G → G//G is the quotient map, and p2 is induced by the n-th power map
g → g n , g ∈ G. The fibered product of p1 and p2 is the variety
(4.17) H ×G//G G//G = {(h, ḡ) ∈ H × G//G | p1 (h) = p2 (ḡ)}.
Theorem 4.11. (i) The action of G on Spec(Z0 ) extend to Spec(Zε ) by a trivial
action on Spec(C[Ω]). Hence, if O is an orbit of G in Spec(Z0 ), the connected
components of τ −1 O are orbits of G in Spec(Zε ).
(ii) Spec(Zε ) is a 3-dimensional affine algebraic variety with singular set D, and is
isomorphic to H ×G//G G//G.
Proof. (i) The claim follows from Proposition 4.4 (ii) and the fact that for all
q we have Ω ∈ Zq , so that Da (Ω) = 0 for all a ∈ Z0 .
122
(ii) According to Theorem 4.9 (iii), Spec(Zε ) is the hypersurface in C4 given in

coordinates (e, y, z, c) by (see (2.16)-(2.17))

n−1

(4.18) (ε − ε−1 )2n c − c±
j = ey + z + z −1 ∓ 2.
j=0
It follows from (4.10) that the singularities of Spec(Zε ) are quadratic and coincide
with D (note, in particular, that (0, 0, ±1, c±
n−1 ) is a smooth point).
Let us prove now the isomorphism with (4.17). By Proposition 4.4 (ii) we have
Zε = Z0 ⊗Z0 ∩C[Ω] C[Ω],
and C[Ω] ∼= C[K + K −1 ]. Let us identify C[K ±1 ] with C[T ], the coordinate ring
of the torus T of H. Then C[Ω] ∼ = C[T ]W . By Proposition 3.1, Z0 ∼ = C[H]. The
Harish-Chandra homomorphism applied to (4.13) gives Z0 ∩ C[Ω] ∼ = C[K n + K −n ].
On another hand, C[K n + K −n ] ∼= C[T /μn ]W , where μn is the subgroup of T of
n-th roots of unity. (Alternatively, Theorem 3.8 gives Z0G = C[T /μn ]W , where
C[T /μn ]W is identified with the subalgebra of Z0 generated by z + z −1 , and so
Z0 ∩ C[Ω] = C[T /μn ]W by (i) above). Hence
Zε ∼
= C[H] ⊗C[T /μn ]W C[T ]W .
Since n is odd, this is equivalent to
Zε ∼
= C[H] ⊗C[T̄ /μn ]W C[T̄ ]W .
The isomorphism of Spec(Zε ) with H ×G//G G//G follows by duality.
Denote by X = Spec(Zε ) \ D the subset of regular central characters. By

Theorem 4.11, it coincides with the smooth part of Spec(Zε ).
Take a point χ ∈ X. Any choice of parameters (λ, a, b) as in (4.8)-(4.9) deter-
mines an isomorphism of Uεχ -modules Vχ ∼ = V (λ, a, b), and an isomorphism of alge-
bras Uεχ ∼
= Mn (C) by identifying V (λ, a, b) with Cn . Both isomorphisms are varying
smoothly with respect to (λ, a, b) over a sufficiently small open neighborhood Oχ
of χ in X. Hence we have coordinate charts gOχ : ρ∈Oχ Uερ → Oχ × Mn (C) with
smooth transition functions
−1
gOχ ◦ gO χ
: (Oχ ∩ Oχ ) × Mn (C) → (Oχ ∩ Oχ ) × Mn (C)
that restrict to algebra automorphisms on the second component. By taking a

locally finite covering of X made of neighborhoods Oχ we thus obtain a trivializing
atlas for a smooth vector bundle of matrix algebras
(4.19) ΞA : Aε → X,

where Aε = χ∈X Uεχ as a set. By a “smooth bundle of algebras with unit” we
mean that a smooth product with unit is defined on sections. For such bundles the
functions on the base can be identified with multiples of the unit section.
Similarly, we have a rank n vector bundle
ΞM : Mε → X
with fibers the Uε -modules Ξ−1M (χ) = Vχ . Clearly, ΞA and ΞM are associated
bundles via the action of Uεχ on Vχ . They are topologically non trivial, since by
21
using Remark 4.6 we √

can find a loop in Spec(Zε ) \ D with non trivial holonomy.
2 −1πd
Namely, put ε := e n . For b = 0, the path in C∗ × C2 given by

√
− 4 −1πdt λ − λ−1
γ : t −→ λe n ,a + t , b , t ∈ [0, 1],
b(ε − ε−1 )
projects to a loop in Spec(Zε ) with holonomy in ΞM the permutation matrix (δi+1,j )
(indices mod n).
Theorem 4.12. The group G ⊂ Aut(Ûε ) acts on ΞA and ΞM by bundle mor-
phisms.
In particular, the orbits of G correspond to symplectic leaves (resp. conjugacy
classes) of Spec(Zε ) \ D (resp. P SL2 C0 ).
Proof. Since by Proposition 3.6 the group G maps the algebra of functions on
Spec(Zε ) into itself, its acts linearly on the fibers of ΞA and ΞM by automorphisms
of algebras and Uε -modules, respectively. More precisely, for any g ∈ G and χ ∈
Spec(Zε ), the action of g −1 on Ûε maps isomorphically the ideal I χ to I g.χ , and
hence the algebra Uεχ to Uεg.χ and its simple module Vχ to Vg.χ .
Remark 4.13. (a) In the terminology of [RVW], Theorem 4.12 reflects the
fact that the pair (Uε , Z0 ) is a Poisson fibered algebra. The infinitesimal action
of G on ΞA defines a morphism of vector bundles D : Ω1 X → Aε with non trivial
curvature, where Aε is the bundle of first order differential operators on ΞA with
symbols Id ⊗ ξ, ξ ∈ T X.
(b) The bundle ΞM extends to the whole of Spec(Zε ), with singular (non simple)
fibers V (±εr , 0, 0), 0 ≤ r ≤ (n − 3)/2, over D.
5. Intertwiners: the category Uε -Mod

The Hopf algebra structure of Uε endows the category Uε -Mod of finite di-
mensional left Uε -modules with a tensor product and a duality. We are going to
describe them in geometric terms when applied to regular Uε -modules, by using the
theorems 4.11 and 4.12. To clarify the picture, after some preliminaries we recall
well-known results on a subcategory based on singular modules and generating the
Reshetikhin-Turaev TQFT.
5.1. A few basic definitions [Ks]. Unless stated otherwise all the modules
we are going to consider will be left modules, finite dimensional over C.
The tensor product vector space V ⊗W of two Uε -modules is naturally a Uε ⊗Uε -
module. It is made into a Uε -module by setting
(5.1) a.(v ⊗ w) = Δ(a).(v ⊗ w)
for all a ∈ Uε , v ∈ V , w ∈ W , where Δ(a) ∈ Uε ⊗ Uε is the coproduct of a. This
action is compatible with the natural tensor product of linear maps of Uε -modules,
so it defines a bifunctor ⊗ : Uε -Mod × Uε -Mod → Uε -Mod.
Consider the trivial action of Uε on C given by the counit, a.z = η(a)z. By
(2.3), for all Uε -modules U, V, W the formula (5.1) turns the canonical isomorphisms
of vector spaces lV : C ⊗ V ∼ = V , rV : V ⊗ C ∼
= V and
∼
=
(5.2) aU,V,W : (U ⊗ V ) ⊗ W −→ U ⊗ (V ⊗ W )
124
into isomorphisms of Uε -modules which are natural with respect to morphisms of

Uε -modules. Hence they define natural isomorphisms of functors l,r and a, with
obvious compatibility relations. In particular
(5.3) a : ⊗ (⊗ × id) −→ ⊗(id × ⊗)
makes the following Pentagonal Diagram commutative:
aU,V,W ⊗idX
(5.4) (U ⊗ (V ⊗ W )) ⊗ X o ((U ⊗ V ) ⊗ W ) ⊗ X
aU ⊗V,W,X

aU,V ⊗W,X (U ⊗ V ) ⊗ (W ⊗ X)
aU,V,W ⊗X
idU ⊗aV,W,X
U ⊗ ((V ⊗ W ) ⊗ X) / U ⊗ (V ⊗ (W ⊗ X)).
The category Uε -Mod endowed with (⊗, a, l, r) is an example of tensor category,

with associativity constraint a and unit C. Note that a explicits the different
module structures at both sides.
The category Uε -Mod has also a (left) duality, that is a pair (b, d) of natural
transformations given on any Uε -module V by morphisms
dV : V∗⊗V −→ C
(5.5)
bV : C → V ⊗ V ∗,
where V ∗ is an Uε -module to be specified, satisfying
(idV ⊗ dV )(bV ⊗ idV ) = (dV ⊗ idV ∗ )(idV ∗ ⊗ bV ) = idV ∗ .
Naturally we put V ∗ = HomC (V, C), the dual linear space, and define dV and bV
as the canonical
pairing of vector spaces between V and V ∗ , and the map taking
1 ∈ C to i vi ⊗ v i , where {vi } and {v i } are dual basis of V and V ∗ . Then a (left)
Uε -module structure is defined on V ∗ by
(5.6) dV ((a.ξ) ⊗ v) = dV (ξ ⊗ (S(a).v))
for all a ∈ Uε , v ∈ V and ξ ∈ V ∗ (recall that the antipode S : Uε → Uε is an
anti automorphism). By using (2.4) one can check that dV and bV are Uε -linear
maps. Any Uε -linear map f : V → W has a transpose f ∗ : W ∗ → V ∗ given by
f ∗ = (dW ⊗ idV ∗ )(idW ∗ ⊗ f ⊗ idV ∗ )(idW ∗ ⊗ bV ),
so that ∗ defines a contravariant functor Uε -Mod → Uε -Mod. As in the case of
linear spaces there are natural bijections HomUε (U ⊗ V, W ) ∼
= HomUε (U, W ⊗ V ∗ )
∗ ∼
and HomUε (U ⊗ V, W ) = HomUε (V, U ⊗ W ). Moreover, we get from (2.5) an
isomorphism of Uε -modules
(5.7) V ∗ ⊗ W∗ ∼
= (W ⊗ V )∗ .
More generally, the vector space HomC (V, W ) is an Uε -module by setting

(5.8) a.f (v) = ai .f (S(ai ).v)
i

for all f ∈ HomC (V, W ) and v ∈ V , where we put Δ(a) = i ai ⊗ ai . Note that by
counitality in (2.3), the action (5.8) reduces to (5.6) when W = C, and (5.1) and
23
(5.6) imply that the canonical C-linear isomorphism

λV,W : W ⊗V∗ −→ HomC (V, W )
(5.9)
w⊗ξ −→ (v → ξ(v)w)
is an isomorphism of Uε -modules. We have λ−1V,W (f ) = (f ⊗ idV ∗ )bV .
Duality allows one to define a trace in the category. For that, we use the
remarkable fact that the square of the antipode S is an inner automorphism: for
all u ∈ Uε we have S 2 (u) = KuK −1 . Then, the linear isomorphism
ΦV : V −→ V ∗∗
v −→ dV ( · ⊗ K.v)
is also Uε -linear. The quantum trace of V is the Uε -linear map
λ−1 ΦV ⊗idV ∗
trq : EndC (V )
V,V
/ V ⊗V∗ / V ∗∗ ⊗ V ∗ dV ∗
/ C.
Explicitly, for all f ∈ EndC (V ) we have

(5.10) trq (f ) = tr (v → K.f (v)) ,
where tr is the usual trace map of linear spaces. The quantum dimension of V is
dimq (V ) = trq (idV ). We say that V has trace zero if for all Uε -linear endomor-
phisms f of V we have
trq (f ) = 0.
Since Δ(K) = K ⊗ K (K is said to be group like), the quantum trace is multiplica-
tive: for all f ∈ EndC (V ), g ∈ EndC (W ) the quantum trace of f ⊗g ∈ EndC (V ⊗W )
is
(5.11) trq (f ⊗ g) = trq (f )trq (g).
Exercise 5.1. (a) Check that all the Uε -linear maps above are indeed Uε -
linear, and that the isomorphism λ(V,W ) : W ∗ ⊗ V ∗ → (V ⊗ W )∗ in (5.7) can be
decomposed as
λ(V,W ) = (dW ⊗ id(V ⊗W )∗ )(idW ∗ ⊗ dV ⊗ idW ⊗ id(V ⊗W )∗ )(idW ∗ ⊗ idV ∗ ⊗ bV ⊗W ).
(b) (Right duality) Show that we have Uε -linear maps dV : V ⊗∗V → C and
bV
: C →∗V ⊗ V analogous to (5.5) but with tensorands permuted, where ∗ V is
defined by (5.6) with dV and S replaced by dV and S −1 . Show that ξ → K −1 .ξ
defines an isomorphism of Uε -modules V ∗ −→∗V , and that trq (f ) = dV (f ⊗idV ∗ )bV .
5.2. The modular category Ūq -Mod. The singular simple Uε -modules Vr±
generate a remarkable subcategory of Uε -Mod which has been described in [RT1]
(in the simply-connected version of Uε , see Remark 2.2).
For all 0 ≤ r ≤ n − 1, the action of Uε on Vr± factors through the restricted
quantum group
(5.12) Ūε = Uε /(E n = F n = 0, K 2n = 1).
The algebra Ūε is finite dimensional, not semisimple, and has for simple modules
the finite set {Vr± , r = 0, . . . , n − 1}. Note that V0+ = C (the trivial module), and
(5.13) if r ≤ n − 2, dimq (Vr± ) = ±[r + 1] = 0.
±
On another hand, dimq (Vn−1 ) = 0. Hence, by Schur’s lemma, trq (f ) = 0 for all
±
Uε -linear endomorphisms f of Vn−1 , which has thus a special status among singular
±
modules, reminiscent of the fact that Vn−1 = V (±ε−1 , 0, 0).
126
Definition 5.2. A color is an integer r such that 0 ≤ r ≤ n − 2.

There is a unique Hopf algebra structure on Ūε such that the quotient map
Uε → Ūε is a morphism of Hopf algebras (see eg. [Ks, Prop. IX.6.1]). Since
Vr− ∼
= V0− ⊗ Vr+ , we can concentrate on the modules Vr := Vr+ . They are self dual,
and satisfy:
Theorem 5.3. [RT1, Th. 8.4.3] For any colors i, j there is a unique trace zero
submodule Zi,j of Vi ⊗ Vj such that
(5.14) ∼ (⊕k Vk ) ⊕ Zi,j ,
V i ⊗ Vj =
where the sum is over all colors k such that the triple (i, j, k) is ε-admissible, that
is:
• i + j + k ∈ 2Z and i + j + k ≤ 2(n − 2);
• |i − j| ≤ k ≤ i + j (the triangle inequalities).
When i + j ≤ n − 2 we have Zi,j = ∅.
The modules Zi,j are built on the highest weight modules V (εi , 0, 0) and some
2n-dimensional extensions thereof. Since their Z0 -characters are trivial, like the
color modules Vr we can consider them as singular Uε -modules. Their duals and
tensor products with color modules Vr decompose into summands of the same form,
so the Vr generate a subcategory of Uε -Mod which is closed under tensor product
and duality.
Remark 5.4. The category Ūε -Mod gives rise to the so called fusion rules
of Wess-Zumino-Witten conformal field theories with gauge group SU (2). In this
context, the admissibility conditions of Theorem 5.3 give a method for counting
dimensions of the space of conformal blocks [Ko, Prop. 1.19] (compare also (5.19)
below and [Ko, Lemma 2.6]).
The associativity constraint (5.2), when applied to color modules and computed
modulo trace zero Ūε -modules, defines the celebrated ε-6j-symbols. Let us recall how
this goes (see [CFS, §4] for details).
Theorem 5.3 implies that for each ε-admissible triple (i, j, k) the space of Uε -
linear embeddings Vk → (Vi ⊗ Vj )/Zi,j has dimension one. There is a natural basis
of lifts, the Clebsch-Gordan operator
(5.15) k
Yi,j : Vk −→ Vi ⊗ Vj ,
defined in terms of the Jones-Wentzl idempotents el (l = i, j, k) of the Temperley-
Lieb algebra; these can be realized as projectors
(5.16) el : V1⊗l → V1⊗l
in the algebra of Uε -linear transformations of V1⊗l , whose image is isomorphic to Vl
[CFS, Prop. 4.3.8].
By using (5.2) and (5.11), we get for all colors a, b, c a Uε -linear isomorphism
relating two splittings
(Va ⊗ Vb ) ⊗ Vc = ⊕(l,k) (Ya,b
l
⊗ idVc )Yl,c
k
(Vk ) ⊕ Z1
j
(5.17) Va ⊗ (Vb ⊗ Vc ) = ⊕(j,k) (idVa ⊗ Yb,c k
)Ya,j (Vk ) ⊕ Z2
where the sums are over all pairs of colors (l, k) (resp. (j, k)), such that (a, b, l) and
(l, c, k) (resp. (b, c, j) and (a, j, k)) are ε-admissible, and Z1 , Z2 are maximal trace
25
zero submodules of Va ⊗ Vb ⊗ Vc . It follows from Exercise 5.6 (c) below that any
Uε -linear map Vk → Va ⊗ Vb ⊗ Vc can be written uniquely as a linear combination
l
of the maps (Ya,b ⊗ idVc )Yl,c
k
, plus a map whose image is contained in a trace zero
summand of Va ⊗ Vb ⊗ Vc . The ε-6j-symbols

a b l
(5.18) ∈C
c k j ε
are thus defined by
j
a b l

(5.19) (idVa ⊗ Yb,c k
)Ya,j = l
(Ya,b ⊗ idVc )Yl,c
k
+S
c k j ε
l
where S maps into a summand of trace zero and the sum is taken over all colors
l such that (a, b, l) and (l, c, k) are ε-admissible. Consider the normalized ε-6j-
symbols

a b f (−1)f Θ(a, b, f )Θ(d, e, f ) a b f
(5.20) =
e d c ε [f + 1] Θ(a, c, d)Θ(b, c, e) e d c ε
where we fix once and for all a square root, and (see (2.9))
a+b+k [ a+b−k
2 ]![ a−b+k
2 ]![ −a+b+k
2 ][ a+b+k
2 + 1]!
Θ(a, b, k) = (−1) 2 .
[a]![b]![k]!
To an abstract tetrahedron with edges colored by a, b, c, d, e, f , let us associate the
scalar (5.20). We have:
Theorem 5.5. ([KR], [RT1]; see [CFS, Th. 4.4.6]) The normalized ε-6j-
symbols (5.20) are well-defined, and satisfy:
• Invariance under full tetrahedral symmetries.
• The Elliot-Biedenharn identity:

c d h b h k
(5.21) . =
g e f ε g a e ε

j b c j j d k c d h
(−1) [j + 1] . . .
f a e ε g a f ε k b j ε
j
Note that the ε-6j-symbols (5.18) are only partially symmetric. The com-
mutativity of the Pentagonal Diagram (5.4) for color modules shows up in the
Elliot-Biedenharn identity.
Exercise 5.6. (a) Why is the algebra Ūε not semisimple ? (Hint: otherwise
every Ūε -module would be semisimple, that is, a sum of simple modules.)
(b) We have claimed the self duality of the modules Vr : determine explicitly
an isomorphism Vr∗ → Vr (use (5.6) and the formulas in Section 4.1 !).
(c) By Theorem 5.3, for any maximal trace zero submodule U of Va ⊗ Vb ⊗ Vc which
is a summand, the complementary submodule W such that Va ⊗ Vb ⊗ Vc = W ⊕ U
is completely reducible. Show that any simple submodule V of Va ⊗ Vb ⊗ Vc such
that V ⊂ U is a color module. Deduce that given any two maximal trace zero
summands Ui (i = 1, 2) of Va ⊗ Vb ⊗ Vc , any simple submodule of U1 is a submodule
of U2 .
128
Hence, for every maximal trace zero submodule U of Va ⊗ Vb ⊗ Vc , the matrix

a b l
(5.22)
c k j ε l,j
of ε-6j-symbols relates the two basis of invariant maps Vk → Va ⊗ Vb ⊗ Vc /U given
j
by {(Ya,b
l
⊗ idVc )Yl,c
k
}l and {(idVa ⊗ Yb,c k
)Ya,j }j .
5.3. Pure regular Clebsch-Gordan decomposition and 6j-symbols.
Similarly to Theorem 5.3 and the change of basis matrix (5.22), the tensor products
of simple regular Uε -modules can be split in different ways, related by morphisms
that we are going to define.
First we consider the duals of regular Uε -modules. Recall the coordinates (4.6)
of the set Spec(Zε ) of central characters of Uε , and the degree n covering map over
the regular ones,
τ : Spec(Zε ) \ D → Spec(Z0 ) \ {±id}.
Let χ ∈ Spec(Zε ) \ D. Denote by Vχ the corresponding simple Uε -module, and let
χ−1 be given by
τ (χ−1 ) = τ (χ)−1 and cχ−1 = cχ
where τ (χ)−1 is the inverse of τ (χ) in the group Spec(Z0 ) ∼ = H. From (3.6) and
(5.6) we get:
Lemma 5.7. The dual module Vχ∗ coincides with Vχ−1 . Hence the duality of
Uε -Mod induces an isomorphism of the bundle Ξ that lifts the inversion map on
Spec(Z0 ) \ {±id}.
It is immediate to check that the regular simple Uε -modules V (λ, a, b) have
vanishing quantum dimension, and so are trace zero modules. Thus, we are in
some sense in a situation opposite to that of Section 5.2, where we dealt with the
color modules Vr . What makes the tensor products of the regular V (λ, a, b)s easy
to handle are the following facts, that we have proved in Lemma 2.8 and Theorem
4.9 (i):
• (a) Z0 is a Hopf subalgebra of Uε , and in particular Δ(Z0 ) ⊂ Z0 ⊗ Z0 ;
• (b) for all χ ∈ Spec(Zε ) \ D, the algebra Uεχ is simple, and isomorphic to
Mn (C).
Let h ∈ Spec(Z0 )\{±Id} be a regular Z0 -character, and I h ⊂ Uε the ideal generated
by Ker(h). For any χ ∈ τ −1 (h), the algebra Uεh = Uε /I h is isomorphic to Uεχ ⊗h Zε .
Then, in virtue of (b) above, it is semisimple, with a direct product decomposition
into complementary ideals

(5.23) Uεh = Uεχ .
χ∈τ −1 (h)
Correspondingly, the unit 1 ∈ Uεh can be written as

(5.24) 1= eχ ,
χ∈τ −1 (h)
where the eχ s are the units of the subalgebras Uεχ ⊂ Uεh , and satisfy Uεχ = Uεh eχ ,
and eχ eχ = 0 for χ = χ [L, Prop. XVII.4.3].
Since Uεh is semisimple, every Uεh -module is semisimple, that is, a sum of simple
submodules. On another hand, because of (a) above, Z0 acts by scalars on any
27
tensor product V of simple Uε -modules, and so V is naturally a Uεh -module for

some h ∈ Spec(Z0 ). In fact, (3.6) shows that h is equal to the product of the
Z0 -characters of the tensorands.
This applies in particular to the Uε -modules Vρ ⊗ Vμ for all ρ, μ ∈ Spec(Zε ) \ D.
They are Uεh -modules, where h = τ (ρ)τ (μ). If h = ±id, Uεh is semisimple, and hence
(5.25) Vρ ⊗ Vμ = 1.(Vρ ⊗ Vμ ) = ⊕χ∈τ −1 (h) eχ .(Vρ ⊗ Vμ ),
where the projectors eχ map Vρ ⊗ Vμ onto a submodule isomorphic to Vχ . We
deduce the following analog of Theorem 5.3 for regular Uε -modules:
Proposition 5.8. (i) Any tensor product of simple Uε -modules which has a
regular Z0 -character is a semisimple Uε -module.
(ii) If ρ, μ ∈ Spec(Zε ) \ D are such that h = τ (ρ)τ (μ) = ±id, then
(5.26) V ρ ⊗ Vμ ∼
= ⊕χ∈τ −1 (h) Vχ .
Remark 5.9. The idempotents eχ of Uεh play in Proposition 5.8 the same role
as the Jones-Wentzl idempotents (5.16) do in Theorem 5.3. In both cases, the
simple summands are distinguished by the Casimir element. For color modules, its
values are determined by the classical or quantum dimension. For regular modules,
these are constantly equal to n and 0, respectively; the Casimir element selects a
nth-root of z (see Theorem 4.11 (ii)).
Definition 5.10. Let h ∈ Spec(Z0 ) \ {±id} be a regular Z0 -character. The
multiplicity module M (h) is the n dimensional vector space spanned by the idem-
potents eχ of Uεh , where χ ∈ τ −1 (h).
A tuple (ρ1 , . . . , ρp ) of regular central characters ρi ∈ Spec(Zε ) \ D is regular if
for all 1 ≤ k ≤ k + l ≤ p, we have
τ (ρk )τ (ρk+1 ) . . . τ (ρk+l ) ∈ Spec(Z0 ) \ {±id}.
By Theorem 4.9 (i) we have a corresponding notion of regular tuples of Uε -
modules.
Remark 5.11. For any two h, h ∈ Spec(Z0 ) \ {±id}, the decomposition (5.24)
provides canonical isomorphisms M (h) ∼ = M (h ). Also, Proposition 5.8 (ii) implies
that any two regular pairs (ρ, μ) and (ν, κ) satisfying h = τ (ρ)τ (μ) = τ (ν)τ (κ) give
isomorphic Uε -modules Vρ ⊗ Vμ and Vν ⊗ Vκ .
We can reorganize the direct sum ⊕χ∈τ −1 (h) Vχ into a tensor product as follows.
Let V be the vector space underlying the modules Vχ . Define an action of Uε on
V ⊗ M (h) by extending linearly the formula
(5.27) a.(v ⊗ eχ ) = (aχ .v) ⊗ eχ
for all a ∈ Uε and v ∈ V , where aχ ∈ Uεχ is the coset of a, with its canonical action
on V . We have a canonical isomorphism ⊕χ∈τ −1 (h) Vχ ∼ = V ⊗ M (h) of Uε -modules,
mapping v ∈ Vχ to v ⊗ eχ . Then, for all regular pairs (ρ, μ) with h = τ (ρ)τ (μ), any
isomorphism of the form (5.26) defines an isomorphism of Uε -modules
F (ρ, μ) : V ⊗V −→
V ⊗ M (h)
(5.28)
v1 ⊗ v2 −→ χ∈τ −1 (h) eχ (v1 ⊗ v2 ) ⊗ eχ
by putting the Uε -module structure of Vρ ⊗Vμ on V ⊗V ; when this structure is clear

from the context, we write F for F (ρ, μ), and similarly for the inverse evaluation
130
map
(5.29) K(ρ, μ) : V ⊗ M (h) −→ V ⊗ V.
Let Δ(ρ,μ) : Uε → Uερ ⊗ Uεμ , a → Δ(a) mod(I ρ ⊗ I μ ). The formula (5.27) shows
that K(ρ, μ) and Δ(ρ,μ) are equivalent data, related by
(5.30) Δ(ρ,μ) (a) = K(ρ, μ)(a ⊗ 1)K(ρ, μ)−1
as operators in Aut(Vρ ⊗ Vμ ).
Remark 5.12. Since the modules Vχ are simple, EndUε (Vχ ) = ∼ C, and so
K(ρ, μ) is uniquely determined up to a scalar factor only. We can reduce this
to an ambiguity modulo nth roots of unity by requiring that det(K(ρ, μ)) = 1.
This produces a degree n polynomial in the coordinates of ρ, μ ∈ Spec(Zε ) \ D.
We are now ready to define the operators and functional relations describing
the isomorphisms
= Vρ ⊗ (Vμ ⊗ Vν ) ∼
(Vρ ⊗ Vμ ) ⊗ Vν ∼ = ⊕χ∈τ −1 (h) (Vχ ⊕ . . . ⊕ Vχ )

n times
for regular triples (ρ, μ, ν), where h = τ (ρ)τ (μ)τ (ν).
Definition 5.13. Let (ρ, μ, ν) be a regular triple of Uε -modules. Put f =

τ (ρ), g = τ (μ), h = τ (ν). The regular 6j-symbol operator of (ρ, μ, ν) is the linear
isomorphism
(5.31) R(ρ, μ, ν) : M (f gh) ⊗ M (gh) −→ M (f g) ⊗ M (f gh)
that makes the following diagram commutative:
id⊗R(ρ,μ,ν)
(5.32) V ⊗ M (f gh) ⊗ M (gh) / V ⊗ M (f g) ⊗ M (f gh)
(id⊗Δ0 )(K) (Δ0 ⊗id)(K)

Vρ ⊗ V ⊗ M (gh) V ⊗ M (f g) ⊗ Vν
id⊗K K⊗id

Vρ ⊗ (Vμ ⊗ Vν ) o aVρ ,V μ,Vν
(Vρ ⊗ Vμ ) ⊗ Vν
where Δ0 (x) = x ⊗ 1 (the standard coproduct). In operator form, we have
(5.33) K12 (ρ, μ)K13 (χr , ν)R23 (ρ, μ, ν) = K23 (μ, ν)K12 (ρ, χl )
for all χr ∈ τ −1 (f g), χl ∈ τ −1 (gh).
Thus, R determines via (5.33) the associativity constraint a of Uε -Mod over

regular modules, like K determines via (5.30) the tensor product of Uε -Mod over
regular modules. The Pentagonal Diagram (5.4) translates as follows:
29
Proposition 5.14. Let (κ, ρ, μ, ν) be a regular 4-tuple of Uε -modules. Set

f = τ (κ), g = τ (ρ), h = τ (μ), k = τ (ν). The following diagram is commutative:
id⊗R(ρ,μ,ν)
M (f ghk) ⊗ M (gh) ⊗ M (ghk) o M (f ghk) ⊗ M (ghk) ⊗ M (hk)
(id⊗Δ0 )(R)

(Δ0 ⊗id)(R) M (f g) ⊗ M (f ghk) ⊗ M (hk)
(τ Δ0 ⊗id)(R)
R(κ,ρ,μ)⊗id
M (f gh) ⊗ M (gh) ⊗ M (f ghk) / M (f g) ⊗ M (f gh) ⊗ M (f ghk)
where τ is the flip map. In operator form, we have the Pentagon Equation:
(5.34) R12 (κ, ρ, μ)R13 (κ, χ1 , ν)R23 (ρ, μ, ν) = R23 (χ2 , μ, ν)R12 (κ, ρ, χ3 )
for all χ1 ∈ τ −1 (gh), χ2 ∈ τ −1 (f g), χ3 ∈ τ −1 (hk).
Proof. Commutativity of the diagram is equivalent to (5.34). The latter is
proved by a straightforward computation using (5.33); details are left as an exercise
(or, compare with the proof of Proposition 6.12 iii)).
Remark 5.15. (The Borel case) The notion of regular module makes sense as
well for any Borel subalgebra Uε b of Uε , say the positive one, generated by K ±1 and
E. Since Uε b has no Casimir element, the isomorphism classes of simple regular
Uε b-modules correspond under the map ϕ of (4.11) to triangular matrices up to
sign in P SL2 C \ {±Id}. Then, for regular pairs (ρ, μ) of simple Uε b-modules, (5.26)
simplifies to a splitting into n isomorphic copies of a single regular Uε b-module
Vρμ . The multiplicity module M (h) becomes the space of equivariant projections
HomUεb (Vρ ⊗ Vμ , Vρμ ), and (5.29) is the map
K : Vρμ ⊗ HomUεb (Vρμ , Vρ ⊗ Vμ ) −→ Vρ ⊗ Vμ
v⊗i −→ i(v).
Exercise 5.16. Let V be a finite dimensional vector space and f , g ∈ End(V )
such that f 2 = f , g 2 = g and f g = gf . Show that R := f ⊗ g satisfies
(5.35) R12 R13 R23 = R23 R12
in End(V ⊗ V ⊗ V ). Show that if R := f ⊗ id or R := id ⊗ f is a solution of (5.35),
then f 2 = f .
6. The regular 6j-symbols as bundle morphisms

The FRT method associates a cobraided Hopf algebra to any finite dimensional
invertible solution of the Quantum Yang-Baxter Equation (see eg. [Ks, Ch. VIII]).
Similarly, any finite dimensional invertible solution of the constant Pentagon Equa-
tion (5.35) is the canonical element of the Heisenberg double of some Hopf algebra
[BS, Mi, D].
We are going to see how this result can be adapted to equation (5.34), which
has the form of a (non-Abelian) 3-cocycle identity over the group P SL2 C via the
map στ : Spec(Zε ) → P SL2 C. We proceed in several steps to identify both the
evaluation map (5.29) and the regular 6j-symbol operator (5.31) as instances of a
same bundle morphism R : Ξ(2) −→ Ξ(2) .
132
6.1. The QUE algebra Uh . The quantum universal enveloping (QUE) alge-
bra Uh = Uh (sl2 ) is the Hopf algebra over C[[h]] topologically generated by three
variables X, Y and H with relations
sinh(hH/2) ehH/2 − e−hH/2
[H, X] = 2X , [H, Y ] = −2Y , [X, Y ] = = .
sinh(h/2) eh/2 − e−h/2
(See eg. [CP, Ch. 6–8] and [Ks, Ch. XVII].) The comultiplication and counit are
determined by
Δ(H) = H ⊗ 1 + 1 ⊗ H,
Δ(X) = X ⊗ ehH/4 + e−hH/4 ⊗ X , Δ(Y ) = Y ⊗ ehH/4 + e−hH/4 ⊗ Y
and
η(H) = η(X) = η(Y ) = 0.
The antipode is
S(H) = −H , S(X) = −ehH/2 X , S(Y ) = −e−hH/2 Y.
The QUE algebra Uh is a topological deformation of the universal enveloping algebra
U sl2 , in the sense that Uh ∼
= U sl2 [[h]] as C[[h]]-modules, and Uh /hUh ∼
= U sl2 as
Hopf algebras, where U sl2 has the Hopf algebra structure determined by
(6.1) Δ(x) = x ⊗ 1 + 1 ⊗ x , η(x) = 0 , S(x) = −x
for all x ∈ sl2 . In particular, Uh is equipped with the h-adic topology, and is a
topologically free C[[h]]-module; the algebraic tensor product Uh ⊗Uh is thus equally
completed in h-adic topology. Then, the coproduct Δ is well-defined, and all the
above structure maps are continuous.
Let us identify Uq with the algebra Uq of Exercise 2.3. The latter is explicitly
defined for all values of q and inherits from Uq a structure of Hopf algebra. There
is an injective morphism of Hopf algebras i : Uq → Uh , given by i(q) = eh/2 and
i(E) = XehH/4 , i(F ) = e−hH/4 Y ,

(6.2)
i(K ±1 ) = e±hH/2 , i(L) = XY − e−hH/4 Y XehH/4 .
Hence, for what regards properties independent of the specific evaluation q = ε, we
will consider Uq as a subHopf algebra of Uh .
6.1.1. The QUE dual Uh◦ . Duality is a delicate question for infinite dimensional
algebras. We say that two Hopf algebras A and A over a ground ring k are dual
if there exists a bilinear pairing , : A ⊗ A → k which is non degenerate in the
sense that for all f ∈ A (resp. u ∈ A), u, f = 0 for all u ∈ A (resp. f ∈ A )
implies f = 0 (resp. u = 0), and
u, f g = Δ(u), f ⊗ g , u ⊗ v, Δ(f ) = uv, f
(6.3)
u, S(f ) = S(u), f , η(f ) = 1, f , η(u) = u, 1
for all u, v ∈ A, f , g ∈ A , where we denote by the same letters the structure maps
of A and A . When suitably interpreted, these formulas give indeed the linear dual
Uh∗ = HomC[[h]] (Uh , C[[h]]) a structure of topological Hopf algebra dual to Uh for
the natural pairing , : Uh ⊗ Uh∗ → C[[h]] [CP, Def. 6.3.3]. However Uh∗ is not a
QUE algebra since there does not exist any Lie algebra g such that Uh∗ /hUh∗ ∼ = Ug
as Hopf algebras (otherwise Uh∗ should be cocommutative up to first order by (6.1)).
31
A way to repair this inconvenience is to consider the space

Uh◦ = h−l I l ,
l≥0

where I = i,j,k (X i Y j H k )∗ Uh∗ is the maximal ideal of Uh∗ , the elements (X i Y j H k )∗
being dual to PBW basis vectors of U sl2 and the sums completed in h-adic topol-
ogy. Recall from Corollary 3.2 and Remark 3.4 that the Poisson-Lie structure of
H ∼= Spec(Z0 ) given by the bracket ψ∗ { , } is dual to the so called standard one
on P SL2 C. We have:
Proposition 6.1. [ES, Prop. 10.3] Uh◦ is a Hopf algebra dual to Uh under
the completion of the natural pairing , : Uh ⊗ Uh∗ → C[[h]]. It is isomorphic to
the QUE algebra Uh (h), where h is the Lie algebra of H with bialgebra structure
tangent to ψ∗ { , }.
We call Uh◦ = Uh h the QUE dual of Uh .
6.1.2. The finite dual. There is another notion of dual Hopf algebra that we
will meet in the sequel. Its definition makes sense for arbitrary Hopf algebras (see
[Mo, § 9.1] or [KS, Ch. 3, Prop. 1.1.3]):
Definition 6.2. The finite dual of a k-algebra A is the subspace A• of the
linear dual A∗ defined by
A• = {f ∈ A∗ | f (I) = 0 for some ideal I such that dimC (A/I) < ∞}.
We have:
Proposition 6.3. If A is a Hopf algebra, the finite dual A• is a Hopf algebra,
with structure maps defined by (6.3) under the natural pairing , : A ⊗ A∗ → k.
Moreover, A• is the largest subspace V of A∗ with coproduct in V ⊗ V .
Note that the coproduct of A always gives A∗ an algebra structure; only the
product may cause some trouble, as its adjoint for , may not map into the
subspace A∗ ⊗ A∗ ⊂ (A ⊗ A)∗ .
The finite dual A• can be equivalently defined as the subalgebra of A∗ generated
by the matrix elements of all finite dimensional A-modules V , that is, by the linear
functionals
cVl,v : A −→ C
(6.4)
a −→ l(a.v)
where v ∈ V , l ∈ V ∗ . In these terms, Proposition 6.3 follows from the fact that the
category of finite-dimensional A-modules is closed under taking duals, direct sums
and tensor products.
A topological interpretation of A• is as follows. Let
J = {Ker(ρ) | ρ : A → End(V ) is a finite dimensional representation}.
Define the J -adic topology on A by taking as a base of neighborhoods of a ∈ A the
sets {a + J | J ∈ J }. Similarly, define a topology on A ⊗ A by taking as a base of
neighborhoods of a⊗b ∈ A⊗A the sets {a⊗b+L | L = A⊗J +K ⊗A and J, K ∈ J }.
Then A is a topological algebra (all the structure maps are continuous), and if the
ground ring k of A is given the discrete topology, A• is the set of continuous k-linear
maps A → k. For A = Uh , by taking representations ρ on free C[[h]]-modules of
finite rank, the finite dual Uh• is the set of C[[h]]-linear maps Uh → C[[h]] which are
134
continuous for the J -adic topology. It is usually called the quantized function ring,
and denoted by SLh (2) [CP, Th. 7.1.4].
The finite dual Uq• is defined also for all values of q ∈ C, q = −1, 0, 1. At
q = ε it is very big, as it contains the representative functions on the bundle ΞA of
Theorem 4.12 (compare eg. with Theorem 6.7 below). So Uq• is usually defined in
±
the literature by restricting the matrix elements (6.4) to the Uq -modules Vr,q ; then
it coincides with the rational form SLq (2) of SLh (2) (see [CP, Ch. 13], [Ks, Ch.
VII], [KS, Ch. 3]).
6.2. The Heisenberg double Hh = H(Uh ). The Heisenberg double of Uh is

a topological C[[h]]-module isomorphic to Uh ⊗ Uh◦ , with an algebra structure given
by a smash product. Let us recall this notion (see [Mo]).
Definition 6.4. Let A be a Hopf k-algebra and V a left A-module which is
simultaneously an algebra. We say that V is an A-module algebra if both its product
V ⊗ V → V and unit k → V are morphisms of A-modules. The smash product
V A of A and an A-algebra V is the algebra isomorphic to V ⊗ A as a vector
space, with product given by
(u a)(v b) = u(a(1) .v) a(2) b
for all u, v ∈ V and a, b ∈ A, where we put Δ(a) = a(1) ⊗ a(2) (Sweedler’s sigma
notation).
The smash product V A encodes the commutation relations between the linear
operators induced by the left action of V on itself, and the linear operators induced
by the action of A on V . Indeed, one checks without difficulty that:
Lemma 6.5. The map
λ: V A −→ End(V )
(6.5)
v a −→ (u → v(a.u))
defines a representation of V A (the Heisenberg representation).
Definition 6.6. Let , : A ⊗ A → k be a duality of Hopf k-algebras A and
A . The Heisenberg double H(A) is the smash product A A , where A is made
into an A -module via the left regular action
: A ⊗ A −→ A
(6.6)
x ⊗ a −→ x a = a(1) a(2) , x.
For Heisenberg doubles, the fact that (6.5) defines a representation follows from
the commutation relation
(6.7) (1 x)(a 1) = (a(1) 1)(1 a(2) , x(1) x(2) )
for all a ∈ A, x ∈ A . In particular, the Heisenberg double Hh = H(Uh ) is the
topological algebra isomorphic to Uh ⊗ Uh◦ as a C[[h]]-module, with product given
by
(u x)(v y) = u(x(1) v) x(2) y = v(2) , x(1) uv(1) x(2) y
for all u, v ∈ Uh , x, y ∈ Uh◦ .
33
6.2.1. A classical example: the cotangent bundle T ∗ G. The Heisenberg double

Hh may be understood as a deformation of the following classical geometric situa-
tion. Let G be an affine algebraic group over C. There are three “classical” Hopf
algebras associated to G:
• The group algebra CG, where Δ(g) = g ⊗ g, η(g) = 1, and S(g) = g −1 for
all g ∈ G.
• The coordinate ring C[G] ⊂ CG• (as the subset of algebraic maps).
• The universal enveloping algebra U g, with structure maps (6.1).
Note that the adjoint action of G on g extends linearly to give an action of CG
on U g by Hopf algebra automorphisms. Denote by e the identity element of G,
and identify U g with the space of all left invariant differential operators on G. For
simplicity, assume that G is connected, simply connected, and semisimple. We
have:
Theorem 6.7. (See [A, Th. 4.3.13]) The pairing

, : U g × C[G] −→ C
(6.8)
X ⊗f −→ (X.f )(e)
induces an injective morphism U g → C[G]• and an isomorphism C[G] → U g• . The

Hopf algebra C[G]• is generated by U g and the evaluation maps f → f (g), g ∈ G.
More precisely, C[G]• ∼
= U g ⊗ CG as a coalgebra, and its product and antipode are
given by
(6.9) (X ⊗ f )(Y ⊗ g) = X(f(1) .Y ) ⊗ f(2) g,
S(X ⊗ f ) = S(f(1) ).S(X) ⊗ S(f(2) ).
In the situation of Theorem 6.7, the action of U g on C[G] ∼

= U g• by left
invariant derivations takes a form dual to (6.6), since
d
(6.10) (X.f )(a) = f (aetX ) |t=0 = f(1) (a)X, f(2)
dt
for all X ∈ g, f ∈ C[G] and a ∈ G, where , is as in (6.8). Hence, by dualiz-
ing Definition 6.6 and considering C[G] as an U g-module via (6.10), we find that
H(C[G]) = C[G] U g coincides with the algebra of all differential operators on G,
with its usual action (6.5) on C[G]. On another hand, the symmetrization map
Sg −→
Ug
x1 . . . x r −→ σ∈Sr xσ(1) . . . xσ(r)
yields vector space isomorphisms C[g∗ ] ∼ = Sg ∼ = U g that allow one to identify

H(C[G]) with the space of functions on the cotangent bundle T ∗ G ∼ = G g∗ . The
algebra structure is determined by the pairing (6.8), and hence by the canonical
symplectic structure on T ∗ G via the map assigning to each function its Hamiltonian
vector field.
6.2.2. Hh as a deformation of C[T ∗ G]. The cotangent bundle T ∗ G is an exam-
ple of Drinfeld double Lie group. It is associated to the trivial Poisson-Lie structure
of G. By starting with the standard Poisson-Lie structure of P SL2 C the general
theory provides us with a Drinfeld double D(P SL2 C) diffeomorphic to P SL2 C × H
136
in a neighborhood of the identity element e. It carries a Poisson bracket { , }s

which is non degenerate at e, and given by (see [KS, Prop. 6.1.16])

(6.11) {f1 , f2 }s = (∂i f1 ∂ i f2 − ∂ i f1 ∂i f2 ) + (∂i f1 (∂ i ) f2 − (∂ i ) f1 ∂i f2 ).
i i
Here, ∂i and ∂ i (resp. ∂i and (∂ i ) ) are dual basis of right (resp. left) invariant vector
fields on P SL2 C and H, respectively, considered as vector fields on D(P SL2 C). The
function space C[D(P SL2 C)] inherits from { , }s a structure of Heisenberg double
algebra H(C[P SL2 C]) that may be deformed to H(Uh• ) ([STS, Prop. 3.3], [Lu1,
Th. 3.10]). More precisely, by working dually and letting d denote the Lie algebra
of D(P SL2 C), one obtains:
Proposition 6.8. The Heisenberg double Hh = H(Uh ) is a topological algebra
deformation of U d over C[[h]], and is a quantization of (D(P SL2 C), { , }s ).
The last claim means that the Poisson bracket { , }s determines a bivector
πs ∈ d ⊗ d such that r = 1 + hπs mod(h2 ), where
(6.12) r = λ−1 (idUh )
and λ is the canonical map λ : Uh ⊗ Uh◦ −→ HomC[[h]] (Uh ).
Remark 6.9. As suggested by T ∗ G and D(G) above, the quantum Heisenberg
doubles (resp. (6.12)) are closely related to Fourier duality and quantum Drinfeld
doubles (resp. universal R-matrices). We refer to [STS] and [Lu2, CR] for results
in this direction.
6.3. The canonical morphism R : Ξ(2) → Ξ(2) . Recall the bundle ΞM of
Theorem 4.12. Let Spec(Zε )2reg denote the set of regular pairs of central characters.
Define a new bundle
(6.13) Ξ(2) : Mε(2) → Spec(Zε )2reg
by restricting the base space of the product bundle ΞM × ΞM and regarding each
fiber as a tensor product of Uε -modules; hence, for any (ρ, μ) ∈ Spec(Zε )2reg , the
fiber over (ρ, μ) has the form Vρ ⊗ Vμ . By Remark 5.11, the fibers over any two
pairs (ρ, μ), (ν, κ) ∈ Spec(Zε )2reg such that τ (ρ)τ (μ) = τ (ν)τ (κ) are isomorphic
Uε -modules.
Proposition 6.10. The coproduct induces an action ΔG of G on Ξ(2) by bundle
morphisms.
Proof. As in Theorem 4.12 it is enough to show that the coproduct induces a
homomorphism G → Aut(Ûε ⊗ Ûε ) whose image preserves Zε ⊗ Zε . For all x ∈ Uε
one computes in Uq that
n n
E E En
Δ( ,u ) = ⊗1+K ⊗ n
+ . . . , Δ(u)
[n]! [n]! [n]!
n n
E K
= , u(1) ⊗ u(2) + , u(1) ⊗ E n u(2) +
[n]! [n]!
n
E
+ u(1) K ⊗
n
, u(2) + . . .
[n]!
35
where we put Δ(u) = u(1) ⊗ u(2) , and the dots refer to elements which are vanishing
at q = ε. By specializing at q = ε and using (3.3) we get
(6.14) Δ(De (u)) = (De ⊗ 1 + z ⊗ De − Dz ⊗ zx)Δ(u).
The right hand side is a derivation of Ûε ⊗ Ûε preserving Zε ⊗ Zε . As in the proof
of Proposition 3.6, it can be integrated to a 1-parameter group of automorphisms
Δ(exp(tDe )). Similar facts hold true for Δ(Df (u)) as well. We let ΔG be generated
by the actions of Δ(exp(tDe )) and Δ(exp(tDf )), t ∈ C.
We wish to let the element (6.12) act on the left on Ξ(2) , or equivalently, by
conjugation on the bundle of algebras obtained from Ξ(2) by replacing ΞM with
ΞA . In order for this to make sense, we realize it as an element of Hh ⊗ Hh :
Definition 6.11. The canonical element of Hh is R = τ ◦ (i ⊗ i )(r)21 ∈
Hh ⊗ Hh , where the subscript “21 ” means that the tensorands are permuted, and
i : Uh → Hh and i : Uh◦ → Hh are the natural linear embeddings in Hh ∼ = Uh ⊗ Uh◦ .

By writing r = i ei ⊗ ei in dual topological basis {ei } and {ei } of Uh and Uh◦ ,
we have

(6.15) R= (1 ei ) ⊗ (ei 1).
i
The next results sums up the fundamental properties of R (compare with [Ks, Th.
IX.4.4] and [BS, Prop. 1.2]). To simplify notations, we will often identify u ∈ Uh
with i(u) = u 1 and x ∈ Uh◦ with i (x) = 1 x.
Proposition 6.12. i) We have R−1 = (S ⊗ id)(R) and (id ⊗ Δ)(R) = R12 R13 .
ii) The following identities hold true:
(6.16) R(u ⊗ 1) = Δ(u)R , (1 ⊗ x)R = RΔ(x)
for all u ∈ Uh and x ∈ Uh◦ , and
(6.17) R12 R13 R23 = R23 R12 .
iii) Denote by μ the product of Uh . The image R = (λ ⊗ λ)(R) ∈ AutC[[h]] (Uh ⊗ Uh )
of the canonical element under the Heisenberg representation (6.5) is given by
(6.18) R = (id ⊗ μ)(Δ ⊗ id).

Proof. i) We have R(S ⊗ id)(R) = i,j (1 ei (ej ◦ S)) ⊗ (ei ej 1). Then, by
using (6.3) we see that for all x, y ∈ Uh◦ and u, v ∈ Uh we have

R(S ⊗ id)(R), x ⊗ u ⊗ y ⊗ v = x(1)η(v) y(ei ej ei (u(1) )ej (S(u(2) )))
i,j,(u)
= x(1)η(v)y(u(1) S(u(2) ))
= x(1)η(v)y(1)η(u)
= 1 ⊗ 1 ⊗ 1 ⊗ 1, x ⊗ u ⊗ y ⊗ v.
Here, η is the counit of Uh , and , the product of the natural duality pairings.
It follows that R−1 = (S ⊗ id)(R). As for the second identity, note that we have
(u, ⊗ id)(R) = u, (id ⊗ , x)(R) = x, and similarly for y. Hence
u ⊗ x ⊗ y, (id ⊗ Δ)(R) = x ⊗ y, Δ(u).
138
On another hand u ⊗ x ⊗ y, R12 R13 = (id ⊗ , x)(R)(id ⊗ , y)(R), u = xy, u.

The result follows again from (6.3).
ii) For all u ∈ Uh and x ∈ Uh◦ , we have (id ⊗ , x)(R(u ⊗ 1)) = xu, and

(id ⊗ , x)(Δ(u)R) = u(1) ei u(2) ei , x
i

= u(1) ei u(2) , x(1) ei , x(2)
i
= u(1) u(2) , x(1) x(2) .
Together with (6.7) this proves the first identity in (6.16). The second one is similar.
Finally, R23 R12 = (id ⊗ Δ)(R)R23 = R12 R13 R23 by (6.16) and i).
iii) One checks (6.18) by a straightforward computation, writing R as in (6.15).
Details are left as an exercise.
Like (6.2) we have an embedding of the rational form Uq h of Uh◦ = Uh h into

Uh◦ ,
and hence of the smash product Hq = Uq Uq h into Hh . One can check that
the conjugation action of R on Hh ⊗ Hh induces an (outer) automorphism R of the
subalgebra Hq ⊗Hq . Moreover, in complete analogy with the semi-classical situation
considered in [WX], Hq has a double structure of quantum groupoid through which
R factors to define an automorphism of Uq ⊗ Uq . Then, by specializing at q = ε
and considering the resulting left action on tensor products of simple Uε -modules,
using (6.16)-(6.17), (5.30) and (5.33) we get:
Theorem 6.13. The canonical element R of Hh induces a bundle morphism
R : Ξ(2) −→ Ξ(2) such that
(6.19) R12 R13 R23 = R23 R12
and R(u ⊗ id) = Δ(u)R for all u ∈ Uε . Hence R coincides at (ρ, μ) ∈ Spec(Zε )2reg
with the evaluation map K(ρ, μ), and the regular 6j-symbol operator R(ρ, μ, ν) is
also a value of R.
Note that the Pentagon equation (5.34) is equivalent to (6.19). By identifying
K with R each multiplicity module M (h) gets a natural structure of Uε -module,
such that R is Uε ⊗ Uε -linear at points (ρ, μ) where τ (ρ)τ (μ) = h.
Finally, let us go back to the quantum coadjoint action. By passing the co-
products on the left in (6.16), we get two actions of Uε and Uε h on the set of bundle
morphisms of Ξ(2) , which leave R invariant. By using these actions one can show
that:
Theorem 6.14. The G-action on ΞM extends to End(Ξ(2) ) by preserving R.
In particular, R is constant along the ΔG -orbits in Ξ(2) . Since the G-orbits in
Spec(Zε ) \ D cover the orbits of the adjoint action in P SL2 C0 , Theorem 4.11 (ii)
and Remark 5.11 show that R descends to a morphism of a vector bundle of rank
n2 over a covering of P SL2 C//P SL2 C of degree 2n.
6.4. Matrix dilogarithms. According to Theorem 6.14 one can compute R

by restricting to pairs (ρ, μ) such that ϕ(ρ), ϕ(μ) ∈ P B+ . So we could have devel-
oped the whole theory by starting with the Heisenberg double H(Uε b) of a Borel
subalgebra Uε b ⊂ Uε . This has been done in [BB1, BB2]. Explicit formulas have
37
shown that R satisfies tetrahedral symmetry relations, and produced an elemen-

tary form of (6.19) reminiscent of the five-term identities satisfied by the classical
dilogarithm functions. In order to state it we need a few preparation.
Define a QH tetrahedron Δ(b, w, f, c) as an oriented abstract tetrahedron Δ
endowed with:
• a branching b, defined as an orientation of the edges inducing an ordering
of the vertices vi by stipulating that an edge points towards the greater
of its endpoints. The 2-faces δi are then ordered as the opposite vertices,
and the edges of δ3 are denoted by ej , where the source vertices of e0 and
e1 are v0 and v1 , respectively.
• a triple w = (w0 , w1 , w2 ), where wj ∈ C \ {0, 1} is associated to ej and
the opposite edge, and wj+1 = 1/(1 − wj ); hence w corresponds to the
cross-ratio moduli of an ideal hyperbolic tetrahedron.
• a flattening f = (f0 , f1 , f2 ) and a charge c = (c0 , c1 , c2 ), where fj , cj ∈ Z
are associated to ej and the opposite edge and satisfy:
The flattening condition: l0 + l1 + l2 = 0, where
√
(6.20) lj = lj (b, w, f ) = log(wj ) + −1πfj ;
The charge condition:
(6.21) c0 + c1 + c2 = 1.
The branching endows Δ with a b-orientation, positive and denoted by ∗b = 1 when
the 2-face δ3 inherits from the orientation of its edges the opposite of the boundary
orientation. In Figure 1 we show the two branched tetrahedra for ∗b = ±1 (up to
global symmetries), together with the 1-skeletons of the cell decompositions dual
to the interiors.
1 1
2, i 0, j 0, j 2, i
3 3
0 2 2 0
1, l 3, k 3, k 1, l
∗b = 1 ∗b = −1
Figure 1. Branched tetrahedra.
Recall that we denote by n the order of ε. The nth root cross-ratio moduli of a
QH tetrahedron Δ(b, w, f, c) are defined by
1 √
(6.22) wj = exp(lj,n ), where lj,n = log(wj ) + π −1(n + 1)(fj − ∗b cj ) .
n
n of the
The pairs (w0 , w1 ) ∈ C2 define global coordinates of the Riemann surface C
map
C \ {0, 1} −→ C2
(6.23) 1
(w0n , (1 − w0 )− n ).
1
w0 −→
140
Let us put m := (n − 1)/2, and

− xn
n−1
−1 1
[x] := n , g(x) := (1 − xζ −j )j/n , h(x) := g(x)/g(1)
1−x j=1

n
v
ω(u , v |n) := , (u )n + (v )n = 1, n ∈ N,
j=1
1 − u ζ j
with ω(u , v |0) := 1 by convention, and x1/n := exp(log(x)/n) is extended to

01/n := 0 by continuity. The function ω is n-periodic in its integer argument, and
g is analytic over C \ {rζ k , r ≥ 1, k = 1, . . . , n − 1}.
Definition 6.15. The (nth) matrix dilogarithm of a branched oriented tetra-

hedron Δ(b) is the regular map
Rn (Δ, b) : n
Z2 × C −→ Aut(Cn ⊗ Cn )
n−1
(c0 , c1 , w0 , w1 ) −→ (w0 )−c1 (w1 )c0 2 (Ln )∗b (w0 , (w1 )−1 )
where
Ln (u , v )i,j
ω(u , v |i − k) δn (i + j − l)
2
kj+(m+1)k
k,l = h(u ) ζ
k,l [u ] −kj−(m+1)k2 δn (i + j − l)
Ln (u , v )−1 i,j = ζ ,
h(u ) ω(u /ζ, v |i − k)
We will write Rn (Δ, b)(c0 , c1 , w0 , w1 ) = Rn (Δ, b, w, f, c). Note that the branch-
ing b associates an index of Ln (w0 , w1 )±1 to each 2-face of Δ by the rule indicated
in Figure 1.
Consider a move T → T between triangulated hexahedra T and T , as shown
in Figure 2. Assume that we have QH tetrahedra on both sides, having branchings
that coincide at every common edge; in Figure 2 we have fixed one such global
branching, but there are five other possible choices up to global symmetries.
(1-x)/(1-y)
y
x
y(1-x)/x(1-y) y/x
Figure 2. An instance of transit.
Define

(6.24) WT (e) = w (h)∗b , CT (e) = c(h)
h→e h→e
where:
39
• “h → e” means that h is an edge of a QH tetrahedron of T that is identified

with the edge e in T , and the products are over all such edges;
• ∗b = ±1 according to the b-orientation of the QH tetrahedron that con-
tains h;
• w (h) is the nth root cross-ratio modulus (6.22) at h, and c(h) its charge.
The same notions are defined for T . We say that T → T is a 2 → 3 transit if at
every common edge e we have
(6.25) WT (e) = WT (e) , CT (e) = CT (e).
The transit conditions are very restrictive; for instance, in Figure 2 we have shown
the relations between the cross-ratio moduli w0 of the five QH tetrahedra.
To each QH tetrahedron of T or T , a matrix dilogarithm Rn (Δ, b, w, f, c) is
associated. Define an n-state of T or T as a function that gives every 2-simplex
an index, with values in {0, . . . , n − 1}. By the rule of Figure 1, every n-state
determines a matrix dilogarithm entry. As two adjacent tetrahedra induce opposite
orientations on a common face, an index is down for the matrix dilogarithm of one
tetrahedron when it is up for the other. By summing over repeated indices we get
a tensor

(6.26) Rn (T ) = s Δ⊂T Rn (Δ, b, w, f, c)s
where the sum is over all n-states of T , and Rn (Δ, b, w, f, c)s stands for the matrix
dilogarithm entry determined by s.
We can now state the analog of Theorem 5.5 for regular Uε -modules. By
comparing formulas it makes explicit the relationship between their 6j-symbols and
the matrix dilogarithms. The first claim is essentially a consequence of Theorem
6.14. The rest is proved in [BB1, BB2]. Recall the map ϕ : Rep(Uε ) → P SL2 C0
in (4.11), and the isomorphism P SL2 C ∼ = Aut(P1 ).
Theorem 6.16. 1) Let (ρ, μ, ν) be a triple of Uε -modules which is regular and
cyclic. Put f = ϕ(ρ), g = ϕ(μ), h = ϕ(ν) ∈ P SL2 C0 . The 6j-symbol operator
R(ρ, μ, ν) coincides up to conjugacy with the map
R: n
C −→ Aut(Cn ⊗ Cn )
(w0 , w1 ) −→ Ln (w0 , (w1 )−1 ),
where w0 squared is the cross-ratio [0 : f (0) : f g(0) : f gh(0)] of the points in P1 ,
and the nth roots w0 and w1 are determined by the Casimir coordinates of Ξ(ρ),
Ξ(μ), Ξ(ν) ∈ Spec(Zε ).
2) The matrix dilogarithm Rn (Δ, b) satisfies:
• Invariance under full tetrahedral symmetries, up to a determined projec-
tive action of SL2 Z on the source and target spaces of Aut(Cn ⊗ Cn ).
• For any 2 → 3 transit T → T we have a five term identity
(6.27) Rn (T ) = Rn (T ).
Note that the tetrahedral symmetries of the matrix dilogarithms depend on the
flattening and charge conditions (6.20)–(6.21). They are necessary to get the five
term identities for all the 2 → 3 transits.
The cross-ratio w0 is a complex number and is distinct from 0 and 1 because
the triple (ρ, μ, ν) is regular and cyclic. The nth root modulus w0 describes via
Theorem 4.11 ii) a coset of Spec(Zε ) mod G. The 3-cocycloid identities (5.34) or
142
(6.19) coincide with (6.27) exactly for the 2 → 3 transits T → T with the branching
of Figure 2. In this form, it is a non Abelian deformation of the celebrated five
term identity
1 − x−1 1−x
(6.28) L(x) − L(y) + L(y/x) − L( ) + L( )=0
1 − y −1 1−y
which x, y are real, 0 < y < x < 1, and L is the dilogarithm function

π2 1 x log(t) log(1 − t)
(6.29) L(x) = − − + dt.
6 2 0 1−t t
References
[A] E. Abe, Hopf algebras, Cambridge Tr. Math. 74, Camb. Univ. Press (1980)
[AMcD] M.F. Atiyah, I.G. MacDonald, Introduction to Commutative Algebra, Addison-Wesley
(1969)
[Bai] H. Bai, Quantum Teichmüller spaces and Kashaev’s 6j-symbols, Alg. Geom. Topol. 7 (2007)
1541–1560
[BB1] S. Baseilhac, R. Benedetti, Quantum hyperbolic invariants of 3-manifolds with P SL(2, C)-
characters, Topology 43 (6) (2004) 1373–1423
[BB2] S. Baseilhac, R. Benedetti, Classical and quantum dilogarithmic invariants of flat
P SL(2, C)-bundles over 3-manifolds, Geom. Topol. 9 (2005) 493–570
[BB3] S. Baseilhac, R. Benedetti, Quantum hyperbolic geometry, Alg. Geom. Topol. 7 (2007)
845–917
[BBL] H. Bai, F. Bonahon, X. Liu, Local representations of the quantum Teichmuller space,
arXiv:0707.2151
[BFK] D. Bullock, C. Frohman, J. Kania-Bartoszynska, Understanding the Kauffman bracket
skein module, J. Knot Theory Ramifications 8 (3) (1999) 265–277
[BHMV] C. Blanchet, N. Habegger, G. Masbaum, P. Vogel, Topological quantum field theories
derived from the Kauffman bracket, Topology 34 (4) (1995) 883–927
[BL] F. Bonahon, X. Liu, Representations of the quantum Teichmuller space and invariants of
surface diffeomorphisms, Geom. Topol. 11 (2007) 889–937
[BS] S. Baaj, G. Skandalis, Unitaires multiplicatifs et dualité pour les produits croisés de C ∗ -
algèbres, Ann. Sc. Ec. Norm. Sup. sér. 4 (26) (1993) 425–488
[CFS] J.S. Carter, D.E. Flath, M. Saito, The Classical and Quantum 6j-symbols, Math. Notes
43, Princeton Univ. Press (1995)
[CP] V. Chari, A. Pressley, A Guide to Quantum Groups, Cambridge Univ. Press (1994)
[CR] C. Cibils, M. Rosso, Hopf bimodules are modules, J. Pure App. Alg. 128 (1998) 225–231
[D] A.A. Davydov, Pentagon equation and matrix bialgebras, Commun. Alg. (29) (6) (2001) 2627–
2650
[DCK] C. De Concini, V. G. Kac, Representations of quantum groups at roots of 1, in “Operator
Algebras, Unitary Representations, Enveloping Algebras, and Invariant Theory (Paris, 1989)”,
Birkhäuser, Progr. Math. 92 (1990) 471–506
[DCKP] C. De Concini, V. G. Kac, C. Procesi, Quantum coadjoint action, J. Amer. Math. Soc.
5 (1) (1992) 151–189
[DCP] C. De Concini, C. Procesi, Quantum Groups, in “D-modules, Representation Theory and
Quantum Groups (Venice 1992)”, Springer Verlag, Lect. Notes Math. 1565 (1993) 31–140
[DCPRR] C. De Concini, C. Procesi, N. Reshetikhin, M. Rosso, Hopf algebras with trace and
representations, Invent. Math. 161 (2005) 1–44
[ES] P. Etingof, O. Schiffman, Lectures on Quantum Groups, Int. Press (1998)
[GP] N. Geer, B. Patureau-Mirand, Topological invariants from non-restricted quantum groups,
ArXiv:math.GT/0112769
[J] J. C. Jantzen, Lectures on Quantum Groups, AMS Grad. Stud. Math. 6 (1996)
[Ka] R.M. Kashaev, The hyperbolic volume of knots from the quantum dilogarithm, Lett. Math.
Phys. 39 (1997) 269–275
[KaRe] R. Kashaev, N.Y. Reshetikhin, Braiding for the quantum gl2 at roots of unity,
ArXiv:math/0410182
41
[Ks] C. Kassel, Quantum Groups, Grad. Texts Math. 155, Springer Verlag, New-York (1995)
[Ko] T. Kohno, Conformal Field Theory and Topology, AMS Iwanami Ser. Mod. Math. Trans.
Math. Monogr. 210 (2002)
[KR] A.N. Kirillov, N.Y. Reshetikhin, Representations of the algebra Uq (sl(2)), q-orthogonal poly-
nomials and invariants of links, reprinted in Kohno “New Developments in the Theory of
Knots”, World Sc. Publishing (1989)
[KS] L.I. Korogodski, Y.S. Soibelman, Algebras of Functions on Quantum Groups: Part I, AMS
Math. Surv. Monogr. 56 (1998)
[L] S. Lang, Algebra, Grad. Texts Math. 211, 3rd edition, Springer Verlag (2002)
[Lu1] J.-H. Lu, Moment maps at the quantum level, Commun. Math. Phys. 157 (2) (1993) 389–404
[Lu2] J.-H. Lu, On the Drinfeld double and the Heisenberg double of a Hopf algebra, Duke Math.
J. 74 (3) (1994) 763–776
[Ma] J. Marché, Geometric approach of simplicial formulas for the Chern-Simons invariant,
preprint
[MN] J. Marché, M. Narimannejad, Some asymptotics of topological quantum field theory via
Skein theory, Duke Math. J. 141 (3) (2008) 573–587
[Mi] G. Militaru, The Hopf module category and the Hopf equation, Commun. Alg. 26 (10) (1998)
3071–3097
[Mo] S. Montgomery, Hopf Algebras and their Actions on Rings, AMS Conf. Board Math. Sc. 82
(1993)
[N] W.D. Neumann, Extended Bloch group and the Cheeger-Chern-Simons class, Geom. Topol. 8
(2004) 413–474
[Prob] Invariants Of Knots And 3-Manifolds (Kyoto 2001). A Collection Of Papers And Prob-
lems, T. Ohtsuki ed., Geom. Topol. Monogr. 4 (2002-4)
[PS] J. Przytycki, A. Sikora, On skein algebras and Sl2 (C)-character varieties, Topology 39 (1)
(2000) 115–148
[Sh] I.R. Shafarevitch, Basic Algebraic Geometry I, Springer Verlag, Berlin Heidelberg (1994)
[RT1] N. Reshetikhin, V.G. Turaev, Invariants of 3-manifolds via link polynomials and quantum
groups, Invent. Math. 103 (1991) 547–597
[RVW] N. Reshetikhin, A. Voronov, A. Weinstein, Semiquantum geometry, J. Math. Sc. 82 (1)
(1996) 3255–3267
[STS] M. A. Semenov-Tian-Shansky, Poisson-Lie groups, quantum duality principle, and the
quantum double, Contemp. Math. 175 (1994) 219–248
[T] V.G. Turaev, Quantum Invariants of 3-Manifolds, De Gruyter Stud. Math. 18 (1994)
[WX] A. Weinstein, P. Xu, Classical solutions of the quantum Yang-Baxter equation, Commun.
Math. Phys. 148 (1992) 309–343
Institut Fourier, Université de Grenoble, 100 rue des Maths BP 74, 38402 Saint-
Martin-d’Hères Cedex, France
Current address: Université Montpellier 2, Institut de Mathématiques et de Modélisation,
Case Courrier 051, Place Eugène Bataillon, 34095 Montpellier Cedex, France
E-mail address: stephane.baseilhac@math.univ-montp2.fr
Volume 541, 2011
What is a sequence of Nilsson type?
Stavros Garoufalidis
Abstract. Sequences of Nilsson type appear in abundance in Algebraic Ge-

ometry, Enumerative Combinatorics, Mathematical Physics and Quantum To-
pology. We give an elementary introduction on this subject, including the
definition of sequences of Nilsson type and the uniqueness, existence, and ef-
fective computation of their asymptotic expansion.
1. Sequences of Nilsson type: definition

Sequences of Nilsson type are the ones that are asymptotic to power series
in powers of 1/n and log n. They appear in abundance Analysis (in asymptotic
expansions of integrals), in Mathematical Physics and in Algebraic Geometry (in
relation to the Gauss-Manin connection); see for example [An2, M1, M2, Ph, Sa].
They also appear in Enumerative Combinatorics (see [FS, WZ2, Ga2]) and in
Quantum Topology. For instance, the Witten-Reshetikhin-Turaev invariant of a
closed 3-manifold is a sequence of complex numbers that depends on the level,
and it is expected to be of Nilsson type; see [Wi, FG, Ga3, Ro, LR, AH]. In
addition, the Kashaev invariant of a knot is expected to be a sequence of Nilsson
type; see [KT, AH, CG]. The quantum spin network evaluation at a fixed root of
unity is known to be a sequence of Nilsson type; see [GV1, GV2]. For a general
discussion of perturbative and non-perturbative invariants of knotted objects that
are expected to be sequences of Nilsson type, see [Ga1].
There is a close connection between sequences of Nilsson type and multivalued
analytic functions with quasi-unipotent monodromy; see for example Theorem 4.1
below.
Several people familiar with the ideas of Quantum Topology have asked for
a self-contained definition of sequences of Nilsson type and their asymptotics, its
uniqueness, existence and effective computation.
Asymptotics of sequences is a well-studied subject of analysis that goes back at
least to Poincare; see for example [O, C, M1]. Since we could not find a reference
for sequences of Nilsson type in the existing literature, we decided to write this
The author was supported in part by NSF.
1991 Mathematics Classification. Primary 57N10. Secondary 57M25.

Key words and phrases: asymptotics, sequence, Nilsson type, holonomic sequence, G-function,
quasi-unipotent monodromy.
2011
c 0000
American
c Mathematical
(copyright Society
holder)
1
145
146
introductory article. It concerns the asymptotic expansion of sequences which are

relevant in Quantum Topology, and may serve as an elementary introduction to
asymptotics. We claim no original results in this survey paper.
In order to define sequences of Nilsson type, we need to introduce Nilsson
monomials hω (n) indexed by a well-ordered set Ω, and a finite set Λ of complex
numbers of equal magnitude.
For a natural number d ∈ N, a finite subset S of the rational numbers consider
the well-ordered set Ω = (S + N) × {0, 1, . . . , d} indexed by (α, β) < (α , β ) if and
only if α < α or α = α and β < β. Ω has the order type of the natural numbers.
In particular, for every ω ∈ Ω, the set of elements strictly smaller than ω is finite.
Consider the Ω-indexed family of monomials of Nilsson type given by:
(log n)β
(1) hω (n) =
nα
for ω = (α, β) ∈ Ω. It is easy to see that limn hω (n)/hω (n) = 0 (abbreviated by
hω (n)/hω (n) = o(1), and also by hω (n) hω (n)) if and only if ω < ω . This and
all limits below are taken when n goes to infinity.
Fix a finite set Λ of nonzero complex numbers of magnitude r > 0. Let cω,λ be
a collection of complex numbers indexed by Ω × Λ.
Definition 1.1. (a) With the above notation, for a complex-valued sequence
(an ) the expression

(2) an ∼ hω (n) cω,λ λn
ω∈Ω λ∈Λ
means that
• for every ω ∈ Ω we have:
⎛ ⎞
1
(3) ⎝an r −n − hω (n) cω ,λ (λr −1 )n ⎠ = o(1)
hω (n)
ω ≤ω λ∈Λ
• cω,λ = 0 for some (ω, λ) ∈ Ω × Λ.

(b) We say that a sequence (an ) is of Nilsson type if there exist Ω, Λ and cω,λ such
that (2) holds.
We will say that an asymptotic expansion (2) is Ω × Λ-minimal if
• For every λ ∈ Λ there exists ω ∈ Ω such that cω,λ = 0.
• For every ω ∈ Ω there exists λ ∈ Λ such that cω,λ = 0.
By considering a subset of Λ or Ω if necessary, it is easy to see that every asymptotic
expansion has a minimal representative.
2. Uniqueness
Our first task is to show that a sequence of Nilsson type uniquely determines
Ω, Λ and the coefficients cω,λ . The key idea is the following elementary lemma.
Lemma 2.1. If Λ is a finite subset of the unit circle and

(4) cλ λn = o(1)
λ∈Λ
holds for some complex numbers cλ , then cλ = 0 for all λ ∈ Λ.

WHAT IS A SEQUENCE OF NILSSON TYPE? 147
3
Proof. Divide (4) by λn1 for some λ1 ∈ Λ. Then we have

cλ1 + cλ (λ/λ1 )n = o(1)
λ=λ1
where λ/λ1 = 1. So,

⎛ ⎞
1
n
⎝cλ1 + cλ (λ/λ1 )k ⎠ = o(1).
n
k=1 λ=λ1
By averaging, it follows that

1 1 − (λ/λ1 )n+1
cλ 1 + cλ = o(1).
n 1 − λ/λ1
λ=λ1
Thus, cλ1 = 0. Since λ1 was an arbitrary element of Λ, the result follows.

Lemma 2.2. If (an ) satisfies (2) then
(5) lim sup |an |1/n = r.
n
Proof. Since cω,λ = 0 for some (ω, λ) ∈ Ω × Λ, without loss of generality

assume that cω0 ,λ = 0 for some λ ∈ Λ where ω0 is the smallest element of Ω.
Equation (2) for ω = ω0 gives that
an r −n
(6) − cω0 ,λ (λr −1 )n = o(1)
hω0 (n)
λ∈Λ
−1
Now λr are on the unit circle. It follows that

an r −n

hω (n) < C
0
for some C > 0. Since limn hω (n) 1/n

= 1 for all ω ∈ Ω, it follows that
lim sup |an |1/n ≤ r
n
−1
Sincesome cω0 ,λ is nonzero and λr are on the unit circle, Lemma 2.1 implies that
limn λ∈Λ cω0 ,λ (λr −1 )n = 0. Since the sequence is bounded, it follows that there
exists a subsequence nk such that

lim cω0 ,λ (λr −1 )nk = C = 0
nk
λ∈Λ
Combined with Equation (6), it follows that

lim |ank |1/nk = r
nk
The result follows.

Lemma 2.2 implies that sequences of Nilsson type satisfy lim supn |an | 1/n
> 0.
Proposition 2.1. Suppose that

(7) an ∼ hω (n) cω,λ λn
ω∈Ω λ∈Λ
and

(8) an ∼ hω (n) cω ,λ λn
ω ∈Ω λ ∈Λ
148
are Ω × Λ-minimal and Ω × Λ -minimal asymptotic expansions. Then Ω = Ω ,

Λ = Λ . Moreover, for all (ω, λ) ∈ Ω × Λ we have cω,λ = cω,λ .
Proof. Let ω0 and ω0 denote the smallest elements of Ω and Ω . Lemma 2.2
implies that r = r where r and r are the magnitudes of the elements of Λ and Λ
respectively. Equation (3) for ω0 and ω0 implies that
an r −n an r −n
(9) − cω0 ,λ (λr −1 )n = o(1), − cω0 ,λ (λ r −1 )n = o(1)
hω0 (n) hω0 (n)
λ∈Λ λ ∈Λ
If ω0 =ω0 ,we may assume that ω0 < ω0 .In that case, observe that hω0 (n)/hω0 (n) =
o(1). Multiply the second equation above by hω0 (n)/hω0 (n) and subtract from the
first. It follows that
hω (n)
− cω0 ,λ (λr −1 )n + 0 c (λ r −1 )n = o(1)
hω0 (n) ω0 ,λ
λ∈Λ λ ∈Λ
Since hω0 (n)/hω0 (n) = o(1), it follows that

cω0 ,λ (λr −1 )n = o(1)
λ∈Λ
Lemma 2.1 implies that cω0 ,λ = 0 for all λ contrary to our minimality assumption
of (7). It follows that ω0 = ω0 . Subtracting, Equation (9) implies that

− cω0 ,λ (λr −1 )n + cω0 ,λ (λ r −1 )n = o(1)
λ∈Λ λ ∈Λ
Lemma 2.1 implies that if cω0 ,λ = 0 for some λ ∈ Λ, then λ ∈ Λ and moreover
cω0 ,λ = cω0 ,λ .
An easy induction on ω ∈ Ω proves the following statement. For every ω ∈ Ω,
the following holds. If cω,λ = 0 for some λ ∈ Λ, then λ ∈ Λ and ω ∈ Ω and
cω,λ = cω,λ .
The minimality assumption and the above statement implies that Ω = Ω and
Λ = Λ and cλ,ω = cλ,ω for all (ω, λ) ∈ Ω × Λ.
Remark 2.2. Proposition 2.1 proves uniqueness in a non-effective way. We

will come back to the problem of computing cω,λ later on.
3. Alternative expression for sequences of Nilsson type

If (an ) is a sequence of Nilsson type, we can write (2) in the form:

(10) an ∼ λn nα (log n)β Sλ,α,β gλ,α,β (1/n)
λ,α,β
where
• the summation in (10) is over a finite set,
• the growth rates λ are complex numbers numbers of equal magnitude,
• the exponents α are rational numbers and the nilpotency exponents β are
natural numbers,
• the Stokes constants Sλ,α,β are complex numbers,
• gλ,α,β (x) ∈ C[[x]] are formal power series in x with complex coefficients
and leading term 1.
5
Remark 3.1. In the definition of a sequence of Nilsson type, we may addition-

ally require that
• Λ is a set of algebraic numbers,
• the formal power series gλ,α,β (x) is Gevrey-1, i.e., that the coefficient of
xk in gλ,α,β (x) is bounded by C k k! for all k, where C depends on gλ,α,β ,
• the coefficients of the formal power series gλ,α,β (x) lie in a fixed number
field K,
These additional requirements hold for the evaluations of classical spin networks,
see [GV1], as well as Sections 4 and 6.1 below.
Example 3.1. For example, if d = 1 and S = {1, 3/2}, then Ω = (1 + N) ∪

(3/2 + N) and we have:
log n 1 log n 1 log n 1
3/2 3/2 2
2 ...
n n n n n n
If Λ = {κ, μ, ν}, the asymptotic expansion (10) of a sequence of Nilsson type be-
comes:

log n n 1 log n n 1
an ∼ λ Sλ,1 gλ,1 + 3/2 λj Sλ,2 gλ,2
n n n n
λ∈Λ λ∈Λ

1 1 1 1
+ λnj Sλ,3 gλ,3 + 3/2 λnj Sλ,4 gλ,4
n n n n
λ∈Λ λ∈Λ
where gλ,j (x) ∈ C[[x]] are formal power series in x and Sλ,j are complex numbers.
4. Existence
In this section we will prove that a sequence is of Nilsson type, under some
analytic continuation assumptions of its generating series. This is a well-known
argument (see for example, [C, CG, FS, GM, GIKM, M1]) that consists of the
following parts:
• apply Cauchy’s theorem to give an integral representation of the sequence,
• deform the contour of integration to localize the computation near the
singularities of the generating function,
• analyse the local computation using the local monodromy assumption of
the generating function.
Let us give the details of the above existence proof. Since sequences of Nilsson
type are exponentially bounded (as follows from Lemma 2.2), fix an exponentially
bounded sequence (an ) and consider its generating series
∞

(11) G(z) = an z n
n=0
G(z) is an analytic function for all complex numbers z that satisfy |z| < 1/R.
Suppose now that G has analytic continuation on a disk of radius r with singularities
at finitely many points κ, λ, μ, ν, . . . . Suppose also that G has further analytic
continuation on a disk of radius r + minus finitely many segments emanating from
150
the singularities radially as in the following figure.
Assume in addition that G has quasi-unipotent local monodromy at each singularity

λ, μ, ν, κ on the circle of radius r (i.e., the eigenvalues of the local monodromy are
complex roots of unity).
Theorem 4.1. Under the above assumptions, the sequence (an ) is of Nilsson
type.
∞ n
Corollary 4.1. Suppose that G(z) = n=0 an z is a multivalued analytic
function on C \ Λ (where Λ ⊂ C is a finite set) which is regular at z = 0, and has
quasi-unipotent local monodromy. Then, (an ) is a sequence of Nilsson type.
Remark 4.2. We know of at least three ways to show that a germ G(z) of an
analytic function can be analytically continued to the complex plane, namely
(a) G satisfies a linear differential equation, see for example [Ga2, Thm.1]
reviewed in Section 5.1 below. For examples that come from Quantum
Topology (specifically, spin networks) see [GV1, GV2].
(b) G satisfies a nonlinear differential equation. See for example the instanton
solutions of Painlevé I studied in detail in [GIKM] and the matrix models
of [GM].
(c) G is resurgent. See for example the Kontsevich-Zagier series studied in
detail in [CG], and more generally the arithmetic resurgence conjecture
of [Ga1] for sequences that appear in Quantum Topology.
Proof. (of Theorem 4.1) We begin by applying Cauchy’s theorem to give an

integral representation of (an ). If γ is a circle of radius less than 1/R that contains
the origin, then we have:

1 G(z)
(12) an = dz
2πi γ z n+1
We can deform γ to a contour C which consists of a contour Hλ around each

singularity λ and finitely many arcs γr+ of the circle of radius 1/(r + ) as in the
7
following figure.
(13)
The contours Hλ are known as Hankel contours in Analysis (see [O]) and Lefschetz
thimbles in Algebraic Geometry (see [Ph, Sa]). Cauchy’s theorem implies that

1 G(z) 1 G(z)
(14) an = n+1
dz + n+1
dz
2πi Hλ z 2πi γr+ z
λ∈Λ
The above expression is exact, and decomposes the sequence (an ) into a finite sum
of sequences (one per singularity) and an extra term. Of course, there is nothing
canonical about this decomposition, since the size of the Hankel contour and γr+
depends on . One could make the decomposition nearly canonical by using Hankel
contours that extend to infinity, but even so there are choices of directions to be
made, and we will not use them here.
The integral over γr, can be estimated by O((r + )−n ) since G is uniformly
bounded on the arcs γr, . Since we assume that the local monodromy of G(z)
around a singularity is quasi-unipotent, it follows (see [M1]) that modulo germs of
holomorphic functions at zero, G(λ + z) has a local expansion of the form

G(λ + z) = z α (log z)β hα ,β (z)
α ,β
where the summation is over a finite set, α ∈ Q, β ∈ N and hα ,β (z) are germs of
functions analytic at z = 0. For a germ f (z) of a multi-valued analytic function at
z = 0, let Δ0 f denote its variation defined by
Δ0 f (z) = lim f (e2πi z) − lim f (e2πi z)
→1 →0
(see [M1]) when z is restricted on a line segment [0, ). The variation of the building
blocks z α and (log z)β are given by
⎧
⎪
⎨(e
2πiα
− 1)z α α ∈ Q \ Z
α
Δ0 (z ) = δ0 α∈Z\N , Δ0 (log z) = 2πi
⎪
⎩
0 α∈N
where δ0 is the Dirac delta function (really, a distribution). For a singularity λ of
G(z), let Δλ G(z) denote the variation of G(λ + z). It follows that for z in the line
segment of Figure (13), we have
∞

(15) Δλ G(z) = z α (log z)β cα,β,λ,k z k−1
α,β k=0
152
∞
where the sum is over a finite set {α, β}, α ∈ Q\N, β ∈ N and ∞ k=0 k=0 cα,β,λ,k z
k
are germs of analytic functions at z = 0. When α ∈ Z\N, we can deform the Hankel
contour into a small circle centered around λ and apply Cauchy’s theorem. For the
remaining cases α ∈ Q \ Z, a change of variables z → λ(1 + z) centers the Hankel
contour at zero and implies that

G(z) −n G(λ(1 + z)) −n Δλ G(λz)
(16) n+1
dz = λ n+1
dz = λ dz
Hλ z H0 (1 + z) 0 (1 + z)n+1
A Beta-integral calculation gives that
∞
z γ−1 Γ(γ)Γ(n + 1 − γ)
n+1
dz =
0 (1 + z) Γ(n + 1)
and therefore

z γ−1 Γ(γ)Γ(n + 1 − γ)
dz = (1 + O((r + )−n ))
0 (1 + z)n+1 Γ(n + 1)
More generally, for a natural number β let us define
∞
z γ−1
(17) Iγ,β (n) = (log z)β dz
0 (1 + z)n+1
Then, we have
Γ(γ)Γ(n + 1 − γ)
(18) Iγ,β (n) = pβ (γ, n)
Γ(n + 1)
where pβ (γ, n) is a polynomial in the variables ψ (k) (n + 1 − γ) and ψ (l) (γ) with
rational coefficients, where ψ(z) = Γ (z)/Γ(z) is the logarithmic derivative of the
Γ-function. For example, we have:
p0 (n) = 1
p1 (n) = −ψ(n + 1 − γ) + ψ(γ)
p2 (n) = ψ(n + 1 − γ)2 + ψ (1) (n + 1 − γ) − 2ψ(γ)ψ(n + 1 − γ) + ψ(γ)2 + ψ (1) (γ)
Compare also with [M1, Eqn.I.4.2] and [M2, Eqn.7.5]. What is important is not
the exact evaluation of Iγ,β (n) given in (18), but the fact that the sequence Iγ,β (n)
is of Nilsson type. This follows from the fact that we have an asymptotic expansion
(see [O]):

Γ(n + 1 − γ) 1 γ2 − γ 3γ 4 − 2γ 3 − 3γ 2 + 2γ
(19) ∼ γ 1+ + + ...
Γ(n + 1) n 2n 24n2
Alternatively, one may show that the sequence Iγ,β (n) is of Nilsson type by a change
of variables z = et − 1 which gives
∞ ∞
z γ−1 β
(log z) dz = e−nt tγ−1 Aγ,β (t)dt
0 (1 + z)n+1 0
where
γ−1 t

β
et − 1 e −1
Aγ,β (t) = log − log t dt
t t
is a function which can be expanded into a polynomial of log t with coefficients
functions which are analytic at t = 0. Expand Aγ,β (t) into power series at t = 0
and interchange summation and integration by applying Watson’s lemma (see [O])
to conclude that Iγ,β (n) is of Nilsson type.
9
Replace Δλ G(λz) by (15) in (16) and interchange summation and integration

by applying Watson’s lemma (see [O]). It follows that
1 ∞

1 G(z) −n 1
dz ∼ λ (log n) β
cα,β,λ,k k
2πi Hλ z n+1 nα n
α,β k=0
Equation (14) cocnludes that (an ) is of Nilsson type. Strictly speaking, the above
analysis works only when (α) > −1. This is a local integrability assumption of
the Beta-integral. The asymptotic expansion (2) remains valid as stated even when
(α) ≤ −1 as follows by first integrating G(z) a sufficient number of times, and
then applying the analysis. This is exactly what was done in [CG, Sec.7] at the
cost of complicating the notation.
Remark 4.3. Since the sequence (cα,β,λ,k ) as a function of k is exponentially

bounded and the asymptotic expansion (19) is Gevrey-1, it follows that the sequence
(cα,β,λ,k ) is Gevrey-1. Moreover, if the sequence (cα,β,λ,k ) lies in a number field K,
then we can write the asymptotic expansion of (an ) in the form (10) where Sα,β,λ
are polynomials (with rational coefficients) of values of logarithmic derivatives of
the Gamma function at rational numbers.
5. G-functions
5.1. G-functions: examples of sequences of Nilsson type. In [Ga2,
Thm.1] it was proven that balanced multisum sequences (which appear in abun-
dance in Enumerative Combinatorics) are sequences of Nilsson type. The proof uses
the theory of G-functions which verifies that the generating series of balanced mul-
tisum sequences satisfies the hypothesis of Corollary 4.1. Let us give the definition
of a balanced multisum sequence, a G-function and an example.
Definition 5.1. (a) A term tn,k in variables (n, k) where k = (k1 , . . . , kr ) is
an expression of the form:

r
J
(20) tn,k = C0n Ciki Aj (n, k)!j
i=1 j=1
where Ci ∈ Q for i = 0, . . . , r, j = ±1 for j = 1, . . . , J, and Aj are integral linear

forms in the variables (n, k) such that for every n ∈ N, the set
(21) supp(tn,• ) := {k ∈ Zr | Aj (n, k) ≥ 0, j = 1, . . . , J}
is finite. We will call a term balanced if in addition it satisfies the balance condition:

J
(22) j Aj = 0.
j=1
(b) A (balanced) multisum sequence (an ) is a sequence of complex numbers of the

form

(23) an = tn,k
k∈supp(tn,• )
where t is a (balanced) term and the sum is over a finite set that depends on t.
154
For example, the following sequence is a balanced multisum

n
2 n + k
3 n + l
(n + k)!3 (n + l)!
(24) an = =
k+l k l k! l!n!2 (k + l)!2 (n − k − l)!2
3
k,l k,l
where the summation is over the set of pairs of integers (k, l) that satisfy 0 ≤ k, l
and k + l ≤ n.
Let us now recall what us a G-function. The latter were introduced by Siegel
in [Si] with motivation being arithmetic problems in elliptic integrals, and tran-
scendence problems in number theory. For further information about G-functions
and their properties, see [An1, An2].

Definition 5.2. We say that series G(z) = ∞ n
n=0 an z is a G-function if
(a) the coefficients an are algebraic numbers,
(b) there exists a constant C > 0 so that for every n ∈ N the absolute value
of every conjugate of an is less than or equal to C n ,
(c) the common denominator of a0 , . . . , an is less than or equal to C n ,
(d) G(z) is holonomic, i.e., it satisfies a linear differential equation with coef-
ficients polynomials in z.
G-functions satisfy the hypothesis of Corollary 4.1; see [An1, An2]. Indeed,
they satisfy a linear differential equation which analytically continues them in the
complex plane. Moreover, the arithmetic hypothesis ensures that the local mon-
odromy is quasi-unipotent. We can now state the main result of [Ga2].
Theorem 5.3. [Ga2] ∞ (a) If (an ) is a balanced multisum sequence, its gener-
ating function G(z) = n=0 an z n is a G-function.
(b) In that case, it follows that (an ) is a sequence of Nilsson type.
The reader may have noticed that we defined the notion of a sequence of Nilsson
type only when lim sup |an |1/n > 0. In case the generating series is a G-function,
the remaining case is taken care by the following lemma.
∞
Lemma 5.4. If G(z) = n=0 an z n is a G-function and lim sup |an |1/n = 0,
then an = 0 for all but finitely many n.
Proof. The assumption implies that G(z) is an entire G-function. Since those
are regular-singular at infinity, it follows that G(z) is a polynomial; see also [An1,
An2]. The result follows.
5.2. Classical spin networks: examples of G-functions. In [GV1, GV2]

it was proven that the evaluation of a quantum spin network at a fixed root of unity
is a balanced multisum sequence, and consequently it is a sequence of Nilsson type.
6. Effective computations
6.1. Exact computations. Proposition 2.1 is a uniqueness statement about
the asymptotics of a sequence of Nilsson type, and Theorem 4.1 is an existence
statement which is not effective. There are two types of effective computations,
exact and numerical. The exact computations use as an input a linear recursion
relation of the sequence. The following proposition is elementary and is discussed
in detail for example in [FS, WZ2].
11
Proposition 6.1. Given a linear recursion relation for a sequence (an ) of

Nilsson type, one can compute exactly λ, α, β and the power series gα,b,λ (x) that
appear in Equation (10).
In particular, a linear recursion relation computes exactly the asymptotics of a

sequence of Nilsson type, up to a finite number of unknown Stokes constants.
To apply Proposition 6.1 one needs to find a linear recursion for a sequence
(an ). This comes from the fundamental theorem of Wilf-Zeilberger which states
that a balanced multisum sequence is holonomic, i.e., satisfies a linear recursion with
coefficients polynomials in n; see [Z, WZ1, PWZ]. The proof of the above theorem
has been computer implemented and works well for single sums and reasonably well
for double sums; see [PWZ, PR1, PR2]. As an example, consider the following
sequence from [GV1, Sec.10]
n!6
4n
(−1)k (k + 1)!
an =
(3n + 1)!2 (k − 3n)!4 (4n − k)!3
k=3n
Using the language of [GV1], (an ) is the evaluation of the tetrahedron spin network
(also known as 6j-symbol) when all edges are equal to n. The command
zb.m
loads the package of [PR2] into Mathematica. The command
teucln, k : n ^ 6 3n 1 ^ 2 1 ^ k k 1 1 4n k ^ 3k 3n ^ 4
defines the summand of the sequence (an ), and the command
Zbteucln, k, k, 3n, 4n, n, 2
computes the following second order linear recursion relation for the sequence (an )
2 2
9 1 n 2 3 n 4 3 n 451 460 n 115 n2 an
3 2 n 319 212 1 427 658 n 2 578 232 n2 2 423 109 n3 1 255 139 n4 340 515 n5 37 835 n6 a1 n
2 2
9 2 n 5 3 n 7 3 n 106 230 n 115 n2 a2 n 0
This linear recursion has two formal power series solutions of the form
√ √ √
1 n −432 ± 31i 2 109847 ∓ 22320i 2 −18649008 ± 4914305i 2
a±,n = Λ ± 1 + + +
n3/2 576n 331776n2 573308928n3
√ √
14721750481 ± 45578388960i 2 −83614134803760 ± 7532932167923i 2
+ +
660451885056n4 380420285792256n5
√
−31784729861796581 ∓ 212040612888146640i 2 1
+ + O
657366253849018368n6 n7
where
√
329 ∓ 460i 2
Λ± = = e∓i6 arccos(1/3)
729
are two complex numbers of absolute value√1. The coefficients of the formal power
series a±,n are in the number field K = Q( −2).
156
6.2. Numerical computations. When a sequence (an ) is given by a multi-

dimensional balanced sum, the computed implemented WZ method may not ter-
minate. In that case, one may develop numerical methods for finding λ, α, β as in
Equation (10). An example of this method is the asymptotics of the evaluation of
the Cube Spin Network that appears in the Appendix of [GV1]. Effective methods
for numerical computations of asymptotics have been developed by several authors,
and have also been studied by Zagier.
6.3. Acknowledgment. The idea of the present paper was conceived during
the New York Conference on Interactions between Hyperbolic Geometry, Quantum
Topology and Number Theory in New York in the summer of 2009. The final writing
of the paper occurred in Oberwolfach in the summer of 2010. The author wishes to
thank the organizers of the New York Conference, A. Champanerkar, O. Dasbach,
E. Kalfagianni, I. Kofman, W. Neumann and N. Stoltzfus, and the organizers of
the Oberwolfach Conference, P. Gunnells, W. Neumann, A. Sikora, D. Zagier, for
their hospitality. In addition, the author wishes to thank E. Croot, D. Zagier, D.
Zeilberger for stimulating conversations and R. van der Veen for a careful reading
of the manuscript.
References
[AH] J.E. Andersen and S.K. Hansen, Asymptotics of the quantum invariants for surgeries
on the figure 8 knot, J. Knot Theory Ramifications 15 (2006) 479–548.
[An1] Y. André, Séries Gevrey de type arithmétique. I. Théorèmes de pureté et de dualité,
Ann. of Math. (2) 151 (2000) 705–740.
[An2] , G-functions and geometry, Aspects of Mathematics, E13 Friedr. Vieweg, 1989.
[CG] O. Costin and S. Garoufalidis, Resurgence of the Kontsevich-Zagier power series, Ann.
Inst. Fourier, in press.
[C] , Asymptotics and Borel summability, Monographs and Surveys in Pure and
Applied Mathematics, 141 CRC Press, 2009.
[FG] D.S. Freed and R.E. Gompf, Computer calculation of Witten’s 3-manifold invariant,
Comm. Math. Phys. 141 (1991) 79–117.
[FS] P. Flajolet and R. Sedgewick, Analytic combinatorics, Cambridge University Press,
2009.
[GL] S. Garoufalidis and T.T.Q. Le, The colored Jones function is q-holonomic, Geom. and
Topology, 9 (2005) 1253–1293.
[Ga1] , Chern-Simons theory, analytic continuation and arithmetic, Acta Math. Viet-
nam. 33 (2008) 335–362.
[Ga2] , G-functions and multisum versus holonomic sequences, Advances in Mathe-
matics 220 (2009) 1945–1955.
[Ga3] , thesis, University of Chicago 1992.
[GM] and M. Mariño, Universality and asymptotics of graph counting problems in
non-orientable surfaces, Journal of Combinatorial Theory A 117 (2010) 715-740.
[GV1] and R. van der Veen, Asymptotics of classical spin networks, preprint 2009
arXiv:0902.3113.
[GV2] and R. van der Veen, Asymptotics of quantum spin networks at a fixed root of
unity, preprint 2010 arXiv:1003.4954.
[GIKM] , A. Its, A. Kapaev and M. Marino, Asymptotics of the instantons of Painlevé
I, preprint 2010 arXiv:1002.3634.
[KT] R.M. Kashaev and O. Tirkkonen, A proof of the volume conjecture on torus knots,
Zap. Nauchn. Sem. S.-Peterburg. Otdel. Mat. Inst. Steklov. (POMI) 269 (2000), Vopr.
Kvant. Teor. Polya i Stat. Fiz. 16, 262–268, 370; translation in J. Math. Sci. 115 (2003),
2033–2036.
[LR] R. Lawrence and L. Rozansky, Witten-Reshetikhin-Turaev invariants of Seifert mani-
folds, Comm. Math. Phys. 205 (1999) 287–314.
13
[M1] B. Malgrange, Introduction aux travaux de J. Écalle, Enseign. Math. 31 (1985) 261–282.
[M2] , Intégrales asymptotiques et monodromie, Ann. Sci. École Norm. Sup. 7 (1974)
405–430.
[O] F. Olver, Asymptotics and special functions, Reprint. AKP Classics. A K Peters, Ltd.,
Wellesley, MA, 1997.
[PR1] P. Paule and A. Riese, A Mathematica q-Analogue of Zeilberger’s Algorithm Based
on an Algebraically Motivated Approach to q-Hypergeometric Telescoping, in Special
Functions, q-Series and Related Topics, Fields Inst. Commun., 14 (1997) 179–210.
[PR2] , qZeil Mathematica software available from RISC www.risc.uni-linz.ac.at
[PWZ] M. Petkovšek, H.S. Wilf and D.Zeilberger, A = B, A.K. Peters, Ltd., Wellesley, MA
1996.
[Ph] F. Pham, La descente des cols par les onglets de Lefschetz, avec vues sur Gauss-Manin,
Differential systems and singularities, Astérisque No. 130 (1985) 11–47.
[Ro] L. Rozansky, Residue formulas for the large k asymptotics of Witten’s invariants of
Seifert manifolds. The case of SU(2), Comm. Math. Phys. 178 (1996) 27–60.
[Sa] C. Sabbah, An explicit stationary phase formula for the local formal Fourier-Laplace
transform, in Singularities I, Contemp. Math., 474 (2008) 309–330.
[Si] C.L. Siegel, Über einige Anwendungen diophantischer Approximationen, Abh. Preuss.
Akad. Wiss. 1 (1929) 1–70. Reprinted in Gesammelte Abhandlungen, vol. 1, no 16 (1966)
209–266.
[St] R.P. Stanley, Enumerative Combinatorics, Volume 1, Cambridge University Press
(1997).
[WZ1] H. Wilf and D. Zeilberger, An algorithmic proof theory for hypergeometric (ordinary
and q) multisum/integral identities, Inventiones Math. 108 (1992) 575–633.
[WZ2] J. Wimp and D. Zeilberger, Resurrecting the asymptotics of linear recurrences, J. Math.
Anal. Appl. 111 (1985) 162–176.
[Wi] E.Witten, Quantum field theory and the Jones polynomial, Commun. Math. Physics.
121 (1989) 360–376.
[Z] D. Zeilberger, A holonomic systems approach to special functions identities, J. Comput.
Appl. Math. 32 (1990) 321–368.
School of Mathematics, Georgia Institute of Technology, Atlanta, GA 30332-0160,

USA, http://www.math.gatech.edu/∼stavros
Volume 541, 2011
From angled triangulations to hyperbolic structures
David Futer and François Guéritaud
Abstract. This survey paper contains an elementary exposition of Casson

and Rivin’s technique for finding the hyperbolic metric on a 3–manifold M with
toroidal boundary. We also survey a number of applications of this technique.
The method involves subdividing M into ideal tetrahedra and solving a
system of gluing equations to find hyperbolic shapes for the tetrahedra. The
gluing equations decompose into a linear and non-linear part. The solutions to
the linear equations form a convex polytope A. The solution to the non-linear
part (unique if it exists) is a critical point of a certain volume functional on
this polytope. The main contribution of this paper is an elementary proof of
Rivin’s theorem that a critical point of the volume functional on A produces
a complete hyperbolic structure on M .
1. Introduction
Around 1980, William Thurston showed that “almost every” 3–manifold M ,
whether closed or bounded, admits a complete hyperbolic metric [31]. When the
boundary of M consists of tori, this metric is unique up to isometry [20, 24].
Thurston introduced a method for finding this unique metric. The idea was to
subdivide the interior of M into ideal tetrahedra (tetrahedra whose vertices are
removed), and then give those tetrahedra hyperbolic shapes that glue up coherently
in M . The shape of a hyperbolic ideal tetrahedron can be completely described by
a single complex number, namely the cross–ratio of its four vertices on the sphere
at infinity. Thurston wrote down a system of gluing equations in those complex
parameters, whose solution corresponds to the complete hyperbolic metric on the
interior of M [30].
The difficulty with Thurston’s approach is that this non-linear system of equa-
tions is very difficult to solve in practice. Even proving the existence of a positively
oriented solution, for a given triangulation, often turns out to be a daunting task.
In the 1990s, Andrew Casson and Igor Rivin discovered a powerful technique
for solving Thurston’s gluing equations. Their main idea (which builds on a result
of Colin de Verdière [3]) was to separate the system into a linear part and a non-
linear part. The linear part of the equations corresponds geometrically to a study
2010 Mathematics Subject Classification. 57M50.

Futer is supported in part by NSF Grant No. DMS–1007221.
Guéritaud is supported in part by the ANR program ETTT (ANR-09-BLAN-0116-01).
0000
c 2011
c (copyright holder)
The Authors
1
159
160
2 DAVID FUTER AND FRANÇOIS GUÉRITAUD
of dihedral angles of the individual tetrahedra. In order for the tetrahedra to fit

together in M , the dihedral angles at each edge of M must sum to 2π; in order
for the tetrahedra to be positively oriented, all the angles must be positive. For
a given triangulation τ , the space of solutions to this linear system of equations
and inequalities is a convex polytope A(τ ), with compact closure A(τ ). A point of
A(τ ), corresponding to an assignment of tetrahedron shapes that satisfies the angle
conditions, is called an angle structure on τ . See Definition 3.1 for details.
An angle structure on τ is weaker than a hyperbolic metric, because the linear
equations that define A(τ ) impose a strictly weaker condition on tetrahedra than
do Thurston’s gluing equations. Nevertheless, Casson proved
Theorem 1.1. Let M be an orientable 3–manifold with boundary consisting
of tori, and let τ be an ideal triangulation of M . If A(τ ) = ∅, then M admits a
complete hyperbolic metric.
In other words, the existence of an angle structure implies the existence of a
hyperbolic structure. Finding this hyperbolic structure requires solving the non-
linear part of the gluing equations. To do this, Casson and Rivin introduced a
volume functional V : A(τ ) → R, which assigns to every angle structure on τ the
sum of the hyperbolic volumes of the tetrahedra in this structure. See Definition 5.2
for a precise definition. As we shall see in Section 5, the function V has a number of
pleasant properties: it is smooth and strictly concave down on A(τ ), and extends
continuously to A(τ ). Casson and Rivin independently proved
Theorem 1.2. Let M be an orientable 3–manifold with boundary consisting of
tori, and let τ be an ideal triangulation of M . Then a point p ∈ A(τ ) corresponds
to a complete hyperbolic metric on the interior of M if and only if p is a critical
point of the functional V : A(τ ) → R.
All told, the Casson–Rivin program can be summarized as follows. Solving
the linear part of Thurston’s gluing equations produces the convex polytope A(τ ).
Solving this linear system is straightforward and algorithmic; by Theorem 1.1,
the existence of a solution implies that a hyperbolic structure also exists. By
Theorem 1.2, solving the non-linear part of the gluing equations amounts to finding
a critical point (necessarily a global maximum) of the functional V : A(τ ) → R. In
practice, the search for this maximum point can be accomplished by gradient–flow
algorithms. It is worth emphasizing that a critical point of V does not always exist:
the maximum of V over the compact closure A(τ ) will occur at a critical point in
A(τ ) if and only all the tetrahedra of τ are positively oriented in the hyperbolic
metric on M .
1.1. Where to find proofs. Proofs of Theorem 1.1 are readily available in
the literature. A nice account appears in Lackenby [13, Corollary 4.6], and a slightly
more general result appears in Futer and Guéritaud [5, Theorem 1.1]. Here is a
brief summary of the argument. By Thurston’s hyperbolization theorem [31], the
existence of a hyperbolic structure is equivalent to the non-existence of essential
spheres, disks, tori, and annuli in M . Any such essential surface S can be moved
into a normal form relative to the triangulation τ ; this means that every component
of intersection between S and a tetrahedron is a disk in one of several combinato-
rial types. The dihedral angles that come with an angle structure permit a natural
FROM ANGLED TRIANGULATIONS TO HYPERBOLIC STRUCTURES 161
3
measure of complexity for normal disks, which mimics the area of hyperbolic poly-
gons. As a result, one can show that every surface S with non-negative Euler
characteristic must have non-positive area, which means it is inessential. Hence M
is hyperbolic.
By contrast, direct proofs of Theorem 1.2 are harder to find. The standard
reference for this result is Rivin [26]. However, the focus of Rivin’s paper is some-
what different: he mainly studies the situation where M is an ideal polyhedron,
subdivided into ideal tetrahedra that meet at a single vertex. The notion of an
angle space A(τ ) and a volume functional V still makes sense in this context, and
Rivin’s proof of the analogous result [26, Lemma 6.12 and Theorem 6.16] extends
(with some effort by the reader) to manifolds with torus boundary.
One reference that contains the above formulation of Theorem 1.2, together
with a direct proof, is Chan’s undergraduate honors thesis [1, Theorem 5.1]. How-
ever, this thesis is not widely available. In addition, Chan’s argument relies on
a certain symplectic pairing introduced by Neumann and Zagier [22], a layer of
complexity that is not actually necessary.
The central goal of this paper is to write down an elementary, self-contained
proof of Theorem 1.2. The argument is organized as follows. In Section 2, we
recall the definition of Thurston’s gluing equations. In Section 3, we give a rigorous
definition of the polytope of angle structures A(τ ) and compute its dimension. In
Section 4, we introduce a natural set of tangent vectors, called leading–trailing
deformations, which span the tangent space Tp A(τ ). In Section 5, we focus on the
volume functional V : A(τ ) → R. We will show that the gluing equations for τ
are in 1–1 correspondence with the leading–trailing deformations; in particular, the
non-linear part of a gluing equation is satisfied if and only if the derivative of V
vanishes along the corresponding deformation. It will follow that the full system of
gluing equations is satisfied at p ∈ A(τ ) if and only p is a critical point of V.
1.2. Extending and applying the method. Section 6, at the end of the
paper, surveys some of the ways in which the Casson–Rivin program has been
generalized and applied. Among the generalizations are versions of Theorem 1.2
for manifolds with polyhedral boundary (Theorem 6.2) and Dehn fillings of cusped
manifolds (Theorem 6.1). Another generalization of the method considers angles
modulo 2π: we discuss this briefly while referring to Luo’s paper in this volume
[14] for a much more thorough treatment.
Among the applications are explicit constructions of the hyperbolic metric on
several families of 3–manifolds, including punctured torus bundles [11], certain link
complements [9], and “generic” Dehn fillings of multi-cusped manifolds [12]. For
each of these families of manifolds, the volume associated to any angle structure
gives useful lower bounds on the hyperbolic volume of the manifold: see Theorem
6.3. For each of these families of manifolds, certain angle inequalities obtained while
proving that V : A(τ ) → R has a critical point also imply that the combinatorially
natural triangulation τ is in fact the geometrically canonical way to subdivide M .
See Section 6.5 for details.
Finally, the behavior of the volume functional V : A(τ ) → R gives a surprising
and short proof of Weil’s local rigidity theorem for hyperbolic metrics [32]. Because
V is strictly concave down (Lemma 5.3), any critical point of V on A(τ ) must be
unique. As a result, there is only one complete metric on M in which all the
162
tetrahedra of τ are positively oriented. This turns out to imply Weil’s local rigidity
theorem, as we shall show in Theorem 6.4.
1.3. A reader’s guide. For a relative novice to this subject, we recommend

the following strategy. Skim over Section 2, as needed, to recall the crucial defini-
tions of gluing equations and holonomy. Read carefully the definitions and opening
discussion of Sections 3 and 4, but skip the proofs in those sections entirely. Then,
proceed directly to Section 5, where the computation of Proposition 5.4 is par-
ticularly important. This computation almost immediately implies the forward
direction of Theorem 1.2. On the other hand, the linear–algebra results in Sections
3 and 4 involve completely different ideas compared to the rest of the paper, and
are only needed for the reverse direction of Theorem 1.2 (complete metric implies
critical point of V) and for various applications in Section 6.
For a practical example of the method, the reader may also study the proof of
[11, Lemma 6.2], which proves Theorem 1.2 in the special case of punctured–torus
bundles.
1.4. Acknowledgements. The proof of Theorem 1.2 was developed while

the two authors were preparing to give a mini-course at Osaka University in Jan-
uary 2006. This argument was then tested and refined in front of three workshop
audiences containing many graduate students: at Osaka University in 2006, at Zhe-
jiang University in 2007, and at Columbia University in 2009. We thank all the
participants of those workshops for their suggestions. We are also grateful to the
organizers of the Columbia workshop, “Interactions Between Hyperbolic Geometry,
Quantum Topology and Number Theory,” for their encouragement to write up this
material for the workshop proceedings. Finally, this paper benefited from a num-
ber of conversations with Marc Culler, Feng Luo, Igor Rivin, Makoto Sakuma, and
Louis Theran.
2. Gluing equations
For the length of this paper, M will typically denote a compact, connected,
orientable 3–manifold whose boundary consists of tori. For example, such a mani-
fold can be obtained from S 3 by removing an open tubular neighborhood of a knot
or link. An ideal triangulation τ is a subdivision of M ∂M into ideal tetrahedra,
glued in pairs along their faces. One way to recover the compact manifold M is
to truncate all the vertices of the tetrahedra: then, the triangles created by the
truncation will fit together to tile ∂M .
As discussed in the Introduction, our eventual goal is to find a complete hy-
perbolic metric on M by gluing together metric tetrahedra. The metric models for
the tetrahedra in M come from ideal tetrahedra in H3 :
Definition 2.1. A hyperbolic ideal tetrahedron T is the convex hull in H3 of
four distinct points on ∂H3 . The four points on ∂H3 are called ideal vertices of T ,
and are not part of the tetrahedron. The tetrahedron T is called degenerate if it
lies in a single plane, and non-degenerate otherwise.
Recall that an isometry of H3 is completely determined by its action on three
points on ∂H3 . Thus we may assume, for concreteness, that three vertices of T
lie at 0, 1, ∞ in the upper half-space model. If T is non-degenerate, we may also
5
β
α γ
γ α
β
z
0 1
Figure 1. Left: the shape of an ideal tetrahedron is defined by

three dihedral angles. Right: the gluing of tetrahedra, seen from
an ideal vertex.
assume that the fourth vertex lies at z ∈ C, with Im(z) > 0. This number z ∈ C
determines T up to isometry.
Suppose we move T by an orientation–preserving isometry of H3 , so that three
vertices again lie at 0, 1, ∞, the fourth vertex again lies in the upper half-plane,
and edge e is mapped to the edge 0∞. Then the fourth vertex of T will be sent to
one of
z−1 1
(2.1) z, z = or z = .
z 1−z
The corresponding number z, z , or z is called the shape parameter of e. Notice
that opposite edges of T have the same shape parameter, and that zz z = −1.
The arguments
arg z > 0, arg z > 0, arg z > 0
represent the dihedral angles of T , or equivalently the angles of the Euclidean
triangle with vertices 0, 1, z. If one of the vertices of T is truncated by a horosphere,
the intersection will be a Euclidean triangle in precisely this similarity class. This
Euclidean triangle of intersection between T and a horosphere is called a boundary
triangle of T . See Figure 1, left.
The tiling of ∂M by boundary triangles of the tetrahedra provides a way to
understand what conditions are required to glue the tetrahedra coherently.
Definition 2.2. Let C be an oriented surface with a specified triangulation
(for example, a boundary torus of M with the tessellation by boundary triangles).
A segment in C is an embedded arc in one triangle, which is disjoint from the
vertices of C, and whose endpoints lie in distinct edges of C. A normal closed
curve σ ⊂ C is an immersed closed curve that is transverse to the edges of C, such
that the intersection between σ and a triangle is a union of segments.
Definition 2.3. Let τ be an ideal triangulation of M , and let C be one torus
component of ∂M . Then τ induces a tessellation of C by boundary triangles. If
each tetrahedron of τ is assigned a hyperbolic shape, then every corner of each
boundary triangle can be labeled with the corresponding shape parameter.
Let σ be an oriented normal closed curve in C. Then every segment of σ in
a triangle of C cuts off a single vertex of a triangle, labeled with a single shape
164
z2 z4 z5 z7
z1 z6 z8
z3
Figure 2. A normal path σ passing through the tessellation of a

torus by boundary triangles.
parameter (see Figure 2). Let z1 , . . . , zk be the sequence of shape parameters cor-
responding to the segments of σ. Then the holonomy of σ is defined to be

k
(2.2) H(σ) = i log(zi ),
i=1
where i = 1 for corners of triangles to the left of σ and i = −1 for corners
of triangles to the right of σ. We always choose the branch of the log where
0 < arg(zi ) < π for Im(zi ) > 0.
Definition 2.4. Thurston’s gluing equations for a triangulation τ require, in
brief, that the holonomy of every curve σ ⊂ ∂M should be trivial. More pre-
cisely, the system consists of edge equations and completeness equations. The edge
equation for an edge e of τ requires that all the shape parameters adjacent to e
satisfy

k
(2.3) log(zi ) = 2πi.
i=1
The completeness equations require that, if ∂M consists of k tori, there should be a
collection of simple closed normal curves σ1 , . . . , σ2k spanning H1 (∂M ), such that
(2.4) H(σj ) = 0 ∀j.
As the name suggests, the completeness equations ensure we obtain a complete
metric.
Proposition 2.5. Let τ be an ideal triangulation of M . Suppose that each
ideal tetrahedron of τ is assigned a non-degenerate hyperbolic shape with positive
imaginary part, as above. If the shape parameters of the tetrahedra satisfy the edge
gluing equations, these metric tetrahedra can be glued together to obtain a (possibly
incomplete) hyperbolic metric on M ∂M . This metric will be complete if and only
if the completeness equations are satisfied.
Proof. Because all hyperbolic ideal triangles are isometric, there is no ob-
struction to gluing the tetrahedra isometrically along the interiors of their faces.
If the edge equation is satisfied for every edge of τ , then this isometric gluing
along faces of τ extends continuously across the edges. Thus we obtain a (possibly
incomplete) hyperbolic metric on M ∂M and a developing map D : M → H3 .

If the metric on M is complete, then D : M → H is an isometry. It is well-
3
known that the deck transformations corresponding to a component of ∂M must be

parabolic isometries of H3 . In particular, if a cusp of M is lifted to ∞ in the upper
7
half-space model of H3 , the parabolic isometries preserving that cusp are Euclidean
translations of C. But a simple closed curve σ ⊂ ∂M realizes a translation of C
precisely when its holonomy is H(σ) = 0. Thus equation (2.4) holds for all cusps
of M .
Conversely, suppose that the completeness equations (2.4) are satisfied for a
basis of H1 (∂M ). Then, for every cusp torus C ⊂ ∂M , the boundary triangles of C
can be developed to tile a horosphere in H3 . In other words, the developing image
of a collar neighborhood of C is a horoball in H3 , which is complete. But if we
remove the collar neighborhoods of every boundary torus in M , the remaining set
is compact, hence also complete. Thus the metric on M is complete.
3. The polytope of angle structures

As we have seen in the last section, the shape of a hyperbolic ideal tetrahedron
is completely determined by its dihedral angles. The notion of an angle structure
is that these dihedral angles should fit together coherently.
Definition 3.1. An angle structure on an ideal triangulation τ is an assign-
ment of an (internal) dihedral angle to each edge of each tetrahedron, such that
opposite edges carry the same dihedral angle, and such that
(1) all angles lie in the range (0, π),
(2) around each ideal vertex of a tetrahedron, the dihedral angles sum to π,
(3) around each edge of M , the dihedral angles sum to 2π.
The set of all angle structures on τ is denoted A(τ ).
Conditions (1) and (2) in the definition above are precisely what is needed to
specify a non-degenerate hyperbolic ideal tetrahedron up to isometry. For con-
creteness, if the three dihedral angles meeting at a vertex of T are labeled α, β, γ
in clockwise order, then the shape parameter corresponding to α is
sin γ iα
(3.1) z(α) = e ,
sin β
by the law of sines. Meanwhile, condition (3) in the definition is nothing other than
the imaginary part of the edge equation (2.3).
The notion of an angle structure can be summarized by saying that the tetra-
hedra of τ carry genuine hyperbolic shapes, but the conditions these shapes must
satisfy are much weaker than the gluing equations. The completeness equations
are discarded entirely1, and the real part of the edge equations is also discarded. If
we attempt to glue the metric tetrahedra coming from an angle structure, we can
encounter shearing singularities at the edges of τ , as in Figure 1.
One may separate the gluing equations of Definition 2.4 into a real part and
an imaginary part. The real part of the equations is non-linear, because the shape
parameters within one tetrahedron are related in a non-linear way in equation (2.1).
On the other hand, the imaginary part of the equations is linear in the dihedral
angles, and the set of angle structures A(τ ) is defined by a system of linear equations
and strict linear inequalities. This leads to
1There is an alternative version of an angle structure, in which the angles must also satisfy
the imaginary part of the completeness equations. This corresponds to taking a linear slice of the
polytope A(τ ). We will not need this version for the main part of the paper — but see Section
6.1 for variations on this theme.
166
Proposition 3.2. Let τ be an ideal triangulation of M , containing n tetrahe-

dra. The set of all ways to assign a real number to each pair of opposite edges of each
tetrahedron is naturally identified with R3n . Then A(τ ) is a convex, finite–sided,
bounded polytope in R3n . If A(τ ) = ∅, its dimension is
dim A(τ ) = |τ | + |∂M |,
where |τ | = n is the number of tetrahedra and |∂M | is the number of boundary tori
in M .
Proof. Condition (1) of Definition 3.1 is a system of strict inequalities that
constrains the coordinates of A(τ ) to the open cube (0, π)3n . Meanwhile, conditions
(2) and (3) impose a system of linear equations, whose solution set is an affine
subspace of R3n . The intersection between this affine subspace and (0, π)3n will be
a bounded, convex, finite–sided polytope.
We claim that the number of edges in the triangulation τ is n, the same as
the tetrahedra. To see this, observe that each tetrahedron gives rise to 4 boundary
triangles that lie in ∂M . Thus ∂M is subdivided into 4n boundary triangles. These
triangles have a total of 12n sides, glued in pairs; thus ∂M has 6n edges. Since
every component of ∂M is a torus,
χ(∂M ) = 0 = 4n − 6n + 2n,
hence there are 2n vertices on ∂M . Since every edge of τ accounts for two vertices
on ∂M , the number of edges in τ is n, as claimed. It will be convenient to think of
the tetrahedra as numbered 1 to n, and the edges as numbered n + 1 to 2n.
Definition 3.1 involves 2n equations and 3n unknowns. This system can be
encoded in a (2n × 3n) matrix A, as follows. Each column of A corresponds to a
pair of opposite edges of one tetrahedron. For 1 ≤ i ≤ n, the entry aij records
whether or not the ith tetrahedron contains the j th edge pair. In other words, aij
will be 1 when 3i − 2 ≤ j ≤ 3i, and 0 otherwise. For n + 1 ≤ i ≤ 2n, the entry aij
records how many edges out of the j th edge pair become identified to the ith edge
of the glued-up manifold. Thus the entries in the bottom half of A can be 0, 1, or
2. With this setup, the system of equations is
A v = [π, . . . , π, 2π, . . . , 2π]T ,
where v ∈ R3n is the vector of angles. Then dim A(τ ) = 3n − rank(A).
Lemma 3.3. rank(A) = 2n − |∂M |.
This result is due to Neumann [21]. Our proof is adapted from Choi [2, Theo-
rem 3.7].
Proof. Since the matrix A has 2n rows, it suffices to show that the row null
space has dimension |∂M |. We will do this by constructing an explicit basis for the
row null space, with basis vectors in 1 − 1 correspondence with the cusps of M .
Let c be a cusp of M . Associated to c, we construct a row vector rc ∈ R2n .
For 1 ≤ i ≤ n, the ith entry of rc is minus the number of ideal vertices that the
ith tetrahedron has at cusp c. For n + 1 ≤ i ≤ 2n, the ith entry of rc is plus the
number of endpoints that the ith edge has at cusp c. Thus rc records incidence,
with tetrahedra counted negatively and edges counted positively.
Claim: For every cusp c, rc A = 0.
9
Let vj be a column of A, and recall that vj corresponds to one pair of opposite

edges in one tetrahedron. Then the dot product rc · vj counts minus the number
of times that one of these edges has an endpoint at c, plus the number of times
that one of these edges is identified to an edge of M that has an endpoint at c.
Naturally the sum is 0.
Claim: The collection of vectors rc is linearly independent.

Suppose that λc rc = 0, and let ei be the ith edge in M . The ith entry of
rc will be 0, except when c is one of the endpoints of ei . Thus, if the edge ei
has endpoints at cusps a and b, and the ith entry of λc rc is 0, we must have
λa + λb = 0.
Now, let Δ be an ideal triangle of τ , whose ideal vertices are at cusps a, b, and
c. The argument above, applied to the three edges of Δ, gives
λa + λb = 0, λa + λc = 0, λb + λc = 0.
The only way that these three equalities can hold simultaneously is if λa = λb =
λc = 0. Thus all coefficients λc must vanish.
Claim: The collection of vectors rc spans the row null space of A.
Let
q = [q1 , . . . , q2n ] be a vector such that q A = 0. Our goal is to show that
q = λk rk . To that end, let Ti , (1 ≤ i ≤ n), be one tetrahedron of τ . Let a, b, c, d
be the cusps of M corresponding to the four ideal vertices of Ti . (Some of these
cusps may coincide.) A pair of letters from the collection {a, b, c, d} determines
an edge of Ti , which is identified to some edge of M . For notational convenience,
suppose that edge ab from tetrahedron Ti is identified to the edge i(ab) in M , and
similarly for the other letters.
Now, let vj be the j th column vector of A. This column of A corresponds to
a pair of opposite edges in one tetrahedron, which will be tetrahedron Ti if and
only if 3i − 2 ≤ j ≤ 3i. After relabeling {a, b, c, d}, we may assume that j = 3i − 2
corresponds to opposite edges i(ab) and i(cd), j = 3i − 1 corresponds to i(ac) and
i(bd), and j = 3i corresponds to i(ad) and i(bc). By the definition of the matrix
A, the only non-zero entries of v3i−2 are in row i (corresponding to the tetrahedron
Ti ) and rows i(ab), i(cd) (corresponding to edges of M ). The analogous statement
holds for j = 3i − 1 and j = 3i.
Since q A = 0, we must have q · vj = 0 for every j. Applying this to 3i − 2 ≤
j ≤ 3i gives
q · v3i−2 = qi + qi(ab) + qi(cd) = 0,
q · v3i−1 = qi + qi(ac) + qi(bd) = 0,
q · v3i = qi + qi(ad) + qi(bc) = 0,
which implies
(3.2) qi(ab) + qi(cd) = qi(ac) + qi(bd) = qi(ad) + qi(bc) = −qi .

We are now ready to find coefficients λk such that q = λk rk . Let Δ be an
ideal triangle in Ti , whose ideal vertices are at cusps a, b, c and whose sides are at
edges i(ab), i(ac), i(bc). Define
(3.3)
qi(ab) +qi(ac) −qi(bc) qi(ab) +qi(bc) −qi(ac) qi(ac) +qi(bc) −qi(ab)
λa = , λb = , λc = .
2 2 2
A priori, this definition depends on the ideal triangle Δ. To check this is well-
defined, let Δ be another ideal triangle of Ti , for example the triangle with ideal
168
vertices at b, c, d. Note that equation (3.2) implies that

qi(cd) − qi(bd) = qi(ac) − qi(ab) .
Thus the definition of λc coming from Δ will be
qi(bc) + qi(cd) − qi(bd) qi(bc) + qi(ac) − qi(ab)
λc = = .
2 2
Therefore, triangles Δ and Δ that belong to the same tetrahedron and share a
vertex at cusp c induce the same definition of λc . Since any pair of ideal triangles
that meet the cusp c are connected by a sequence of tetrahedra, it follows that λc
is well-defined.
It remains to check that this definition of λc satisfies q = λk rk . Recall that
for n + 1 ≤ i ≤ 2n, the ith entry of rk is the number of endpoints that the ith edge
has at cusp k. If the ends of this edge are at cusps a and b, i.e. if this is the edge
we have referred to as i(ab), then the ith entry of λk rk is
(3.4) λa + λb = qi(ab) ,
using equation (3.3). This is exactly the ith entry of q. Similarly, for 1 ≤ i ≤ n, the
ith entry of rk is minus the number of ideal vertices that Ti has at cusp k. Thus, if
th
Ti has ideal
vertices at cusps a, b, c, d (where some of these may coincide), the i
entry of λk rk is
−(λa + λb + λc + λd ) = −qi(ab) − qi(cd) = qi ,

using equations (3.4) and (3.2). Thus q = λk rk , as desired.
This completes the proof that the vectors rc , indexed by the cusps of M , form
a basis for the row null space of A. Therefore, rank(A) = 2n − |∂M |
We conclude that dim A(τ ) = 3n − rank(A) = n + |∂M |, completing the proof
of Proposition 3.2.
4. Leading–trailing deformations
Given a point p of the angle polytope A(τ ), let Tp A(τ ) be the tangent space
to A(τ ) at the point p. There is a particularly convenient choice of spanning vec-
tors for Tp A(τ ). These leading–trailing deformations, which were probably folklore
knowledge to experts for several years, are the main innovation in our direct proof
of Theorem 1.2. In a certain sense, they serve as a more concrete reformulation of
Neumann and Zagier’s symplectic pairing [22]. To the best of the authors’ knowl-
edge, this is the first place where they are described in print.
Definition 4.1. Let C be a cusp torus of M , with a tessellation by boundary
triangles coming from τ . Let σ ⊂ C be an oriented normal closed curve (see
Definition 2.2), consisting of segments σ1 , . . . σk . Every oriented segment σi lies
in a boundary triangle Δi . We define the leading corner of Δi to be the corner
opposite the side where σi enters Δi , and mark it with a +. We define the trailing
corner of Δi to be the corner opposite the side where σi leaves Δi , and mark it
with a
. See Figure 3.
For every oriented segment σi , we define a vector w(σi ) ∈ R3n , where as above
each coordinate of R3n corresponds to one pair of opposite edges in one of the n
tetrahedra. Every corner of Δi corresponds to one such edge pair in a tetrahedron.
The vector w(σi ) will have a 1 in the coordinate corresponding to the leading corner
11

+
+
+
+

+
+
+
+
Figure 3. The leading–trailing deformation along σ increases the

angles marked + and decreases the angles marked
.
of Δi , a −1 in the coordinate corresponding to the trailing corner of Δi , and 0’s

otherwise.
Finally, the leading-trailing deformation corresponding to σ is the vector w(σ) =
i w(σi ).
Example 4.2. Let e be an edge of M , and suppose for simplicity that all
tetrahedra adjacent to e are distinct. Let σ be a simple closed curve on ∂M , running
counterclockwise about one endpoint of e. The boundary triangles intersected by σ
have angles αi , βi , γi , labeled clockwise with αi inside σ. Then the leading–trailing
deformation w(σ) adds 1 to every βi and subtracts 1 from every γi , keeping the
angle sum in each tetrahedron equal to π. In addition, the dihedral angle marked
βi is adjacent to the same edge of M as the dihedral angle marked γi+1 . Thus the
angle sum at each edge is unchanged. (See Figure 4.) Observe that w(σ) has no
effect at all on the dihedral angles αi adjacent to e.
+

α2 α1 +
e α3 α5
α3 α4 α4

+ +

+
+
+
Figure 4. The leading–trailing deformation about a single edge e

of M . Every increase to an angle in a tetrahedron is canceled out
by a decrease to an adjacent angle.
In the more general setting, we will prove in Lemma 4.5 that leading–trailing
deformations are always tangent to A(τ ). To do that, we need a better understand-
ing of the interaction between different deformations.
Definition 4.3. Let τ be an ideal triangulation of M , and let ρ, σ be oriented
normal closed curves on ∂M that intersect transversely (if at all). Define the signed
intersection number ι(ρ, σ) to be the number of times that σ crosses ρ from right to
170
σ σ
⇒
σ

σ σ σ
Figure 5. Decomposing σ into smaller curves. Left: when two

segments of σ pass through the same boundary triangle, we may
cut and rejoin to form two normal curves σ , σ . Right: a local
isotopy in the complement of the vertices may be required to ensure
that σ , σ are again normal curves.
left, minus the number of times that σ crosses ρ from left to right. This definition
has a few immediate properties:
• It is anti-symmetric: ι(ρ, σ) = −ι(σ, ρ).
• It depends only on the homology classes of σ and ρ in H1 (∂M ).
• By considering a transverse pushoff of σ, one obtains ι(σ, σ) = 0.
Lemma 4.4. Let ρ, σ be oriented normal closed curves on ∂M that intersect
transversely, if at all. Then
∂
Im(H(ρ)) = 2 ι(ρ, σ).
∂w(σ)
Recall that Im(H(ρ)) is the linear, angled part of the holonomy in Definition
2.3.
Proof. The proof involves three steps.
Step 1 introduces several simplifying assumptions with no loss of generality.
First, it will help to assume that every tetrahedron T is embedded in M (that
is, T does not meet itself along an edge or face). This assumption can always be
met by passing to a finite–sheeted cover of M , since π1 (M ) is residually finite by
Selberg’s lemma. Note that the tetrahedra, angles, boundary curves, holonomies,
and leading–trailing deformations all lift naturally to covers. This assumption also
implies that boundary triangles Δi , Δj meet in either at most one edge or at most
one vertex.
Next, no generality is lost by assuming that σ passes through each boundary
triangle at most once. For, if two segments σi , σj run through the same boundary
triangle Δ, we may cut and rejoin the curve σ into a pair of shorter normal closed
curves σ , σ . See Figure 5. (To ensure that σ and σ are normal curves, it may
be necessary to move them by isotopy in the complement of the vertices, as in the
right panel of the figure.) Note that this operation is natural and topologically
additive: we have w(σ) = w(σ ) + w(σ ), and ι(ρ, σ) = ι(ρ, σ ) + ι(ρ, σ ). Thus, if
13
we prove the lemma for each of σ and σ , the result will follow for σ. In particular,
we may now assume that σ is embedded.
Similarly, we may assume that ρ also passes through each boundary triangle
at most once. For, if ρi , ρj run through the same boundary triangle Δ, we may
decompose ρ into a pair of curves ρ , ρ just as above. This natural operation
ensures H(ρ) = H(ρ ) + H(ρ ), and ι(ρ, σ) = ι(ρ , σ) + ι(ρ , σ). Thus it suffices to
consider each of ρ and ρ in place of ρ.
Given these simplifying assumptions, suppose that the segments σ1 , . . . , σk of
σ are contained in distinct boundary triangles Δ1 , . . . , Δk . A priori, there are two
ways in which w(σ) could affect the angular component of H(ρ):
(1) A segment of ρ lies in the same boundary triangle Δi that contains σi .
Then, w(σi ) changes the angles of Δi , thereby changing Im(H(ρ)).
(2) A segment of ρ lies in a boundary triangle Δi , where Δi = Δi are two
truncated vertices of the same ideal tetrahedron. By changing dihedral
angles in the ambient tetrahedron, w(σi ) also changes the angles of Δi .
We consider the effect of (2) in Step 2 and the effect of (1) in Step 3.
Step 2 is to show that the effect of w(σ) in the boundary triangles “on the
other side” of the ambient tetrahedra, as in (2) above, will never change Im(H(ρ)).
To see why this is true, it helps to group together boundary triangles that share a
common vertex v.
Let v be a vertex of ∂M , and let σv = σ1 ∪ . . . ∪ σj be a maximal union of
consecutive segments in σ, such that the ambient boundary triangles Δ1 . . . , Δj are
all adjacent to v. By our embeddedness assumption, each Δi is adjacent to v in
only one corner. Recall that v is one endpoint of an edge e ⊂ M ; we will investigate
the effect of w(σv ) on boundary triangles Δ1 , . . . , Δj at the other end of edge e.
The arc σv can take one of three forms:
(i) σv = σ, and forms a closed loop about v.
(ii) σv takes a right turn in Δ1 , followed by a sequence of left turns about v,
followed by a right turn in Δj .
(iii) σv takes a left turn in Δ1 , followed by a sequence of right turns about v,
followed by a left turn in Δj .
Scenario (i) is depicted in the right panel of Figure 4. Here, v is in the interior
of a polygon Pv = Δ1 ∪ . . . ∪ Δj . Notice that none of the external angles along ∂Pv
actually change, even though the shapes of constituent triangles are changing. The
same will be true in the boundary polygon Pv = Δ1 ∪ . . . ∪ Δj at the other end of
e. Thus, if ρ passes through Pv , its holonomy will be unaffected by this change.
In Scenario (ii), we once again construct a polygon Pv by gluing Δ1 to Δ2 along
their (unique) shared edge, and so on up to Δj . Let Pv = Δ1 ∪ . . . ∪ Δj be the
polygon formed by the boundary triangles in the same tetrahedra, on the other end
of e. These two polygons are depicted in Figure 6. Notice that although Δi and
Δi are in the same oriented similarity class (both have the same angles αi , βi , γi ,
in clockwise order), the triangles in these similarity classes are rearranged to form
Pv . In particular, w(σv ) does not change any of the external angles along ∂Pv ,
even though the shapes of constituent triangles Δi are changing. Thus, if ρ passes
through Pv , its holonomy will be unaffected.
Scenario (iii) is the mirror image of (ii). Once again, w(σv ) does not change
any of the external angles along ∂Pv .
172
+ α4
α1
v

+

+
α3 α4 + α3 α
α2 2

+
+
α1 Pv
Pv +
Figure 6. Left: the boundary polygon Pv comprised of boundary

triangles Δ1 , . . . , Δj adjacent to v. In each Δi , the smallest angle
is αi . Right: in the boundary polygon Pv = Δ1 ∪ . . . ∪ Δj , at
the other end of the edge that starts at v, none of the total angles
along ∂Pv change.
We conclude that for every vertex v in a boundary triangle visited by σ, the

polygon Pv on the other end of the same edge will have all its external angles
unaffected by w(σv ). Since each segment σi of σ belongs to three different maximal
arcs σv (corresponding to the three distinct vertices of Δi ), we have v w(σv ) =
3w(σ). Thus, since each w(σv ) does not affect the holonomy of ρ “on the other
side” of the ambient tetrahedra, neither does w(σ). This completes Step 2.
Step 3 considers the holonomy of ρ in the same boundary triangles that are
also visited by σ. Consider a maximal consecutive string of segments σ1 , . . . , σj ,
contained in Δ1 , . . . , Δj , such that consecutive segments ρ1 , . . . , ρj run through the
same boundary triangles. Depending on orientations, we can have either ρ1 ⊂ Δ1
or ρ1 ⊂ Δj .
In the special case where Δ1 ∪ . . . ∪ Δj contains all of σ and all of ρ, the two
curves cut off exactly the same corners of boundary triangles (possibly in opposite
cyclic order). Then, since ∂M is orientable, we must have ι(ρ, σ) = 0. In this case,
the angles that go into computing the angular holonomy Im(H(ρ)) are exactly the
angles that are unaffected by w(σ). Thus, in this special case, ∂Im(H(ρ))/∂w(σ) =
0 = 2 ι(ρ, σ).
In the general case, we may assume that σ and ρ do not run in parallel through
Δ1 or through Δj (otherwise, the sequence 1, . . . , j is not maximal). As above, it
helps to construct a polygon P by gluing Δ1 to Δ2 along their (unique) shared
edge, and so on up to Δj . By Definition 4.1, each Δi has one mark of + in the
leading corner relative to σi , and one mark of
in the trailing corner. Altogether,
the polygon P contains 2j markers, with j pluses and j minuses. We consider how
many of these markers are to the left and right of ρ.
If σ enters polygon P to the right of ρ and leaves to the left of ρ, the interior of
P contributes +1 to ι(ρ, σ). Also, the part of P to the left of ρ will have a surplus
of two +’s. Thus w(σ1 ∪ . . . ∪ σj ) increases the angular component of H(ρ) by 2.
See Figure 7.
If σ enters polygon P to the left of ρ and leaves to the right of ρ, the interior of
P contributes −1 to ι(ρ, σ). Also, the part of P to the left of ρ will have a surplus
of two
’s. Thus w(σ1 ∪ . . . ∪ σj ) increases the angular component of H(ρ) by −2.
15
ρ
+
+
σ
+
+
Figure 7. When σ crosses ρ from right to left, the part of the

polygon P to the left of ρ contains an excess of two +’s.
Finally, if σ enters and leaves P on the same side of ρ, the interior of P con-
tributes 0 to ι(ρ, σ). Also, the part of P to the left of ρ will have the same number
of +’s and
’s, hence w(σ1 ∪ . . . ∪ σj ) does not affect the angular component of
H(ρ).
Summing these contributions over the (disjoint) polygons of intersection, we
conclude that ∂Im(H(ρ))/∂w(σ) = 2 ι(ρ, σ), as desired.
We can now show that every leading–trailing deformation w(σ) is tangent to

A(τ ).
Lemma 4.5. Let p ∈ A(τ ) be an angle structure, and let σ be an oriented
normal curve on a cusp of M . Then the vector w(σ) is tangent to A(τ ). In other
words, for all sufficiently small ε > 0, p + εw(σ) ∈ A(τ ).
Proof. Let us check the conditions of Definition 3.1. By condition (1) of the
definition, p lies in the open set (0, π)3n . Thus, for sufficiently small ε, we have
p + εw(σ) ∈ (0, π)3n also.
Next, observe that in Definition 4.1, every vector w(σi ) only affects a single
tetrahedron. In this tetrahedron, εw(σi ) adds ε to the dihedral angle on one (lead-
ing) pair of opposite edges, and adds −ε to the dihedral angle on another (trail-
ing) pair of opposite edges. Thus the deformation of angles coming from w(σi )
keeps theangle sum equal to π in every tetrahedron. Clearly, the same is true for
w(σ) = i w(σi ).
Finally, we check condition (3) of Definition 3.1. Let e be an edge of M , and
let ρ be a normal closed curve on ∂M that encircles one endpoint of e. Because
ρ is homotopically trivial, we have ι(ρ, σ) = 0. Thus, by Lemma 4.4, deforming
the dihedral angles along w(σ) does not change the imaginary part of H(ρ). But
Im(H(ρ)) is nothing other than the sum of dihedral angles about edge e; this sum
stays constant, equal to 2π.
There is a convenient choice of leading–trailing deformations that will span

Tp A(τ ).
Proposition 4.6. For every edge ei of M , where 1 ≤ i ≤ n, choose a normal
closed curve ρi about one endpoint of ei . In addition, if M has k cusps, choose
simple closed normal curves σ1 , . . . , σ2k that will span H1 (∂M ). Then the vectors
w(ρi ) and w(σj ) span the tangent space Tp A(τ ).
174
Notice that the curves ρ1 , . . . , ρn , σ1 , . . . , σ2k are exactly the ones whose holo-
nomy is being considered in Definition 2.4.
Proof. By Proposition 3.2, dim Tp A(τ ) = n + k. Thus it suffices to show that

w(ρi ), w(σj ) span a vector space of this dimension.
First, we claim that the vectors w(σ1 ), . . . , w(σ2k ) are linearly independent
from one another, and from each of the w(ρi ). Suppose, after renumbering, that σ1
and σ2 are homology basis curves on the same torus of ∂M . Then ι(σ1 , σ2 ) = ±1,
while ι(σj , σ2 ) = 0 for every j = 1 (including j = 2). Similarly, since each ρi is
homotopically trivial, ι(ρi , σ2 ) = 0 for every i. Thus, by Lemma 4.4, w(σ1 ) is the
only deformation among the ρi and σj that affects Im(H(σ2 )). Therefore, w(σ1 )
is independent from all the other vectors in the collection. Similarly, each w(σj ) is
independent from all the other vectors in the collection.
To complete the proof, it will suffice to show that w(ρ1 ), . . . , w(ρn ) span a
vector space of dimension n − k. Let B be the (n × 3n) matrix whose ith row is
w(ρi ). Then, we claim that rank(B) = n − k. The proof of this claim is virtually
identical to the proof of Lemma 3.3. For every cusp c, we define a row vector
rc ∈ Rn , whose ith entry is the number of endpoints that the ith edge ei has at cusp
c. (This is the second half of the vector rc from Lemma 3.3.) Then, by the same
argument as in that lemma, one checks that the vectors rc form a basis for the row
null space of B. Therefore, rank(B) = n − k, and the w(ρi ), w(σj ) span a vector
space of dimension (n − k) + 2k, as required.
5. Volume maximization
In this section, we show how volume considerations give a way to turn an angle
structure into a genuine hyperbolic metric on M . To compute the volume of an
ideal hyperbolic tetrahedron, recall the Lobachevsky function L : R → R. Its
definition is
x
(5.1) L(x) = − log |2 sin t| dt.
0
Lemma 5.1. The Lobachevsky function L(x) is well defined and continuous
on R (even though the defining integral is improper), and periodic with period π.
Furthermore, if T is a hyperbolic ideal tetrahedron with dihedral angles α, β, γ, its
volume satisfies
vol(T ) = L(α) + L(β) + L(γ).
Proof. See, for example, Milnor [19].
Following Lemma 5.1, we may define the volume of an angle structure in a

natural way.
Definition 5.2. Let τ be an ideal triangulation of M , containing n tetrahedra.
Let A(τ ) ⊂ R3n be the polytope of angle structures on M . Then we define a volume
functional V : A(τ ) → R, by assigning to a point p = (p1 , . . . , p3n ) the real number
V(p) = L(p1 ) + . . . + L(p3n ).
By Lemma 5.1, V(p) is equal to the sum of the volumes of the hyperbolic tetrahedra
associated to the angle structure p.
17
Lemma 5.3. Let p = (p1 , . . . , p3n ) ∈ A(τ ) be an angle structure on τ , and let
w = (w1 , . . . , w3n ) ∈ Tp A(τ ) be a nonzero tangent vector at p. Then the first two
derivatives of V(p) satisfy
∂V 3n
∂2V
= −wi log sin pi and < 0.
∂w i=1
∂w2
In particular, V is strictly concave down on A(τ ).

Proof. Since the definition of V is linear over the tetrahedra in τ , it suffices
to consider the volume of one tetrahedron. Thus, suppose that a tetrahedron T
has angles p1 , p2 , p3 > 0, which are changing at rates w1 , w2 , w3 . Note that, since
the tangent vector w must preserve the angle sum in each tetrahedron, we have
w1 + w2 + w3 = 0. We may also assume that at least one (hence, at least two) of
the wi are nonzero. Then, by equation (5.1) and Lemma 5.1,
∂vol(T ) 3 3
= −wi log |2 sin pi | = −wi log sin pi ,
∂w i=1 i=1

because all sines are positive and wi log 2 = 0. This completes the computation
of the first derivative.
To compute the second derivative, assume by symmetry that p1 , p2 < π/2.
Differentiating vol(T ) a second time, we get
∂ 2 vol(T )
− = w12 cot p1 + w22 cot p2 + w32 cot p3
∂w2
1 − cot p1 cot p2
= w12 cot p1 + w22 cot p2 + (w1 + w2 )2
cot p1 + cot p2
(w1 + w2 ) + (w1 cot p1 − w2 cot p2 )2
2
=
cot p1 + cot p2
≥ 0.
In fact, the numerator in the next-to-last line must be strictly positive. For the
numerator to be 0, we must have w1 = −w2 , hence cot p1 = − cot p2 , which is im-
possible when p1 , p2 ∈ (0, π/2). Thus ∂ 2 vol(T )/∂w2 < 0, and the volume functional
V is also strictly concave down.
By Lemma 5.3, the only potential critical point of V is a global maximum. As

the next proposition shows, derivatives of V are closely connected to the holonomy
of curves on ∂M .
Proposition 5.4. Let C be a cusp torus of M , with a tessellation by boundary
triangles coming from τ . Let σ ⊂ C be an oriented normal closed curve. Recall the
holonomy H(σ) from Definition 2.3 and the tangent vector w(σ) ∈ Tp A(τ ) from
Definition 4.1.
Then for every point p ∈ A(τ ), we have
∂V
= Re(H(σ)).
∂w(σ)
Proof. Let σ1 , . . . , σk be the segments of σ, with σi contained in boundary
triangle Δi . For each i, label the angles of Δi as αi , βi , γi , in clockwise order, such
176
that αi is the angle cut off by σi . Recall that, in Definition 2.3, we defined

1, if αi is to the left of σi ,
i =
−1, if αi is to the right of σi .
Comparing this with Definition 4.1 and Figure 3, we see that the vector w(σi )
increases angle βi at rate i , and increases γi at rate − i . By Lemma 5.3,
∂V
k
∂V
=
∂w(σ) i=1
∂w(σi )

k
= (− i log sin βi + i log sin γi )
i=1

k
sin γi
= i log
i=1
sin βi

k
= i log |zi | by equation (3.1)
i=1

k
= Re i log zi ,
i=1
as desired.
Recall, from Section 3, that an angle structure on τ corresponds to solving the
imaginary part of the edge gluing equations. By Proposition 5.4, solving the real
part of each edge equation amounts to having vanishing derivative in the direction
of the corresponding deformation. This turns out to be the crucial step in the proof
of Theorem 1.2.
tori, and let τ be an ideal triangulation of M . Then a point p ∈ A(τ ) corresponds
to a complete hyperbolic metric on the interior of M if and only if p is a critical
point of the functional V : A(τ ) → R.
Proof. For one direction of the theorem, suppose that p ∈ A(τ ) is a critical
point of V. This angle structure defines a shape parameter on each tetrahedron
of τ . By Proposition 2.5, proving that these shape parameters give a complete
hyperbolic metric on M amounts to checking the edge and completeness equations
of Definition 2.4.
First, consider the edge equation about an edge e. Note that the imaginary part
of the gluing equation about edge e is automatically satisfied for any angle structure.
To check the real part of the gluing equation, let σ ⊂ ∂M be a normal closed curve
encircling one endpoint of e. Since p is a critical point of V, Proposition 5.4 implies
that Re(H(σ)) = 0, as desired. Thus the edge gluing equations are satisfied.
To check completeness, let C be a boundary torus of M , and let σ1 , σ2 be a pair
of simple closed normal curves that span π1 (C) = H1 (C). Recall that C is tiled by
boundary triangles that truncate the tips of ideal tetrahedra. The angle structure p
gives each of these triangles a Euclidean shape, well-defined up to similarity. If Δ is
a triangle in which σ1 and σ2 intersect, we may place the corners of Δ at 0, 1, z ∈ C,
and develop the other triangles of C from there. Note that, since the edge gluing
19
equations are satisfied, the boundary triangles fit together correctly around every
vertex of C.
Let d1 and d2 be the deck transformations of C corresponding to σ1 and σ2 .
Consider the complex numbers
q = d1 (0), r = d2 (0), s = d1 (r) = d2 (q).
The four points 0, q, r, s form four corners of a fundamental domain for C. By the
definition of holonomy (Definition 2.3) and the solution to the edge equations,
s−q s−r
H(σ1 ) = log , H(σ2 ) = log .
r−0 q−0
By Proposition 5.4, the real part of each of these holonomies is 0. Thus we have
|s − q| = |r| and |s − r| = |q|, hence the fundamental domain of C is a parallelo-
gram. Therefore, H(σ1 ) = H(σ2 ) = 0, and the completeness equations are satisfied
for cusp C.
For the converse implication of the theorem, suppose that p ∈ A(τ ) defines a
complete metric. Then the tetrahedron shapes corresponding to p must satisfy all
the edge and completeness equations of Definition 2.4: we must have H(σ) = 2πi
for every closed curve σ encircling an endpoint of an edge, and H(σj ) = 0 for
a collection of simple closed curves σ1 , . . . , σ2k that form a basis of H1 (∂M ). In
particular, the real part of each of these holonomies is 0. But, by Proposition
4.6, the leading–trailing deformations that correspond to these closed curves span
Tp A(τ ). Thus p is a critical point of V, as desired.
6. Extensions, applications, generalizations

This final section of the paper surveys several ways in which the Casson–Rivin
program has been generalized and extended, as well as several infinite families of
manifolds to which the program has been successfully applied.
Theorem 1.2 concerns manifolds with non-empty boundary that consists of tori.
It is natural to ask whether similar methods can be applied to treat manifolds with
more general boundary, or closed manifolds that have no boundary at all. Indeed,
there has been considerable progress in these areas.
6.1. Closed manifolds via Dehn filling. The most straightforward way to
extend Theorem 1.2 to closed manifolds is via Dehn surgery. If C is a boundary
torus of M , and μ, λ are simple closed normal curves that form a basis for H1 (C),
then M (p/q) is the manifold obtained by attaching a solid torus to C, such that
the boundary of the meridian disk is mapped to pμ + qλ. In terms of the gluing
equations of Definition 2.4, attaching a disk to the closed curve pμ+qλ is equivalent
to solving the holonomy equation
(6.1) p H(μ) + q H(λ) = 2πi.
Just as above, the imaginary part of equation (6.1) is linear in the angles of τ .
Imposing this linear equation corresponds to taking a codimension–1 linear slice of
the angle space A(τ ).
Theorem 6.1. Let M be a manifold with boundary a single torus C, and let
τ be an ideal triangulation of M . Choose a pair (p, q) of relatively prime integers,
and let Ap/q (τ ) ⊂ A(τ ) be the set of all angle structures that satisfy the imaginary
178
part of equation (6.1). Then a critical point of the volume functional V on Ap/q (τ )
yields a complete hyperbolic structure on M (p/q), the p/q Dehn filling of M .
The analogous statement holds for fillings along multiple boundary tori.
The proof follows the same outline as Theorem 1.2; see [1, Theorem 6.2] for
more details.
6.2. Manifolds with polyhedral boundary. There is also an analogue of

Theorem 1.2 for ideal triangulations where not every face of the tetrahedra is glued
to another face. Given such a partial gluing of tetrahedra, one obtains a 3–manifold
N whose boundary is subdivided into ideal triangles. If the tetrahedra carry di-
hedral angles, then every edge along ∂N will also carry a prescribed angle. The
simplest case of this is when N is an ideal polyhedron, but one may also consider
cases where N has more complicated topology.
It is worth asking exactly when a 3–manifold N with polyhedral boundary, and
with a fixed assignment of convex dihedral angles, carries a complete hyperbolic
metric with ideal vertices that realizes those angles. The following combinatorial
condition was suggested by Rivin [25]:
(∗) For every simple closed normal curve σ ⊂ ∂N that bounds a disk in N , the
sum of exterior angles along σ is at least 2π, with equality iff σ encircles
an ideal vertex.
Theorem 6.2. Let N be a 3–manifold with polyhedral boundary and prescribed
(convex) dihedral angles along every edge of ∂N . Suppose that N is irreducible and
atoroidal, and furthermore that N is a 3–ball, or a solid torus, or has incompressible
boundary. Then N carries a complete hyperbolic metric realizing the prescribed
angles if and only if condition (∗) holds. Furthermore, any hyperbolic realization is
unique up to isometry.
The case when N is a polyhedron is due to Rivin [27, 28]; the case of solid tori,
to Guéritaud [8]. In both cases, the argument works by first proving that N has
an ideal triangulation with a non-empty angle space A(τ ), and then proving that
the volume functional V has a critical point in A(τ ). This critical point gives the
hyperbolic realization of N . In the preprint [29], Schlenker gives a more analytic
and general argument in the case of incompressible boundary. Conjecturally, the
special hypotheses of all these papers are not needed: all that should be necessary
is that N is irreducible and contains no incompressible tori, and that the angle
assignments satisfy (∗). See [5, Conjecture 2.4].
6.3. Generalized angle structures. All of the angle structures discussed so

far have involved strictly positive dihedral angles. One natural generalization of
the definitions in Section 3 would be to allow negative angles, or more generally, to
consider angles mod 2π.
A generalized angle structure is an assignment of a real number to every pair
of opposite edges in a tetrahedron, so that equations (2) and (3) of Definition 3.1
are satisfied, but the inequalities are discarded. The set of all such assignments
is denoted GA(τ ). Luo and Tillmann showed that for any ideal triangulation τ of
a manifold with torus boundary, GA(τ ) is always non-empty [18]; in other words,
there is no analogue of Theorem 1.1. Furthermore, they establish a linear–algebraic
duality between angle structures and normal surfaces (see also Rivin [28]). For
21
example, the obstruction to finding a non-empty positive polytope A(τ ) ⊂ GA(τ )

is a certain branched normal surface with non-negative Euler characteristic.
Generalizing further, an S 1 –valued angle structure on a triangulation τ is an
assignment of a real number (mod 2π) to every pair of opposite edges in a tetrahe-
dron, such that
(1) Around each ideal vertex of a tetrahedron, the dihedral angles sum to π
(mod 2π),
(2) Around each edge of M , the dihedral angles sum to 0 (mod 2π).
The set of all S 1 –valued angle structures on τ is denoted SA(τ ).
Just as with real–valued solutions in GA(τ ), Luo showed that the existence
of S 1 –valued solutions is extremely general. For any triangulated closed pseudo–
manifold (i.e., any cell complex obtained by gluing tetrahedra in pairs along all of
their faces, whatever the link of a vertex), he showed that SA(τ ) = ∅, and is a
closed smooth manifold [15, Proposition 2.6].
Even though the existence of S 1 –valued angle structures does not distinguish
the class of hyperbolic manifolds, studying the volume of such a structure can still
yield geometric information. Recall from Lemma 5.1 that the Lobachevsky function
L is π–periodic; as a result, Definition 5.2 of the volume functional V extends in a
natural way to SA(τ ). Because SA(τ ) = ∅ is a compact manifold, V must achieve
a maximum. The work of Luo [15] and Luo–Tillmann [17] uses this maximum
point to either solve a generalized version of the gluing equations (which yields a
representation from a double branched cover of M into P SL(2, C)), or find certain
highly restrictive normal surfaces in M . See Luo’s survey paper in this volume [14]
for more details.
6.4. Volume estimates. Recall that, by Lemma 5.3, the volume functional
V is concave down on A(τ ). As a result, any critical point of V must actually be
the global maximum of the function over the compact closure A(τ ). Thus Theorem
1.2 has the following corollary.
tori, and let τ be an ideal triangulation of M . Suppose that V : A(τ ) → R has a
critical point at p ∈ A(τ ). Then, for any point q ∈ A(τ ),
V(q) ≤ vol(M ),
with equality iff q = p (i.e., iff q gives the complete hyperbolic metric on M ).
In fact, the analogous statement also holds in the settings of Section 6.1 (Dehn
filling) or Section 6.2 (polyhedral boundary). The uniqueness of a critical point
of V turns out to be the key idea in Rivin’s proof of the uniqueness statement of
Theorem 6.2.
Theorem 6.3 allows to compute effective, combinatorial volume estimates for
hyperbolic 3–manifolds. In certain settings, the combinatorics of a 3–manifold nat-
urally guides a choice of triangulation τ , and the same combinatorial data provide
a convenient point q ∈ A(τ ). Then, Theorem 6.3 says that V(q) is a lower bound
on the volume of M . This approach is illustrated by the first family of manifolds
to which the Casson–Rivin method was successfully applied [11].
A decade after Colin de Verdière, Casson, and Rivin developed the theory of
volume maximization in the early 1990s, it was Guéritaud who first applied the
method to find the hyperbolic metrics on an infinite family of manifolds, namely
180
punctured torus bundles [11]. In an appendix to the same paper, Futer extended
the method to two–bridge links [11, Appendix]. For both of these families of
manifolds, the existence of hyperbolic metrics was well-known, but the volume
estimates coming from Theorem 6.3 were both new and sharp. Combined with Dehn
surgery techniques, the volume estimates from [11] also give explicit, combinatorial
bounds for the volume of several families of knot and link complements [6, 7], as
well as of a number of closed manifolds [23].
It is worth asking whether the existence of a critical point of V in A(τ ) is actu-
ally necessary for the volume inequality of Theorem 6.3. Casson conjectured that
this inequality holds for any angled triangulation τ , whether or not the tetrahedra
of this triangulation can be given positively oriented shapes in the hyperbolic met-
ric on M . (Note that, by Theorem 1.1, the hyperbolic metric must exist.) If proved
true, this conjecture would provide a practical tool for finding volume estimates on
many more families of 3–manifolds.
6.5. Canonical triangulations. Nearly everything discussed thus far in this

paper has depended on the choice of triangulation. As it turns out, many of the
methods already discussed can show that a particular triangulation is geometrically
canonical for M .
Given a hyperbolic 3–manifold M with k cusps, let H1 , . . . , Hk be disjoint
horospherical neighborhoods of the cusps. Then the Ford–Voronoi domain F is the
set of all points in M that have a unique shortest path to the union of the Hi . This
is an open set in M , whose complement L = M F is a compact 2–complex, called
the cut locus. The dual to L is an ideal polyhedral decomposition P of M ; the n–
cells of P are in bijective correspondence with the (3 − n)–cells of L. This is called
the canonical polyhedral decomposition of M , relative to the cusp neighborhoods
Hi .
The combinatorics of L —and therefore, of P— depends only on the relative
volumes of the Hi . In particular, if M has only one cusp, there are no choices what-
soever, and P is completely determined by the hyperbolic metric. If the horoballs
in H3 obtained by lifting the cusp neighborhoods Hi are in “general position,” ev-
ery vertex of L will have exactly four closest horoballs. Thus each vertex of L will
meet four edges, and the dual polyhedral decomposition P will generically be a
triangulation.
If τ is a given triangulation (or, more generally, a given polyhedral decompo-
sition), proving that τ is canonical amounts to verifying finitely many inequalities
about nearest horoballs. Epstein and Penner found a way to translate these inequal-
ities into convexity statements in the Minkowski space R3+1 , where H3 is modeled
by a hyperboloid [4]. More recently, Guéritaud discovered that the convexity in-
equalities that imply canonicity (once translated to inequalities involving dihedral
angles of the polyhedra) can be verified using information obtained in the course
of showing that V : A(τ ) → R has a critical point [9].
To date, the method of angled triangulations has found both the hyperbolic
metric and the canonical polyhedral decomposition of several families of 3–manifolds:
punctured–torus bundles [9, Theorem 1.11.1], two-bridge links [9, Theorem 2.1.7],
and certain special arborescent links [9, Theorem 2.3.1]. Guéritaud used the
same ideas to find the canonical triangulations for convex cores of quasi-Fuchsian
punctured–torus groups [10]. All of these families of manifolds actually have closely
23
related combinatorial features, with the structure of the triangulation effectively de-
termined by the combinatorics of SL(2, Z) and continued fractions.
There is a recent result that brings together several themes from this section.
Suppose that M is a hyperbolic 3–manifold with k cusps, and the canonical de-
composition of M is indeed a triangulation. If we perform Dehn filling along one
of these cusps, the result will (generically) be another hyperbolic manifold M (p/q),
with a new geometry. In [12], Guéritaud and Schleimer use the canonical triangu-
lation of M to completely describe the canonical triangulation of its generic Dehn
fillings. Their argument uses Theorems 6.1 and 6.2 to construct a triangulated solid
torus with the right shape and glue it into M (p/q).
6.6. Weil rigidity. We close this paper with an application whose statement
has nothing to do with triangulations. The following rigidity theorem, due to Weil
[32], precedes Mostow–Prasad rigidity by a dozen years.
Theorem 6.4. Let M be a 3–manifold with boundary consisting of tori. Then
any complete hyperbolic metric on the interior of M is locally rigid: there is no
local deformation of the metric through other complete hyperbolic metrics.
We thank Marc Culler and Feng Luo for a fruitful discussion that produced the
following extremely short proof.
Proof. Suppose that M admits a complete hyperbolic metric. Choose horoball
neighborhoods H1 , . . . , Hk about the cusps of M . Then, as in Section 6.5, this
choice of cusp neighborhoods determines a decomposition P of M into ideal poly-
hedra. Luo, Schleimer, and Tillmann showed that M has a finite–sheeted cover N ,
in which the lift of P decomposes into positively oriented ideal tetrahedra [16]. Let
τ be this positively oriented ideal triangulation of N .
By Theorem 1.2, the complete hyperbolic metric on N (which was obtained by
lifting the metric on M ) represents a critical point of V : A(τ ) → R. By Lemma
5.3, this critical point is unique. But any local deformation of the complete metric
on M would lift to a deformation of the metric on N , which would violate Lemma
5.3.
References
[1] Ken Chan, Constructing hyperbolic 3-manifolds, Undergraduate thesis with Craig Hodgson,
University of Melbourne, 2002.
[2] Young-Eun Choi, Positively oriented ideal triangulations on hyperbolic three-manifolds,
Topology 43 (2004), no. 6, 1345–1371.
[3] Yves Colin de Verdière, Un principe variationnel pour les empilements de cercles, Invent.
Math. 104 (1991), no. 3, 655–669.
[4] David B. A. Epstein and Robert C. Penner, Euclidean decompositions of noncompact hyper-
bolic manifolds, J. Differential Geom. 27 (1988), no. 1, 67–80.
[5] David Futer and François Guéritaud, Angled decompositions of arborescent link complements,
Proc. London Math. Soc. 98 (2009), no. 2, 325–364.
[6] David Futer, Efstratia Kalfagianni, and Jessica S. Purcell, On diagrammatic bounds of knot
volumes and spectral invariants, Geom. Dedicata 147 (2010), 115–130.
[7] , Cusp areas of Farey manifolds and applications to knot theory, Int. Math. Res.
Notices (to appear), doi: 10.1093/imrn/rnq037, arXiv:0808.2716.
[8] François Guéritaud, Deforming ideal solid tori, arXiv:0911.3067.
[9] , Géométrie hyperbolique effective et triangulations idéales canoniques en dimension
3., Ph.D. thesis, Université de Paris XI (Orsay), 2006.
[10] , Triangulated cores of punctured-torus groups, J. Differential Geom. 81 (2009), no. 1,
91–142.
182
[11] François Guéritaud and David Futer (appendix), On canonical triangulations of once-
punctured torus bundles and two-bridge link complements, Geom. Topol. 10 (2006), 1239–
1284.
[12] François Guéritaud and Saul Schleimer, Canonical triangulations of Dehn fillings, Geom.
Topol. 14 (2010), 193–242.
[13] Marc Lackenby, Word hyperbolic Dehn surgery, Invent. Math. 140 (2000), no. 2, 243–282.
[14] Feng Luo, Triangulated 3–manifolds: from Haken’s normal surfaces to Thurston’s algebraic
equations, This volume.
[15] , Volume optimization, normal surfaces and Thurston’s equation on triangulated 3–
manifolds, arXiv:0903.1138.
[16] Feng Luo, Saul Schleimer, and Stephan Tillmann, Geodesic ideal triangulations exist virtu-
ally, Proc. Amer. Math. Soc. 136 (2008), no. 7, 2625–2630.
[17] Feng Luo and Stephan Tillmann, Volume optimization on minimally triangulated 3-
manifolds, Preprint.
[18] , Angle structures and normal surfaces, Trans. Amer. Math. Soc. 360 (2008), 2849–
2866.
[19] John Milnor, Hyperbolic geometry: the first 150 years, Bull. Amer. Math. Soc. (N.S.) 6
(1982), no. 1, 9–24.
[20] George D. Mostow, Strong rigidity of locally symmetric spaces, Princeton University Press,
Princeton, N.J., 1973, Annals of Mathematics Studies, No. 78.
[21] Walter D. Neumann, Combinatorics of triangulations and the Chern-Simons invariant for
hyperbolic 3-manifolds, Topology ’90 (Columbus, OH, 1990), Ohio State Univ. Math. Res.
Inst. Publ., vol. 1, de Gruyter, Berlin, 1992, pp. 243–271.
[22] Walter D. Neumann and Don Zagier, Volumes of hyperbolic three-manifolds, Topology 24
(1985), no. 3, 307–332.
[23] Carlo Petronio and Andrei Vesnin, Two-sided bounds for the complexity of cyclic branched
coverings of two-bridge links, Osaka J. Math. 46 (2009), no. 4, 1077–1095.
[24] Gopal Prasad, Strong rigidity of Q-rank 1 lattices, Invent. Math. 21 (1973), 255–286.
[25] Igor Rivin, On geometry of convex ideal polyhedra in hyperbolic 3-space, Topology 32 (1993),
no. 1, 87–92.
[26] , Euclidean structures on simplicial surfaces and hyperbolic volume, Ann. of Math.
(2) 139 (1994), no. 3, 553–580.
[27] , A characterization of ideal polyhedra in hyperbolic 3-space, Ann. of Math. (2) 143
(1996), no. 1, 51–70.
[28] , Combinatorial optimization in geometry, Adv. in Appl. Math. 31 (2003), no. 1,
242–271.
[29] Jean-Marc Schlenker, Hyperbolic manifolds with polyhedral boundary, arXiv:math/0111136.
[30] William P. Thurston, The geometry and topology of three-manifolds, Princeton Univ. Math.
Dept. Notes, 1979.
[31] , Three-dimensional manifolds, Kleinian groups and hyperbolic geometry, Bull. Amer.
Math. Soc. (N.S.) 6 (1982), no. 3, 357–381.
[32] André Weil, On discrete subgroups of Lie groups, Ann. of Math. (2) 72 (1960), 369–384.
Department of Mathematics, Temple University, Philadelphia, PA 19122, USA

E-mail address: dfuter@temple.edu
Laboratoire Paul Painlevé, CNRS UMR 8524, Université de Lille 1, 59650 Ville-
neuve d’Ascq, France
E-mail address: Francois.Gueritaud@math.univ-lille1.fr
Volume 541, 2011
Triangulated 3-Manifolds: from Haken’s normal surfaces to

Thurston’s algebraic equation
Feng Luo
Abstract. We give a brief summary of some of our work and our joint work
with Stephan Tillmann on solving Thurston’s equation and Haken equation on
triangulated 3-manifolds in this paper. Several conjectures on the existence
of solutions to Thurston’s equation and Haken equation are made. Resolu-
tions of these conjecture will lead to a new proof of the Poincaré conjecture
without using the Ricci flow. We approach these conjectures by a finite dimen-
sional variational principle so that its critical points are related to solutions to
Thurston’s gluing equation and Haken’s normal surface equation. The action
functional is the volume. This is a generalization of an earlier program by
Casson and Rivin for compact 3-manifolds with torus boundary.
1. Introduction
This paper is based on several talks given by the author at the conference “Inter-
actions Between Hyperbolic Geometry, Quantum Topology and Number Theory”
at Columbia University in 2009 and a few more places. The goal of the paper
is to give a quick summary of some of our work [19] and our joint work with
Stephan Tillmann [20], [21] on triangulated 3-manifolds. Our work is an attempt
to connect geometry and topology of compact 3-manifolds from the point of view
of triangulations. We will recall Haken’s normal surface theory, Thurston’s work
on construction of hyperbolic structures, Neumann-Zagier’s work, the notion of an-
gle structures introduced by Casson, Rivin and Lackenby, and the work of several
other people. One important point we would like to emphasize is the role that the
Neumann-Zagier Poisson structure plays in these theories. It is conceivable that
the Neumann-Zagier Poisson structure will play an important role in discretization
and quantization of SL(2,C) Chern-Simons theory in dimension three.
A combination of the recent work of Segerman-Tillmann [32], Futer-Guéritaud
[8], Luo-Tillmann [21] and [19] has prompted us to make several conjectures on
the solutions of Thurston’s equation and Haken’s normal surface equations. The
1991 Mathematics Subject Classification. Primary 57M50; Secondary 52C26 .

Key words and phrases. 3-manifolds, volume, Haken’s equation, Thurston’s equation, normal
surfaces, hyperbolic metrics .
The author was supported in part by an NSF Grant.
c 2011
XXXX
1
183
184
2 FENG LUO
resolution of some of these conjectures will produce a new proof of the Poincaré
conjecture without using the Ricci flow method.
Let us begin with a recall of closed triangulated pseudo 3-manifolds. Take a
disjoint union of tetrahedra. Identify codimension-1 faces of tetrahedra in pairs
by affine homeomorphisms. The quotient space is a triangulated closed pseudo 3-
manifold. (See §2.1 for more details). In particular, closed triangulated 3-manifolds
are closed triangulated pseudo 3-manifolds and ideally triangulated 3-manifolds are
pseudo 3-manifolds with vertices removed. Given a closed triangulated oriented
pseudo 3-manifold, there are linear and algebraic equations associated to the tri-
angulation. Besides the homology theories, the most prominent ones are Haken’s
equation of normal surfaces [11] and Thurston’s algebraic gluing equation for con-
struction of hyperbolic metrics [35] using hyperbolic ideal tetrahedra. Haken’s
theory is topological and studies surfaces in 3-manifolds and Thurston’s equation
is geometric and tries to construct hyperbolic metrics from the triangulation. In
the most general setting, Thurston’s equation tries to find representations of the
fundamental group into P SL(2, C) ([39]). Much work has been done on both nor-
mal surface theory and Thurston’s equation with fantastic consequences in the past
fifty years.
Haken’s normal surface equation is linear. A basis for the solution space was
found recently by Kang-Rubinstein [17]. In particular, there are always solutions
to Haken’s equation with non-zero quadrilateral coordinates. The situation for
solving Thurston’s equation is different. The main problem which motivates our
investigation is the following.
MAIN PROBLEM. Given a closed oriented triangulated pseudo 3-manifold (M, T),
when does there exist a solution to Thurston’s gluing equation?
The most investigated cases in solving Thurston’s equation are associated to
ideal triangulated 3-manifolds with torus boundary so that the complex numbers z
are in the upper-half plane (see for instance [35], [34], [6], [27] and many others).
These solutions are closely related to the hyperbolic structures. However, we intend
to study Thurston’s equation and its solutions in the most general setting of closed
oriented triangulated pseudo 3-manifolds, in particular, on closed triangulated 3-
manifolds. Even though a solution to Thurston’s equation in the general setting
does not necessarily produce a hyperbolic structure, one can still obtain important
information from it. For instance, it was observed in [39] (see also [25], [32]) that
each solution of Thurston’s equation produces a representation of the fundamen-
tal group of the pseudo 3-manifold with vertices of the triangulation removed to
P SL(2, C). A simplified version of a recent theorem of Segerman-Tillmann [32]
states that
Theorem 1.1. (Segerman-Tillmann) If (M, T) is a closed triangulated oriented
3-manifold so that the triangulation supports a solution to Thurston’s equation, then
each edge in T either has two distinct end points or is homotopically essential in
M.
In particular, their theorem says any one-vertex triangulation of a simply con-

nected 3-manifold cannot support a solution to Thurston’s equation. A combina-
tion of theorem 1.1 and a result of [38] gives an interesting solution to the main
problem for closed 3-manifold. Namely, a closed triangulated 3-manifold (M, T)
FROM HAKEN’S NORMAL SURFACES TO THURSTON’S ALGEBRAIC EQUATION 185
3
supports a solution to Thurston’s equation if and only if there exists a representa-

tion ρ : π1 (M ) → P SL(2, C) so that ρ([e]) = 1 for each edge e having the same
end points. The drawback of this solution is that the representation ρ has to be a
priori given.
Our recent work [19] suggests another way to resolve the main problem using
Haken’s normal surface equation. To state the corresponding conjecture, let us
recall that a solution to Haken’s normal surface equation is said to be of 2-quad-type
if it has exactly one or two non-zero quadrilateral coordinates. A cluster of three
2-quad-type solutions to Haken’s equation consists of three 2-quad-type solutions
x1 , x2 and x3 so that there is a tetrahedron containing three distinct quadrilaterals
q1 , q2 , q3 with xi (qi ) = 0 for i = 1, 2, 3. A triangulation of a 3-manifold is called
minimal if it has the smallest number of tetrahedra among all triangulations of the
3-manifold.
The main focus of our investigation will be around the following conjecture.
We thank Ben Burton and Henry Segerman for providing supporting data which
helped us formulating it in the current form.
CONJECTURE 1. (Haken-Thurston Alternative) For any closed irreducible
orientable minimally triangulated 3-manifold (M, T), one of the two holds:
(1) there exists a solution to Thurston’s equation associated to the triangula-
tion, or
(2) there exists a cluster of three 2-quad-type solutions to Haken’s normal
surface equation.
Using a theorem of Futer-Guéritaud, we proved in [19] the following result
which supports conjecture 1.
Theorem 1.2. Suppose (M, T) is a closed triangulated oriented pseudo 3-
manifold. Then either there exists a solution to the generalized Thurston equation
or there exists a cluster of three 2-quad-type solutions to Haken’s normal surface
equation.
In our joint work with Tillmann [21], using Jaco-Rubinstein’s work [21], we
proved the following theorem concerning the topology of 3-manifolds satisfying part
(2) of conjecture 1.
Theorem 1.3. ([21]) Suppose (M, T ) is a minimally triangulated orientable
closed 3-manifold so that there exists a cluster of three 2-quad-type solutions to
Haken’s equation. Then,
(a) M is reducible, or
(b) M is toroidal, or
(c) M is a Seifert fibered space, or
(d) M contains the connected sum #3i=1 RP 2 of three copies of the projective
plane.
Using theorems 1.1 and 1.3, one can deduce the Poincaré conjecture from con-
jecture 1 (without using the Ricci flow) as follows. Suppose M is a simply connected
closed 3-manifold. By the Kneser-Milnor prime decomposition theorem, we may
assume that M is irreducible. Take a minimal triangulation T of M . By the work
of Jaco-Rubinstein on 0-efficient triangulation [14], we may assume that T has
186
4 FENG LUO
only one vertex, i.e., each edge is a loop. By Segerman-Tillmann’s theorem above,
we see that (M, T) cannot support a solution to Thurston’s equation. By conjec-
ture 1, there exists a cluster of three 2-quad-type solutions to Haken’s equation.
By theorem 1.3, the minimality of T and irreducibility of M , we conclude that
M = S3 .
Theorem 1.2 is proved in [19] where we proposed a variational principle associ-
ated to the triangulation to approach conjecture 1. In this approach, 2-quad-type
solutions to Haken’s equation arise naturally from non-smooth maximum points.
We generalize the notion of angle structures introduced by Casson, Lackenby [18]
and Rivin [28] (for ideally triangulated cusped 3-manifolds) to the circle-valued an-
gle structure (or S1 -angle structure or SAS for short) and its volume for any closed
triangulated pseudo 3-manifold. It is essentially proved in [20] and more specifically
in [19] that an SAS exists on any closed triangulated pseudo 3-manifold (M, T).
The space SAS(T) of all circle-valued angle structures on (M, T) is shown to be
a closed smooth manifold. Furthermore, each circle-valued angle structure has a
natural volume defined by the Milnor-Lobachevsky function. The volume defines a
continuous but not necessarily smooth volume function vol on the space SAS(T).
In particular, the volume function vol achieves a maximum point in SAS(T). The
two conclusions in theorem 1.2 correspond to the maximum point being smooth or
not for the volume function.
More details of the results obtained so far and our approaches to resolve the
conjecture 1 will be discussed in sections 4 and 5. We remark that conjecture 1
itself is independent of the angle structures and there are other ways to approach
it.
There are several interesting problems arising from the approach taken here.
For instance, how to relate the critical values of the volume function on SAS(T)
with the Gromov norm of the 3-manifold. The Gromov norm of a closed 3-manifold
is probably the most important topological invariant for 3-manifolds. Yet its com-
putation is not easy. It seems highly likely that for a triangulation without a cluster
of three 2-quad-type solutions to Haken’s equation, the Gromov norm of the mani-
fold (multiplied by the volume of the regular ideal tetrahedron) is among the critical
values of the volume function on SAS(T). In our recent work with Tillmann and
Yang [22], we have solved this problem for closed hyperbolic manifolds. An affir-
mative resolution of this problem for all 3-manifolds may provide insights which
help to resolve the Volume Conjecture for closed 3-manifolds.
Futer and Guéritaud have written a very nice paper [7] on volume and angle
structures which is closely related to the material covered in this paper.
We remark that this is not a survey paper on the subject of triangulations of
3-manifolds. Important work in the field, in particular the work of Jaco-Rubinstein
[14] on efficient triangulations of 3-manifolds, is not discussed in the paper.
The paper is organized as follows. In section 2, we will recall the basic material
on triangulations and Haken’s normal surface theory. In section 3, we discuss
Neumann-Zagier’s Poisson structures and Thurston’s gluing equation. In section
4, we discuss circle valued angle structures, their volume, some of our work and a
theorem of Futer-Guéritaud. In section 5, we introduce a Z2 version of Thurston’s
equation (Z2 -taut structure).
5
ACKNOWLEDGEMENT. The work is supported in part by the NSF. We

thank the editors of the conference proceedings for inviting us to write the paper
and S. Tillmann and the referee for suggestions on improving the writing of this
paper. We would like to thank in particular David Futer and Francois Guéritaud
for allowing us to present their unpublished theorem. The proof of this theorem
was also supplied by them.
2. Triangulations and normal surfaces

The normal surface theory, developed by Haken in the 1950’s, is a beautiful
chapter in 3-manifold topology. In the late 1970’s, Thurston introduced the notion
of spun normal surfaces and used it to study 3-manifolds.
We will revisit the normal surface theory and follow the expositions in [12] and
[33] closely in this section. Some of the notations used in this section are new.
The work of Tollefson, Kang-Rubinstein, Tillmann, and Jaco on characterizing the
quadrilateral coordinates of normal surfaces will be discussed.
2.1. Some useful facts about tetrahedra. The following lemma will be
used frequently in the sequel. The proof is very simple and will be omitted. To
start, suppose σ = [v1 , .., v4 ] is a tetrahedron with vertices v1 , ..., v4 and edges
eij = {vi , vj }, i = j. We call ekl the opposite edge of eij if {i, j, k, l} = {1, 2, 3, 4}.
Lemma 2.1. Given a tetrahedron σ, assign to each edge eij a real number
aij ∈ R, called the weight of eij . Assume {i, j, k, l} = {1, 2, 3, 4}.
(a) If the sum of weights of opposite edges is a constant, i.e., aij + akl is
independent of indices, then there exist real numbers b1 , .., b4 (weights at vertices)
so that
aij = bi + bj .
(b) If the sum of weights of the edges from each vertex is a constant, i.e.,
aij + aik + ail is independent of indices, then weights of opposite edges are the
same, i.e.,
aij = akl .
(c) If the tetrahedron σ is oriented and edges are labelled by a, b, c so that
opposite edges are labelled by the same letter (see figure 1(a)), then the cyclic order
a → b → c → a is independent of the choice of the vertices and depends only on the
orientation of σ.
2.2. Triangulated closed pseudo 3-manifolds and Haken’s normal sur-

face equation. Let X be a union of finitely many disjoint oriented Euclidean
tetrahedra. The collection of all faces of tetrahedra in X is a simplicial complex
T∗ which is a triangulation of X. Identify codimension-1 faces in X in pairs by
affine orientation-reversing homeomorphisms. The quotient space M is a closed
oriented pseudo 3-manifold with a triangulation T whose simplices are the quo-
tients of simplices in T∗ . Let V, E, F, T (and V ∗ , E ∗ , F ∗ and T ∗ ) be the sets of all
vertices, edges, triangles and tetrahedra in T (in T∗ respectively). The quotient
of a simplex x ∈ T∗ will be denoted by [x] in T. We call x ∈ T∗ the unidentified
simplex and [x] the quotient simplex. Since the sets of tetrahedra in T ∗ and T are
bijective under the quotient map, we will identify a tetrahedron σ ∈ T ∗ with its
quotient [σ], i.e., σ = [σ] and T = T ∗ .
188
6 FENG LUO
If x, y ∈ V ∪ E ∪ F ∪ T (or in T∗ ), we use x > y to denote that y is a face of

x. We use |Y | to denote the cardinality of a set Y .
Note that in this definition of triangulation, we do not assume that simplices
in T are embedded in M . For instance, it may well be that |V | = 1. Furthermore,
the non-manifold points in M are contained in the set of vertices.
According to Haken, a normal surface in a triangulated pseudo 3-manifold
M is an embedded surface S ⊂ M so that for each tetrahedron σ, topologically
the intersection S ∩ σ consists of a collection of planar quadrilaterals and planar
triangles, i.e., inside each tetrahedron, topologically the surface S looks like planes
cutting through the tetrahedron generically. Haken’s theory puts this geometric
observation into an algebraic setting. According to [11], a normal arc in X is an
embedded arc in a triangle face so that its end points are in different edges and a
normal disk in X is an embedded disk in a tetrahedron so that its boundary consists
of 3 or 4 normal arcs. These are called normal triangles and normal quadrilaterals
respectively. A normal isotopy is an isotopy of X leaving each simplex invariant.
Haken’s normal surface theory deals with normal isotopy classes of normal disks
and normal surfaces. For simplicity, we will interchange the use of normal disk with
the normal isotopy class of a normal disk.
Figure 1. t, t are normal triangles and q, q are normal quadrilaterals
The projections of normal arcs and normal disks from X to M constitute normal
arcs and normal disks in the triangulated space (M, T). For each tetrahedron, there
are four normal triangles and three normal quadrilaterals inside it up to normal
isotopy. See figure 1(b). Note that there is a natural one-one correspondence
between normal disks in T ∗ and T . In the sequel, we will not distinguish normal
disks in T or T∗ and we will use , to denote the sets of all normal isotopy
classes of normal triangles and quadrilaterals in the triangulation T and also T∗ .
The set of normal arcs in T∗ and T are denoted by A∗ and A respectively.
There are relationships among the sets V, E, F, T, , , A. These incidence
relations, which will be recalled below, are the basic ingredients for defining Haken’s
and Thurston’s equations.
Take t ∈ , a ∈ A, q ∈ , and σ ∈ T . The following notations will be used. We
use a < t (and a < q) if there exist representatives x ∈ a, y ∈ t (and z ∈ q) so that
x is an edge of y (and z). We use t ⊂ σ and q ⊂ σ to denote that representatives
of t and q are in the tetrahedron σ. In this case, we say the tetrahedron σ contains
t and q.
7
As a convention, we will always use the letters σ, e and q to denote a tetrahe-

dron, an edge and a quadrilateral in the triangulation T respectively.
The normal surface equation is a system of linear equations defined in the space
R × R , introduced by W. Haken [11]. It is defined as follows. For each normal
arc a ∈ A, suppose σ, σ are the two tetrahedra adjacent to the triangular face
which contains a. (Note that σ may be σ .) Then there is a homogeneous linear
equation for x ∈ R × R associated to a:
(2.1) x(t) + x(q) = x(q ) + x(t )

where t, q ⊂ σ, t , q ⊂ σ and t, t , q, q > a. See figure 2(a).
Recall that we identify the set of edges E with the quotient of E ∗ , i.e., E =
{[y]|y ∈ E ∗ } where [y] = {y ∈ E ∗ |y ∼ y }. The index i : E ∗ × → Z is defined
as follows: i(y, q) = 1 if y, q lie in the same tetrahedron σ ∈ T∗ so that y ∩ q = ∅,
= 0 in all other cases. The index i : E × → Z is defined to be
and i(y, q)
i(e, q) = y∈e i(y, q). See figure 2(b) for a picture of i(e, q) = 1, 2. For simplicial
triangulations, i(e, q) = 1 means that the quadrilateral q faces the edge e in a
tetrahedron, i.e., q ∩ e = ∅ and e, q ⊂ σ. In general, i(e, q) ∈ {0, 1, 2}. However, for
simplicial triangulations, i(e, q) = 0, 1.
Figure 2. incident indices
2.3. Normal surfaces and tangential angle structures. Given x ∈ R ×

R , we will call x(t) (t ∈ Δ) and x(q) the t-coordinate and q-coordinate (triangle
and quadrilateral coordinates) of x. Haken’s normal surface equation addresses the
following question. Given a finite set of normal triangles and normal quadrilaterals
in a triangulation T, when can one construct a normal surface with these given
triangles and quadrilaterals as its intersections with the tetrahedra? Haken’s equa-
tion (2.1) is a set of necessary conditions. Spun normal surface theory addresses
the following question, first investigated by Thurston [35]. Suppose we are given a
finite set of quadrilaterals in each tetrahedron. When can one construct a normal
surface whose quadrilateral set is the given one? We can phrase it in terms of the
normal coordinates as follows. Given a vector z ∈ R , when does there exist a
solution to Haken’s equation (2.1) whose projection to R is z? The question was
completely solved in [37], [17], [33] and [13]. We will interpret their results in
terms of angle structures.
DEFINITION. A tangential angle structure on a triangulated pseudo 3-manifold
(M, T) is a vector x ∈ R so that,
190
8 FENG LUO
for each tetrahedron σ ∈ T ,

(2.2) x(q) = 0,
q∈,q⊂σ
and for each edge e ∈ E,

(2.3) i(e, q)x(q) = 0.
q∈
The linear space of all tangential angle structures on (M, T) is denoted by

T AS(T) or T AS.
Recall that a (Euclidean type) angle structure, introduced by Casson, Rivin
[28] and Lackenby [18], is a vector x ∈ R
>0 so that for each tetrahedron σ ∈ T ,

(2.4) x(q) = π,
q∈,q⊂σ
and for each e ∈ E,

(2.5) i(e, q)x(q) = 2π.
q∈
These two conditions (2.4) and (2.5) have very natural geometric meaning.
Suppose a hyperbolic manifold admits a geometric triangulation by ideal hyperbolic
tetrahedra. The first equation (2.4) says that a normal triangle in a hyperbolic ideal
tetrahedron is Euclidean and the second equation (2.5) says that the sum of the
dihedral angles around each edge is 2π. By definition, a tangential angle structure
is a tangent vector to the space of all angle structures.
The following is a result proved by Tollefson (for closed 3-manifolds), Kang-
Rubinstein and Tillmann for all cases. The result was also known to Jaco [13]. Let
Sns be the space of all solutions to Haken’s homogeneous linear equations (2.1).
Given a finite set X, the standard basis of RX will be denoted by X ∗ ={x∗ ∈
R |x ∈ X} so that x∗ (t) = 0 if t ∈ X − {x} and x∗ (x) = 1. We give RX the
X
standard inner product ( , ) so that X ∗ forms an orthonormal basis.

Theorem 2.2. ([37], [17], [33]) For a triangulated closed pseudo 3-manifold
(M, T), let P roj : R × R → R be the projection. Then
(2.6) P roj (Sns ) = T AS(T)⊥ ,
where R has the standard inner product so that {q ∗ |q ∈ } is an orthonormal
basis.
For a short proof of this theorem, see [19]. This result is very important for us
to relate normal surfaces to critical points of the volume function on the space of
all circle-valued angle structures.
3. Neumann-Zagier Poisson structure and Thurston’s gluing equation

The Neumann-Zagier Poisson structure on R , introduced in [25], is of fun-
damental importance for studying triangulated 3-manifolds and in particular for
Thurston’s gluing equation. We will recall its definition and derive some of its
properties in this section. See also [3] and [4] for different proofs.
9
3.1. The Neumann-Zagier Poisson structure. Recall that our triangu-

lated pseudo 3-manifolds (M, T) are oriented so that each tetrahedron has the
induced orientation. Since a pair of opposite edges {e, e } in a tetrahedron σ is
the same as a normal quadrilateral q ⊂ σ with i(e, q) = 0, by lemma 2.1, for each
tetrahedron σ in T, there exists a natural cyclic order on the three quadrilaterals
q1 , q2 , q3 in σ. We denote the cyclic order by q1 → q2 → q3 → q1 , and write q → q
in if q, q are in the same tetrahedron and q → q in the cyclic order. Define
a map w : × → R by w(q, q ) = 1 if q → q , w(q, q ) = −1 if q → q and
w(x, y) = 0 otherwise. The Neumann-Zagier skew symmetric bilinear form, still
denoted by w : R × R → R, is defined to be:

w(x, y) = w(q, q )x(q)y(q ).
q,q ∈
From the definition, it is evident that w(x, y) = −w(y, x).
The following was proved in [25],
Proposition 3.1. (Neumann-Zagier). Suppose (M, T) is a triangulated, ori-
ented closed pseudo 3-manifold.
Then
(a) for any q ∈ , q∈ w(q, q ) = 0,
(b) for any pair of edges e, e ∈ E,

i(e, q)i(e , q )w(q, q ) = 0.
q,q ∈

Let Z be the linear subspace {x ∈ R | for all σ ∈ T , q⊂σ x(q) = 0}. Then
the Neumann-Zagier symplectic 2-form is the restriction of w to Z 2 . It provides an
identification between Z and the dual space Z ∗ .
A simple property of the Neumann-Zagier form is the following identity. For
any q1 , q2 ∈ ,
⎧
⎨ 0, q1 , q2 not in a tetrahedron
(3.1) w(q1 , q)w(q, q2 ) = −2, q1 = q2
⎩
q∈ 1, q1 = q2 and q1 , q2 ⊂ σ
If y ∈ Z, then

(3.2) w(q1 , q)w(q, q2 )y(q2 ) = −3y(q1 ).
q,q2 ∈
Indeed, by (3.1), the left-hand-side of (3.2) is equal to −2y(q1 ) + y(q3 ) + y(q4 )

where q1 , q3 , q4 are three quadrilaterals in a tetrahedron. Since y(q1 ) + y(q3 ) +
y(q4 ) = 0 by definition of Z, equation (3.2) follows. For any q ∈ , the vector
y(q) = w(q, q ) is an element in Z by proposition 3.1(a). Putting this y = y(q2 ) =
w(q2 , q4 ) into identity (3.2), we obtain, for any q1 , q4 ∈ ,

(3.3) w(q1 , q2 )w(q2 , q3 )w(q3 , q4 ) = −3w(q1 , q4 ).
q2 ,q3 ∈
We will identify the dual space (RX )∗ with RX via the standard inner product
( , ) where X ∗ is an orthonormal basis.
For a triangulated pseudo 3-manifold (M, T), define the linear map A : Z →
RE by
192
10 FENG LUO

(3.4) A(x)(e) = i(e, q)x(q).
q
Note that the space of all tangential angle structures T AS is exactly equal to
ker(A).
Lemma 3.2. Suppose (M, T) is oriented. The dual map A∗ : RE → Z, where
the dual spaces of RE and Z are identified with themselves via the standard inner
product (, ) on RE and w on Z, is
1
A∗ (x)(q) = W (e, q)x(e),
3
e∈E
where
W (e, q) = i(e, q )w(q , q).
q ∈
Proof. We need to show for any x ∈ RE and y ∈ Z,

(A(y), x) = w(y, A∗ (x)).
Indeed, the left-hand-side of it is

A(y)(e)x(e) = x(e)i(e, q)y(q).
e e,q
The right-hand-side of it is

y(q )w(q , q )A∗ (x)(q )
q ,q ∈
1
= y(q )w(q , q )W (e, q )x(e)
3
q ,q ,e
1
= y(q )w(q , q )i(e, q)w(q, q )x(e)
3
q ,q ,e,q
1
=− i(e, q)x(e) y(q )w(q , q )w(q , q)
3 e,q
q ,q

= i(e, q)x(e)y(q).
e,q
Here the last equation comes from (3.3). This ends the proof.

Let B : RE → RV be the map B(x)(v) = e>v x(e). If both end points of
e are v, then the edge e is counted twice in the summation e>v x(e). The dual

map B ∗ : RV → RE is given by B ∗ (y)(e) = v<e y(v).
Theorem 3.3. (Neumann-Zagier) For any oriented triangulated closed pseudo
3-manifold (M, T), the sequences
A B
Z−
→ RE −→ RV −
→0
and
B∗ A∗
→ RV −−→ RE −−→ Z
0−
11
are exact. Furthermore, if x, y ∈ RE , then

w(A∗ (x), A∗ (y)) = 0.
Proof. (See also [3].) Since the second sequence is the dual of the first, it suffices
to prove that one of them is exact. First, BA = 0 follows from the definition of Z.
Furthermore, it is easy to see that B ∗ is injective. Indeed, if B ∗ (y) = 0 for some
y ∈ RV , then by definition, y(v) + y(v ) = 0 whenever v, v form the end points of
an edge. Now for any v ∈ V , take a triangle in T with vertices v1 = v, v2 , and v3 .
Then equations y(vi ) + y(vj ) = 0 for i = j in {1, 2, 3} imply that y(vi ) = 0, i.e.,
y(v) = 0. It remains to prove that ker(A∗ ) ⊂ Im(B ∗ ). Suppose x ∈ RE so that
A∗ (x) = 0, i.e., for all q ∈ ,
1
A∗ (x)(q) = W (e, q)x(e) = 0.
3 e
Spelling out the details of the above equation, we see that it is equivalent to
x(e1 ) + x(e2 ) = x(e3 ) + x(e4 )

whenever {e1 , e2 } and {e3 , e4 } are two pairs of opposite edges in a tetrahedron σ in
T. Fix a tetrahedron σ, and consider x(e) as weights on the edges of σ. By lemma
2.1, there exists a map y : {(v, σ)|v < σ, v ∈ V } → R so that

(3.5) x(e) = y(v, σ).
v<e
We claim that the above equation implies that y(v, σ) = y(v, σ ) for any other

tetrahedron
σ > v. Assuming this claim, and taking y(v) = y(v, σ), then we have
x(e) = v<e y(v), i.e., x = B ∗ (y), or x ∈ Im(B ∗ ).
To see the claim, let us first assume that σ and σ share a common triangle face
which has v as a vertex. Say the three vertices of the triangle face are v1 = v, v2 ,
and v3 . Then equation (3.4) says that
y(vi , σ) + y(vj , σ) = y(vi , σ ) + y(vj , σ ),
for i = j in {1, 2, 3}. The common value is xij = x({vi , vj }). This system of three
equations has a unique solution, namely y(vi , σ) = y(vi , σ ) = (xik + xij − xjk )/2
for {i, j, k} = {1, 2, 3}. Now in general, if σ and σ are two tetrahedra in T which
have a common vertex v, by the definition of pseudo 3-manifolds, there exists a
sequence of tetrahedra σ1 = σ, σ2 , ..., σn = σ so that for each index i, σi , σi+1
share a common triangle face which has v as a vertex. Thus, by repeating the same
argument just given, we see that y(v, σ) = y(v, σ ).
To see the last identity,

w(A∗ (x), A∗ (y)) = w(q1 , q2 )A∗ (x)(q1 )A∗ (y)(q2 )
q1 ,q2
1
= w(q1 , q2 )W (e1 , q1 )x(e1 )W (e2 , q2 )y(e2 )
9q
1 ,q2 ,e1 ,e2
1
= w(q1 , q2 )i(e1 , q3 )w(q3 , q1 )i(e2 , q4 )w(q4 , q2 )x(e1 )y(e2 )
9q
1 ,q2 ,q3 ,q4 ,e1 ,e2
194
12 FENG LUO
1
=− i(e1 , q3 )i(e2 , q4 )x(e1 )y(e2 ) w(q3 , q1 )w(q1 , q2 )w(q2 , q4 ).
9 q ,q ,e ,e q ,q
3 4 1 2 1 2
By (3.3) and proposition 3.1(b), the above is equal to

1
= i(e1 , q3 )i(e2 , q4 )x(e1 )y(e2 )w(q3 , q4 ) = 0.
3 q ,q ,e ,e
3 4 1 2
This ends the proof.

3.2. Thurston’s equation. Let us recall briefly Thurston’s gluing equation
[35] on a triangulated closed oriented pseudo 3-manifold (M, T). Assign each edge
in each tetrahedron in the triangulation T a complex number z ∈ C − {0, 1}. The
assignment is said to satisfy the generalized Thurston algebraic equation if
(a) opposite edges of each tetrahedron have the same assignment;
(b) the three complex numbers assigned to three pairs of opposite edges in each
1
tetrahedron are z, 1−z and z−1
z subject to an orientation convention; and
(c) for each edge e in the triangulation, if {z1 , ..., zk } is the set of all complex
numbers assigned to the edge e in the various tetrahedra adjacent to e, then

k
(3.6) zi = ±1.
i=1
If the right-hand-side of (3.6) equals 1 for all edges, we say that the assignment
satisfies Thurston algebraic equation.
Since a pair of opposite edges in a tetrahedron is the same as the normal isotopy
class of a quadrilateral, we see that Thurston’s equation is defined on C . To be
more precise, given z ∈ C , we say z satisfies the generalized Thurston equation,
if the following assertions are satisfied:
(1) if q → q in , then z(q ) = 1−z(q)
1
, and
(2) if e ∈ E, then

(3.7) z(q)i(e,q) = ±1.
q
If the right-hand-side of (3.7) equals 1 for all edges, we say z satisfies Thurston’s
equation.
This equation was introduced by Thurston in [35] in 1978. He used it to
construct the complete hyperbolic metric on the figure-eight knot complement.
Since then, many authors have studied Thurston’s equation. See for instance [25],
[26], [3], [10], [39], [34] and others. This equation was originally defined for ideal
triangulated 3-manifolds with torus boundary, i.e., closed triangulated pseudo 3-
manifolds (M, T) so that each vertex link is a torus. We would like to point out
that Thurston’s equation (3.6) is defined on any closed triangulated oriented pseudo
3-manifold. It was first observed by Yoshida [39], a solution to Thurston’s equation
produces a representation of the fundamental group π1 (M − T(0) ) to P SL(2, C)
where T(0) is the set of all vertices. Thus, in the broader setting, solving Thurston’s
equation amounts to find P SL(2, C) representations of the fundamental group. The
recent work of [16] seems to have rediscovered equation (3.6) independently while
working on TQFT.
13
Let D(T) be the space of all solutions to Thurston’s equation defined in C .

By definition, D(T) is an algebraic set. There are several very nice results known
for D(T). Let H = {w ∈ C| im(w) > 0} be the upper-half-plane.
Theorem 3.4. (Choi [3]). The set D(T) ∩ H is a smooth complex manifold.
Her proof makes an essential use of Neumann-Zagier’s symplectic form (theorem
3.3).
Another result on Thurston’s equation is in the work of Tillmann [34] and
Yoshida [39] relating degenerations of solutions of Thurston’s equation to normal
surface theory. See also the work of [15] and [31]. The geometry behind their
construction was first observed by Thurston [36]. Though this work does not
address conjecture 1 in the introduction, it does indicate a relationship between
Thurston’s equation and Haken’s equation.
Here is Tillmann’s construction. Suppose zn ∈ D(T) is an unbounded sequence
of solutions to Thurston’s equation (3.7) so that for each q ∈ ,
ln |zn (q)|
u(q) = lim
n→∞
1 + q ∈ (ln |zn (q )|)2
exists in R.
Take
the logarithm of equation (3.7) for zn , divide the resulting equation by
1 + q ∈ (ln |zn (q )|)2 , and let n → ∞. We obtain, for each edge e ∈ E,

(3.8) i(e, q)u(q) = 0.
q
By definition, u(q) = 0 unless limn→∞ zn (q) = 0, or ∞. Furthermore, if

limn zn (q) = 1 and q → q → q , then limn zn (q ) = limn 1−z1n (q) = ∞ and
limn zn (q ) = limn znz(q)−1
n (q)
= 0 so that limn zn (q )zn (q ) = −1. This implies that
u(q ) + u(q ) = 0 and u(q ) ≥ 0. Let I = {q ∈ | limn zn (q) = 1} and for q ∈ I, let
aq = u(q ) ≥ 0 where q → q . Then

u= aq w(q, q )(q )∗ ∈ R.
q∈I q
Substitute into (3.8), we obtain for each e ∈ E,

(3.9) v(q)W (e, q) = 0
q∈I

where v = q∈I aq q ∗ . Equation (3.9) appeared in the work of Tollefson [37] in
which he proved that, if (M, T) is a closed 3-manifold, then (3.9) gives a complete
characterization of the quadrilateral coordinates of solutions to Haken’s equation.
Namely, if M is closed, a vector v ∈ R is in P roj (Sns ) if and
only if (3.9) holds
for all e ∈ E. Thus, by Tollefson’s theorem, the specific v = q∈I aq q ∗ belongs to
P roj (Sne ). As a consequence, one has,
Theorem 3.5. (Tillmann) For a closed triangulated 3-manifold (M, T), the
logarithmic limits of D(T) correspond to solutions of Haken’s normal surface equa-
tion.
196
14 FENG LUO
We remark that Tillmann’s theorem in [34] is more general and works for
all pseudo 3-manifolds. We state it in the above form for simplicity. Furthermore,
Tillmann observed in [34] that the solution v has the property that there is at most
one non-zero quadrilateral coordinate in each tetrahedron. Thus if all coefficients aq
are non-negative integers, then the vector v produces an embedded normal surface
in the manifold.
It follows from the definition that for each e ∈ E, the vector

(3.10) ue = w(e, q)q ∗
q
is in T AS(T). What Tollefson proved, using the language of TAS, is that for a
closed triangulated 3-manifold (M, T), the set {ue |e ∈ E} generates the linear
space T AS(T). A generating set for T AS(T) for all closed triangulated pseudo
3-manifolds (M, T) was found in the work of Kang-Rubinstein [17] and Tillmann
[33].
In the recent work [38], Yang is able to construct many solutions of Thurston’s
equation on closed triangulated 3-manifolds (M, T) with the property that each
edge has distinct end points.
4. Circle valued angle structures and maximization of volume

Following Casson, Rivin [28] and Lackenby [18], we introduced the following
notion in [19],
DEFINITION. An S1 -angle structure (or SAS for simplicity) on a closed trian-
is a function x : (T) → S so that
1
gulated pseudo 3-manifold (M, T)
(5) for each tetrahedron σ, q⊂σ x(q) = −1; and

(6) for each edge e ∈ E, q∈ x(q)i(e,q) = 1.
Let SAS(T) be the set of all S1 -angle structures on the triangulation T. If
x ∈ SAS(T) and v ∈ T AS(T), then xeiv , defined by xeiv (q) = x(q)eiv(q) , is still
in SAS(T). We use this to identify the tangent space of SAS(T) with T AS(T).
The Lobachevsky-Milnor volume (or simply the volume) of an S1 -angle structure x
is defined to be:
vol(x) = Λ(arg(x(q)))
q∈

t
where arg(w) is the argument of a complex number w and Λ(t) = − 0 ln |2 sin(s)|ds.
The volume formula is derived from the volume of an ideal hyperbolic tetrahedron.
See Milnor [23]. It is well known that Λ(t) : R → R is a continuous function with
period π. Thus,
vol : SAS(T) → R
is a continuous function. Our goal is to relate the critical points of vol with the
topology and geometry of the 3-manifold.
Using volume maximization to find geometric structures based on angle struc-
tures for manifolds with cusps was introduced by Casson [2] and Rivin [29]. In
a recent work [9], Guéritaud used the tool to prove the existence of hyperbolic
metrics on the once-punctured torus bundle over the circle with Anosov holonomy.
Our approach follows the same path in a more general setting.
15
4.1. Existence of SAS and critical points of volume. In [20], we proved

a general theorem on the existence of real-valued prescribed-curvature angle struc-
tures on a triangulated pseudo 3-manifold. One can check that the proof in [20]
implies the following proposition. Also see [19] for a proof.
Proposition 4.1. ([20], [19]) For a closed triangulated pseudo 3-manifold
(M, T), the space SAS(T) is non-empty and is a smooth closed manifold of dimen-
sion χ(M ) + |T |. In particular, the volume vol : SAS(T) → R has a maximum
point.
Following [20], we give a short proof of it for real-valued angle structures on ide-
ally triangulated 3-manifolds with torus boundary (i.e., closed triangulated pseudo
3-manifolds so that each vertex link is a tours). The main idea of the proof for the
general case is the same.
Suppose otherwise that such a manifold (M, T) does not support a real-valued

linear map h : R → RT × RE so that h(x)(σ) =
T E
angle
structure. Consider the
q⊂σ x(q) and h(x)(e) = q i(e, q)x(q). Let α ∈ R × R be α(σ) = π and
α(e) = 2π. Then the assumption that (M, T) does not support a real-valued angle
structure means α ∈ / h(R ). Therefore, there exists a vector f ∈ RT × RE so that
f is perpendicular to the image h(R ) and (f, α) = 0. This means that
1
(f, α) = f (σ) + 2 f (e) = 0
π σ e
and
(h(x), f ) = (x, h∗ (f )) = 0,
for all x ∈ R , i.e.,
h∗ (f ) = 0,
∗
where h is the transpose
of h.
Since h∗ (f )(q) = σ,q⊂σ f (σ) + e i(e, q)f (e), it follows that if e, e are two
opposite edges in σ, then
f (e) + f (e ) = −f (σ).
In particular, the sum of the values of f at opposite edges in σ is independent of
the choice of the edge pair. By lemma 2.1 (b), we see that there is a map g defined
on the pairs (v, σ) with v < σ so that
f (e) = g(v, σ) + g(v , σ)

where v, v < e. By the same argument as the one we used in the proof of theorem
3.3, we see that g(v, σ) is independent of the choices of tetrahedra, i.e., g : V → R
so that

f (e) = g(v)
v<e
and
f (σ) = − g(v).
v<σ
Now we express
f (σ) + 2 f (e)
σ e
198
16 FENG LUO

=− g(v) + 2 g(v)
σ,v<σ e,v<v

=− g(v)[ 1−2 1]
v∈V σ>v e>v

= g(v)[−|{ triangles in lk(v)}| + 2|{vertices in lk(v)}|] = 0
v∈V
The last equality is due to the fact that the number of vertices of a triangulation
of the torus is equalto half of the
number of triangles in the triangulation. Also,
in the summations σ>v 1 and e>v 1, we count σ and e with multiplicities, i.e.,
if σ (or e) has k vertices which are v, then σ (or e) is counted k times in the sum.
This ends the proof for manifolds with torus boundary.
Proposition 4.1 guarantees that critical points for the volume function always
exist. Here the concept of critical point of the non-smooth function vol has to be
clarified. It can be shown ([19]) that for any point p ∈ SAS(T) and any tangent
itv
vector v of SAS(T) at p, the limit limt→0 vol(pe t)−vol(p) always exists as an element
in [−∞, ∞]. A point p ∈ SAS(T) is called a critical point of the volume if the above
limit is 0 for all tangent vectors v at p.
The main focus of our research is to extract topological and geometric infor-
mation from critical points of the volume function on SAS(T).
Pursuing in this direction, we have proved the following [19].
Theorem 4.2. Let (M, T) be an oriented triangulated closed pseudo 3-manifold.
Suppose x is a critical point of the volume function vol on SAS(T).
(a) If the critical point x is a non-smooth point for the volume function, so
x(q ) = ±1 for some q ∈ , then there exists a solution y to Haken’s normal
surface equation defined on T, which has exactly one or two non-zero quadrilateral
coordinates with y(q ) = 0;
(b) If the critical point x is a smooth point (i.e., x(q) = ±1 for all q), then x
produces a solution to the generalized Thurston equation.
Here are the key steps in the proof of theorem 4.2. Given x ∈ SAS(T), we say
a tetrahedron σ ∈ T is flat with respect to x if x(q) = ±1 for all q ⊂ σ and partially
flat if x(q) = ±1 for one q ⊂ σ. Let U be the set of all partially flat but not flat
tetrahedra and W = {q|x(q) = ±1, q ⊂ σ, σ ∈ U }. By analyzing the derivative of
t
0
ln |2 sin(s)|ds, we obtain the following main identity. For u ∈ T AS(T),
d
(4.1) vol(xeiut ) = − u(q) ln |t| − u(q) ln |u(q)|−
dt
q∈W q,x(q)=±1

u(q) ln | sin(arg(x(q)))| + o(t)
q,x(q)
=±1
Now at a smooth point x, equation (4.1) becomes
d
vol(xeiu )|t=0 = − u(q) ln | sin(arg(x(q)))|.
dt q
By taking u = ue given by (3.10) and assuming x is a smooth critical point, we
obtain a solution z ∈ C to the generalized Thurston equation, where
17
sin(arg(x(q )))
z(q) = x(q),
sin(arg(x(q )))
and q → q → q . This argument was known to Casson [2] and Rivin. One may
find a detailed argument in [19] or [7].
If x is a non-smooth critical point, then we deduce from (4.1) two equations
for all u ∈ T AS(T),

(4.2) u(q) = 0,
q∈W
and

(4.3) u(q) ln |u(q)| = g(u),
q,x(q)=±1
where g(u) is a linear function in u. Now we use the following simple lemma.
Lemma 4.3. Suppose V is a finite dimensional vector space over R and f1 , ..., fn , g
are linear functions on V so that for all x ∈ V,

n
fi (x) ln |fi (x)| = g(x).
i=1
Then for each index i there exists j = i and λij ∈ R so that
fi (x) = λij fj (x).
Using lemma 4.3 for (4.3) where the vector space V is T AS(T) and the linear
functions are u(q) with x(q) = ±1 and g, we conclude that for each q with x(q) =
±1, there exist q1 and λ ∈ R so that u(q) = λu(q1 ) for all u ∈ T AS(T). This
shows that for all u ∈ T AS(T), the inner product (u, q ∗ − λq1∗ ) = 0. By theorem
2.2, q ∗ − λq1∗ is in P roj (Sns ). Thus theorem 4.1 (a) follows.
4.2. Futer-Guéritaud’s Theorem. In an unpublished work [8], David Futer
and Francois Guéritaud proved a very nice theorem concerning the non-smooth
maximum points of the volume function. The proof given below is supplied by
Futer and Guéritaud. We are grateful to Futer and Guéritaud for allowing us to
present their proof in this paper.
Theorem 4.4. (Futer-Guéritaud) Suppose (M, T) is an oriented triangulated
closed pseudo 3-manifold. If x is a non-smooth maximum point of the volume
function on SAS(T), then there exists a non-smooth maximum volume point y ∈
SAS(T) so that all partially flat tetrahedra in y are flat.
Proof(Futer-Guéritaud). Suppose x is a non-smooth maximum volume point in

SAS(T). Let U be the set of all partially flat but not flat tetrahedra in x and
W = {q ⊂ σ|σ ∈ U , x(q) = ±1} as above. Note that, by assumption, for each
tetrahedron σ, there is at most one quadrilateral in W contained in σ.

Claim. Define the vector v = q ∈W q∈ w(q , q)q ∗ , i.e., v(q) = q ∈W w(q , q).
Then v ∈ T AS(T).
200
18 FENG LUO
To see the claim, we must verify two conditions

for v: (1) for tetrahedron σ ∈ T,

v(q) = 0, and (2) for each edge e ∈ E, i(e, q )v(q ) = 0.
q⊂σ
q ∈
The first condition follows from the fact that for any q ∈ , q ∈ w(q , q) = 0.

Indeed, for each tetrahedron σ, q⊂σ v(q) = q ∈W q⊂σ w(q , q) = 0.
To see the second condition, we use (4.2).
By (3.9), for e ∈ E, the vector

ue = W (e, q)q ∗ ,
e
i.e., ue (q) = W (e, q), is in T AS(T). Taking this ue to be the vector u in (4.2), we
obtain

W (e, q) = 0
q∈W
i.e.,

i(e, q )( w(q , q)) = 0.
q ∈ q∈W

The last equation says q i(e, q )v(q ) = 0. This verifies the claim.
Now back to the proof of the theorem. For each point p ∈ SAS(T), let N (p)
be the number of partially flat but not flat tetrahedra in p. For the maximum point
x, we may assume N (x) > 0. We will produce a new maximum point y so that
N (y) < N (x) as follows. Let v be the tangential angle structure constructed in the
claim above. Consider the smooth path
r(t) = xeitv ∈ SAS(T).
Note, by definition, for |t| small, N (r(t)) = N (x). Take |t0 | be the smallest number
so that
N (r(t)) = N (x)
for all |t| < |t0 | and
N (r(t0 )) < N (x).
Furthermore, due to the basic property of the Lobachevsky function that
Λ(a) + Λ(b) + Λ(c) = 0
for a + b + c ∈ πZ and one of a, b, c is in πZ, we have
vol(r(t)) = vol(x)
for |t| ≤ |t0 |. Take y = r(t0 ). Then we have produced a new maximum point y with
smaller N (y). Inductively, we produce a new maximum point so that all partially
flat tetrahedra are flat. This ends the proof.
Combining theorem 4.2 with the theorem of Futer-Guéritaud, we obtain theo-
rem 1.2,
THEOREM 1.2 Suppose (M, T) is a closed triangulated oriented pseudo 3-
manifold. Then there either exists a solution to the generalized Thurston equation
or there exists a cluster of three 2-quad-type solutions to Haken’s normal surface
equation.
19
Indeed, by Futer-Guéritaud’s theorem, we can produce a non-smooth maximum

point y so that there are three distinct quadrilaterals q1 , q2 , q3 in a tetrahedron with
y(qi ) = ±1. Now we use theorem 4.2 to produce the corresponding 2-quad-type
solutions xi , one for each qi with xi (qi ) = 0. Note that we do not assume that
x1 , x2 , x3 are pairwise distinct.
A stronger version of conjecture 1 is the following.
CONJECTURE 2. Suppose (M, T) is a minimally triangulated closed ir-
reducible 3-manifold so that all maximum points of the volume function vol :
SAS(T) → R are smooth for vol. Then Thurston’s equation on T has a solu-
tion.
4.3. Minimal triangulations with a cluster of three 2-quad-type so-

lutions. Our recent joint work with Stephan Tillmann shows the following.
THEOREM 1.3. ([21]) Suppose (M, T ) is a minimally triangulated orientable

closed 3-manifold so that there are three 2-quad-type solutions x1 , x2 , x3 of Haken’s
equation with xi (qi ) = 0 for three distinct quadrilaterals q1 , q2 , q3 inside a tetrahe-
dron. Then,
(a) M is reducible, or
(b) M is toroidal, or
(c) M is a Seifert fibered space, or
(d) M contains the connected sum #3i=1 RP 2 of three copies of the projective
plane.
By the work of W. Thurston and others, it is known, without using the Ricci
flow method, that manifolds in class (d) but not in cases (a), (b), (c) above are either
Haken or hyperbolic. See for instance [24]. Indeed, an irreducible, non-Haken,
atoroidal, non-Seifert-fibered 3-manifold containing #3i=1 RP 2 has a two fold cover
which is a closed 3-manifold of Heegaard genus at most 2. Such a manifold admits
a Z2 action with 1-dimensional fixed point set. By Thurston’s Orbifold theorem
[1], or [5] one concludes that the manifold is hyperbolic.
The proof of theorem 1.3 makes essential uses of Jaco-Rubinstein’s work on
0-efficient triangulations. We analyze carefully the cluster of three 2-quad-type
solutions of Haken’s normal surface equation constructed from theorem 1.2.
Theorem 1.3 takes care of the topology of closed minimally triangulated 3-
manifolds which have non-smooth maximum volume points.
We don’t know if theorem 1.3 can be improved by using only one 2-quad-type
solution instead of a cluster of three 2-quad-type solutions. Such an improvement
will help in reproving the Poincaré conjecture. For instance, one can weaken con-
jecture 1 by replacing the cluster of three 2-quad-type solutions by one 2-quad-type
solution. Another related conjecture is the following,
CONJECTURE 3. Suppose (M, T) is a minimally triangulated closed ori-
entable 3-manifold so that one edge of T has the same end points and is null
homotopic in M . Then there exists a cluster of three 2-quad-type solutions on T.
By theorem 1.3, one sees that conjecture 3 implies the Poincaré conjecture
without using the Ricci flow.
202
20 FENG LUO
5. Some open problems

Another potential approach to conjecture 1 is to use volume optimization on
a space closely related to SAS(T). Let W (T) be the space {z ∈ C | so that if
q → q , then z(q ) = 1/(1 − z(q)) and for each edge e, the right-hand-side of (3.7) is
a positive real number}. The volume function vol : W (T) → R is still defined. The
maximum points of the volume are related to the solutions of Thurston’s equation.
In fact, a critical point of the volume function in the set W (T) ∩ (C − R) gives a
solution to Thurston’s equation.
It is conceivable that the following holds.
CONJECTURE 4. Suppose (M, T) is a closed orientable triangulated 3-manifold
so that W (T) = ∅. Then sup{vol(z)|z ∈ W (T)} ≤ v3 ||M || where v3 is the volume
of the regular ideal hyperbolic tetrahedron.
The first step to carry out this approach is to find conditions on the triangu-
lation T so that W (T) is non-empty. To this end, we consider solving Thurston’s
equation over the real numbers, i.e., z ∈ R . Here is a step toward producing a
real-valued solution to Thurston’s equation.
DEFINITION. Let Z2 be the field of two elements {0, 1}. A Z2 -taut structure
on a triangulated closed pseudo 3-manifold (M, T) is a map f : → {0, 1} so that
(a) if q1 , q2 , q3 are three quadrilaterals in each tetrahedron σ, then exactly one
of f (q1 ), f (q2 ), f (q3 ) is 1, and
(b) for each edge e in T, q∈ i(e, q)f (q) = 0.
The motivation for the definition comes from taut triangulations and real-
valued solutions to Thurston’s equation. Indeed, if z is a real-valued solution to
Thurston’s equation, then there is an associated Z2 -taut structure f defined by:
f (q) = 0 if z(q) > 0 and f (q) = 1 if z(q) < 0. Another motivation comes from
taut triangulations. Suppose T is a taut triangulation,
i.e., there is a map g :
→ {0, π} so that for each tetrahedron σ, q⊂σ g(q) = π and for each edge e,
1
q i(e, q)g(q) = 2π. Then one defines a Z2 -taut structure by f (q) = π g(q). A very
interesting question is to find condition on T so that Z2 -taut structures exist. Is it
possible that the non-existence of Z2 -taut structures implies the existence of some
special solutions to Haken’s normal surface equation?
Tillmann and I observed that the equations for Z2 -taut structures are non-
linear but quadratic in f (q). Indeed, a vector f ∈ Z2 is a Z2 -taut structure if and
only if condition (b) in definition 5.1 holds and for each tetrahedron σ

(5.1) f (q) = 1,
q⊂σ
and
f (q)f (q ) = 0.
q
=q ,q,q ⊂σ
The condition (b) in definition 5.1 and (5.1) should be considered as the defi-
nition of a Z2 -angle structure.
We end the paper with several questions.
QUESTION 1. Given a triangulated pseudo 3-manifold (M, T), when does there
exist a Z2 -taut structure? Can one relate the non-existence of Z2 -taut structure to
some special solutions to Haken’s equation?
21
QUESTION 2. When is a critical point of the volume function of Morse type (i.e.,
when is the Hessian matrix non-degenerated) and when is the volume function a
Morse function?
Let v3 be the volume of the ideal regular hyperbolic tetrahedron.
QUESTION 3. Is the Gromov norm of a closed 3-manifold multiplied by v3 among
the critical values of the volume function?
QUESTION 4. Is it possible to produce a Floer type homology theory associated
to the volume function on SAS(T) which will be a topological invariant of the
3-manifold?
References
[1] Boileau, Michel; Leeb, Bernhard; Porti, Joan; Geometrization of 3-dimensional orbifolds. Ann.
of Math. (2) 162 (2005), no. 1, 195–290.
[2] Casson, Andrew; private communication.
[3] Choi, Young-Eun; Positively oriented ideal triangulations on hyperbolic three-manifolds.
Topology 43 (2004), no. 6, 1345–1371.
[4] Choi, Young-Eun; Neumann and Zagier’s symplectic relations. Expo. Math. 24 (2006), no. 1,
39–51.
[5] Cooper, Daryl; Hodgson, Craig D.; Kerckhoff, Steven P. Three-dimensional orbifolds and cone-
manifolds. With a postface by Sadayoshi Kojima. MSJ Memoirs, 5. Mathematical Society of
Japan, Tokyo, 2000. x+170 pp.
[6] Dunfield, Nathan; Cyclic surgery, degrees of maps of character curves, and volume rigidity for
hyperbolic manifolds. Invent. Math. 136 (1999), no. 3, 623–657.
[7] Futer, D.; Guéritaud, F.; From angled triangulations to hyperbolic structures, in this volume.
[8] Futer, D.; Guéritaud, F., private communication.
[9] Guéritaud, F.; On canonical triangulations of once-punctured torus bundles and two-bridge
link complements. With an appendix by David Futer. Geom. Topol. 10 (2006), 1239–1284
(electronic).
[10] Hodgson, Craig D.; Kerckhoff, Steven P.; The shape of hyperbolic Dehn surgery space. Geom.
Topol. 12 (2008), no. 2, 1033–1090.
[11] Haken, Wolfgang; Theorie der Normalflächen. Acta Math. 105 1961 245–375.
[12] Jaco, William; Tollefson, Jeffrey L.; Algorithms for the complete decomposition of a closed
3-manifold. Illinois J. Math. 39 (1995), no. 3, 358–406.
[13] Jaco, William; private communication, 2008.
[14] Jaco, William; Rubinstein, J. Hyam 0-efficient triangulations of 3-manifolds. J. Differential
Geom. 65 (2003), no. 1, 61–168.
[15] Kabaya, Yuichi; A method to find ideal points from ideal triangulations, arXiv:0706.0971,
preprint 2007.
[16] Kashaev Rinat, Korepanov Igor, Martyushev Evgeniy; A finite-dimensional TQFT for three-
manifolds based on group PSL(2, C) and cross-ratios, arXiv: 0809.4239, 2008.
[17] Kang, Ensil; Rubinstein, J. Hyam; Ideal triangulations of 3-manifolds. I. Spun normal surface
theory. Proceedings of the Casson Fest, 235–265 (electronic), Geom. Topol. Monogr., 7, Geom.
Topol. Publ., Coventry, 2004.
[18] Lackenby, Marc; Word hyperbolic Dehn surgery. Invent. Math. 140 (2000), no. 2, 243–282.
[19] Luo, Feng; Volume optimization, normal surface and Thurston equation, preprint,
arXiv:0903.1138, 2009.
[20] Luo, Feng; Tillmann, Stephan; Angle structures and normal surfaces. Trans. Amer. Math.
Soc. 360 (2008), no. 6, 2849–2866.
[21] Luo, Feng; Tillmann, Stephan; Volume optimization on minimally triangulated 3-manifolds,
preprint 2010.
[22] Luo, Feng; Tillmann, Stephan; Yang, Tian; Thurstons Spinning Construction and Solutions
to the Hyperbolic Gluing Equations for Closed Hyperbolic 3-Manifolds, preprint, 2010.
[23] Milnor, John; Computation of volume, chapter 7 in Thurston’s note, 1978.
[24] Morgan, John W.; The Smith conjecture. (New York, 1979), 3–6, Pure Appl. Math., 112,
Academic Press, Orlando, FL, 1984.
204
22 FENG LUO
[25] Neumann, Walter D.; Zagier, Don; Volumes of hyperbolic three-manifolds. Topology 24
(1985), no. 3, 307–332.
[26] Petronio, Carlo; Porti, Joan; Negatively oriented ideal triangulations and a proof of
Thurston’s hyperbolic Dehn filling theorem. Expo. Math. 18 (2000), no. 1, 1–35.
[27] Petronio, Carlo; Weeks, Jeffrey R.; Partially flat ideal triangulations of cusped hyperbolic
3-manifolds. Osaka J. Math. 37 (2000), no. 2, 453–466.
[28] Rivin, Igor; Combinatorial optimization in geometry. Adv. in Appl. Math. 31 (2003), no. 1,
242–271.
[29] Rivin, Igor; Euclidean structures on simplicial surfaces and hyperbolic volume. Ann. of Math.
(2) 139 (1994), no. 3, 553–580.
[30] Culler, Marc; Shalen, Peter B.; Varieties of group representations and splittings of 3-
manifolds. Ann. of Math. (2) 117 (1983), no. 1, 109–146.
[31] Segerman, Henry; A generalisation of the deformation variety arXiv:0904.1893, preprint,
2009.
[32] Segerman, Henry; Tillmann, Stephan; Pseudo-developing maps for ideal triangulations I, in
preparation.
[33] Tillmann, Stephan; Normal surfaces in topologically finite 3-manifolds. L’Ens. Math. 54
(2008) 329-380.
[34] Tillmann, Stephan; Degenerations of ideal hyperbolic triangulations. math.GT/0508295.
[35] Thurston, William P.; Three-dimensional geometry and topology. 1979-1981, Princeton Uni-
versity Mathematics Dept.
[36] Thurston, William P.; Hyperbolic structures on 3-manifolds. I. Deformation of acylindrical
manifolds. Ann. of Math. (2) 124 (1986), no. 2, 203–246.
[37] Tollefson, Jeffrey L.; Normal surface Q-theory. Pacific J. Math. 183 (1998), no. 2, 359–374.
[38] Yang, Tian; On the Hyperbolic Gluing Equations and Representations of Fundamental
Groups of Closed 3-Manifolds, preprint, 2010.
[39] Yoshida, Tomoyoshi; On ideal points of deformation curves of hyperbolic 3-manifolds with
one cusp. Topology 30 (1991), no. 2, 155–170.
Department of Mathematics
Rutgers University
New Brunswick, NJ 08854, USA
email: fluomath.rutgers.edu
Department of Mathematics, Rutgers University, New Brunswick, NJ 08854, USA
E-mail address: fluo@math.rutgers.edu
Volume 541, 2011
An introduction to fully augmented links
Jessica S. Purcell
Abstract. In this article we summarize information on the class of fully aug-

mented links. These links are geometrically explicit, and therefore provide a
large class of examples of hyperbolic links for which geometric information can
be computed.
1. Introduction
The purpose of this paper is primarily expository, to introduce students and
researchers in 3–manifolds, hyperbolic geometry, knot theory, and related fields to
a class of hyperbolic links which is geometrically explicit, but seems to be under-
utilized.
The class which we study is that of fully augmented links. These are obtained
from diagrams of links in S 3 as follows. Let K be a link with diagram D(K). Regard
D(K) as a 4–valent graph in the plane. A string of bigon regions of the complement
of this graph arranged end to end is called a twist region, as is a vertex adjacent to
no bigons. We assume throughout that the diagram is alternating in a twist region,
else it can be replaced by a diagram with fewer crossings in an obvious way. To
form a fully augmented link, we encircle each twist region with a single unknotted
component, called a crossing circle. The complement of the result is homeomorphic
to the link obtained by removing all full–twists, i.e. pairs of crossings, from each
twist region. See Figure 1. A diagram of the fully augmented link therefore contains
a finite number of crossing circle components, each perpendicular to the projection
plane and encircling exactly two strands of the link. The other link components
are embedded on the projection plane, except possibly for a finite number of single
crossings adjacent to crossing circles. These single crossings we call half–twists.
The original link complement can be obtained from the fully augmented link
by performing (p, 1) Dehn filling on the crossing circles, for an appropriate choice
of p. This process appears, for example, in Rolfsen’s book [15].
Geometric properties of augmented links seem to have been studied first by
Adams [2]. Rather than adding a crossing circle to each twist region, or augmenting,
Adams considered diagrams where some collection of twist regions are augmented.
He showed that if the original link has a nonsplit prime alternating diagram, and
is not a (2, q) torus link, then any augmentation is a hyperbolic link.
1991 Mathematics Subject Classification. Primary 57M25, 57M50.

Supported in part by NSF grants DMS-0704359 and DMS-1007437.
1 2011
205
206
2 JESSICA S. PURCELL
Figure 1. Left to right: A link diagram. The augmented link

diagram. A diagram of the corresponding fully augmented link.
These ideas were used by Lackenby to show that families of alternating knots
admit no exceptional Dehn fillings [8], and to determine information on the volumes
of alternating knots [9]. In an appendix to [9], I. Agol and D. Thurston investigated
the geometry of fully augmented links, using a decomposition of the link comple-
ment into ideal polyhedra to improve Lackenby’s upper bound on volume. As far as
I am aware, this appendix is the first place the polyhedral decomposition appeared
in print.
We describe Agol and Thurston’s decomposition of fully augmented links into
totally geodesic, right angled ideal polyhedra below. Manipulating these polyhedra
allows us to determine geometric information on the links, including bounds on
volume, cusp shape and cusp area. These have been used to bound exceptional
surgeries on knots [7], volumes of knots [6], and cusp shapes [11]. Moreover, the
polyhedral decomposition enabled Chesebro, DeBlois, and Wilton to show that fully
augmented links satisfy the virtually fibered conjecture [3]. Restricting to an even
more geometrically explicit subset of fully augmented links, R. Van der Veen was
able to show that such links satisfy a version of the volume conjecture [18]. Thus
the geometric properties of these links allow us to prove several interesting facts
about the links and their Dehn fillings.
In this paper, we present many of the above geometric results. Most of the
results here are not new, with the possible exception of Proposition 3.8, which does
not seem to appear elsewhere in the literature. Those results that have appeared
before, however, seem to be scattered throughout many different papers. We present
them together in this article to give a (more) complete picture of the geometric
properties of these links.
While I have made every attempt to refer to the appropriate papers where these
results have appeared or have been applied, I may have missed some references that
should belong in this expository paper. I apologize for any such omission.
1.1. Acknowledgements. This paper was written following the workshops

on hyperbolic geometry, quantum topology, and number theory held at Columbia
University in June 2009. We thank the organizers for arranging the workshops,
as well as the NSF for its support of the workshops through an FRG grant. In
addition, work on this paper was partially funded by NSF grant DMS-0704359.
2. Polyhedral decomposition and circle packings

Given a fully augmented link, there is an associated polyhedral decomposi-
tion of the link complement into two isometric, totally geodesic, ideal hyperbolic
polyhedra, described in Agol and Thurston’s appendix to Lackenby’s paper [9].
These polyhedra correspond to a circle packing of S 2 , which in turn corresponds
AN INTRODUCTION TO FULLY AUGMENTED LINKS 207
3
to a triangulation of S 2 . All this information can be read off a diagram of a fully

augmented link. We review it in this section.
To create the polyhedral decomposition, first note that a fully augmented link
contains many geodesic surfaces.
Lemma 2.1. Let L be a hyperbolic fully augmented link, with no half–twists. The
following surfaces are embedded totally geodesic surfaces in the hyperbolic structure
on the link complement.
• Each twice–punctured disk bounded by a crossing circle, punctured by two
link components lying on the projection plane.
• Each component of the projection plane.
Proof. Any essential 2–punctured disk embedded in a hyperbolic 3–manifold

is totally geodesic [1]. We argue that any 2–punctured disk E bounded by a crossing
circle must be essential. For suppose there is a compressing disk D with ∂D ⊂ E.
Since ∂D is an essential closed curve on E, it must encircle one or two punctures
of E. If only one puncture, then the union of D and the disk in S 3 bounded by
∂D (with subset E) forms a sphere in S 3 met by the link exactly once. This is
impossible. Thus ∂D must bound a 2–punctured disk E on E. Then E \ E ∪ D
is a boundary compressing disk for the crossing circle, contradicting the fact that
L is hyperbolic.
Similarly, a boundary compressing disk D for E will have boundary consisting
of two arcs, one of which is an essential arc α on E with endpoints on the same
component of ∂E. For each boundary component of E, there is exactly one pos-
sibility for α with endpoints on that boundary component, up to isotopy. In each
case, α will separate E into two components, at least one containing exactly one
puncture. Let A ⊂ E be the annulus with one boundary component on a puncture
of E, and the other boundary component consisting of two arcs: one arc α, and the
other an arc on another boundary component of E. Now attach D to A along α.
Because D is disjoint from E, this gives an embedded annulus in S 3 \ L. The an-
nulus cannot be essential, since L is hyperbolic. Compressing along a compression
disk or boundary compression disk for the annulus again would give a boundary
compressing disk for S 3 \ L. Thus the annulus must be boundary parallel. Since
one component of ∂A is a meridian, the other must be as well, and the boundary
of D must be isotopic to a closed curve embedded on E encircling both punctures.
As before, this is a contradiction.
As for the projection plane, notice that reflection through the projection plane
preserves the link complement, fixing the plane pointwise. It is a consequence of
Mostow–Prasad rigidity that such a surface must be totally geodesic.
To create the polyhedral decomposition of the link complement S 3 \ L, we cut
along totally geodesic surfaces.
Proposition 2.2 (Polyhedral decomposition of S 3 \ L). Let L be a hyperbolic
fully augmented link. There is a decomposition of S 3 \ L into two identical totally
geodesic polyhedra. In addition, these polyhedra have the following properties.
• Faces of the polyhedra can be checkerboard colored, with shaded faces all
triangles corresponding to 2–punctured disks, and white faces correspond-
ing to components of the projection plane.
• Ideal vertices are all 4–valent.
208
Figure 2. Left to right: Steps 1, 2, and 3.
• The dihedral angle at each edge is π/2.
Proof. First, assume those components of L that are not crossing circles lie
flat on the projection plane, i.e. that there are no half–twists. Create the polyhedra
by the following procedure, illustrated in Figure 2:
Step 1. Cut S 3 \L along the projection plane, slicing it into two identical pieces, one
above and one below the projection plane. Note this slices each of the 2–punctured
disks bounded by a crossing circle in half.
Step 2. For each of the two pieces resulting from Step 1, slice up the middle of
the halves of 2–punctured disks, opening half disks out as in Figure 2.
Step 3. Collapse each component of the link to a single ideal vertex.
This decomposes S 3 \ L into two identical polyhedra.
The edges of the polyhedra come from the intersections of the 2–punctured disks
with the projection plane. We shade the faces coming from 2–punctured disks and
leave the components of the projection plane white. Note that each edge bounds
one shaded face and one white face. Hence the polyhedra can be checkerboard
colored. Moreover, note that each shaded face meets exactly three edges, hence all
are triangles.
Ideal vertices correspond to components of the link after cutting, slicing, and
flattening (steps 1 and 2 above). Each component coming from a link component
embedded in the projection plane will meet exactly four edges: two for each of
the 2–punctured disks in which that link component terminates. Each component
coming from a crossing circle will also meet exactly four edges: two for each of the
two triangles corresponding to the half of the 2–punctured disk it bounds in each
polyhedron. So ideal vertices are 4–valent, as claimed.
Finally, note that reflection in the white faces preserves the link complement.
Hence shaded faces must be orthogonal to the white faces, and hence the dihedral
angle at each edge is exactly π/2.
To obtain S 3 \ L from the polyhedra, reverse the slicing procedure. First glue
pairs of shaded triangles across a vertex corresponding to a crossing circle, then
glue corresponding white faces in the two polyhedra.
Finally, if L has half–twists, we modify the procedure slightly. Remove all
half–twists from the diagram, then repeat Steps 1, 2, and 3 above to obtain two
checkerboard colored polyhedra, with properties as claimed in the statement of the
lemma. To obtain S 3 \ L from these polyhedra, change the gluing at each half–
twist. Rather than glue shaded triangles across their common vertex on a single
polyhedron, glue each triangle of one polyhedron to the opposite triangle of the
other. See Figure 3. The result is homeomorphic to S 3 \ L.
5
Figure 3. Left: Gluing shaded triangles in same polyhedron

across their common vertex gives a crossing circle with no half–
twist. Left: Gluing each shaded triangle in one polyhedron to
opposite triangle in other polyhedron inserts a half–twist.
Notice that the polyhedra of the polyhedral decomposition of Proposition 2.2

agree for fully augmented links with or without half–twists. Only the gluing of the
polyhedra changes when half–twists are present.
By a circle packing, we mean a finite collection of Euclidean circles in R2 or S 2
which meet only in points of tangency. The nerve of a circle packing is the graph
obtained by taking one vertex for each circle, and an edge between two vertices for
which the corresponding circles are tangent.
Lemma 2.3. Let L be a hyperbolic fully augmented link. Then the polyhedral
decomposition of S 3 \ L corresponds a circle packing on S 2 whose nerve is a trian-
gulation of S 2 . Moreover, the nerve satisfies the following two properties.
• Each edge of the nerve has distinct endpoints.
• No two vertices are joined by more than one edge.
Proof. The circle packing of the lemma is given by considering the hyperbolic
structure on the polyhedra of Proposition 2.2. Each white face is totally geodesic.
Hence it extends to the boundary at infinity, S∞ 2
of H3 to give a Euclidean circle
2
on S∞ . Since none of the white faces overlap, but meet only at ideal vertices,
the corresponding circles will meet only in points of tangency corresponding to
associated ideal vertices.
The fact that the nerve of the circle packing is a triangulation follows from the
fact that the shaded faces of the polyhedra are all triangles. White faces meet at
ideal vertices, so corresponding to each ideal vertex is an edge of the nerve. Two
triangular shaded faces also meet at each ideal vertex. These group ideal vertices
into triples. Corresponding edges of the nerve form a triangle. See Figure 4.
Finally, notice that since the circles of our circle packing are geometric circles
2
on S∞ , a circle cannot be tangent to itself, so each edge has distinct endpoints.
Moreover, if two circles are tangent, then they may only be tangent in a single
point. So no two vertices are joined by more than one edge.
The circle packing of Lemma 2.3 is obtained by extending white faces to the
boundary at infinity of H3 . Notice that the dual circle packing is given by extending
shaded faces to the boundary at infinity of H3 . To obtain each polyhedron, take
the intersection of appropriate half spaces in H3 with boundary the hemispheres
corresponding to these circles.
210
Figure 4. An example of a circle packing and its nerve.
Figure 5. Each red edge (shown here dashed) is replaced by a

crossing circle.
There is a converse to Lemma 2.3.

Lemma 2.4. Let γ be a triangulation of S 2 such that each edge has distinct
ends and no two vertices are joined by more than one edge. Choose a collection of
edges of γ and paint them red, such that each triangle of γ meets exactly one red
edge. Then associated to this painted graph is a hyperbolic fully augmented link, the
nerve of which is γ.
Proof. By a corollary of Andreev’s theorem noted by W. Thurston [16, Chap-
ter 13], associated to γ is a circle packing, unique up to Möbius transformation,
whose nerve is isotopic to γ. The circles in the circle packing extend to give hemi-
spheres in H3 . Circles dual to our circle packing also extend to give hemispheres in
H3 . The intersection of half–spaces in H3 bounded by all these hemispheres, lying
between the hemispheres gives an ideal polyhedron in H3 .
We may construct the complement of a hyperbolic fully augmented link by
gluing together faces of this polyhedron and an identical copy of this polyhedron.
In particular, red edges of γ will correspond to crossing circles. First glue together
triangular shaded faces across each of these vertices in each polyhedron. Then glue
corresponding white faces across the two polyhedra.
We need to ensure this actually is homeomorphic to the complement of a fully
augmented link. To see the diagram of the fully augmented link, let Γ be the graph
dual to γ. It is a trivalent graph on S 2 with one of every three edges colored red.
Replace each red edge by a crossing circle, as in Figure 5. Then one easily checks
that the circle packing associated with this augmented link agrees with that given
by Lemma 2.3.
Figure 5 gives us a combinatorial way of associating a triangulation of S 2 with
the diagram of a fully augmented link without stepping through the polyhedral
decomposition. Namely, given any augmented link, replace crossing circles (and
any half–twists) with red edges, connecting the strands entering into the crossing
circle into a trivalent vertex before and after the crossing circle, as in the figure.
7
A or B A or B
A B ⇒ A B ⇒
..
.
Figure 6. Left: A prime diagram. Right: A twist reduced diagram.
This gives a trivalent planar graph on S 2 . Its dual graph, a triangulation of S 2 , is

the nerve of the circle packing of Lemma 2.3.
A choice of red edge for each triangle above is a choice of dimer for the graph
γ. Thus enumerating all (not necessarily distinct) fully augmented links with a
given number of crossing circles amounts to enumerating triangulations of S 2 with
a fixed number of triangles, along with a choice of dimer for each triangulation.
Additionally, there is the choice at each red edge as to whether to insert a half–
twist or not.
The hyperbolicity of a fully augmented link L obtained by augmenting the
diagram of a link K is known to be related to the diagram of K, as well. Recall that a
diagram is prime if whenever a simple closed curve on the projection plane intersects
the diagram in exactly two points, the curve bounds a portion of the projection
plane containing no crossings. A diagram is twist reduced if whenever a simple
closed curve on the projection plane intersects the diagram in four points, adjacent
to two crossings, the curve bounds a region of the projection plane containing a
(possibly empty) twist region. See Figure 6.
The following theorem follows easily from work of Adams [2].
Theorem 2.5. A fully augmented link is hyperbolic if and only if the associated
knot or link diagram is nonsplittable, prime, twist reduced, with at least two twist
regions.
The more difficult direction can also be shown using Andreev’s theorem, which
was done in [11, Theorem 6.1]. As for the other direction, if the diagram is not
prime or not twist reduced the corresponding fully augmented link will contain an
obvious incompressible annulus. One twist region (or fewer) gives a Seifert fibered
augmented link. We omit the full details here.
3. Geometric properties of fully augmented links

3.1. Cusp shape. Restriction of the hyperbolic metric on the complement of
a fully agumented link to a horospherical torus about each link component gives
a Euclidean structure on the torus [17]. For more general hyperbolic manifolds,
it is often difficult to determine the Euclidean structure about a cusp. For fully
augmented links, however, the cusp shape can be read off of the circle packing of
Lemma 2.3 and the gluing of the polyhedra.
Lemma 3.1. Any cusp of a fully augmented link is tiled by rectangles, each
determined by a circle packing corresponding to a vertex of the ideal polyhedra of
Proposition 2.2. Moreover, the circle packing for any such vertex always consists of
two parallel white lines, both tangent to a pair of white circles as in Figure 7, with
all additional circles interior to these four.
212
Figure 7. Result of taking any ideal vertex to infinity is a rectangle.
Proof. Any cusp of a fully augmented link will correspond to a collection of

ideal vertices in the polyhedral decomposition. Proposition 2.2 implies that ideal
vertices are rectangles: each ideal vertex is 4–valent and dihedral angles are π/2. To
obtain the circle packing, consider points of tangency of circles in the circle packing
of Lemma 2.3. Select one of these points of tangency, call it p. The circle packing
lies on the boundary at infinity of H3 , i.e. on the Riemann sphere S∞ 2
= C ∪ {∞}.
Therefore there exists a Möbius transformation, an isometry of H , taking the point
3
p of tangency of the two circles to the point ∞ in C ∪ {∞}.

Consider the effect of this Möbius transformation on the circle packing. The
two circles tangent at p will go to parallel lines through ∞. Since the nerve of the
circle packing is a triangulation, on either side of p there is a closest circle tangent
to both circles mapped to parallel lines. Each of these two circles will be mapped
to circles tangent to both parallel lines, as in Figure 7. All other circles will be
mapped under the Möbius transformation to circles lying between these four.
In Figure 8, we show several examples of fully augmented links and the associ-
ated “rectangular” circle packing, seen with an ideal vertex at infinity.
We remark that many additional beautiful examples of fully augmented links
appear in Chesebro, DeBlois, and Wilton [3, Section 7]. They show the examples
of fully augmented links together with a trivalent graph they call the crushtacean
of the link. In our terminology, this crushtacean is exactly the dual graph to the
nerve of the circle packing associated with fully the augmented link.
To obtain the shapes of the cusps of augmented links, we need to look at the
tilings of cusps by the rectangles of Proposition 2.2. This is done by walking through
the gluing of polyhedra, forming fundamental domains for the cusps, keeping track
of curves corresponding to meridians and longitudes. This is done very carefully in
[7]. We reproduce the argument here for crossing circles, and leave shapes of other
cusps as an exercise for the reader. The following is part of Lemma 2.3 of [7].
Proposition 3.2. Let C be a cusp corresponding to a crossing circle in a

hyperbolic fully augmented link. Then the following hold:
(1) A fundamental region for C consists of two rectangles coming from vertices
of the polyhedra of Proposition 2.2.
(2) A longitude of C is parallel to the curve given by a shaded face intersected
with the cusp boundary, and it intersects white faces twice. See Figure 9.
(3) In the case C corresponds to a crossing circle with no half–twist, a merid-
ian is parallel to the curve given by a white face intersected with the cusp
boundary.
9
Figure 8. Examples of fully augmented links with 2, 3, and 4

crossing circles, their associated trivalent graphs, and their circle
packings. For the third example, combinatorics of polyhedral de-
composition are also shown.
(4) If C corresponds to a crossing circle with a half–twist, then a meridian

connects the corner of one of the rectangles of Figure 7 to the opposite
corner.
Proof. The result is obtained by analyzing the gluing across the rectangles
of Figure 7. A rectangle corresponding to a crossing circle cusp is shown for an
example on the left in Figure 9. Note that there are exactly two such rectangles,
one on one of the two polyhedra in the decomposition, and one on the other. The
reflection of S 3 \ L in the white faces of the polyhedra glues these two rectangles
together along a side corresponding to the intersection of a white face with the
cusp. These are the only rectangles corresponding to this cusp, so part (1) holds.
From the diagram of the fully augmented link, we see that a longitude of a
crossing circle is parallel to the shaded disk bounded by the crossing circle. In the
polyhedral decomposition, this shaded disk becomes shaded faces. The intersection
of the shaded disk with the boundary of the cusp therefore projects to the longitude.
Since the two rectangles of part (1) are glued along white faces, a longitude must
run over both rectangles and intersect white faces twice, for example as in the
middle in Figure 9. This gives part (2).
When there are no half–twists, a meridian can be seen in the diagram of the
fully augmented link. It runs along the projection plane, which is one of the totally
geodesic faces of Lemma 2.1, and becomes a white face of the polyhedral decompo-
sition. Thus the intersection of a white face with the cusp boundary projects to a
meridian. Note its endpoints are on shaded faces which glue to each other, hence it
steps along just one rectangle of Figure 7, as shown for example in Figure 9. This
gives part (3).
214
Figure 9. Left: Ideal vertex corresponding to crossing circle.

Middle: Fundamental region of associated cusp of S 3 \ L. With-
out half–twists, λ is a longitude, μ a meridian. Right: Half–twists
cause shearing.
Finally, if C corresponds to a crossing circle with a half–twist, then a shaded

triangle on one polyhedron is glued to a shaded triangle opposite the vertex cor-
responding to C on the other polyhedron. We see this as a shear in the universal
cover, as in the right in Figure 9. The meridian travels a step along a white side,
followed by a step along a shaded side (in a direction corresponding to the direction
of the half–twist). This gives part (4).
3.2. Canonical polyhedral decomposition. In this section, we show that
edges of the polyhedra obtained in Proposition 2.2 are actually canonical, as defined
by Epstein and Penner [5]. That is, these edges are geometric duals to faces of
a Ford domain. Recall that when a 3–manifold has more than one cusp, as in
the case of fully augmented link complements, a Ford domain corresponds to a
choice of horoball neighborhoods about all cusps. We will see there is a choice of
horoball neighborhoods with disjoint interiors such that the horoball neighborhoods
are tangent across every edge of the polyhedra of Proposition 2.2. It follows that
each edge is in the geometric dual of this Ford domain.
Definition 3.3. Given any edge e of a hyperbolic ideal triangle T , define the
midpoint of e to be the point on e where a geodesic from the vertex of T opposite
e meets e at a right angle.
In the complement of a fully augmented link, shaded faces form triangles. Thus
for every edge of every shaded face, we may find a midpoint. Two shaded faces are
glued to each other by a reflection through a white face. Since angles are preserved
under reflection, the midpoint of an edge on one shaded face will agree with the
midpoint of the same edge in an adjacent shaded face. Thus each edge of the
polyhedral decomposition of a fully augmented link has a well–defined midpoint.
The following is Theorem 3.8 of [7].
Theorem 3.4. Let L be a fully augmented link. Then there exists a horoball
expansion about the cusps of S 3 \ L such that the midpoint of every edge is a point
of tangency of horospherical tori.
Proof. Let P be one of the (identical) ideal polyhedra in the decomposition of
S 3 \ L. First, lift P to H3 so that the (Euclidean) width of the rectangle of Figure 7
11
is exactly 1. That is, the two circles forming sides of the rectangle of Figure 7 have
diameter 1. There are exactly four edges meeting the ideal vertex at infinity. Note
the midpoint of each of them is at (Euclidean) height 1 in H3 . Thus a horizontal
plane of (Euclidean) height 1 in H3 is a horosphere meeting all the midpoints of
edges through the vertex at infinity.
Suppose we cannot expand a horoball neighborhood about all cusps to the
midpoints of the edges. Expand cusps as much as possible, without expanding any
cusp beyond the midpoints of the adjacent edges. In the universal cover, we will
see a tiling of H3 by copies of P which we may assume have Euclidean width 1. By
assumption, we may apply an isometry so that the horoball about infinity cannot
be expanded to (Euclidean) height 1. Thus there is a horoball H of diameter h > 1
tangent to the horoball about infinity, which projects to some embedded cusp of
S 3 \ L which has not been expanded beyond midpoints of edges. Let z ∈ C be the
center of the horoball H of diameter h > 1.
First, notice that z cannot lie strictly between the two white vertical planes
of the rectangles of Lemma 3.1, as follows. If so, since the vertical white planes
are (Euclidean) distance 1 apart, and H has (Euclidean) diameter h > 1, H must
intersect at least one of the two vertical white planes. Since a reflection in the
white planes preserves the link complement, the isometry given by reflection in the
plane meeting H descends to an isometry of S 3 \ L. But under this isometry, H is
taken to a horoball intersecting H. This contradicts the fact that our horoball was
embedded. Thus z must lie on one of the vertical white planes forming top and
bottom sides of the rectangle of Figure 7, and the reflection takes H to itself.
Next, notice that z cannot lie at the endpoint of an edge of the polyhedral
decomposition running down from infinity, for if so, H contains the midpoint of
this edge in its interior. This is a contradiction: we assumed that H was not
expanded beyond any midpoints.
Finally, we prove that z cannot lie in the interior of one of the vertical white
faces, disjoint from an edge through infinity. For if so, say z lies on the vertical
white plane V , since z is an ideal vertex of a copy of the polyhedron P , there exists
a white plane in H3 tangent to V at the point z, so that V and this white plane are
both boundary faces of P . The white plane has boundary a circle C on C. Consider
the projection of C to V . That is, consider the set of geodesics through points of C
meeting V at right angles. The endpoints of these arcs on V define a (Euclidean)
circle C of the same diameter as C, tangent to z. Since C lies between two vertical
planes, it has diameter less than 1. Thus C is contained in the interior of H.
Now apply an isometry taking z to infinity, taking V to itself, and taking the
plane with boundary C to a vertical plane of distance 1 from V . The circle C
on the vertical plane is taken to a horizontal line on V of height exactly 1, which
must still lie in the interior of the image of H. But this is impossible: we assume
that H was not expanded beyond the midpoints of the edges meeting it. This final
contradiction finishes the proof.
Corollary 3.5. The edges of the ideal polyhedra of Proposition 2.2 are canon-
ical edges for the link complement, in the sense of Epstein and Penner [5].
Note that there may be additional canonical edges in the canonical polyhedral
decomposition.
216
Figure 10. Central subdivision.
3.3. Volume bounds for fully augmented links. We discussed volumes of

fully augmented links in [6]. The following is Proposition 3.1 of that paper.
Proposition 3.6. Let L be a hyperbolic fully augmented link with c crossing
circles. Then its volume is at least 2v8 (c − 1), where v8 = 3.66386 . . . is the volume
of a regular ideal octahedron. Moreover, the volume is exactly 2v8 (c − 1) if and only
if S 3 \ L decomposes into regular ideal octahedra.
Proof. By work of Adams [1], the volume of the complement of L agrees
with that of the augmented link with no half–twists, so we will assume L has no
half–twists. Cut S 3 \ L along the projection plane, dividing it into two isometric
hyperbolic manifolds with totally geodesic boundary, where the isometry is given
by reflection in the projection plane.
Miyamoto showed that if N is a hyperbolic 3–manifold with totally geodesic
boundary, then vol(N ) ≥ −v8 χ(N ) [10], with equality exactly when N decomposes
into regular ideal octahedra. In our case, each cut half of S 3 \ L is a ball with a
tube removed for each crossing circle, hence it is a genus c handlebody, and has
Euler characteristic 1 − c. The result now follows.
Definition 3.7. Given a triangle T , define the central subdivision of T to be
the subdivision obtained by inserting a vertex in the center of T , then adding three
edges running from the new vertex to one of the three vertices of T . See Figure 10.
The following proposition gives all examples of fully augmented links for which
the polyhedra of Proposition 2.2 decompose into regular ideal octahedra. Hence
for these links, the estimate of Proposition 3.6 is sharp.
Proposition 3.8. Let L be a fully augmented link with polyhedral decomposi-
tion into two polyhedra isometric to P , and let N be the nerve associated with the
circle packing of L. Then P is obtained by gluing regular ideal octahedra if and only
if N is obtained by central subdivision of the complete graph on four vertices. In
this case, there are c − 1 such octahedra, where c is the number of crossing circles
in the diagram of L.
Proof. Suppose the polyhedral decomposition of L gives two polyhedra which
are obtained by gluing regular ideal octahedra. Apply a Möbius transformation
taking a vertex of one of the polyhedra to infinity. This vertex is an ideal vertex
of at least one of the regular octahedra. Under the Möbius transformation, any
octahedron with vertex taken to infinity will give rise to a collection of circles of
the circle packing of the form in Figure 11, on the left.
If multiple octahedra share the vertex at infinity, then we claim they must be
glued together in a linear manner as shown on the right of Figure 11. This can be
seen by an inductive argument. Given a single octahedron, a second octahedron
13
...
Figure 11. Left: Top down view of a regular ideal octahedron.

Right: such octahedra sharing the ideal vertex at infinity must be
glued in a linear manner, as shown. (Dashed lines show location
of dual circles, giving shaded faces.)
...
Figure 12. Left: Nerve of circles of octahedron is complete graph

on four vertices. Right: For multiple octahedra glued at vertex at
infinity, nerve is central subdivision of graph on the left.
meeting at the vertex at infinity will be attached to the first either along a white
face or one of the two shaded faces. If it is attached to a white face, then the two
shaded faces on the ends become quadrilaterals, contradicting Proposition 2.2. By
induction, the n-th octahedron must be attached along one of the shaded faces as
well, else again we have a contradiction to Proposition 2.2.
Consider the portion of the nerve of the circle packing corresponding to the
circles of octahedra sharing this vertex at infinity. If we have just one ideal octa-
hedron, the nerve is the complete graph on four vertices, as shown on the left of
Figure 12. If there are more octahedra, the result is a central subdivision of the
complete graph on four vertices, subdividing the triangle enclosing the point at
infinity, as on the right of Figure 12.
The above holds for any ideal vertex. When we move a different ideal vertex to
infinity, we apply a Möbius transformation. This will change the nerve by moving
a new edge to infinity. (In Figure 12 we pushed this edge off infinity slightly to
obtain the curved edge in that figure.) While the vertices and edges of the nerve
may move around, each triangle before the Möbius transformation corresponds to
a triangle after, with vertices and edges the images of vertices and edges before.
Thus the triangle containing infinity in the nerve before the Möbius transformation,
which we saw was subdivided by central subdivision once for every octahedron
meeting infinity, will be mapped to a triangle which has been subdivided by central
subdivision. Thus the entire nerve is obtained by central subdivision of the complete
graph on four vertices.
Now suppose the nerve of the fully augmented link L is obtained by successive
central subdivision of the nerve of the connected graph on four vertices. We induct
on the number of times we must subdivide to obtain the nerve of L. If 0 times, then
the nerve of L is the graph shown on the left of Figure 12, and the corresponding
polyhedron is a single regular ideal octahedron shown on the left of Figure 11.
218
Figure 13. Left: Subdivide a triangle in the nerve. Middle: add

a circle to the corresponding circle packing. Right: Attach a poly-
hedron bounded by the hemispheres of these circles, and their du-
als.
Next suppose that we have subdivided n times, each time attaching a new
regular ideal octahedron to the previous polyhedron. At the (n + 1)-st subdivision,
we add a vertex and three edges to some triangle of the nerve, as on the left of Figure
13. This corresponds to adding a circle to a triangle of the circle packing, as in
the middle of Figure 13. In the 3–dimensional polyhedron, we attach a polyhedron
with faces bounded by circles as shown on the right of Figure 13. In particular, note
that if we take any of the points of tangency of circles to infinity in that Figure, we
obtain exactly the circles on the left of Figure 11. Thus this polyhedron is exactly
a regular ideal octahedron.
Finally, we relate the number of octahedra to the number of crossing circles
c. Suppose the polyhedron of Proposition 2.2 is obtained by gluing n regular ideal
octahedra. Then the nerve is formed by central subdivision of the complete graph
on four vertices. From the proof above, the first octahedron corresponds to that
complete graph on four vertices, containing 6 edges, and then each successive sub-
division, adding 3 edges, gives exactly one more octahedron. Thus the number of
edges E in the nerve is E = 3n + 3.
On the other hand, the nerve is a triangulation of S 2 . For each triangle, exactly
one edge is painted red, i.e. corresponds to an ideal vertex associated with a crossing
circle. A single red edge borders two triangles. Thus the total number of red edges
is equal to half the total number of triangles, T , in the nerve. Since each red edge
corresponds to a distinct crossing circle, we have c = T /2. Because the nerve is a
triangulation of S 2 , 3T = 2E = 6n + 6. Hence c = n + 1, or there are exactly c − 1
octahedra, as claimed.
4. Restricting and extending results

In this section, we study two examples of classes of links related to fully aug-
mented links. The first class is obtained by restricting to a subset of fully augmented
links, obtaining a class of links even more geometrically explicit. The second class
is obtained by extending to a class of links containing fully augmented links, losing
some of the geometric structure as we relax the definitions.
4.1. Octahedral fully augmented links. The geometry of fully augmented

links is completely determined by a circle packing, which can be computed from
a nerve, for example in simple cases by K. Stevenson’s program CirclePack [4].
However, for extremely complicated nerves or families of nerves it may be more
15
difficult to obtain circle packings. Also, additional geometric estimates such as

volume may be harder to obtain exactly.
To remedy this problem, we may restrict to a subclass of fully augmented links:
those whose polyhedra are a union of regular ideal octahedra. For these links, which
we will call octahedral augmented links, the volume is given exactly by Proposition
3.6. The circle packing can be constructed by hand from the nerve (provided your
pencil tip is sharp enough). The rectangles of Lemma 3.1 are all squares, or strings
of squares as in Figure 11. The geometry is completely explicit. Indeed, among
these links are many links for which the volume conjecture is known, due to work
of R. Van der Veen [18].
4.2. Generalized fully augmented links. We may generalize the class of

fully augmented links by relaxing conditions of their definition. In a generalized
augmented link, we allow crossing circles to encircle more than two strands per
twist region. For these links, we lose the triangulation and circle packing. Results
on hyperbolicity and cusp shapes are not as clean [14, 12]. However, there is still a
reflection in the projection plane, giving a totally geodesic surface when these links
are hyperbolic. The existence of this surface and the reflection allows us to obtain
some information on volumes and geometry of these links. These are explored in
[13] and in [12].
References
1. Colin C. Adams, Thrice-punctured spheres in hyperbolic 3-manifolds, Trans. Amer. Math.
Soc. 287 (1985), no. 2, 645–656. MR 768730 (86k:57008)
2. , Augmented alternating link complements are hyperbolic, Low-dimensional topology
and Kleinian groups (Coventry/Durham, 1984), London Math. Soc. Lecture Note Ser., vol.
112, Cambridge Univ. Press, Cambridge, 1986, pp. 115–130. MR 903861 (89f:57003)
3. Eric Chesebro, Jason DeBlois, and Henry Wilton, Some virtually special hyperbolic 3–manifold
groups, arXiv:0903.5288.
4. Tomasz Dubejko and Kenneth Stephenson, Circle packing: experiments in discrete analytic
function theory, Experiment. Math. 4 (1995), no. 4, 307–348. MR MR1387696 (97f:52027)
5. D. B. A. Epstein and R. C. Penner, Euclidean decompositions of noncompact hyperbolic man-
ifolds, J. Differential Geom. 27 (1988), no. 1, 67–80. MR MR918457 (89a:57020)
6. David Futer, Efstratia Kalfagianni, and Jessica S. Purcell, Dehn filling, volume, and the Jones
polynomial, J. Differential Geom. 78 (2008), no. 3, 429–464. MR MR2396249 (2009c:57010)
7. David Futer and Jessica S. Purcell, Links with no exceptional surgeries, Comment. Math.
Helv. 82 (2007), no. 3, 629–664. MR MR2314056 (2008k:57008)
8. Marc Lackenby, Word hyperbolic Dehn surgery, Invent. Math. 140 (2000), no. 2, 243–282.
MR 1756996 (2001m:57003)
9. , The volume of hyperbolic alternating link complements, Proc. London Math. Soc. (3)
88 (2004), no. 1, 204–224, With an appendix by Ian Agol and Dylan Thurston. MR 2018964
(2004i:57008)
10. Yosuke Miyamoto, Volumes of hyperbolic manifolds with geodesic boundary, Topology 33
(1994), no. 4, 613–629. MR 1293303 (95h:57014)
11. Jessica S. Purcell, Cusp shapes under cone deformation, J. Differential Geom. 80 (2008),
no. 3, 453–500. MR MR2472480
12. , Slope lengths and generalized augmented links, Comm. Anal. Geom. 16 (2008), no. 4,
883–905. MR MR2471374
13. , Hyperbolic geometry of multiply twisted knots, Comm. Anal. Geom. 18 (2010), no. 1,
101–120. MR 2660459
14. , On multiply twisted knots that are Seifert fibered or toroidal, Comm. Anal. Geom.
18 (2010), no. 2, 219–256.
15. Dale Rolfsen, Knots and links, Publish or Perish Inc., Berkeley, Calif., 1976, Mathematics
Lecture Series, No. 7. MR MR0515288 (58 #24236)
220
16. William P. Thurston, The geometry and topology of three-manifolds, Princeton Univ. Math.
Dept. Notes, 1979.
17. , Three-dimensional geometry and topology. Vol. 1, Princeton Mathematical Series,
vol. 35, Princeton University Press, Princeton, NJ, 1997, Edited by Silvio Levy. MR 1435975
(97m:57016)
18. Roland van der Veen, The volume conjecture for augmented knotted trivalent graphs, Algebr.
Geom. Topol. 9 (2009), no. 2, 691–722. MR MR2496886
Department of Mathematics, Brigham Young University, Provo, UT 84602

E-mail address: jpurcell@math.byu.edu
Volume 541, 2011
Orbifolds and commensurability
Genevieve S. Walsh
Abstract. This paper is based on a series of talks that the author gave at the
“Interactions between hyperbolic geometry and quantum groups” conference
held at Columbia University in June of 2009. We describe the structure of
orbifolds, and show that they are very useful in the study of commensurability
classes. We also survey some recent results in the area.
1. Background on hyperbolic manifolds and orbifolds

In understanding manifolds and their commensurability classes, we will find it
extremely helpful to employ orbifolds. A manifold is an object locally modeled on
open sets in Rn , and an orbifold O is locally modeled on open sets in Rn modulo
finite groups of Euclidean isometries. That is, each point x ∈ O has a neighborhood
modeled on Ũ /G, where G is a finite subgroup of SO(n) and Ũ is an open ball in Rn .
A geometric orbifold is the quotient of a simply connected Riemannian manifold X
by a discrete subgroup Γ of Isom(X), and we say that O = X/Γ is an X-orbifold.
In this case the orbifold fundamental group is the group Γ. There are non-geometric
orbifolds, but we will only be concerned with geometric ones here. We will describe
the structure of orbifolds through some examples, see [5] for a good description of
the details.
1: The “football” is an S 2 -orbifold which is the quotient of the S 2 by the
group Z/3Z generated by a rotation of 2π/3 which fixes the north and
south poles. The ramification locus of an orbifold O is the set of points
where any neighborhood is modeled on an open set in Rn modulo a non-
trivial group. The ramification locus in this case is two points, which we
label 3, since that is the order of the local group. The underlying space |O|
of an orbifold O is the space obtained from O by forgetting the orbifold
structure, which is S 2 in this case. The two ramification points are both
modeled on disks in R2 modulo a group of rotations, and we call these
cone points. The football is commonly denoted as S 2 (3, 3). In general,
M 2 (r1 , ...rn ) is a 2-orbifold with underlying space the 2-manifold M 2 and
n cone points of orders ri .
2: A common Euclidean orbifold is S 2 (2, 2, 2, 2), which is the quotient of R2
by the group generated by the translations (x, y) → (x + 1, y), (x, y) →
1991 Mathematics Subject Classification. Primary 57M25, 57M10, 57M60.

Key words and phrases. Orbifolds, Commensurability, Hyperbolic Three-manifolds.
The author was supported in part by NSF Grant #0805908.
1 2011
221
222
2 GENEVIEVE S. WALSH
(x, y + 1) and rotation by π about (0, 1/2). Note that this will generate
rotations by π in all half-integer lattice points. Exercise: What is a fun-
damental domain? Find a rotation of order 4 on R2 such that the group
generated by this rotation and the above generators yields the orbifold
S 2 (2, 4, 4).
An orbifold covering f : Q → Q is a continuous map between the underlying spaces
|Q | → |Q|. We further require that if a point x ∈ |Q| has a neighborhood whose
orbifold structure is U = Ũ /G then each component Vi of f −1 (U ) is isomorphic
to Ũ /Gi where Gi < G and f |Vi : Vi → U is Ũ /Gi → Ũ /G. See [5, Section 2.3]
for elaboration. Example 1 above gives a cover of S 2 (3, 3) by S 2 , and the exercise
in Example 2 gives a cover of S 2 (2, 4, 4) by S 2 (2, 2, 2, 2). Note that S 2 (2, 4, 4) also
covers itself with non-trivial degree.
3: We regard hyperbolic 2-space H2 as the upper-half-space {z|Im(z) > 0}
of the complex
a b plane and declare Isom+ (H2 ) to be PSL(2, R), where
the
matrix c d acts by z → cz+d . Then the group generated by 21 11 and
az+b
1 1 2
1 2 is a free group of rank 2. The quotient of H by this group 0 is1 a
punctured torus. If we add a generator that rotates by π about i, −1 0 ,
the quotient orbifold is S (2, 2, 2, ∞) (the ∞ denotes a cusp). This is a
2
quotient of the punctured torus by an involution which fixes three points

on the punctured torus and takes the puncture to itself. It is perhaps
easier to understand the quotient by considering the induced action on
the unit disk. Use the transformation U (z) = zi+1z+i to take the upper-half
space to the unit disk. Then the rotation by π about i becomes rotation
by π about the origin. This illustrates S 2 (2, 2, 2, ∞) as a hyperbolic 2-
orbifold.
Here we will mainly consider hyperbolic 3-orbifolds. These are the quotient
of H3 (regarded as {(z, t)|z ∈ C, t ∈ R, t > 0}) by a discrete finitely generated
subgroup of Isom+ (H3 ) ∼ = PSL(2, C). As above,
PSL(2, C) acts on the sphere
at infinity {(z, 0), z ∈ C ∪ ∞} such that ac db acts by z → az+b cz+d . The action
on H3 is by Poincaré extension, where hemispheres perpendicular to the sphere at
infinity are mapped to hemispheres perpendicular to the sphere at infinity. (See [14,
Chapter 1].) Volume will be discussed in other lectures here, but for our purposes
all subgroups of PSL(2, C) will be discrete and have finite co-volume, meaning
that the quotient is a finite-volume hyperbolic orbifold. A discrete subgroup of
PSL(2, C) is called a Kleinian group. Let Γ1 and Γ2 be two finite co-volume Kleinian
groups. Mostow-Prasad rigidity states that if Γ1 and Γ2 are isomorphic, then
they are conjugate. This means that if O1 and O2 are finite volume hyperbolic
orbifolds H3 /Γ1 and H3 /Γ2 , then any isomorphism of their fundamental groups is
determined by a unique isometry of the hyperbolic orbifolds. Therefore, invariants
which depend only on the conjugacy class of the representation of the fundamental
group into PSL(2, C) are topological invariants. We will describe two here, the
trace field and the cusp field. Others are discussed in Section 2.
Let H3 /Γ be a hyperbolic orbifold. Then the field generated by the traces
of elements in Γ is invariant under conjugation of Γ and this is the trace field of
Γ. If there is a subgroup of Γ which fixes some point on the sphere at infinity,
S∞2
= C ∪ ∞, then we can conjugate so that this subgroup fixes ∞. Then the
quotient of the plane {(z, )|z ∈ C} in H3 will be a Euclidean 2-orbifold Q in
ORBIFOLDS AND COMMENSURABILITY 223
3
H3 /Γ for some , and we call the quotient a cusp cross-section. The quotient of
{(z, t)|t > } is Q2 × R+ and we call this a cusp.
If the cusp
cross-section is a torus,
we can conjugate so that the generators are 10 11 and 10 g1 . The cusp field of the
cusp is Q(g). Note that g coincides with the shape of the torus in the identification
of the Teichmuller space of T 2 with upper half-space. If the cusp-cross section is a
compact Euclidean orbifold which is not a manifold, then we take a cover of H3 /Γ
where this cusp cross-section lifts to a torus and take the cusp field of some cusp
in the pre-image. Using a geometrical interpretation of the invariant trace field,
one can show that the cusp field is a sub-field of the invariant trace field, see [13,
Section 5.5]. Furthermore, the invariant trace field is just the trace field for knot
complements [17]. Note that the cusp field of a particular cusp is left unchanged
after taking finite covers.
4: The figure-8 knot complement can be realized as a hyperbolic manifold
1 0 √
−1+ 3i
H3 /Γk where Γk is generated by 10 11 and −w 1 , where w = 2 .
(This is not obvious, see Thurston’s notes [20, Section
4.3].) There
is a
subgroup Z ⊕ Z which fixes infinity generated by 10 11 and 10 g1 where
√ √
g =√2 + 4 −3. Then the cusp field is Q( −3) and the trace field is also
Q( −3). For most, (but not all!) hyperbolic knots in S 3 the trace field
is the same as the cusp field of the unique cusp.
Question 1.1. (See [16].) What are necessary or sufficient conditions for a
hyperbolic knot complement to have its trace field strictly larger than its cusp field?
Currently, there are four knots known that have this property, the two dodecahedral
knots of Aitchison and Rubinstein, 12n706 (discovered by D. Boyd) and 15n132539
(discovered by N. Dunfield). See [16, 8, 3].
A knot K in S 3 is strongly invertible if there is an order two involution t
of (S 3 , K) such that the fixed point set of t intersects the knot twice. A strong
involution induces an order two involution on the knot complement where the fixed
point set intersects each cusp cross-section 4 times. The figure-8 knot is strongly
invertible and we can take the quotient of the figure-8 knot complement by the
strong inversion to
obtain
an hyperbolic 3-orbifold. This corresponds to adjoining
the element τ = 0i −i 0
to the group Γk in example 4 above. Note the action of τ
on the cusp subgroup yields a cusp S 2 (2, 2, 2, 2). Also, τ takes each generator of
the knot group to its inverse.
Let K be a strongly invertible hyperbolic knot. The quotient of S 3 by a strong
inversion is S 3 with an unknotted circle labeled two. The quotient of N (K), a
regular neighborhood of the knot, is a ball with two unknotted arcs labeled 2.
Therefore, the quotient of S 3 \ N (K) by the inversion has underlying space a ball
and two arcs of the ramification locus which are both labeled two. All of the
information about this orbifold is contained in the ramification locus, since its
underlying space is topologically trivial. This orbifold is the minimal orbifold in
the commensurability class when the S 3 \ K is non-arithmetic, the strong inversion
is the only symmetry, and K does not admit hidden symmetries. See Sections 2
and 3 below for more details.
2. Commensurability
Definition 2.1. Two orbifolds are commensurable if they admit homeomor-
phic finite-sheeted covers.
224
We will require that the homeomorphisms are orientation-preserving. Accord-

ingly, we consider the commensurator C + (Γ) to be a subgroup PSL(2, C) in Defini-
tion 2.2 below. We claim that the relation ∼ on n-dimensional orbifolds defined by
commensurability is an equivalence relation. Indeed, clearly O ∼ O, and if O ∼ P,
P ∼ O. Now assume that O ∼ S with a common finite index cover X, and S ∼ P
with common finite index cover Y , so that X and Y both cover S. To show O ∼ P,
we will construct a finite index orbifold cover of X and Y . This is the orbifold
fibered product X ×S Y :
X ×S Y

X Y

O S P
Let f : X → S and g : Y → S be orbifold covering maps. Then each point
w ∈ S has a neighborhood Uw = Ũ /G such that a pre-image of Uw in X, respectively
Y , has the form Ũ /G1 , respectively Ũ /G2 for some Gi < G. It is helpful to think of
Ũ /G as the set of orbits {Gy|y ∈ Ũ }. We define fg : Ũ → Ũ /G1 × Ũ /G2 by fg (y) =
(G1 gy, G2 y). This map factors through Ũ /(g −1 G1 g ∩ G2 ) since y and xy where
x ∈ g −1 G1 g∩G2 have the same image. We get maps from Ũ /(g −1 G1 g∩G2 ) to Ũ /Gi
by composing fg with the projections. Thus we define the fiber product of Ũ /G1
and Ũ /G2 over Ũ /G as the disjoint union of the orbifolds Ũ /(g −1 G1 g ∩ G2 ) where
g is taken over representatives of the double cosets of G1 \G/G2 . (The map fg1 gg2
differs from fg only by the action of G on Ũ .) We can patch these neighborhoods
together to get an orbifold which covers X and Y . See [20, Section 13.2] for an
example of why this is the correct way to extend the definition of the fibered product
for manifolds. This exhibits X ×S Y as an orbifold cover of O, S, and P so ∼ is
transitive which shows the claim.
Aside from the intrinsic reason that this equivalence relation is a method of
organizing manifolds, there are many reasons to study commensurability classes. In
particular, there are lots of properties which are preserved by commensurability. For
example, if a manifold X is virtually fibered, and Y is commensurable with X, then
Y is also virtually fibered. This is because a cover of a manifold which fibers over the
circle fibers over the circle. Similarly, the properties of being virtually Haken, having
fundamental group containing a subgroup which maps onto a free group of rank
at least two, and containing an immersed geodesic surface are all properties of the
commensurability class. In addition, if we restrict ourselves to 3-manifolds which
admit a finite-volume geometry, this equivalence relation on 3-manifolds preserves
the geometric type. Thus we can regard commensurability classes of geometric 3-
manifolds as a refinement of geometrization. This is more useful for some geometries
than for others. For example, all spherical orbifolds are commensurable, but there
are infinitely many commensurability classes of hyperbolic orbifolds.
Commensurability classes are particularly relevant for the study of finite-volume
hyperbolic 3-manifolds and orbifolds, where classification has often been centered
around notions of volume. Complementing this, commensurability classes are trans-
verse to volume. Selberg’s lemma states that a finitely generated subgroup of
GL(n, C) has a torsion-free subgroup of finite index. Therefore, all hyperbolic
orbifolds are finitely covered by manifolds and there are manifolds in every com-
mensurability class. We define the volume of a hyperbolic orbifold to be 1/d the
5
volume of a d-fold cover which is a manifold. By rigidity, volume is a topological in-

variant and volumes of commensurable manifolds are rationally related. There has
also been recent progress in understanding the commensurator of infinite volume
hyperbolic manifolds [11].
Henceforth, we will mainly be concerned with commensurability classes of finite
volume hyperbolic 3-orbifolds. In this case, we can characterize commensurability
of hyperbolic orbifolds by commensurability of subgroups of PSL(2, C). By Mostow-
Prasad rigidity, M1 ∼ = H3 /Γ1 and M2 ∼ = H3 /Γ2 are commensurable if and only if
Γ1 and a conjugate of Γ2 share a finite index subgroup. Following [13, Definition
1.3.4] we say that the groups Γ1 and Γ2 are commensurable in the wide sense in
PSL(2, C).
Let Γ be a finitely generated Kleinian group with finite co-volume. Then by
rigidity, the trace field of Γ is a topological invariant of the hyperbolic orbifold
H3 /Γ, as mentioned in Section 1. It is also a finite degree extension of Q [13,
Theorem 3.1.2]. It is not, however, an invariant of the commensurability class. In
the above example of the figure-8 knot complement and the orbifold which is the
quotient by the strong inversion, it can be seen that the trace field of the quotient
√
orbifold contains i where the trace field of the figure-8 knot complement is Q( −3),
which does not contain i. However, the invariant trace field kΓ = Q(tr(γ 2 )|γ ∈ Γ)
is an invariant of the commensurability class [13, p 117]. Often we will restrict
to hyperbolic knot complements and in this case the trace field is the same as the
invariant trace field [17].
An equivalent way to define the invariant trace field for a cusped finite-volume
hyperbolic manifold with an ideal triangulation is the field generated by the cross-
ratios of all ideal tetrahedra, or the shapes of all the tetrahedra [13, Section 5.5].
This suggests that the cusp field is a subfield of the invariant trace field and indeed
this is the case.
Probably the most useful invariant of a commensurability class is the commen-
surator. Two subgroups A and B of a group are commensurable if their intersection
is finite index in each.
Definition 2.2. The commensurator of a Kleinian group Γ is
C + (Γ) = {g ∈ PSL(2, C)| gΓg −1 and Γ are commensurable}.
The geometry of the commensurator is dramatically different for arithmetic

and non-arithmetic Kleinian groups.
A Kleinian group Γ is arithmetic if it is commensurable with a k-embedding
into M2 (C) of the group of norm 1 elements of an order of a quaternion algebra
A over k where A ramifies at all real places and where k is a number field with
one complex place. (See [13, Definition 8.2.1] and the discussion therein.) A useful
characterization is that a non-cocompact Kleinian group Γ is arithmetic if and only
if a conjugate of Γ is commensurable with a Bianchi√group. These are the groups
PSL(2, Od ), where Od is the ring of integers in Q( −d). We will say that the
orbifold H3 /Γ is arithmetic when Γ is arithmetic. For example, the complement
of the figure-8 knot described above is arithmetic and furthermore this is the only
arithmetic knot complement in S 3 [18].
Let Γ be a finitely generated Kleinian group with finite co-volume. Margulis’s
theorem (which holds in more generality) implies that either Γ is finite index in
226
C + (Γ), or C + (Γ) is dense in PSL(2, C). Furthermore, C + (Γ) is dense exactly when
Γ is arithmetic.
Lemma 2.3. For finite co-volume Kleinian groups Γ1 and Γ2 , the orbifolds
H3 /Γ1 and H3 /Γ2 are commensurable if and only if the commensurators of Γ1 and
Γ2 are conjugate in PSL(2, C).
Proof. The fact that commensurable orbifolds have conjugate commensura-
tors follows directly from Mostow-Prasad rigidity and the definition. For non-
arithmetic orbifolds, Margulis’s theorem says that Γ1 and Γ2 are finite index in their
commensurators, so if their commensurators are conjugate, the orbifolds H3 /Γ1 and
H3 /Γ2 are commensurable. For the arithmetic case, the commensurators of Γ1 and
Γ2 are the set of invertible elements in the quaternion algebra defined by the trace
field. (See [13].) Therefore we may assume, after conjugation, that Γ1 and Γ2 are
contained in two orders O1 and O2 in the same quaternion algebra. Furthermore,
the intersection of two orders is an order so we may assume that O1 ⊂ O2 . Since we
are considering groups up to commensurability and conjugation, the result follows
from the fact that if O1 and O2 are the groups of elements of norm 1 in each order,
then O1 is finite index in O2 .
Thus, for a non-arithmetic finite-covolume Kleinian group Γ, the commensura-
tor C + (Γ) is the maximal element in the commensurability class of Γ. In this case,
the commensurator corresponds to an orbifold H3 /C + (Γ) which is the minimal
element in the commensurability class of O = H3 /Γ. In other words every orbifold
commensurable with O finitely covers H3 /C + (Γ). Thus two non-arithmetic finite-
volume hyperbolic orbifolds have a finite sheeted-cover exactly when they finitely
cover a common orbifold.
On a related note, it is shown in [4] that a hyperbolic fibered commensurability
class contains a unique minimal element.
3. Hidden symmetries
Symmetries of a hyperbolic manifold M and symmetries between finite covers
of M will play a very important role in understanding the commensurability class
of M . Recall that for a Kleinian group Γ, the normalizer of Γ is
N + (Γ) = {g ∈ PSL(2, C)|gΓg −1 = Γ}.

Any self-isomorphism of a finite volume hyperbolic manifold M = H3 /Γ yields
an automorphism of Γ which by Mostow-Prasad rigidity is realized by conjugation.
The fundamental group Γ ≤ N + (Γ) and these conjugations are realized by a change
of base point. Therefore, Isom+ (H3 /Γ) ∼= N + (Γ)/Γ.
3 +
The hyperbolic orbifold H /N (Γ) may or may not be minimal in the commen-
surability class of H3 /Γ, and we distinguish between the two cases below. Clearly,
N + (Γ) ≤ C + (Γ).
Definition 3.1. Let Γ be a finite co-volume Kleinian group. If N + (Γ) is
strictly smaller than C + (Γ), then Γ (and H3 /Γ) are said to have hidden symmetries.
If S 3 \ K ∼
= H3 admits hidden symmetries, then we also say that K admits
hidden symmetries. The elements of the commensurator correspond to isometries
between finite-sheeted covers of H3 /Γ. Indeed, if gΓg −1 ∩ Γ is a finite-index sub-
group of Γ, then conjugation by g −1 gives an isomorphism between the subgroups
7
gΓg −1 ∩ Γ and Γ ∩ g −1 Γg, which is an isometry between the corresponding finite-

sheeted covers. If φ : H3 /Γ1 → H3 /Γ2 is an isometry between finite-sheeted covers,
then it is realized by conjugation by g, for some g ∈ PSL(2, C) by Mostow-Prasad
rigidity. Γ and gΓg −1 have the same finite co-volume, and their intersection has
finite covolume. Therefore, Γ and gΓg −1 are commensurable.
Any isometry of H3 /Γ will permute the finite-sheeted covers of H3 /Γ. Thus, if
the commensurator of Γ is strictly larger than the normalizer, there is an isometry
between finite-sheeted covers of H3 /Γ which is not realized by an isometry of H3 /Γ.
Hence the term “hidden symmetries”. In the non-arithmetic case, by Margulis’s
theorem above, Γ is finite-index in C + (Γ). Therefore, for every g ∈ C + (Γ), there is a
k such that g k ∈ Γ. Hence Γ = Γ∩gΓg −1 ∩g 2 Γg −2 ...∩g (k−1) Γg −(k−1) is normalized
by g for some k. Thus the cover corresponding to Γ exhibits a hidden symmetry.
In the arithmetic case, there are infinitely many hidden symmetries, see [9]. In the
non-arithmetic case, there is a finite-sheeted cover of H3 /Γ which exhibits all the
hidden symmetries of H3 /Γ, namely H3 /Γ , where Γ is the intersection of all the
conjugates of Γ in C + (Γ).
Goodman, Heard and Hodgson [8] have recently obtained a characterization
of the commensurator of cusped hyperbolic manifolds. Their characterization is
as follows. A horoball packing is a collection of disjoint horoballs in H3 . A cusp
neighborhood is any neighborhood of the cusp that lifts to a horoball packing. They
show [8, Lemma 2.3] that two hyperbolic cusped orbifolds of finite volume cover a
common orbifold if and only if they admit choices of cusp neighborhoods lifting to
isometric horoball packings. Furthermore, the commensurator of a finite-covolume
non-arithmetic Kleinian group Γ is the maximal symmetry group of a horoball
packing, amongst those which are lifts of cusp neighborhoods in H3 /Γ. Equiva-
lently, this is the maximal symmetry group of a tiling of H3 amongst those which
are obtained by lifting canonical cell decompositions of H3 /Γ. Using this, they
are able to compute the commensurators for a large number of cusped hyperbolic
orbifolds and to detect the presence of hidden symmetries, by comparing the com-
mensurator with the normalizer. In particular, their computations yield that out
of all hyperbolic knots up to 12 crossings, there are only three whose complements
admit hidden symmetries. These are the two dodecahedral knots of Aitchison and
Rubinstein [1], and the figure-8 knot, whose complement is arithmetic. The do-
decahedral knots are the only two knots whose complements decompose into two
regular ideal hyperbolic dodecahedra. They are commensurable, have the same
volume, and one is fibered while the other is not. Also the knots have different
genus and are amphichiral. This never happens for commensurable knots which do
not admit hidden symmetries, see [2].
A non-arithmetic hyperbolic 3-orbifold H3 /Γ admits hidden symmetries if and
only if it non-normally covers a finite orbifold, since in this case the orbifold
H3 /N + (Γ) is not the minimal element in the commensurability class. We can
understand the commensurator orbifold H3 /C + (Γ) in part by looking at its cusp
cross section, which is a Euclidean 2-orbifold. The symmetry group of a hyper-
bolic knot complement is finite, and any symmetry of the complement will take a
minimal genus Seifert surface (along with the canonical longitude) to a minimal
genus Seifert surface. Thus by the solution to the Smith conjecture, the symmetry
group Isom+ (S 3 \ K) is either cyclic or dihedral. Therefore, any normal covering of
an orbifold by a knot complement has a torus or a S 2 (2, 2, 2, 2) cusp cross-section.
228
Conversely, if a knot complement covers an orbifold with a torus or S 2 (2, 2, 2, 2)

cusp cross-section, the covering is normal, [18, 7]. The orientable Euclidean orb-
ifolds which are not a torus or S 2 (2, 2, 2, 2) are S 2 (2, 4, 4), S 2 (3, 3, 3) and S 2 (2, 3, 6).
These are rigid, meaning that their moduli spaces are trivial. Thus Neumann and
Reid show:
Lemma 3.2. [16, Proposition 9.1] The following are equivalent for a hyperbolic
knot complement H3 /Γ which is not the figure-8 knot complement.
(1) The knot complement admits hidden symmetries.
(2) H3 /C + (Γ) has a rigid Euclidean cusp cross-section.
(3) The knot complement non-normally covers some orbifold.
Since a rigid Euclidean 2-orbifold cannot be deformed, the cusp√field of an
orbifold with one cusp which has a rigid cusp cross-section is either Q( −3) (when
the cusp cross-section is S 2 (3, 3, 3) or S 2 (2, 3, 6)) or Q(i) (when the cusp cross-
section is S 2 (2, 4, 4)). Furthermore, any hyperbolic orbifold which covers an orbifold
with one cusp with a rigid cusp-cross section has the same cusp field. Therefore,
√
a knot complement which admits hidden symmetries has cusp field Q( −3) or
Q(i). Although it is not known exactly how cusp fields are distributed amongst
hyperbolic knot complements, it is suspected that these cusp fields are not any more
prevalent than others. This, along with the experimental results above, suggests
that hyperbolic knots with hidden symmetries are extremely rare.
4. Knot complements
Although understanding commensurability classes for arbitrary hyperbolic 3-
manifolds is quite difficult, the situation for hyperbolic knot complements appears
to be much easier. If the knot complements S 3 \ K and S 3 \ K are commensurable,
we say that the knots are commensurable and write K ∼ K . There are finitely
many knots K with K ∼ K when K (equivalently K ) does not admit hidden
symmetries. In this case, the commensurator quotient orbifold will have a flexible
cusp, and any filling of the knot complement will cover a filling of the commensura-
tor quotient. In particular, the S 3 filling of the knot complement will cover a filling
of the orbifold which necessarily has finite fundamental group. There are finitely
many fillings of a hyperbolic orbifold which yield orbifolds with finite fundamental
group, hence there can be only finitely many knot complements in the commensu-
rability class of a knot which does not admit hidden symmetries. As of this writing,
it is not known if there can be infinitely many knots in the commensurability class
of a knot which admits hidden symmetries, or even if the commensurability class of
the two dodecahedral knots contains infinitely many knot complements, or if there
are infinitely many different commensurability classes containing hyperbolic knot
complements with hidden symmetries.
However, we have the following result which sheds some light on this situation:
Theorem 4.1. (Also [2, Corollary 4.6]) If O is the minimal element of a non-
arithmetic commensurability class which contains a knot complement S 3 \ K then
the underlying space of O is either an open ball or the complement of a knot in a
lens space.
Proof. Let Ô denote the corresponding orbifold with boundary obtained by
truncating the cusp. Since ∂ Ô is an orientable Euclidean orbifold, it is either a torus
9
or has underlying space S 2 . When ∂ Ô is a torus, the covering S 3 \ N (K) → Ô

is normal by [7] and [18, Lemma 4]. Analysis of the action of the covering group
in [2, Lemma 4.5] yields that the underlying space of Ô is the exterior of knot
in a lens space. The case where ∂ Ô has underlying space S 2 is an observation of
M. Kapovich. The fundamental group S 3 \ N (K) is normally generated by the
meridian and S 3 \ N (K) surjects the underlying space of Ô. The underlying space
of ∂ Ô has trivial fundamental group, so the image of π1 (S 3 \ N (K)) is trivial in the
map S 3 \ N (K) → Ô → |Ô|. Then if Ô has any non-trivial cover (for example its
universal cover) this cover has more than one boundary component and the map
from S 3 \ N (K) to |Ô| lifts, a contradiction. Thus the underlying space of the
orbifold Ô is a ball since there are no fake 3-balls by Perelman [15]. Therefore, in
this case, the underlying space of O is an open ball.
There are several known ways for hyperbolic knots to be commensurable. The
first is the classic (but rare) situation when a knot complement S 3 \ K admits a
Dehn filling such that the resulting manifold S 3 \ N (K)(r) is a lens space. In this
case the universal cover of the filled manifold is S 3 , with cyclic covering group.
Since a cyclic cover of a knot complement is unique and has one cusp, the pre-
image of K in the covering S 3 → S 3 \ N (K)(r) has one component K and this
induces a covering S 3 \ K → S 3 \ K. Thus, if the covering and the knot are
both non-trivial, then K ∼ K . In [7] it is proven that a knot complement covers
another knot complement if and only if the covered knot admits a cyclic surgery.
By the cyclic surgery theorem [6], a hyperbolic knot can admit at most three
cyclic surgeries. By volume considerations, different cyclic surgeries will correspond
to different knot complements in the commensurability class by the construction
above. Thus, we can only obtain three knot complements in a commensurability
class using this method. There is also a family of commensurable pairs of knot
complements given by Walter Neumann. (See [8, 2.2] for a description.) Each pair
turns out to cyclically cover a common orbifold. In addition, there is one example
of a pair of knot complements that non-normally cover a common orbifold, the pair
of dodecahedral knots discussed above. This led to the following conjecture. Let
K be a hyperbolic knot and let C(K) denote the set of knots commensurable with
K.
Conjecture 4.2. [19] |C(K)| ≤ 3.
This conjecture has been verified in a number of cases, including all hyperbolic
two-bridge knot complements [19], (−2, 3, n)-pretzel knot complements [12], and
for an infinite family of hyperbolic knot complements constructed by Hoffman [10]
each of which has exactly three knots in its commensurability class. In each of these
cases, the knots in question
√ are shown not to admit hidden symmetries by proving
that neither Q(i) nor Q( −3) can be a subfield of the invariant trace field. This
implies that the knots do not admit hidden symmetries via the Neumann and Reid
characterization discussed above.
In [2], conjecture 4.2 is proven in the “generic” case. Namely,
Theorem 4.3. [2, Theorem 1.2] If K is a hyperbolic knot whose complement
does not admit hidden symmetries then |C(K)| ≤ 3.
To prove this, it is shown that in the case when K and K are commensurable
and do not admit hidden symmetries, then the knot complements cyclically cover a
230
common orbifold (as in Walter Neumann’s examples). This generalizes the situation
where a knot complement cyclically covers another knot complement, as in the case
of Berge knots. An orbi-lens space is the quotient of S 3 by a finite cyclic group. The
orbifold which is cyclically covered by the two knot complements is the complement
of a knot in an orbi-lens space. As another example of an orbifold, we give an
example of an orbi-lens space. Here the ramification locus is the cores of a genus 1
Heegaard splitting of the underlying space, which is a lens space.
5: We consider S 3 as the unit 3-sphere in C2 . Then let G be the group of
2πi 4πi
isometries of C2 generated by φ where φ(z, w) = (e 6 z, e 15 w). Then G
is cyclic of order 30 and leaves the unit three-sphere invariant. φ6 fixes the
z axis of S 3 , φ15 fixes the w axis and φ10 acts freely. Thus the quotient
orbifold L has underlying space |L| a lens space with fundamental group
of order 3. The ramification locus is two circles labeled 5 and 2 which are
the cores of a genus 1 Heegaard splitting of |L|.
We conclude by remarking that understanding commensurability and hidden
symmetries can lead to information about the symmetries and properties of hyper-
bolic knot complements. For example, we have the following:
Lemma 4.4. [2] If a hyperbolic knot K does not admit hidden symmetries and
|C(K)| > 1, then K is not amphichiral and S 3 \ K fibers over the circle. If in
addition K is periodic, then it must be strongly invertible.
Acknowledgements I learned much of this material from working with my
collaborators, particularly Alan Reid, Steven Boyer and Michel Boileau. I thank
them for their efforts. I am also very grateful for the hospitality of the Harvard
mathematics department and for support from the NSF.
References
[1] I. R. Aitchison and J. H. Rubinstein, Combinatorial cubings, cusps, and the dodecahedral
knots, in Topology ’90, Ohio State Univ. Math. Res. Inst. Publ. 1, pp. 17-26, de Gruyter
(1992).
[2] M. Boileau, S. Boyer, R. Cebanu, and G. S. Walsh, Knot commensurability and the Berge
conjecture, arXiv:1008.1034.
[3] D. Boyd. The A-polynomials of families of symmetric knots, on-line notes,

http://www.math.ubc.ca/ boyd/Apoly.symm.pdf.
[4] D. Calegari, H. Sun, and S. Wang, On fibered commensurability arXiv:1003.0411.
[5] D. Cooper, C. Hodgson, and S. Kerckhoff, Three-dimensional Orbifolds and Cone-

Manifolds, Math. Soc. of Japan, MSJ Memoirs 5 (2000).
[6] M. Culler, C. McA. Gordon, J. Luecke, and P.B. Shalen, Dehn surgery on knots, Ann. of
Math. 125 (1987), pp. 237–300.
[7] F. González-Acuña and W. C. Whitten, Imbeddings of three-manifold groups, Mem. Amer.

Math. Soc. 474 (1992).
[8] O. Goodman, D. Heard, and C. Hodgson, Commensurators of cusped hyperbolic manifolds,

Exp. Math. 17 (2008), pp. 283 – 306.
11
[9] B. Farb and S. Weinberger, Hidden symmetries and arithmetic manifolds in Geometry,
spectral theory, groups, and dynamics, Contemp. Math., 387 (2005), pp. 111–119.
[10] N. Hoffman, Commensurability classes containing three knot complements, Alg. & Geom.
Topol. 10 (2010), 663-677.
[11] C. Leininger, D. D. Long, and A. W. Reid, Commensurators of non-free finitely-generated

Kleinian groups, arXiv:0908.2272.
[12] M. L. Macasieb and T. W. Mattman, Commensurability classes of (-2,3,n) pretzel knot

complements, Algebraic and Geometric Topology 8 (2008) 1833–1853.
[13] C. Maclachlan and A. W. Reid, The Arithmetic of Hyperbolic 3-manifolds, Graduate

Texts in Mathematics 219, Springer-Verlag (2003).
[14] A. Marden, Outer Circles: An Introduction to Hyperbolic 3-manifolds, Cambridge

University Press (2007).
[15] J. W. Morgan and G. Tian, Ricci flow and the Poincaré Conjecture, preprint 2007,
math.DG/0607607.
[16] W. D. Neumann and A. W. Reid, Arithmetic of hyperbolic manifolds, in Topology ’90,

Ohio State Univ. Math. Res. Inst. Publ. 1 273–310, de Gruyter (1992).
[17] A. W. Reid, A note on trace-fields of Kleinian groups, Bull. London Math. Soc. 22 (1990),
349–352.
[18] A. W. Reid, Arithmeticity of knot complements, J. London Math. Soc. 43 (1991), 171–184.
[19] A. W. Reid and G. S. Walsh, Commensurability classes of two-bridge knot complements,

Algebraic and Geometric Topology, 8 (2008) 1031 – 1057.
[20] W. Thurston, The Geometry and Topology of 3-manifolds Princeton University lecture
notes, 1980. Electronic version 1.1: http://www.msri.org/publications/books/gt3m/.
Department of Mathematics, Tufts University, Medford, MA 02155

Current address: Department of Mathematics, Harvard University, Cambridge, MA 02138
E-mail address: genevieve.walsh@tufts.edu
Volume 541, 2011
Realizing arithmetic invariants of hyperbolic 3–manifolds
Walter D. Neumann
These are notes based on the course of lectures on arithmetic invariants of

hyperbolic manifolds given at the workshop associated with the last of three “Vol-
ume Conferences,” held at Columbia University, LSU Baton Rouge, and Columbia
University respectively in March 2006, May/June 2007, June 2009.
The first part of the lecture series was expository, and since most of the material
is readily available elsewhere, we move rapidly over it here (the very first lecture
was a rapid introduction to algebraic number theory, here compressed to less than
2 pages, but hopefully sufficient for the topologist who has never had a course in
algebraic number theory). Section 2 on arithmetic invariants has some new mate-
rial, while Section 3 describes a question that Alan Reid and the author first asked
about 20 years ago, and describes a very tentative approach. It is here promoted
to a conjecture, in part because the author believes he is safe from contradiction in
his lifetime.
In its simplest form the conjecture says:
Conjecture 1. Every non-real concrete number field k and every quaternion al-
gebra over it arise as the invariant trace field and invariant quaternion algebra of
some hyperbolic manifold.
With an excess of optimism, one might add to the conjecture that, moreover,
every set of primes of Ok arises as the set of primes in denominators in the invariant
trace ring of one of these hyperbolic manifolds.
Section 3 describes the already mentioned tentative first step for a program of
proof, which the author has revisited over many years without significant advance.
We discuss also the question whether the Bloch invariants of manifolds with a
given invariant trace field k generate the Bloch group B(k) for that number field,
or even whether their extended Bloch invariants generate K3ind (k).
1. Notation and terminology for algebraic number theory

1.1. Number fields. A number field K is a finite extension of Q. That is,
K is a field containing Q, and finite-dimensional as a vector space over Q. This
2000 Mathematics Subject Classification. 57M27.

This work was supported by the NSF. Useful conversations and correspondence with Alan
Reid and Christian Zickert are gratefully acknowledged.
2011
c 0000
c Mathematical
holder)
1
233
234
2 WALTER D. NEUMANN
dimension d, denoted d = [K : Q], is the degree of the number field. K has exactly
d embeddings into the complex numbers,
θi : K → C, i = 1, . . . , d = r1 + 2r2 ,
where r1 is the number of them with real image, and the remaining embeddings
come in r2 complex conjugate pairs. Indeed, the “Theorem of the Primitive Ele-
ment” implies that K is generated over Q by a single element, from which if follows
that K ∼ = Q[x]/(f (x)) with f (x) an irreducible polynomial of degree d; the embed-
dings K → C arise by mapping the generator x of K to each of the d zeros in C of
f (x).
A concrete number field is a number field K with a chosen embedding into C,
i.e., K given as a subfield of C. The union of all concrete number fields is the field
of algebraic numbers in C, which is the concrete algebraic closure Q ⊂ C of Q.
An algebraic integer is a zero of a monic polynomial with rational integer coef-
ficients. The algebraic integers in K form a subring OK ⊂ K, the ring of integers of
K. It is a Dedekind domain, which is to say that any ideal in Ok factors uniquely as
a product of prime ideals. Each prime ideal p (or “prime” for short) of OK is a di-
visor of a unique ideal (p) with p ∈ Z a rational prime (determined by |OK /p| = pe
for some e > 0). The factorization of (p) as a product (p) = pf11 . . . pfkk of primes of
OK follows patterns which can be found in any text on algebraic number theory.
In particular, the exponents fi are 1 for all but a finite number of primes p of OK ,
which are called ramified.
For the ring of integers OQ = Z of Q, every ideal is principal, and the factoriza-
tion of the ideal (n) into a product of ideals (pi ) expresses the familiar unique prime
factorization of rational integers. In general OK is a unique factorization domain
(UFD) if and only if it is a PID (every ideal is principal), which is somewhat rare.
It is presumed to happen infinitely often, but this is not proven.
Given a prime p of Ok , there is a multiplicative norm ||.||p defined for a ∈ OK by
||a||p := c−r , where pr is the largest power of p which “divides” a (i.e., contains a)
and c > 1 is some constant1; the norm is then determined for arbitrary elements of
K by the multiplicative property ||ab||p = ||a||p ||b||p . This norm determines a trans-
lation invariant topology on K and the completion of K in this topology is a field de-
noted Kp . The unit ball around 0 in Kp is its ring of integers OKp , and the open unit
ball is the unique maximal ideal in this ring. The norm ||.||p is non-Archimedean,
i.e., it satisfies the strong triangle inequality ||a + b||p ≤ max(||a||p , ||b||p ). Up to
equivalence (norms are equivalent if one is a positive power of the other), the only
non-Archimedean multiplicative norms are the ones just described, and the only
other multiplicative norms on K are the norms ||a||θ := |θ(a)| given by absolute
value in C for an embedding θ : K → C. The completion of K in the topology
induced by one of these is R or C according as the image of θ lies in R or not.
The fields R, C, Kp arising from completions are local fields 2. The name is
geometrically motivated: one thinks of OK as a ring of functions on a “space” with
a “finite point” for each prime ideal, plus r1 + r2 “infinite points” corresponding to
the embeddings in R and C; “local” means focusing on an individual point. One
therefore refers to an embedding of K into Kp as a “finite place” and an embedding
1The value of c is unimportant for topological considerations but is standardly taken as
c = N (p) := |OK /p|
2The definition of local field is: non-discrete locally compact topological field. The ones
mentioned here are all that exist in characteristic 0.
REALIZING ARITHMETIC INVARIANTS OF HYPERBOLIC 3–MANIFOLDS 235
3
into R or C as an “infinite place,” and if an object A associated with K (e.g., an

algebra A over K) has corresponding objects associated to each place (e.g., A ⊗ Kp ,
A ⊗ R, A ⊗ C) then a “property of A at the (finite or infinite) place” means that
property for the associated object. We stress that an “infinite place” refers to the
embedding of K in C up to conjugation (even though conjugate embeddings may
have different images), so there are r1 real places and just r2 complex places.
1.2. Quaternion algebras. References for this section are [25] and [5]. A
quaternion algebra over a field K is a simple algebra over K of dimension 4 and with
center K. The simplest example is the algebra M2 (K) of 2 × 2 matrices over K.
This is the only quaternion algebra up to isomorphism for K = C. For K = R there
are exactly two, namely M2 (R) and the Hamiltonian quaternions. The situation for
the non-Archimedean local fields Kp is similar: there are exactly two quaternion
algebras over each of them, one being the trivial one M2 (Kp ) and the other being a
division algebra. In each case the trivial quaternion algebra M2 is called unramified
and the division algebra is called ramified. For a number field K the classification
of quaternion algebras over K is as follows:
Theorem 1.1 (Classification). A quaternion algebra E over K is ramified at only
finitely many places (i.e., only finitely many of the E ⊗ Kp and E ⊗ R’s are division
algebras) and is determined up to isomorphism by the set of these “ramified places.”
The number of ramified places is always even, and every set of places of K of even
size arises as the set of ramified places of a quaternion algebra over K.
A quaternion algebra E over K can always be given in terms of generators and
relations in the form
E = Ki, j : i2 = α, j 2 = β, ij = −ji ,

with α, β ∈ K ∗ . The Hilbert symbol notation α,β refers to this quaternion algebra.
−1,−1 K 1,β
For example, R is Hamilton’s quaternions, and K = M2 (K) for any K.
The Hilbert symbol for a given quaternion algebra is far from unique, but computing
the ramification—and hence the isomorphism class—of a quaternion algebra from
the Hilbert symbol is not hard, and is described in [25], see also [5] for a description
tailored to 3-manifold invariants.
In terms of the above presentation, the map i
→ −i, j
→ −j, ij
→ −ij of a
quaternion algebra E to itself is an anti-automorphism called conjugation, and the
norm of x = a + ib + jc + ijd ∈ E is defined as N (x) := xx̄ = a2 + αb2 + βc2 + αβd2 .
1.3. Arithmetic subgroups of SL(2, C) and PSL(2, C). For a quaternion

algebra E over K the set OE of integers of E (elements which are zeros of monic
polynomials with coefficients in OK ) does not form a subring. One considers instead
an order in E: any subring O of E, contained in OE and containing OK and of
rank 4 over OK . E has infinitely many orders; we just pick one of them.
The subset O1 ⊂ O of elements of norm 1 is a subgroup. At any complex place,
E becomes E ⊗ C = M2 (C) and O1 becomes a subgroup of SL(2, C), while at an
unramified real place E becomes E ⊗ R = M2 (R) and O1 becomes a subgroup of
SL(2, R). We thus get an embedding of Γ := O1 /{±1}
u

r2
r1
Γ⊂ PSL(2, C) × PSL(2, R) ,
i=1 j=1
236
4 WALTER D. NEUMANN
where r1u is the number of unramified real places of K. This subgroup is a lattice
(discrete and of finite covolume).
If r1u = 0 and r2 = 1 this gives an arithmetic subgroup of PSL(2, C), and
similarly for r1u = 1, r2 = 0 and PSL(2, R). Up to commensurability this group only
depends on E and not on the choice of order O. Any subgroup commensurable
with an arithmetic subgroup—i.e., sharing a finite index subgroup with it up to
conjugation—is, by definition, also arithmetic.
The general definition of an arithmetic group is in terms of the set of Z-points
of an algebraic group which is defined over Q. But Borel shows in [1] that all
arithmetic subgroups of PSL(2, C) (and PSL(2, R)) can be obtained as above.
Arithmetic orbifolds (orbifolds H3 /Γ with Γ arithmetic) are very rare—there
are only finitely many of bounded volume—but surprisingly common among the
manifolds and orbifolds of smallest volume.
2. Arithmetic invariants of hyperbolic manifolds

2.1. Invariant trace field and quaternion algebra. A Kleinian group Γ
is a discrete subgroup of PSL(2, C) = Isom+ (H3 ) for which M = H3 /Γ is finite
volume (M may be an orbifold). Let Γ ⊂ SL(2, C) be the inverse image of Γ under
the projection SL(2, C) → PSL(2, C).
Definition 2.1. The trace field of Γ (or of M = H3 /Γ) is the field tr(Γ) generated
by all traces of elements of Γ. We also write tr(M ).
The invariant trace field is the field k(Γ) := tr(Γ(2) ) where Γ(2) is the group gen-
erated by squares of elements of Γ. It can also be computed as k(Γ) = Q({(tr(γ))2 |
γ ∈ Γ}) ([23], see also [19]). We also write k(M ).
The invariant quaternion algebra is the k(Γ)-subalgebra of M2 (C) (2 × 2 ma-
(2)
trices over C) generated over k(Γ) by the elements of Γ . It is denoted A(Γ) or
A(M ).
Theorem 2.2. k(Γ) and A(Γ) are commensurability invariants of Γ.
If Γ is arithmetic, then k(Γ) and A(Γ) equal the defining field and defining
quaternion algebra of Γ, so they form a complete commensurability invariant. They
are not a complete commensurability invariant in the non-arithmetic case.
An obvious necessary condition for arithmeticity is that k(Γ) have only one non-
real complex embedding (it always has at least one). Necessary and sufficient is
that in addition all traces should be algebraic integers and A(Γ) should be ramified
at all real places of k. See [23]. Equivalently, each γ ∈ Γ trace(γ 2 ) should be an
algebraic integer whose absolute value at all real embeddings of k is bounded by 2.
These invariants are already quite powerful invariants of a hyperbolic manifold.
For example, if a hyperbolic manifold M is commensurable with an amphichiral
manifold N (i.e., N has an orientation reversing self-homeomorphism) then k(M ) =
k(M ) and A(M ) = A(M ) (complex conjugation).
If M has cusps then the invariant quaternion algebra is always unramified,
so it gives no more information than the invariant trace field, but for closed M
unramified invariant quaternion algebras are uncommon; for example among the
almost 40√ manifolds in the Snappea closed census [26] which have invariant trace
field Q( −1), only two have unramified quaternion algebra.
5
2.2. The PSL-fundamental class. For details on what we discuss here see
[18, 21, 22] or the expository article [17].
2.2.1. PSL-fundamental class of a hyperbolic manifold. The PSL-fundamental
class of M is a homology class
[M ]P SL ∈ H3 (PSL(2, C)δ ; Z) ,
where the superscript δ means “with discrete topology”.
This class is easily described if M is compact. Write M = H3 /Γ with Γ ⊂
PSL(2, C). The PSL-fundamental class is the image of the fundamental class of
M under the map H3 (M ; Z) = H3 (Γ; Z) → H3 (PSL(2, C)δ ; Z), where the first
equality is because M is a K(Γ, 1)-space. If M has cusps one obtains first a class
in H3 (PSL(2, C)δ , P ; Z), where P is a maximal parabolic subgroup of PSL(2, C)δ .
One then uses a natural splitting of the map
H3 (PSL(2, C)δ ; Z) → H3 (PSL(2, C)δ , P ; Z)
to get [M ]PSL . This was described in [18] and proved carefully by Zickert in [8],
who shows that the class in H3 (PSL(2, C)δ , P ; Z) depends on choices of horoballs
at the cusps, but the image [M ]PSL ∈ H3 (PSL(2, C)δ ; Z) does not.
The group Γ ⊂ PSL(2, C) can be conjugated to lie in PSL(2, K) for a number
field K (which can always be chosen to be a quadratic extension of the trace field,
but there is generally no canonical choice), so the PSL-fundamental class is then
defined in H3 (PSL(2, K); Z).
The following theorem, which holds also with PSL replaced by SL, summarizes
results of various people, see [22] and [27] for more details.
Theorem 2.3. H3 (PSL(2, C); Z) is the direct sum of its torsion subgroup, isomor-
phic to Q/Z, and an infinite dimensional Q vector space.
If k ⊂ C is a number field then H3 (PSL(2, k); Z) is the direct sum of its torsion
subgroup and Zr2 , where r2 is the number of conjugate pairs of complex embeddings
of k. Moreover, the map H3 (PSL(2, k); Z) → H3 (PSL(2, C); Z) has torsion kernel.
The Rigidity Conjecture, which is about 30 years old (see [17] for a discussion),
posits that each of the following equivalent statements is true:
Conjecture 2. (1) H3 (PSL(2, C)δ ; Z) is countable.
(2) H3 (PSL(2, Q)δ ; Z) = H3 (PSL(2, C)δ ; Z)
(3) H3 (PSL(2, C)δ ; Z) is the union of the images of the maps H3 (PSL(2, K); Z) →
H3 (PSL(2, C)δ ; Z), as K runs through all concrete number fields.
2.3. Invariants of the PSL-fundamental class. There is a homomorphism
ĉ : H3 (PSL(2, C); Z) → C/π 2 Z
called the “Cheeger-Simons class” ([3]) whose real and imaginary parts give Chern-
Simons invariant and volume:
ĉ([M ]P SL ) = cs(M ) + i vol(M ) .
The Chern-Simons invariant here is the Chern-Simons invariant of the flat connec-
tion, which is defined for any complete hyperbolic manifold M of finite volume.
If M is closed the Riemannian Chern-Simons invariant CS(M ) ∈ R/2π 2 is also
defined; it reduces to cs(M ) mod π 2 . See [18] for details.
238
6 WALTER D. NEUMANN
We denote the homomorphisms given in the obvious way by the real and imag-
inary parts of ĉ by:
cs : H3 (PSL(2, C); Z) → R/π 2 Z , vol : H3 (PSL(2, C); Z) → R .
The homomorphism cs is injective on the torsion subgroup of H3 (PSL(2, C); Z). A
standard conjecture that appears in many guises in the literature (see [17] for a
discussion) is:
Conjecture 3. The Cheeger-Simons class is injective. That is, volume and Chern-
Simons invariant determine elements of H3 (PSL(2, C); Z) completely.
If k is an algebraic number field and σ1 , . . . , σr2 : k → C are its different complex
embeddings up to conjugation then denote by volj the composition
volj = vol ◦(σj )∗ : H3 (PSL(2, k); Z) → R.
The map
Borel := (vol1 , . . . , volr2 ) : H3 (PSL(2, k); Z) → Rr2
is called the Borel regulator.
Theorem 2.4. The Borel regulator maps H3 (PSL(2, k); Z)/Torsion injectively onto
a full sublattice of Rr2 .
By Theorem 2.3 and the discussion above, cs(M ) ∈ R/Z and Borel([M ]P SL ) ∈
Rr2 (k) determine the PSL-fundamental class [M ]P SL ∈ H3 (PSL(2, C); Z) com-
pletely.
These invariants are computed by the program Snap (see [5]). Snap does this
via a more easily computed invariant which we describe next.
2.4. Bloch group and Bloch invariant. The Bloch group B(C) is the quo-
tient of H3 (PSL(2, C)δ ; Z) by its torsion subgroup. It has the advantage that it has
a simple symbolic description and the image of [M ]PSL in B(C) is readily computed
from an ideal triangulation.
The Bloch group is defined for any field. There are different definitions of
it in the literature; they differ at most by torsion and agree with each other for
algebraically closed fields (see, e.g., [7]). We use the following.
Definition 2.5. Let K be a field. The pre-Bloch group P(K) is the quotient of
the free Z-module Z(K − {0, 1}) by all instances of the following relation:
y 1 − x−1 1−x
[x] − [y] + [ ] − [ −1
]+[ ] = 0,
x 1−y 1−y
called the five term relation. The Bloch group B(K) is the kernel of the map
(1) P(k) → K ∗ ∧Z K ∗ , [z]
→ 2(z ∧ (1 − z)) .
Suppose we have an ideal triangulation of a hyperbolic 3-manifold M using ideal
hyperbolic simplices with cross ratio parameters z1 , . . . , zn . This ideal triangulation
can be a genuine ideal triangulation of a cusped 3-manifold, or a deformation of
such a one as used by Snap and SnapPea to study Dehn filled manifolds, but it
may be more generally any “degree one triangulation”; see [22].
n
Definition 2.6. The Bloch invariant β(M ) is the element 1 ±[zj ] ∈ P(C) with
signs as explained below. It lies in B(C) by [22].
7
The cross-ratio parameter of an ideal simplex depends on a chosen ordering of

the vertices, and the sign in the above sum reflects whether or not this ordering
orients the simplex as it is oriented as part of the degree one triangulation.
If the zj ’s all belong to a subfield K ⊂ C, we may consider β(M ) as an element
of B(K). But it is then necessary to assumethat the vertex orderings of the
simplices match on common faces. If not, then n1 ±[zj ] may differ from β(M ) by
a torsion element (of order dividing 12; this torsion issue does not arise in B(C),
which is torsion-free). Not every triangulation has compatible vertex-orderings for
the simplices, although a triangulation can always be refined to one which does.
Theorem 2.7. If M has cusps then β(M ) is actually defined in B(k), for the
invariant trace field k, while if M is closed this holds for 2β(M ).
This was known in the cusped case (the simplex parameters of an ideal trian-
gulation then lie in k, see [19]) but it was only known up to a higher power of 2 in
the closed case ([17, 22]). The proof, joint with Zickert (but mostly Zickert), is at
the end of this subsection.
Since β(M ) only loses torsion information over [M ]PSL , the Borel regulator
Borel(M ) can be computed from β(M ). It is computed from the simplex parameters
zi as follows. The zi generate a field K which contains the invariant trace field k
of M . The j-th component volj ([M ]P SL ) of Borel(M ) is

n
Borel(M )j = ±D2 (τj (zi )),
i=1
where τj : K → C is any complex embedding which extends σj : k → C. Here the

signs are as above, and D2 is the “Wigner dilogarithm function”
D2 (z) = Im ln2 (z) + log |z| arg(1 − z), z ∈ C − {0, 1},
where ln2 (z) is the classical dilogarithm function. D2 (z) can also be defined as the
volume of the ideal simplex with parameter z.
Recall that k is a concrete number field, i.e., it comes as a subfield of C.
The component of Borel(M ) corresponding to this embedding in C is ± vol(M ),
and it has maximal absolute value among the components of Borel(M ) (see [22]).
This restricts which elements of B(k) can be the Bloch invariant of a hyperbolic
3-manifold. A related (and conjecturally equivalent) restriction is in terms of the
Gromov norm, which is defined on B(k) (see [22]); the Bloch invariants of hyperbolic
manifolds are constrained to lie in the cone over a single face of the norm ball.
Nevertheless, it is plausible that the Bloch group can be generated by Bloch
invariants of 3-manifolds. No obstructions to this are known, and there is (very
mild) experimental evidence for it for low degree fields which appear as invariant
trace fields of manifolds in the cusped and Snappea closed censuses [2, 26]; some
computations related to this are in [5]. So we ask:
Question 2.8. Is B(k) generated by Bloch invariants of hyperbolic manifolds with
invariant trace field in k; how about B(k) ⊗ Q over Q?
Proof of Theorem 2.7. (See also [28].) Since the theorem is known in the
cusped case we assume M is closed. We will need Suslin’s version of the Bloch
group [24], defined by omitting the factor 2 in the map (1) in the definition above.
We will denote it BS (K). Clearly, BS (K) ⊂ B(K), and the quotient B(K)/BS (K)
240
8 WALTER D. NEUMANN
is of exponent 2 (one can show it is infinitely generated if K is a number field). We

will actually show that β(M ) ∈ BS (k).
We will use Suslin’s theorem that BS (K) is a quotient of K3ind (K) by a finite
cyclic group ([24], see also [28]).
The geometric PSL–representation of π1 (M ) lifts to a representation π1 (M ) →
SL(2, C). The set of such lifts is in one-one correspondence with spin struc-
tures on M ; we just pick one for now. The image Γ of such a lifted represen-
tation lies in the quaternion algebra QΓ, which can be unramified by extend-
ing its scalars to a quadratic extension field K of the trace field K (such a K
can be taken as K(λ) for any eigenvalue λ of a nontrivial element of QΓ; see
e.g., [14]). We get π1 (M ) → GL(2, K ), leading to a “GL–fundamental class”
[M ]GL ∈ H3 (GL(2, K ); Z). There is a natural map H3 (GL(2, K ); Z) → K3ind (K )
(see, e.g., [28]), and we denote the image of [M ]GL by [M ]K ∈ K3ind (K ). Now
the non-trivial element of Gal(K /K) preserves traces of π1 (M ) → GL(2, C), so it
takes this representation to a representation which is equivalent over C. It therefore
fixes the Borel invariant and Chern-Simons invariant of [M ]K . The Chern-Simons
invariant on K3ind takes values in C/4π 2 Z, and this Chern-Simons invariant and
Borel invariant together determine any element of K3ind (K ) (see [28]). Thus the
class [M ]K is invariant under Gal(K /K), and since K3ind satisfies Galois descent
(Merkuriev and Suslin [15]), this class lies in K3ind (K). We note, however, that a
priori [M ]K may depend on which lift π1 (M ) → SL(2, C) we started with.
By [19, Theorem 2.2] the trace field K is a multi-quadratic extension of the
invariant trace field k, with Galois group Gal(K/k) ∼ = (Z/2)r for some r. This
group permutes the lifts π1 (M ) → SL(2, C) and hence acts on the elements [M ]K ∈
K3ind (K) defined by these lifts. In [12] it is shown that [M ]K is changed by at most
the unique element of order 2 (and such a change can occur). Thus 2[M ]K is
invariant under this Galois group, and is hence an invariant of M in K3ind (k) which
is independent of the lift to SL(2, C).
The Bloch invariant 2β(M ) is the image of 2[M ]K under a natural transforma-
tion from K3ind to BS , so the theorem is proved.
Problem 2.9. Find a general explicit way to obtain a representative for β(M ) in
BS (k) if M is closed.
Remark 2.10. In [28] Zickert points out that β(M ) ∈ B(k) may not be in its
subgroup BS (k) if M has cusps, in contrast to the closed case. An example is the
manifold M 009 in the census [2].
2.5. Extended Bloch group. The extended Bloch group B(C) of [18] is de-
fined by replacing C − {0, 1} by a Z × Z–cover in the definition of Bloch group, and
appropriately lifting the 5-term relation and the map λ. There are two different
versions of this defined in [18]; we will write BPSL (C) for the first and B(C)
for the

second (they were denoted B(C) and EB(C) respectively in [18]).
Theorem 2.11. There are natural isomorphisms H3 (PSL(2, C)δ ; Z) ∼ = BPSL (C)
∼ ∼
and H3 (SL(2, C) ; Z) = B(C) = K3 (C). (See [18] and [9] respectively.)
δ ind
The program Snap actually computes the element in BPSL (C), and then prints
the Borel regulator and Chern-Simons invariant, which, as already mentioned, de-
termine this element. In [27] Zickert gives a much simpler way of computing the
9
element of BPSL (C) than the one currently used by Snap. This has now been
implemented in SnapPy [6] by Matthias Goerner.
In [28] Zickert extends the definition of extended Bloch groups to number fields
and shows

Theorem 2.12. There is a natural isomorphism B(K) ∼
= K3ind (K) for any number
field K.
Moreover, as mentioned in the proof of Theorem 2.7, for a closed hyperbolic
manifold with spin structure Zickert shows that there is a natural invariant [M ]K ∈

B(k) = K3ind (k) which lifts the Bloch invariant. If M has cusps he shows that [M ]K

is defined in B(K) ⊗ Z[ 12 ], where K is the trace field, and his arguments show that

8[M ]K is well defined in B(k) = K3ind (k) (it is easily seen that 4[M ]K is well defined
ind
in K3 (C)).
2.6. Mutation. If a hyperbolic manifold M contains an essential 4–punctured

sphere then one can cut along this embedded surface and re-glue by an involution,
and it is well known that this process, called Conway mutation or simply mutation,
yields a hyperbolic manifold which shares many properties with M , for example
it has the same volume and Chern-Simons invariant [11], and if M was a knot
complement, many knot theoretic invariants are preserved too. It is folk knowledge
that the Bloch invariant is preserved (and hence also [M ]PSL ) but there is no proof
in the literature, so we give a direct proof here. There are other types of mutation,
sometimes called “generalized mutation.” For example one can mutate along any
essential embedded 3–punctured sphere.
Theorem 2.13. If M and M are hyperbolic manifolds related by Conway muta-
tion, then [M ]PSL = [M ]PSL .
If they are related by mutation on a 3–punctured sphere then [M ]PSL and
[M ]PSL differ by the element of order 2 in H3 (PSL(2, C)δ ; Z).
For any generalized mutation [M ]PSL and [M ]PSL differ by an element of finite
order. Every torsion element of H3 (PSL(2, C)δ ; Z) arises this way.
Proof. Suppose we have an essential embedded two-sided surface Σ ⊂ M
which has a tubular neighborhood N which admits a (necessarily finite order)
isometry Φ which preserves orientation and sides of Σ. Denoting by φ : Σ → Σ
the restriction of Φ, let M be the manifold obtained from M by cutting along Σ
and re-gluing by φ. This is “generalized mutation.” If Σ is a 3– or 4–punctured
sphere or a closed surface of genus 2 then it can always be positioned to have a
Z/2–symmetry.
Let K be a K(PSL(2, C)δ , 1)–space, so H3 (K; Z) = H3 (PSL(2, C)δ ; Z). The
holonomy map α : π1 (M ) → PSL(2, C) induces a map α : M → K (well defined
up to homotopy), and similarly we have α : M → K. These maps can be chosen
to agree outside the tubular neighborhood of N ∼ = Σ × [0, 1] of Σ. They then give
a map N ∪ −N → K. Thinking of N ∪ −N as Σ × [0, 1] ∪ −Σ × [0, 1], it is glued
on one end by the identity and on the other end by φ, so it is simply the mapping
torus Tφ Σ of φ : Σ → Σ. Its fundamental group is the semidirect product π1 (Σ) Z
and it is represented into PSL(2, C) by the homomorphism which on π1 (Σ) is the
restriction of α and on a generator of Z is an isometry of H3 which restricts to a
lift to Ñ of the isometry Φ : N → N .
242
10 WALTER D. NEUMANN
The images in H3 (K; Z) = H3 (PSL(2, C)δ ; Z) of the fundamental classes of M ,

M and Tφ Σ clearly satisfy [M ]PSL − [M ]PSL = [Tφ Σ]PSL . Since Tφ Σ is n–fold

covered by Σ × S 1 , where n is the order of φ, the element [Tφ Σ]PSL is n–torsion,

and is hence determined by the Chern-Simons invariant. It describes the change of
[M ]PSL under the corresponding mutation. Moreover, it depends only on the finite
order map Φ : N → N and not on the geometry in a neighborhood of N , since if one
deforms the geometry in an equivariant fashion then cs([TΦ Σ]) is a continuously
varying n–torsion element in R/π 2 Z, hence constant. So to compute it one just
needs to compute the change in Chern-Simons invariant for a single example. The
following two examples thus complete the proof of the first two sentences of the
theorem.
The Conway and Kinoshita-Teresaka knots, which are related by Conway mu-
tation, both have Chern-Simons invariant 7.1925796077528967037240463 . . . (mod
π 2 ), while the two orientations of the Whitehead link are related by a three-
punctured mutation and have Chern-Simons invariants ±π 2 /4.
The final sentence of the theorem is by section 3 of [13], in which Meyerhoff
and Ruberman give examples to show any change of Chern-Simons invariant by a
rational multiple of π 2 arises by generalized mutation.
The Riemannian Chern-Simons invariant, defined for a closed hyperbolic man-
ifold M , is a lift of cs(M ) to an invariant defined modulo 2π 2 . It is not uncommon
to see claims in the literature that the Chern-Simons invariant of a cusped manifold
can be defined modulo 2π 2 , but we have the following consequence of the above
theorem:
Theorem 2.14. There is no consistent definition of cs(M ) which is well defined
modulo 2π 2 for cusped manifolds.
Proof. Mutation along a thrice-punctured sphere is an involution which changes
cs by π 2 /2. Such a change cannot lift to a an order two change modulo 2π 2 .
2.7. Scissors Congruence. Two hyperbolic manifolds M1 and M2 are scis-
sors congruent if M1 can be cut into finitely many (possibly partially ideal) poly-
hedra which can be reassembled to form M2 . They are stably scissors congruent
if there is some polyhedron Q such that M1 + Q is scissors congruent to M2 + Q
(disjoint union). If M1 and M2 are either both compact or both non-compact then
stable scissors congruence implies scissors congruence. The following follows easily
from [20] (see Theorem 7.2 of [5]):
Theorem 2.15. Let K be a field that contains the invariant trace fields of M1 and
M2 . Then M1 and M2 are stably scissors congruent if and only if Borel(M1 ) −
Borel(M2 ) is the Borel regulator of an element of B(K ∩ R)
In particular, if K ∩ R is totally real (as is “usually” the case) then scissors congru-
ence class of M is not only determined by Borel(M ) but also determines it.
As discussed in [17], the following conjecture would be a consequence of Con-
jecture 3.
Conjecture 4. The stable scissors congruence class of M is determined by vol(M ).
In view of the above theorem this conjecture is amenable to experimentation
with Snap; all the evidence from this is positive.
11
3. Realizing invariants
We know of no way to generate examples of manifolds with given Bloch in-
variant. The program Snap enables extensive experimentation—basically casting a
fishing line in an ocean of examples—but even for quadratic number fields, only a
few fields allow manifolds of small enough volume that they can easily be caught
this way.
Moreover, to try to realize such fine arithmetic invariants, one must first realize
any non-real number field as an invariant trace field. Here we address this question,
and that of realizing a quaternion algebra. The only general result known in this
direction is a result first observed by Reid and the author, described in [14], that
any non-real multi-quadratic extension of Q can be realized.
Let k be number field, A a quaternion algebra over k, O an order in A, and Γ
a torsion free subgroup of finite index in O 1 . Then, as described in subsection 1.3,
each complex embedding of k induces a map Γ → PSL(2, C), each real embedding
at which A is unramified induces a map Γ → PSL(2, R) and, via these maps, Γ acts
discretely with finite co-volume on the product
u

r2
r1
X := H ×
3
H2
i=1 j=1
of copies of H and H (here

3 2
r1uis the number of real places of k at which A is
unramified). Denote Y = X/Γ. Each projection of X to one of the H3 factors gives
a codimension 3 foliation on X which is preserved by the Γ–action, so Y inherits
a codimension 3 foliation from each of these projections. This is a transversally
hyperbolic foliation: there is a metric on the normal bundle of the foliation which
induces a hyperbolic metric on any local transverse section. Similarly, each projec-
tion to H2 gives a codimension 2 transversally hyperbolic foliation.
Now assume that k is a concrete non-real number field, i.e., it comes with a
particular complex embedding singled out (which we call the concrete embedding).
Pick the corresponding codimension 3 foliation F. Let M 3 → Y be an immersion
of a 3-manifold to Y which is everywhere transverse to F. So M 3 has an induced
hyperbolic metric. If M 3 is compact this metric is, of course, complete of finite
volume. We are interested also in the case that M 3 is not compact, but we require
then that the metric be complete of finite volume (as we will see, this can only
happen if A is unramified over k).
Theorem 3.1. The invariant trace field and quaternion algebra for M 3 embed in k
resp. A (as concrete field and quaternion algebra). Moreover, M has integral traces.
Conversely, up to commensurability, every finite volume hyperbolic 3–manifold
with invariant quaternion algebra in A (and hence invariant trace field in k) and
with integral traces occurs this way.
Proof. Suppose first that M → Y is as described in the theorem. Then M
inherits a hyperbolic metric locally from the metric transverse to the foliation F.
Consider one component M̃ → X of the pullback to the universal cover X of Y .
By assumption the projection X → H3 to the first factor restricts to a proper local
isometry, hence an isometry, of M̃ to H3 . It follows that M = M̃ /Γ0 = H3 /Γ0 ,
where Γ0 is a subgroup of the group Γ. The invariant trace field and invariant
quaternion algebra of Γ0 therefore embed in the invariant trace field k and invariant
quaternion algebra A of Γ.
244
Conversely, suppose M has invariant trace field in k and invariant quaternion

algebra in A and has integral traces. By going to a finite cover if necessary, we
can assume the trace field of M is in k. Write M = H3 /Δ with Δ ⊂ PSL(2, C)
and let Δ be the inverse image of Δ in SL(2, C). Then A is the k–subalgebra of
M2 (C) generated by Δ. The subring O ⊂ A consisting of Ok –linear combinations
of elements of Δ is an order in A, and Δ ⊂ O1 (see the proof of Theorem 8.3.2 in
[14]). Now any two orders in A are commensurable, so by going to a finite cover
of M if necessary we can assume that the order O we have here is contained in the
order we used to construct Y , and therefore that Δ ⊂ Γ.
The developing map M̃ → H3 is Δ–equivariant for the action of Δ = π1 (M )
by covering transformations on M̃ and the given action of Δ on H3 . The latter
is induced from the inclusion of Δ̄ in O1 ⊂ SL(2, C) coming from the concrete
embedding k → C.
Now the non-concrete complex embeddings of k give actions of Δ = π1 (M )
on H3 which are not discrete. But, by Lemma 3.2 below, for each of these we can
construct a smooth Δ–equivariant map M̃ → H3 . Similarly, we construct smooth
equivariant maps M̃ → H2 for each unramified real embedding. Together these
2 r1u
maps give a Δ-equivariant map of M̃ to X = ri=1 H3 × j=1 H2 , where Δ acts on
X as a subgroup of Γ. We thus get an induced map of M̃ /Δ = M to X/Γ = Y ,
which clearly does what is required.
We used the following well known lemma:
Lemma 3.2. If X is a simplicial or CW-complex then for any action of π1 (X)
on a contractible space Y there is a π1 (X)–equivariant map of X̃ to Y , and it is
unique up to equivariant homotopy. Moreover, if X and Y are smooth manifolds
and the action of π1 (X) on Y is by diffeomorphisms then this map can be chosen
to be smooth.
Proof. Indeed, one constructs the map inductively over skeleta of X̃. If X
is smooth one can triangulate X and construct the smooth map inductively over
thin neighborhoods of the skeleta. At the k-th step one chooses a lift of each k–
simplex and first extends the smooth map already defined on a neighborhood of
the boundary of this lifted k–simplex smoothly to a neighborhood of the whole
k–simplex and then defines the map on π1 (X)–images of this neighborhood by
equivariance.
One can extend the theorem to remove the restriction that the hyperbolic
manifold have integral traces. For each prime p ⊂ Ok which one wishes to allow
in denominators of traces one should add the corresponding Bass-Serre tree for p
(see, e.g., [16, Chapter VI]) as a factor on the right side of the product of factors
defining X. Then Y is no longer a manifold, but the foliation is still defined and
any transversal to it will be a manifold.
The theorem applies also to hyperbolic surfaces. Suppose k has a chosen real
place (and is no longer required to have at least one complex place) and A is now
unramified at this chosen real place. We consider the foliation of Y given by pro-
jecting X to the corresponding H2 factor. An immersion M 2 → Y transversal to
this foliation will induce a hyperbolic structure on M 2 with invariant quaternion al-
gebra in the concrete quaternion algebra A, and again, any finite volume hyperbolic
surface with integral traces and invariant quaternion algebra in A occurs this way
13
up to commensurability. As before, the integral trace restriction can be avoided by

allowing also some Bass-Serre tree factors in X.
In the 2-dimensional case the existence of surfaces with given invariant trace
field can often be shown. For example, explicit computation for the character
varieties of “small” surfaces point to the lack of restriction on what fields occur.
And in at least one√ case the existence of the transversals in Y has√ been shown
directly: if k = Q( d) is a real quadratic field and Γ = PSL(2, O( d)) (so A is
totally unramified) then Y = H2 ×H2 /Γ is a Hilbert modular surface and Hirzebruch
and Zagier [10] constructed many Riemann surfaces in Y which are transverse to
both foliations. In this case these surfaces are all arithmetic, so they give nothing
new (they have invariant trace field Q and quaternion algebra ramified at a possibly
√
empty set of finite places, but becoming unramified on extending scalars to Q( d)).
In the 3-dimensional case the best evidence that such transversals might always
exist may be the richness of the collection of fields and quaternion algebras provided
by the 3-manifold census and snap, plus the fact that their existence seems very
likely in the 2-dimensional case.
There is a 3–dimensional foliation of Y transverse to F, provided by the pro-
jection of X to the factors other than the one used to construct F. W. Thurston
(private communication) has suggested that one might seek immersed 3-manifolds
which are everywhere almost tangent to this foliation (and hence transverse to F).
This would be very interesting from the point of view of realizing Bloch invari-
ants, since it would realize Bloch invariants for which the components other than
the “concrete component” (giving vol(M )) of the Borel regulator are small with
respect to volume.
References
[1] A. Borel, Commensurability classes and volumes of hyperbolic 3-manifolds. Ann. Scuola
Norm. Sup. Pisa 8 (1981), 1–33.
[2] P.J. Callahan, M.V. Hildebrand and J.R. Weeks, A census of cusped hyperbolic 3-manifolds,
Mathematics of Computation 68 (1999) 321–332.
[3] J. Cheeger and J. Simons, Differential characters and geometric invariants, Springer Lect.
Notes in Math. 1167 (1985), 50–80.
[4] S. Chern, J. Simons, Some cohomology classes in principal fiber bundles and their application
to Riemannian geometry, Proc. Nat. Acad. Sci. U.S.A. 68 (1971), 791–794.
[5] D. Coulson, O. Goodman, C. Hodgson, W.D. Neumann, Computing arithmetic invariants of
3-manifolds, Experimental Mathematics 9 (2000), 127–152
[6] M. Culler, N. Dunfield, SnapPy (a python user interface to the SnapPea kernel),
http://www.math.uic.edu/~t3m/SnapPy
[7] Johan L. Dupont, Chin Han Sah, Scissors congruences II, J. Pure and App. Algebra 25
(1982), 159–195.
[8] Johan L. Dupont, Christian K. Zickert. A dilogarithmic formula for the Cheeger-Chern-
Simons class. Geom. Topol. 10 (2006), 1347–1372
[9] Sebastian Goette, Christian K. Zickert. The extended Bloch group and the Cheeger-Chern-
Simons class. Geom. Topol. 11 (2007), 1623–1635.
[10] F. Hirzebruch and D. B. Zagier, Intersection numbers of curves on hilbert modular surfaces
and modular forms of nebentypus, Invent. math. 36 (1976), 57–113.
[11] Robert Meyerhoff, Daniel Ruberman. Mutation and the η-invariant. J. Differential Geom. 31
(1990), no. 1, 101–130
[12] Michel Massey, Wolfgang Pitsch, Jérôme Scherer, Generalized orientations and the Bloch
invariant, J. K-Theory, Available on CJO 18 Nov 2009 doi:10.1017/is009009019jkt093
[13] Robert Meyerhoff, Daniel Ruberman. Cutting and pasting and the η-invariant. Duke Math.
J. 61 (1990), 747–761
246
[14] Colin Maclachlan, Alan W. Reid. The arithmetic of hyperbolic 3-manifolds. Graduate Texts
in Mathematics, 219. Springer-Verlag, New York, 2003
[15] A.S. Merkurev, A.A. Suslin, The group K3 for a field. (Russian) Izv. Akad. Nauk SSSR Ser.
Mat. 54 (1990), 522–545; translation in Math. USSR-Izv. 36 (1991), 541–565.
[16] Morgan, J. W. and Bass, H, eds, The Smith Conjecture, (Academic Press 1984).
[17] Walter D. Neumann, Hilbert’s 3rd problem and 3-manifolds, The Epstein Birthday Schrift
(Igor Rivin, Colin Rourke and Caroline Series, editors), Geometry and Topology Monographs
Volume 1 (International Press 1998 and www.maths.warwick.ac.uk/gt), 383–411
[18] Walter D. Neumann, Extended Bloch group and the Cheeger-Chern-Simons class, Geom.
Topol. 8 (2004), 413–474
[19] Walter D. Neumann and Alan W. Reid, Arithmetic of hyperbolic manifolds, In Topology 90,
Proceedings of the Research Semester in Low Dimensional Topology at Ohio State, Walter
de Gruyter Verlag, Berlin - New York, 1992, 273–310.
[20] Walter D. Neumann and Jun Yang, Problems for K-theory and Chern-Simons invariants of
hyperbolic 3-manifolds, L’Enseignement Mathématique 41 (1995), 281–296.
[21] Walter D. Neumann and Jun Yang, Invariants from triangulation for hyperbolic 3-manifolds,
Electronic Research Announcements of the Amer. Math. Soc. 1 (2) (1995), 72–79.
[22] Walter D. Neumann and Jun Yang, Bloch invariants of hyperbolic 3-manifolds, Duke Math.
J. 96 (1999), 29-59
[23] Alan W. Reid, A note on trace-fields of Kleinian groups, Bull. London Math. Soc. 22 (1990),
349–352.
[24] A.A. Suslin, K3 of a field, and the Bloch group. (Russian) Translated in Proc. Steklov Inst.
Math. 4 (1991) 217–239. Galois theory, rings, algebraic groups and their applications (Rus-
sian). Trudy Mat. Inst. Steklov. 183 (1990), 180–199, 229.
[25] M-F Vignéras, Arithmétique des algèbres de Quaternions. L.N.M. 800, Springer-Verlag, 1980.
[26] J. Weeks, Snappea, the program, http://www.geometrygames.org/SnapPea/index.html.
[27] Christian K Zickert, The Volume and Chern-Simons invariant of a representation, Duke Math.
J. 150 (2009), 489–532.
[28] Christian K Zickert, The extended Bloch group and algebraic K-theory, arXiv:0910.4005
Barnard College, Columbia University, New York, NY 10027, USA

E-mail address: neumann@math.columbia.edu
Volume 541, 2011
Fields of definition of canonical curves
D. D. Long and A. W. Reid
Abstract. This note studies the question of which number fields can arise as
canonical components.
1. Introduction
Let k ⊂ C be a field. A complex algebraic set V ⊂ Cn is defined over k if the
ideal of polynomials I(V ) vanishing on V is generated by a subset of k[x1 , . . . , xn ].
We say that a field k is the field of definition of V if V is defined over k, and if for
any other field K ⊂ C with V defined over K, then k ⊂ K (for the existence of
the field of definition, see [14], Chapter III). Note that the field of definition of an
algebraic variety depends on the embedding in a particular Cn . By a curve we will
mean an irreducible algebraic curve unless otherwise stated.
Now let M be an orientable finite volume hyperbolic 3-manifold with cusps, and
let X(M ) (resp. Y (M )) denote the SL(2, C)-character variety (resp. PSL(2, C)-
character variety) associated to π1 (M ) (see for example [7] and [2] for definitions).
In [7] and [2] it is shown that X(M ) and Y (M ) are defined over Q. However, the
fields of definition of irreducible components of X(M ) and Y (M ) may be defined
over other number fields; i.e. subfields of C which are finite extensions of Q.
If we restrict M to have a single cusp, then the work of Thurston [23] shows
that a component of X(M ) and Y (M ) containing the character of a faithful discrete
representation of π1 (M ) is a curve. There may be two such curves in Y (M ),
related by complex conjugation (which corresponds to change of orientation of M )
and several in X(M ), arising from the different lifts of π1 (M ) from PSL(2, C) to
SL(2, C) (see [2] and [9] §2.7 for more on this). Throughout this paper we will
usually simply fix one of these curve components, and denote it by X0 (M ) (resp.
Y0 (M )) or X0 (resp. Y0 ) if no confusion will arise. These are called canonical
components. Notice that, if the field of definition of Y0 is a real field, then Y0 will
be fixed by complex conjugation and so there will be precisely one component in
Y (M ) containing the character of a faithful discrete representation. If the field of
definition of Y0 is a non-real field, there is possibly a second component defined
2000 Mathematics Subject Classification. Primary 57M50.

The N. S. F. .
The N. S. F., The Alfred P. Sloan Foundation and a grant from the Texas Advanced Research
Program.
2011
c 0000
c Mathematical
holder)
1
247
248
2 D. D. LONG AND A. W. REID
over the field obtained by applying complex conjugation (see §5 for more on this).
A similar discussion applies to X0 .
It is known that there are examples of M with a single cusp for which the field
of definition is not Q. For example, in [10], there is an example of a once punctured
torus bundle M where Y0 is defined over Q(i). Little else seems known, and so a
natural question is the following.
Question: Which number fields can arise as fields of definition of the curves X0
and Y0 ?
In this paper we will restrict attention to Y0 . As mentioned, little seems known

about what number fields can arise as fields of definition for canonical curve compo-
nents. In particular, to our knowledge, there is no known obstruction to a number
field being a field of definition.
In this note we provide some constructions of examples of one cusped hyperbolic
3-manifolds for which the fields of definition of Y0 are extensions of Q of degree ≥ 2.
To give an indication of the type of result we shall prove we introduce some notation.
Let p ≥ 2 be an integer, and p be the totally real number field Q(cos π/p).
Theorem 1.1. For every odd integer p ≥ 5, there is a one cusped finite volume
orientable hyperbolic 3-manifold Mp for which Y0 (Mp ) has, as field of definition, a
number field kp containing p .
It is a standard fact that [p : Q] goes to infinity with p and so the fields kp will
be distinct on passage to a subsequence. Moreover, the methods also allow us to
construct examples for which the field of definition has arbitrarily large degree over
any p (p odd), and examples for which the field of definition is a non-real number
field. The methods of proof exploit certain “rigidity” phenomena, in particular the
work in [19].
The discussion and results described above are similar in spirit to those con-
cerning the question as to which number fields arise as (invariant) trace-fields for
finite co-volume Kleinian groups. We refer the reader to the recent survey article
[20] (in these proceedings), and [22] for more on this. Beyond the obvious fact that
the invariant trace-field of a finite volume hyperbolic 3-manifold cannot be a real
field, no other obstructions are known. Some obstructions are known for certain
classes of manifolds; for example once punctured torus bundles [4]. In §5 we prove
a theorem that gives obstructions for a number field to be the field of definition of
Y0 for a hyperbolic knot complement in S 3 . Modulo a conjecture about characters
of real representations, we prove that the field of definition in this case has to be
either real or contain a real subfield of index 2 (see Theorem 5.1).
We conclude the Introduction by remarking that, in §6 we construct examples
of hyperbolic knots in S 3 that have invariant trace-fields with class numbers at least
2 (see §6 for more discussion of this topic).
Remark: Much of this paper was basically written in 1998 and remained stub-
bornly unfinished. In conversations at a recent workshop on character varieties at
Banff International Research Station, it became clear that there is some interest in
the fields of definition of character varieties. It was this, together with the invita-
tion of the organizers/editors of the conference/proceedings “Interactions Between
Hyperbolic Geometry, Quantum Topology and Number Theory” that took place in
FIELDS OF DEFINITION OF CANONICAL CURVES 249
3
Columbia in 2009 to submit a paper that prompted us to finish off the paper.
Acknowledgement: The authors wish to thank Steve Boyer, Neil Hoffman, and
Walter Neumann (over many years) for conversations related to topics discussed in
this paper. We also thank Hoffman for help with SnapPy. We are also very grateful
to Walter Neumann and a referee for several helpful comments that helped clarify
this paper.
2. Fields of definition
It will be convenient to describe some additonal background material on fields
of definition. In this section, k ⊂ C will be a number field.
Suppose that V ⊂ Cm is a complex algebraic variety defined over k: recall that
this means that
I(V ) = {f ∈ C[X1 , . . . , Xm ] : f (z) = 0 ∀z ∈ V }
is generated by polynomials f1 , . . . fr ∈ k[X1 , . . . , Xm ]. If k is the field of definition

then k is the smallest such field for which this can be achieved.
We now fix k to be the field of definition, and k ⊂ C the algebraic closure of
k. Then notice that V is also the vanishing set of the k[X1 , . . . Xm ]-ideal
{f ∈ k[X1 , . . . Xm ] : f (z) = 0 ∀z ∈ V }.
It will be convenient to work with this ideal in the following discussion and
for convenience we will just refer to this ideal as I(V ). Now, as is easy to see,
if G = Gal(k/k) denotes the Galois group of the extension k/k, then G acts
on k[X1 , . . . , Xm ] by application of σ ∈ G to the coefficients of a polynomial in
k[X1 , . . . , Xm ]. Since k is fixed by any σ ∈ G, it follows that I(V ) is preserved by
σ. Briefly, any polynomial in I(V ) is a sum of terms gi (X1 , . . . , Xm )fi (X1 , . . . , Xm )
with gi (X1 , . . . , Xm ) ∈ k[X1 , . . . , Xm ]. Applying σ to this product fixes the coeffi-
cients of fi , and thereby determines a term giσ (X1 , . . . , Xm )fi (X1 , . . . , Xm ) ∈ I(V ).
With this observation we prove the following lemma.
Lemma 2.1. Let V be as above, and assume further that for some fixed j with
1 ≤ j ≤ m, there exists an algebraic number t such that every point of V has xj
co-ordinate equal to t. Then t ∈ k.
Proof: We will assume that j = 1 for convenience. By assumption the polynomial

X1 − t vanishes on V , and hence X1 − t ∈ I(V ). Assume to the contrary that t ∈/ k.
Since t is algebraic we can find an element σ ∈ G such that σ(t) = t.
As noted above, σ preserves I(V ), and so σ(X1 − t) ∈ I(V ), that is to say
X1 −σ(t) ∈ I(V ). Then t−σ(t) ∈ I(V ), which is a non-zero constant. In particular,
this does not vanish on V , which contradicts that all elements of I(V ) must vanish
on V .

3. A lemma
Throughout this section, M will denote a cusped orientable hyperbolic 3-
manifold of finite volume.
250
3.1. It will be convenient to recall some of the construction of X(M ) and

Y (M ) from [7] and [2]. We begin with X(M ). Recall that given a representation
ρ : π1 (M ) → SL(2, C), this determines a character χρ : π1 (M ) → C by χρ (γ) =
tr(ρ(γ)).
It is shown in [7] there exists a finite collection of elements {γ1 , . . . , γm } of
π1 (M ) such that for each γ ∈ π1 (M ), χρ (γ) is determined by the collection (χρ (γi ))1≤i≤m .
The complex algebraic structure on X(M ) is then determined by the embedding:
χρ → (χρ (γ1 ), χρ (γ2 ), . . . , χρ (γm )).
The collection of elements {γi }1≤i≤m can be taken to be any generating set for
π1 (M ) and all double and triple products of these generators ([7], [12]).
We will say that γ ∈ π1 (M ) has constant trace on an irreducible component
X ⊂ X(M ), if χρ (γ) = tr(ρ(γ)) is constant for all χρ ∈ X.
This can be extended to Y (M ) in the following way. Recall from [2] §3,
that Y (M ) is constructed from X(M ) as follows. The group H 1 (π1 (M ); Z/2Z) =
Hom(π1 (M ); {±I}) acts on X(M ) via:
(χρ )(γ) = χ(ρ) (γ) = (γ)χρ (γ),

where ∈ H (π1 (M ); Z/2Z), χρ ∈ X(M ) and γ ∈ π1 (M ). This action is algebraic
1
and Y (M ) is the quotient of X(M ) by this action.

Using this map, as shown in [2] §3, co-ordinates for Y (M ) can be constructed
in a similar way to that described above for X(M ). We can therefore talk about
an element γ ∈ π1 (M ) having constant trace on a component of Y ⊂ Y (M ).
3.2. As remarked in the Introduction, the field of definition of an algebraic
variety is an invariant of an algebraic variety that depends on the embedding in
affine space. In particular, in the context of the character variety, X(M ) (resp.
Y (M )) is computed with respect to a generating set, which determines (via the
finite collection of characters described in §3.1) an embedding in some Cm . However
notice that if < S > and < S > are finite generating sets for π1 (M ), and g ∈ S,
then χρ (g) is a Z-polynomial in the finite collection of characters described in
§3.1 arising from S and similarly, if g ∈ S then χρ (g ) is a Z-polynomial in the
finite collection of characters described in §3.1 arising from S. These integral maps
determine an isomorphism of the different embeddings of X(M ) (resp. Y (M )), and
also for components of the character variety. We summarize this in the following
proposition.
Proposition 3.1. Let M be a cusped hyperbolic 3-manifold, and X (resp. Y )
an irreducible component of X(M ) (resp. Y (M )). Then the field of definition of
X (resp. Y ) does not depend on the generating set used to compute X(M ) (resp.
Y (M )).
3.3. Our constructions depend on the following lemma which is a simple con-
sequence of Lemma 2.1 and Proposition 3.1.
Lemma 3.2. Let {γ1 , . . . , γm } be as above, and assume for some j, γj has
constant trace on an irreducible component Y ⊂ Y (M ), say with value t. Then the
field of definition of Y contains t.
In Lemma 3.2, we dealt with the case of a generator having constant trace.
However, there is no loss in generality in doing this, since if γ is any element of
5
L1
L
2
L3
0/1
K
constant trace, then we can simply adjoin γ to a generating set and work with this.
The discussion in §3.2 and Proposition 3.1 shows that this does not effect the field
of definition.
3.4. It is easy to construct hyperbolic knots in S 3 that have many curve com-
ponents in Y (M ), all defined over an extension of Q, and for which none of these
components contains the character of a faithful discrete representation.
For example, [2] Example 3.2 shows that the free product of two non-trivial fi-
nite cyclic groups of orders p and q has a PSL(2, C)-character variety with [p/2][q/2]
curve components. By Lemma 3.2, it follows that these components are defined over
fields containing cos π/p and cos π/q. Considering these groups as the base orbifold
groups of appropriate torus knot exteriors (assuming p and q are relatively prime),
it is easy to construct examples of hyperbolic knots whose fundamental groups sur-
ject these torus knots groups. In some cases these hyperbolic knots can be made
2-bridge and so can only have curve components in their character varieties (by [5]
Theorem 4.1 for example).
4. Applications
Lemma 3.2 gives a method to construct canonical components which are defined
over extensions of Q. We now describe settings where this can be achieved.
4.1. The examples which prove Theorem 1.1, come from [19]. These examples
exploit a “rigid” totally geodesic surface. We briefly recall the one cusped hyperbolic
3-manifolds Mp constructed in §2 of [19]. These manifolds are p-fold cyclic covers
of the orbifolds obtained by (p, 0), (p, 0), (p, 0) Dehn filling on the components
L1 , L2 , L3 of the manifold shown in the figure (this is Figure 2 of [19]). The odd p
assumption was used in [19] to easily arrange a manifold cover with a single cusp.
By construction, Mp contains an embedded non-separating totally geodesic

surface Σp of genus (p − 1)/2. The arguments of [19] show that because Σp covers
a rigid orbifold (namely H2 modulo the (p, p, p)-triangle group), then Σp remains
rigid under all generalized hyperbolic Dehn surgeries. In particular, we deduce
from this that tr(ρ(γ)) is constant on the component Y0 (Mp ) for all γ ∈ π1 (Σp ).
Now since p is odd, the trace-field and the invariant trace-field of the rigid orbifold
coincide, and equals p . Hence it follows that the trace-field of ρ(π1 (Σp )) is p , and
so Lemma 3.2 now implies Theorem 1.1.
252
4.2. It is easy to modify the construction so that for any fixed odd p, the
field kp has arbitrarily large degree over p . To see this, let Bp denote the 3-
manifold obtained by cutting Mp along Σp . So Bp has a single cusp and a pair of
totally geodesic boundary components, both isometric to Σp . This manifold covers
a hyperbolic 3-orbifold with a single cusp and a pair of totally geodesic orbifolds
isometric to H2 modulo the (p, p, p)-triangle group.
Performing a genuine hyperbolic r-Dehn filling on the cusp of Bp gives a hy-
perbolic 3-manifold Bp (r) with geodesic boundary, which by the aforementioned
rigidity is a pair of surfaces isometric to Σp . An adaptation of an argument of
Hodgson (see also [15]) shows that as we vary r the degree of the trace of the core
curve (denoted by tr ) of the r-Dehn filling goes to infinity. Now form 1-cusped
hyperbolic 3-manifolds Np,r by gluing Bp to Bp (r). Since tr will remain constant
on Y0 (Np,r ), Lemma 3.2 and the construction in §4.1 implies the field of definition
of Y0 (Np,r ) contains both cos π/p and tr . This proves the claim.
4.3. Examples where one does not have a rigid surface can also be constructed
using the methods of [19]. Theorem 2 of [19] provides an example of a 2 cusped
hyperbolic 3-manifold M whose cusps were geometrically isolated from each other.
This manifold was conjectured to have strongly geometrically isolated cusps, in
the sense that performing a genuine topological Dehn filling produced a 1-cusped
hyperbolic 3-manifold containing a closed geodesic γ (the core of the attached solid
torus) with tr(ρ(γ)) being constant on Y0 . This was proved in [3]. Both [19] and
[3] exploit a certain 2-cusped hyperbolic orbifold Q that arises as a 2-fold quotient
of M (see [19] Fig 7 and [3]). Indeed the strong geometric isolation established in
[3] is proved via the orbifold Q.
As was pointed out in [10], the example of [10] mentioned in §1 is constructed
as a filling on the manifold M , which shows why this example has field of definition
Q(i). We describe below an example that is built in a similar way. This example
seems interesting as it appeared in [6] in connection with Bloch group computations.
Example: Let N denote the manifold v3066 of the SnapPea (or the recent updated
version SnapPy [8]) census. This is a 1-cusped hyperbolic 3-manifold of volume
6.2328329776455849 . . .. Using SnapPea (or SnapPy) the manifold N can be seen
to arise from Dehn filling on a double cover of Q. This double cover is distinct from
M.
The core of the Dehn filling has trace t ∈ / R with Q(t) a non-real embedding
of the unique cubic field k of signature (1, 1) and discriminant −59 (t is a root of
the polynomial x3 + 2x2 + 1). This trace will be constant on Y0 and so the field of
definition contains k.
Remark: Note that k is also a subfield of the invariant trace-field since t2 generates
k. It is amusing to play with Snap to construct invariant trace fields of surgeries
on N (or the example of [10]) where one can check (using Pari [21]) inclusions of
number fields to “see” the field k as a subfield.
For example, the invariant trace-field of N is a field generated by a root of the
polynomial x6 − 4x4 + 4x2 + 1 (which has 3 complex places). We can test inclusion
of k using the Pari command nfisincl; this produces an output of 0 if there is no
embedding or returns an embedding of fields we find:
7
nfisincl(x3 + 2x2 + 1, x6 − 4x4 + 4x2 + 1) = [x2 − 2].

This can be repeated for many small filling coefficients. For example, doing
(1, 2)-Dehn filling has k as invariant trace-field.
5. Obstructions on fields for knots in S 3

We now focus on fields of definition of Y0 for the case when M is a hyperbolic
knot complement in S 3 . As far as the authors are aware no example of a non-real
field of definition for Y0 is known in this case. We make the following observations
in this regard.
By [13], every knot in S 3 admits a curve of characters of infinite non-abelian
(non-faithful) representations into SU(2) and SO(3) ∼ = PSU(2). For the case of
a hyperbolic knot it is as yet unknown as to whether the canonical component
contains such characters. However, this seems plausible. Indeed the following con-
jecture is a weak version of this.
Conjecture: Let K ⊂ S 3 be a hyperbolic knot. Then Y0 contains infinitely many

characters which are the characters of irreducible real representations.
Given this we have the following obstruction to certain fields being fields of defini-
tion.
Theorem 5.1. Let K ⊂ S 3 be a hyperbolic knot, and suppose that the above
conjecture holds. Then the field of definition of Y0 = Y0 (S 3 \ K) is either real or
contains a real subfield of index 2.
Proof: Let Y0 denote the canonical component that is obtained by applying com-
plex conjugation to Y0 . The conjecture asserts the existence of infinitely many
characters of irreducible real representations. Let C ⊂ Y0 denote this set. Now C
is fixed by complex conjugation and so Y0 and Y0 meet along C. However, these
are curves and so it follows that Y0 = Y0 ,
Let k denote the field of definition of Y0 . In particular, generators for the ideal
I(Y0 ) are elements of k[x1 , . . . , xm ]. Applying complex conjugation, the coefficients
of these generators lie in k (the field obtained by applying complex conjugation to
k). Thus Y0 is defined over k . Since Y0 = Y0 , it follows that Y0 is defined over k .
Now k is the field of definition of Y0 so we must have k ⊂ k , and therefore k = k .
A field that is fixed by complex conjugation is either real, or contains a real subfield
of index 2.
Remark: By way of contrast with §4, we know of no example of a hyperbolic knot

in S 3 (or even an integral homology 3-sphere) for which the canonical component
is defined over an imaginary field.
6. Class numbers and knot groups

In this section we make an observation regarding class numbers of invariant
trace-fields of hyperbolic knots. We begin with some motivation for this.
Let k be a number field with ring of integers Rk . An enormous amount of
research has been devoted to the study of the class group of Rk . This group mea-
sures the extent to which Rk fails to be a principal ideal domain. It is a classical
254
theorem that the class group is finite (see [24] for example); its order is called the
class number of k (and denoted by hk ). It is an open problem dating back to the
time of Gauss whether there are infinitely many number fields of class number one,
or even if given a constant C, there are infinitely many number fields with class
number at most C.
We now relate this to questions about subgroups of PSL(2, C). Thus, suppose
that Γ < PSL(2, C) (not necessarily discrete). The fixed points of parabolic ele-
ments of Γ are the cusps of Γ. By positioning three of these cusps at 0, 1 and ∞ in
C ∪ {∞}, one computes easily that all the cusps now lie in the invariant trace-field
of Γ. In the special case that Γ = PSL(2, Rk ) one sees that the cusps are precisely
the elements of k ∪ {∞} and the action of the group by fractional linear transfor-
mations gives an action on the field k. It is a theorem of Bianchi and Hurwitz (see
[11] Chapter 7.2) that the number of equivalence classes for this action is hk . Note
that the groups PSL(2, Rk ) are discrete only when k is Q or imaginary quadratic.
If now Γ is a non-cocompact Kleinian group of of finite covolume with Γ <
PSL(2, RkΓ ) (e.g any small hyperbolic knot in S 3 ), if one can prove that every
element of k ∪ {∞} is a parabolic fixed point of Γ, then it follows from the Bianchi-
Hurwitz result that hkΓ = 1.
With this as background, following a good deal of experimental work by the
authors, it seems plausible that infinitely many hyperbolic knot complements in S 3
have invariant trace-fields with class number 1 (for example every 2-bridge twist
knot).
The purpose of this section is to prove the following result in the opposite
direction.
Theorem 6.1. There are hyperbolic knots in S 3 for which the invariant trace-
field has class number at least 2.
6.1. To prove Theorem 6.1 we make use of periodic knots. Recall that a knot
K ⊂ S 3 is said to have period q > 1 if there is an orientation-preserving homeo-
morphism h : S 3 → S 3 of order q mapping K to itself and with fixed point set a
circle disjoint from K. The character variety technology for a 1-cusped orientable
hyperbolic 3-orbifold with a torus cusp is exactly as in the manifold setting (see for
example [16] §2.2).
Lemma 6.2. Let Q = H3 /Γ be a 1-cusped orientable hyperbolic 3-orbifold with
a torus cusp, and γ ∈ Γ an element of order q > 1. Then the field of definition of
Y0 (Q) contains q .
Proof: As discussed §3.1, we can assume without loss of generality that γ is part
of a generating set for Γ. Since under infinitely many hyperbolic Dehn surgeries on
Q, γ remains an element of order q, it follows that γ has constant trace on Y0 (Q),
namely ±2 cos π/q. Applying Lemma 3.2 implies the result.
We now describe a particular family of periodic knots and identify the component
Y0 (Q) (as in Lemma 6.2) for the orbifold quotient that we will make use of below.
Example: Let q be an odd positive integer > 1 and relatively prime to 3. The
(3, q) Turks heads knot Kq is the closure of the 3-braid (σ1 σ2−1 )q , and is a knot
with period q. Let Qq denote the orbifold quotient obtained by quotienting out
by this symmetry. The PSL(2, C) character variety of Qq was basically computed
9
in [18]; this paper deals only with representations, but these representations are
normalized by a choice of conjugacy. We find it more convenient to work with more
traditional trace co-ordinates. The orbifold groups of Qq can be presented as (see
[18] Theorem 5)
< x, λ|λq = 1, [λ, xλxλ−1 x−1 λ−1 xλx] = 1 > .

In this presentation, x is the image of a meridian of Kq . Using the co-ordinates
P = tr(x), R = tr(xλ), a simple mathematica computation shows that the algebraic
set X1 which contains the characters of all irreducible SL(2, C)-representations is
defined by the vanishing locus of
F (P, R) = 1 − (2 cos π/q)P R − 3R2 + P 2 R2 + (2 cos π/q)2 R2 − (2 cos π/q)P R3 + R4 .

As is evident this is defined over q . To determine Y0 (Q) from this it will be
convenient to make the following observations.
Note that H1 (Qq ; Z) ∼= Z ⊕ Z/qZ (see the presentation given above). Now the
discussion in §3 regarding the construction of the PSL(2, C)-character variety still
applies to the orbifold group of Qq . Since q is odd, we deduce that H 1 (Qq ; Z/2Z) ∼
=
Z/2Z and moreover, that Z/2Z acts on X(Qq ) via the image of x. Applying this
to the traces we find that in the (P, R)-co-ordinates this involution (which we will
denote by τ ) acts by
τ : (P, R) → (−P, −R).

Note that P R is invariant under τ , and so to describe the quotient under the
action of τ , we will switch co-ordinates to (X, R) where X = P R. This gives the
polynomial (still defined over q ):
G(X, R) = 1 − (2 cos π/q)X − 3R2 + X 2 + (2 cos π/q)2 R2 − (2 cos π/q)XR2 + R4 .

Taking the quotient by τ describes the PSL(2, C)-image of X1 to be the vanishing
locus of
G(X, Y ) = 1 − (2 cos π/q)X − 3Y + X 2 + (2 cos π/q)2 Y − (2 cos π/q)XY + Y 2 .

We now establish that the plane curve G(X, Y ) is irreducible and so its vanishing
locus will define Y0 (Q) (which is therefore defined over q ).
To that end, notice that, G(X, Y ) is quadratic in both X and Y . Solv-
this quadratic
ing for
Y in terms
of X gives the following roots (we let D =
2X cos q − 2 cos q + 1 :
π 2π

1 π
± D2 − 4 −2X cos 2
+X +1 +D .
2 q
Thus if G(X, R) is reducible, it follows that the term
2
π 2π π
Pq (X) = 2X cos − 2 cos + 1 − 4 −2X cos 2
+X +1
q q q
256
is a square of a polynomial in X. This can be readily checked by seeing whether

Pq (X) and its derivative have a common zero. However, a simple mathematica
calculation computes the appropriate resultant to be:

π 2π
−64 sin 2
2 cos −3
q q
which is non-zero since q = 1. This completes the identification of Y0 (Q).

6.2. We now prove Theorem 6.1.
Consider the knots Kq from §6.1, and for convenience we again assume that q
is odd. As noted in §6.1, H1 (Qq ; Z) ∼
= Z ⊕ Z/qZ, and so the invariant trace-field of
Qq (and therefore Kq ) coincides with the trace field of Qq (see [17] Theorem 4.2.1).
This is only stated for manifolds but the same proof holds in the current situation).
Let Lq denote the invariant trace-field of Kq . This can be obtained by special-
izing P = 2 in F (P, R), and this gives
(−1 − R + (2 cos π/q)R − R2 )(−1 + R + (2 cos π/q)R − R2 ) = 0.

It follows that Lq is generated over q by a root of one of the factors above, and in
either case this determines an imaginary quadratic extension of q .
As we discuss below, the proof is completed by applying Theorem 10.1 of [24]
(which is stated below for convenience) and the fact that there are values of q for
which the class number of q is greater than 1. For example, [1] shows that the
smallest prime q for which the class number is greater than 1 is 257.
We now recall the theorem referred to above.

Theorem 6.3. Suppose that the extension of number fields L/K contains no
unramified abelian subextensions F/K with K = F . Then hK |hL .
In our context, Lq is a quadratic extension of q and so the condition on subex-
tensions is vacuous. Moreover, since Lq is necessarily imaginary it is a ramified
extension of q ; in the case of embeddings of the field, this simply means that the
identity embedding of q lifts to a pair of complex conjugate embeddings. Thus we
can apply Theorem 6.3 to Lq /q to deduce that hLq is greater than one for certain
values of q.

References
[1] N. C. Ankeny, S. Chowla, and H. Hasse, On the class-number of the maximal real subfield
of a cyclotomic field, J. Reine Angew. Math. 217 (1965), 217–220.
[2] S. Boyer and X. Zhang, On Culler-Shalen seminorms and Dehn filling, Annals of Math.
148 (1998), 737–801.
[3] D. Calegari, A note on strong geometric isolation in 3-orbifolds, Bull. Austral. Math. Soc.
53 (1996), 271–280.
[4] D. Calegari, Real places and torus bundles, Geom. Dedicata 118 (2006), 209–227.
[5] D. Cooper and D. D. Long, Derivative varieties and the pure braid group, American J.
Math. 115 (1993), 137–160.
[6] D. Coulson, O. Goodman, C. D. Hodgson, and W.D. Neumann, Computing arithmetic
invariants of 3-manifolds, Experiment. Math. 9 (2000), 127–152.
[7] M. Culler and P. B. Shalen, Varieties of group representations and splittings of 3-
manifolds, Annals of Math. 117 (1983), 109–146.
[8] M. Culler, N. M. Dunfield, and J. R. Weeks, SnapPy, a computer program for studying
the geometry and topology of 3-manifolds, available at http://snappy.computop.org.
11
[9] N. M. Dunfield, Cyclic surgery, degrees of maps of character curves, and volume rigidity
for hyperbolic manifolds, Invent. Math. 136 (1999), 623–657.
[10] N. M. Dunfield, Examples of non-trivial roots of unity at ideal points of hyperbolic 3-
manifolds, Topology 38 (1999), 457–465.
[11] J. Elstrodt, F. Grunewald, J. Mennicke, Groups acting on hyperbolic space. Harmonic
analysis and number theory, Springer Monographs in Mathematics, Springer-Verlag,
Berlin, (1998).
[12] F. Gonzalez-Acuna and J. M. Montesinos, On the character variety of group representa-
tions in SL(2, C) and PSL(2, C), Math. Zeit. 214 (1993), 627–652.
[13] P. B. Kronheimer and T. S. Mrowka, Dehn surgery, the fundamental group and SU(2),
Math. Res. Lett. 11 (2004), 741–754.
[14] S. Lang, Introduction to Algebraic Geometry, Interscience Tracts in Pure and Appl. Math.
5 (1958).
[15] D. D. Long and A. W. Reid, Integral points on character varieties, Math. Annalen 325
(2003), 299–321.
[16] D. D. Long and A. W. Reid, Commensurability and the character variety, Math. Research
Letters 6 (1999), 581–591.
[17] C. Maclachlan and A. W. Reid, The Arithmetic of Hyperbolic 3-Manifolds, Grad. Texts
in Math., 219, Springer-Verlag (2003).
[18] W. Magnus, Two generator subgroups of PSL(2, C), Nachr. Akad. Wiss. Göttingen Math.-
Phys. Kl. II , 7 (1975), 81–94.
[19] W. D. Neumann and A. W. Reid, Rigidity of cusps in deformations of hyperbolic 3-
orbifolds, Math. Annalen 295 (1993), 223–237.
[20] W. D. Neumann, Realizing arithmetic invariants of hyperbolic 3-manifolds, to appear
in Interactions Between Hyperbolic Geometry, Quantum Topology and Number Theory,
Contemporary Math.
[21] The PARI Group, PARI/GP, version 2.3.4, Bordeaux, (2008). Available from
http://pari.math.u-bordeaux.fr/.
[22] J. Sinick, Real places and surface bundles, preprint.
[23] W.P. Thurston, The Geometry and Topology of 3-manifolds, Princeton University
mimeographed notes (1979).
[24] L. C. Washington, Introduction to Cyclotomic Fields, Grad. Texts in Math., 83 Springer-
Verlag, New York, (1982).
Department of Mathematics,University of California, Santa Barbara, CA 93106

E-mail address: long@math.ucsb.edu
Department of Mathematics, University of Texas, Austin, TX 78712

E-mail address: areid@math.utexas.edu
This book is based on a 10-day workshop given by leading experts in hyperbolic geometry,
quantum topology and number theory, in June 2009 at Columbia University. Each speaker
gave a minicourse consisting of three or four lectures aimed at graduate students and
recent PhDs. The proceedings of this enormously successful workshop can serve as an
introduction to this active research area in a way that is expository and broadly accessible
to graduate students.
Although many ideas overlap, the twelve expository/research papers in this volume can be
grouped into four rough categories:
(1) different approaches to the Volume Conjecture, and relations between the main
quantum and geometric invariants;
(2) the geometry associated to triangulations of hyperbolic 3-manifolds;
(3) arithmetic invariants of hyperbolic 3-manifolds;
(4) quantum invariants associated to knots and hyperbolic 3-manifolds.
The workshop, the conference that followed, and these proceedings continue a long tradition
in quantum and geometric topology of bringing together ideas from diverse areas of math-
ematics and physics, and highlights the importance of collaborative research in tackling
big problems that require expertise in disparate disciplines.
CONM/541 AMS on the Web

www.ams.org

(Abhijit Champanerkar, Oliver Dasbach, Efstratia K (B-Ok - Xyz)

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

(Abhijit Champanerkar, Oliver Dasbach, Efstratia K (B-Ok - Xyz)

Încărcat de

Drepturi de autor:

Formate disponibile

CONTEMPORARY

Abhijit Champanerkar, Oliver Dasbach,

American Mathematical Society

Abhijit Champanerkar, Oliver Dasbach,

American Mathematical Society

Library of Congress Cataloging-in-Publication Data

This book is based on a 10-day workshop of minicourses in hyperbolic geometry,

applications of arithmetic invariants to study the problem of commensurability of

Department of Mathematics, College of Staten Island, CUNY, 2800 Victory

Department of Mathematics, Louisiana State University, Baton Rouge, LA

Department of Mathematics, Michigan State University, East Lansing, MI

Department of Mathematics, College of Staten Island, CUNY, 2800 Victory

Department of Mathematics, Barnard College, Columbia University, 2990 Broad-

Department of Mathematics, Louisiana State University, Baton Rouge, LA

Abdelmalek Abdesselam Abhijit Champanerkar

Alexander Ellis Eric Katerman

Heather Molle Neal Stoltzfus

Hitoshi Murakami Sean Thomas

Kei Nakamura Rohit Thomas

Walter Neumann Dylan Thurston

An Introduction to the Volume Conjecture

Abstract. This is an introduction to the Volume Conjecture and its gener-

2000 Mathematics Subject Classiﬁcation. Primary 57M27 57M25 57M50.

2. Link invariant from a Yang–Baxter operator

product of braids β1 and β2 given by putting β2 below β1 . It is well known (see

Figure 2. ith generator of the n-braid group Bn

Figure 3. braid relation

Figure 4. the ﬁgure-eight knot is presented as the closure of a braid

Here the closure of an n-braid is obtained by connecting the n points on the

Figure 6. β (βσn± , respectively) is stabilized (destabilized, respec-

and {e0 , e1 , . . . , eN −1 } is a basis of V .

Figure 7. Replace a generator with the R-matrix.

Figure 8. A braid and the corresponding homomorhism

2.3. Invariant. Let (R, μ, a, b) be an enhanced Yang–Baxter operator on an

T(R,μ,a,b) (σ1 σ2−1 σ1 σ2−1 )

We can show that T(R,μ,a,b) gives a link invariant.

Figure 9. A braid and its invariant

Figure 10. braid relation corresponds to the Yang–Baxter equation

Φ(α) Φ(β) Φ(β)

Figure 11. invariance under a conjugation

To prove the invariance under a stabilization, we ﬁrst note that if a homomor-

then its n-fold trace is given by

Therefore if g is a homomorphism g : V ⊗ V → V ⊗ V given by glk11,l,k2 2 , then we have

Figure 12. invariance under a stabilization

2.4. Quantum (g, V ) invariant. One of the important ways to construct an

The two-dimensional colored Jones polynomial J2 (L; q) is (a version) the orig-

Figure 13. skein triple

Remark 2.12. The original Jones polynomial V (L; q) satisﬁes

Figure 16. Label the ﬁrst (top-left) arc with N − 1.

Label the top-right arc with i with 0 ≤ i ≤ N − 1 (Figure 18).

Figure 18. Label the top-right arc with i (0 ≤ i ≤ N − 1).

Figure 20. Choose j and k. Then the diﬀerence at the second

Now we can calculate the colored Jones polynomial. We have

Figure 23. The label k should coincide with j.

From Figure 14, we have

Using the formula (see [28, Lemma 3.2])

3.1. Statement of the Volume Conjecture. In [15] Kashaev introduced

Figure 25. (2, 1)-cable of the ﬁgure-eight knot

hyperbolic Seifert ﬁbered

Figure 26. The JSJ decomposition of the (2, 1)-cable of the

3.2. Proof of the Volume Conjecture for the ﬁgure-eight knot. We

where we put f (N ; k) := 4 sin2 (kπ/N ). The graph of f (N ; k) is depicted in Fig-

Figure 27. Graph of f (N ; k)

{N − 1 − k}!q:=ξ = (a power of ξN ) × (ξN )k−